BACKGROUND OF THE INVENTION 1. Field of the Invention [0001] The present invention relates to a plant control device, a rolling control device, a method for controlling a plant, and a program for controlling a plant. 2. Description of the Related Art [0002] A rolling mill which is a plant for efficiently producing thin metal materials by rolling metal plates may occasionally cause plate thickness defects attributed to unevenness of hardness in metal plates that are rolling object materials. The unevenness of hardness represents a state of a rolling object material with its hardness not uniform across the rolling object material. The hardness of the rolling object material acts as deformation resistance when being rolled. Accordingly, if the rolling object material contains the unevenness of hardness in a rolling direction being a conveyance direction to convey the rolling object material during the rolling, the degree of compression of the rolling object material varies depending on the location, whereby a variation occurs in the plate thickness after the rolling object material is rolled. [0003] The rolling is generally conducted by feeding a rolling object material several times through the rolling mill so as to bring an original plate thickness that is an initial plate thickness of the metal plate into a product thickness . If the unevenness of hardness is present, the plate thickness variation occurs in some locations due to the unevenness of hardness. In the meantime, a plate thickness deviation occurs anew at every one of the multiple rolling sessions. Although plate thickness control is carried out in the rolling mill in order to improve plate thickness accuracy, such conventional plate thickness control has a difficulty in eliminating the plate thickness variation that occurs due to the unevenness of hardness every time the rolling takes place. [0004] For example, it is possible to suppress a plate thickness variation by: detecting a plate thickness variation attributed to unevenness of hardness, which has occurred in a certain rolling session, by using an entry side plate thickness gauge in the next rolling session; and conducting plate thickness control in a feedforward manner. However, even though the plate thickness control suppresses the existing plate thickness variation, a new plate thickness variation will be brought about by the unevenness of thickness. In this case, a control gain larger than a normal control gain is required in order to suppress the new plate thickness variation. To this end, a plate thickness control method disclosed in Patent Literature 1 determines the presence of unevenness of hardness by a frequency analysis, and changes a control gain of feedforward plate thickness control. [0005] As well as the control gain, a phase shift amount of a control output is important for producing a sufficient control effect in the feedforward control. Accordingly, a plate thickness control device disclosed in Patent Literature 2 aims at achieving a maximum control effect by adjusting the control gain and the phase shift amount based on a phase relation between multiple control state quantities. [0006] [Related Art Documents] [Patent Literatures] Patent Literature 1: JP 2000-33409 A Patent Literature 2: JP 2017-058932 A [0007] According to the technique disclosed in Patent Literature 1, in order to eliminate a variation in deformation resistance in a conveyance direction of the rolling object material based on the unevenness of hardness, the plate thickness variation having occurred in the preceding rolling session is detected in the next rolling session as an entry side plate thickness variation, and is eliminated by the feedforward control. In this technique, the control gain of the feedforward control is changed depending on the presence of the unevenness of hardness. [0008] The feedforward control is proportional control which can maximize its control effect by providing a deviation of a target control state quantity with a control output having a phase and an amplitude which are adapted to the deviation. Here, assuming that a sinusoidal wave is the deviation of the control state quantity of a control target while the control output is obtained by multiplying the deviation of the control state quantity by the control gain, how the phase and the amplitude of the deviation of the control state quantity change as a result of the control is examined. [0009] For example, a sinusoidal wave having a control gain G and a phase shift amount A is formed as the control output with respect to a sinusoidal wave sin (cot) that represents the deviation of the control state quantity. Meanwhile, a control result of the feedforward control is defined as y. In this case, the value y is expressed in Formula (1): [Formula 1] y = sm(cot)-Gsm(cot + A) = Xsm(cot + S) (l) (1) . [0010] Here, an amplitude X and a phase difference 5 in the value y in Formula 1 are expressed in Formulae (2-1) and (2-2), respectively: [Formulae 2] X = -sjl2+G2-2Gcos(-A) (2-i; and 8 = tan l [0011] f -GsinA ^ (2-2; [^l-GcosAJ FIGs. 24A and 24B are graphs showing relations of the phase shift amount A and the phase difference 5 as well as the amplitude X in control state quantity before and after the control, which are included in the control output in the feedforward control. Specifically, FIG. 24A is a graph showing a relation between the phase shift amount A and the phase difference 5, and FIG. 24B is a graph showing a relation between the phase shift amount A and the amplitude X in the control state quantity after the control. As shown in FIG. 24B, the amplitude grows larger as the phase shift amount A to be controlled becomes larger, and depending on the control gain G, the state where the phase shift amount A exceeds 60 degrees positive or negative is likely to bring about not only a failure to achieve any control effect but also an adverse effect. Specifically, it turns out that when the phase shift amount A is included in the control output, the phase of the obtained control result y deviates from the original sinusoidal wave sin (cot) . [0012] In other words, even if the control gain G is increased in the feedforward control which is the proportional control, the control effect is not only reduced but also deteriorated when the phase of the control output deviates from the phase of the control state quantity of the control target, that is, when the phase shift amount A is present (not equal to zero). [0013] Here, when the plate thickness variation occurs due to the unevenness of hardness, the rolling control conducts not only the plate thickness control but also tension control. Hence, a phase relation between the plate thickness variation and the unevenness of hardness deviates as a consequence. The phase relation indicates how many degrees of the angle each waveform peak position deviates from a cycle of 360 degrees. Accordingly, even when the feedforward control is conducted based on an entry side plate thickness deviation of a rolling object material, the control cannot achieve a sufficient control effect due to the deviation in the phase relation with the original unevenness of hardness. [0014] Note that such a situation may occur not only in the unevenness of hardness of the rolling object material in the rolling of a metal material but also in control of a plant in general. Particularly, in a case of obtaining a control result by controlling a control target object that includes a variable factor before control, which occurs based on a variable factor constituting a criterion, it is not possible to achieve a sufficient control effect as with the above-mentioned case if a phase of the variable factor constituting the criterion deviates from a phase of the variable factor before the control. [0015] Patent Literature 2 discloses a technique for favorably adjusting a phase shift amount A of a control output and thereby improving a control effect, the technique being applicable to feedforward control of a plant such as a rolling mill in which a control state quantity of a control target includes multiple variable factors having mutually different phases. According to this technique, a phase difference acquisition unit first acquires a phase difference 5 between a variation in control state quantity before control (a pre-control state quantity) and a variation in control state quantity after control (a post-control state quantity) when a processing treatment such as rolling takes place. Then, based on the phase difference 5, a feedforward adjustment unit determines a phase shift amount A when a measurement result of the pre-control state quantity is reflected in the feedforward control. In this way, it is possible to appropriately a control gain G and the phase shift amount A used in a control output of the feedforward control, and thus to improve a control effect. [0016] However, in the invention disclosed in Patent Literature 2, the phase difference acquisition unit creates time-series tables of the pre-control state quantity and the post-control state quantity, and determines the phase difference 5 while comparing these tables with each other. For this reason, if a lot of frequency components are included in the pre-control state quantity and the post-control state quantity, and waveforms thereof become complicated, then it is hard to identify a frequency corresponding to a plate thickness disturbance (unevenness of hardness) subject to control, and to determine the phase difference 5. As a consequence, this invention turns out to have a problem of a difficulty in accurately determining the phase shift amount A of the control output. SUMMARY OF THE INVENTION [0017] In view of the above-mentioned problems of the conventional techniques, it is an objective of the present invention to provide a plant control device, a rolling control device, a method for controlling a plant, and a program for controlling a plant, which are capable of efficiently obtaining a control timing shift amount (a phase shift amount A) of a control output, which can achieve a larger feedforward control effect even in the case where a pre-control state quantity and a post-control state quantity form complicated waveforms. [0018] To attain the object, a plant control device according to the present invention provides a plant control device, based on a pre-control state quantity being a control state quantity before control at the time of subjecting a processing object to a processing treatment, for performing feedforward control of a post-control state quantity being a control state quantity after the control. The plant control device includes: frequency response measurement means for acquiring a phase difference and an attenuation of the post-control state quantity with respect to the pre-control state quantity based on a result of performing fast Fourier transform on time-series data of the pre-control state quantity and time-series data of the post-control state quantity; and feedforward control parameter adjustment means for determining a control output timing shift amount based on the phase difference and the attenuation acquired, the control output timing shift amount representing delay time to reflect the pre-control state quantity in the feedforward control. [0019] Thus, the present invention provides a plant control device, a rolling control device, a method for controlling a plant, and a program for controlling a plant, which are capable of efficiently obtaining a control timing shift amount (a phase shift amount A) of a control output, which can achieve a larger feedforward control effect even in the case where a pre-control state quantity and a post-control state quantity form complicated waveforms. BRIEF DESCRIPTION OF THE DRAWINGS [0020] FIG. 1 is a diagram showing an example of an overall configuration of a rolling mill and a rolling control device according to an embodiment of the present invention. FIG. 2 is a diagram showing an example of parameters related to a rolling phenomenon and rolling control of a rolling object 5 material by the rolling mill. FIG. 3 is a diagram showing an example of a control model of the rolling phenomenon. FIG. 4 is a diagram showing an example of a basic control configuration of plate thickness control by a plate thickness 10 control device. FIG. 5 is a diagram showing an example of a basic control configuration of tension control by a tension control device. FIG. 6 includes graphs showing an example of a result of simulation obtained when neither the plate thickness control nor 15 the tension control is carried out. FIG. 7 includes graphs showing an example of a result of simulation obtained when the tension control on an entry side and an exit side is carried out by proportional-integral control, and when only feedback control of the plate thickness control on the 20 exit side is carried out. FIG. 8 includes graphs showing an example of a result of simulation obtained when the feedback control of the plate thickness control on an exit side of a preceding stand rolling mill is carried out in addition to the conditions in the case of FIG. 7. 25 FIG. 9 is a diagram showing an example of an extended control configuration including the plate thickness control device and a feedforward control adjustment device according to the embodiment of the present invention. FIG. 10 is a diagram showing an example of a detailed 30 configuration of a control gain and timing shift amount setting device. FIGs. 11A and 11B are diagrams for explaining an outline of 9 a frequency response method, in which FIG. 11A is a diagram showing a time-domain response model and FIG. 11B is a diagram showing a frequency-domain response model. FIGs. 12A and 12B are diagrams showing results of frequency 5 response simulation using the FFT, in which FIG. 12A shows an example of data collection time equal to 10.24 seconds and FIG. 12B shows an example of data collection time equal to 5.12 seconds. FIGs. 13A and 13B are more diagrams showing the results of frequency response simulation using the FFT, in which FIG. 13A shows 10 an example of data collection time equal to 2.56 seconds and FIG. 13B shows an example of data collection time equal to 2.56 seconds with an input signal at a single frequency. FIG. 14 is a diagram showing an example of a sampling period and data quantity search table. 15 FIG. 15 is a diagram showing an example of a configuration of a plate thickness disturbance measurement device. FIG. 16 is a graph showing an example of frequency-dependent characteristics of an entry side plate thickness deviation amplitude and an exit side plate thickness deviation amplitude. 20 FIG. 17 is a diagram showing an example of a configuration of a frequency response estimation device. FIG. 18 includes graphs showing an example of a result of simulation obtained when the tension control on an entry side and an exit side of a fourth stand rolling mill is carried out by 25 proportional-integral control, and when feedback control and feedforward control of the exit side plate thickness of the fourth stand rolling mill is carried out. FIG. 19 includes graphs showing an example of a result of simulation obtained when a control output timing shift amount for 30 the feedforward control is changed in an increasing direction under the same simulation conditions as those in FIG. 18. FIG. 20 includes graphs showing an example of a result of 10 simulation obtained when a control gain for the feedforward control is increased under the same simulation conditions as those in FIG. 19. FIG. 21 includes graphs showing an example of a result of 5 simulation obtained when the control output timing shift amount for the feedforward control is changed in a decreasing direction under the same simulation conditions as those in FIG. 20. FIG. 22 includes graphs showing an example of a result of simulation obtained when the control output timing shift amount for 10 the feedforward control is shifted to the opposite side from that in the example in FIG. 18. FIG. 23 is a diagram showing a hardware configuration of an information processing device constituting the rolling control device according to the embodiment of the present invention. 15 FIGs. 24A and 24B are graphs showing relations of a phase shift amount and a phase difference as well as an amplitude in control state quantity before and after control included in a control output in feedforward control, in which FIG. 24A is a graph showing a relation between the phase shift amount and the phase difference, 20 and FIG. 24B is a graph showing a relation between the phase shift amount and the amplitude in the control state quantity after the control. DETAILED DESCRIPTION OF THE EMBODIMENTS [0021] 25 An embodiment of the present invention will be hereinafter described in detail with reference to the drawings. Note that constituents that are common in the drawings will be designated by the same reference signs and duplicate explanation will be omitted. Moreover, in this specification, a rolling control device for a 30 rolling mill to roll a rolling object material such as a metal material will be described below as a specific example of a plant control device. 11 [0022] «1. Basic Control Configuration» FIG. 1 is a diagram showing an example of an overall configuration of a rolling mill 1 and a rolling control device 2 5 according to an embodiment of the present invention. Here, the rolling mill 1 is a four-stand tandem rolling mill, and the rolling control device 2 mainly performs control for minimizing a plate thickness variation occurring due to unevenness of hardness when rolling a rolling object material 3. 10 [0023] As shown in FIG. 1, the rolling mill 1 (the tandem rolling mill) of this embodiment includes four stand rolling mills 11 to 14 which are arranged in series. The rolling object material 3 is rolled consecutively by these four stand rolling mills 11 to 14. 15 At this time, the rolling object material 3 moves from the left side to the right side in FIG. 1 while being rolled. [0024] Each of the stand rolling mills 11 to 14 includes six rolls arranged vertically. The vertically arranged six rolls are called 20 work rolls, intermediate rolls, and backup rolls from an inner side where the rolling object material 3 is sandwiched. Moreover, one of plate thickness gauges 41 to 44 and one of tensiometers 51 to 54 for acquiring control state quantities necessary for control by the rolling control device 2 are provided on an exit side and the 25 like of the corresponding one of the stand rolling mills 11 to 14, respectively. [0025] In addition, the rolling control device 2 includes motor velocity control devices 21 to 25, roll gap control devices 31 to 30 34, plate thickness control devices 61 to 64, tension control devices 71 to 74, and the like. The plate thickness control devices 61 to 64 and the tension control devices 71 to 74 play important roles 12 in this embodiment, details of which will be described in turn below. [0026] First, a rolling phenomenon of the rolling object material 3 will be explained prior to detailed description of plate thickness 5 control. FIG. 2 is a diagram showing an example of parameters related to a rolling phenomenon and rolling control of the rolling object material 3 by the rolling mill 1. As shown in FIG. 2, the rolling is carried out by crushing the rolling object material 3 in a space between the vertical pair of work rolls of the rolling 10 mill 1. At this time, the rolling object material 3 is pulled by an entry side tension Tb and an exit side tension Tf and is crushed by a rolling load P. As a consequence, an entry side plate thickness H turns into an exit side plate thickness h. A forward slip f and a backward slip b are caused by the above-described rolling 15 phenomenon. When a work roll velocity is VR, an entry side velocity Ve and an exit side velocity Vo are expressed in formulae indicated in FIG. 2, respectively, by using the forward slip f and the backward slip b. [0027] 20 FIG. 3 is a diagram showing an example of a control model of the rolling phenomenon. In the case of the tandem rolling mill, the entry side tension Tb and the exit side tension Tf vary with the entry side velocity Ve as well as the exit side velocity Vo of the stand rolling mill of its own, an entry side velocity of a following 25 stand rolling mill, and an exit side velocity of a preceding stand rolling mill. Such variations in tension bring about variations in rolling load P, exit side plate thickness h, entry side velocity Ve, and exit side velocity Vo. [0028] 30 As shown in FIG. 3, each of the rolling load P, the forward slip f and the backward slip b is expressed as a function dependent on the entry side plate thickness H, the exit side plate thickness 13 h, the entry side tension Tb, the exit side tension Tf, a deformation resistance k, and a friction coefficient μ. Moreover, a parameter L included in formulae written at a lower right part in FIG. 3 represents a distance between two adjacent stands of the stand 5 rolling mills 11 to 14. Furthermore, an input V-1 represents the exit side velocity of the adjacent and preceding stand rolling mill while an input V+1 represents the entry side velocity of the adjacent and following stand rolling mill. [0029] 10 While the rolling phenomenon is the phenomenon which involves the entry side plate thickness H, the work roll velocity VR, and a roll gap S collectively as the input, and involves the entry side tension Tb, the exit side tension Tf, and the exit side plate thickness h collectively as the output as described above, this is 15 a complicated phenomenon at the same time which also relates to rolling phenomena at the preceding and following stand rolling mills by the intermediary of the tension. [0030] Referring to FIG. 1, the motor velocity control devices 21 20 to 24 to control the work roll velocity VR respectively, and the roll gap control devices 31 to 34 to operate the roll gap S that is a gap between the work rolls respectively are provided so as to correspond to the four stand rolling mills 11 to 14, respectively. Since the plate thickness of the rolling object material 3, which 25 is formed into a product, is especially important in terms of product quality in the rolling process, the plate thickness gauges 41 to 44 for measuring the plate thickness of the rolling object material 3 are installed on the exit side of the stand rolling mills 11 to 14, respectively. Moreover, since the tension to be applied to the 30 rolling object material 3 is important in terms of stability of the rolling operation and concerns accuracy in plate thickness as well, the tensiometers 51 to 54 are installed on the exit side of the stand 14 rolling mills 11 to 14, respectively. Furthermore, exit side bridle rolls 15 and the motor velocity control device 25 for controlling a velocity of a motor for driving the exit side bridle rolls 15 are installed on the exit side of the fourth stand rolling mill 14. 5 [0031] In the rolling mill 1 and the rolling control device 2 configured as described above, the plate thickness control device 61 of the first stand rolling mill 11 controls the roll gap S in the first stand rolling mill 11 through the roll gap control device 10 31. Meanwhile, plate thickness control devices 62 to 64 of the second to fourth stand rolling mills 12 to 14 control the work roll velocities VR of the preceding, i.e., the first to third stand rolling mills 11 to 13 through the motor velocity control devices 21 to 23, respectively. 15 [0032] At this time, the plate thickness control devices 62 to 64 of the second stand rolling mill 12 and the following rolling mills carry out feedforward control by using detection results of the plate thickness gauges 41 to 43 on the entry side, and further carry out 20 feedback control by using detection results of the plate thickness gauges 42 to 44 on the exit side. For example, the plate thickness control device 62 carries out the feedforward control by using the detection result of the plate thickness gauge 41 on the entry side, and further carries out the feedback control by using the detection 25 result of the plate thickness gauge 42 on the exit side. [0033] Meanwhile, the tension control devices 71 to 73 of the first to third stand rolling mills 11 to 13 obtain the roll gaps S of the following stand rolling mills 12 to 14 based on the tensions detected 30 with the tensiometers 51 to 55 on the exit side thereof. The roll gap control devices 32 to 34 manipulate positions of the work rolls in accordance with the obtained roll gaps S. For example, the 15 tension control device 71 obtains the roll gap S in the second stand rolling mill 12 based on the tension detected with the tensiometer 51 on the exit side of the first stand rolling mill 11, and the roll gap control device 32 manipulates the positions of the work rolls 5 of the second stand rolling mill 12 based on the result. [0034] In the meantime, the tension control device 73 of the fourth stand rolling mill 14 controls the tension on the exit side of the fourth stand rolling mill 14 by operating the velocities of the 10 exit-side bridle rolls 15 through the motor velocity control device 25. [0035] FIG. 4 is a diagram showing an example of a basic control configuration of the plate thickness control by the plate thickness 15 control device 64. As shown in FIG. 4 (see also FIG. 2), the plate thickness control device 64 performs transfer processing in such a way as to delay an entry side plate thickness deviation ΔH measured with the plate thickness gauge 43 on the exit side of the third stand rolling mill 13 by time TFF that elapses before a measurement position 20 of the rolling object material 3 reaches immediately below the fourth stand rolling mill 14. Here, the measurement result of the entry side plate thickness deviation ΔH is equivalent to a control state quantity before the rolling, which can be regarded as a so-called pre-control state quantity. 25 [0036] Next, the plate thickness control device 64 obtains a feedforward control amount by multiplying a result of the transfer processing by a control gain GFF. Moreover, the plate thickness control device 64 performs integration processing by multiplying 30 an exit side plate thickness deviation Δh, which is measured with the plate thickness gauge 44 on the exit side of the fourth stand rolling mill 14, by a control gain GFB and thus obtains a feedback 16 control amount. The plate thickness control device 64 outputs an amount obtained by adding the feedforward control amount to the feedback control amount, which are acquired as described above, to the motor velocity control device 23 of the third stand rolling mill 5 13. Here, the measurement result of the exit side plate thickness deviation Δh is equivalent to a control state quantity after the rolling, which can be regarded as a so-called post-control state quantity. [0037] 10 Note that a plate thickness variation cannot be detected at a position of occurrence which is immediately below the fourth stand rolling mill 14, but is instead detected with the plate thickness gauge 44 installed at a position away from the fourth stand rolling mill 14. As a consequence, there exits wasted time between the 15 occurrence of the plate thickness variation and the detection thereof, so that an integration control amount is included in the calculation of the feedback control amount. [0038] Configurations of the plate thickness control devices 62 and 20 63 are the same as the configuration of the plate thickness control device 64, and description thereof will be omitted below. On the other hand, the plate thickness control device 61 controls the roll gap S in the first stand rolling mill 11, and the configuration and the control method thereof are different from those of the plate 25 thickness control device 64. Nonetheless, the description of the configuration and the control method of the plate thickness control device 61 will be omitted in this embodiment. [0039] FIG. 5 is a diagram showing an example of a basic control 30 configuration of tension control by the tension control device 73. As shown in FIG. 5 (see also FIG. 2), the tension control device 73 performs proportional-integral control by using a deviation ΔT34 17 between a tension instruction value T34ref and a tension result value T34FB measured with the tensiometer 53 installed between the third stand rolling mill 13 and the fourth stand rolling mill 14. In this integral control, a phase of a control output is shifted by 90 degrees 5 with respect to the control state quantity. Accordingly, in an exit side plate thickness h of the fourth stand rolling mill 14 obtained as a result has a phase shifted by a plate thickness deviation Δh with respect to the original location of unevenness of hardness. [0040] 10 «2. Simulation Based on Basic Control Configuration» Next, results of simulation of the rolling phenomenon in the four-stand tandem rolling mill as shown in FIG. 1 will be described with reference to FIGs. 6 to 8. The simulation calculates how plate thickness variations, tension variations, and load variations of 15 the fourth stand rolling mill 14 change over time in response to the variation in deformation resistance representing the unevenness of hardness. [0041] FIG. 6 includes graphs showing an example of a result of 20 simulation obtained when neither the plate thickness control nor the tension control is carried out. Meanwhile, FIG. 7 includes graphs showing an example of a result of simulation obtained when the tension control on the entry side and the exit side of the fourth stand rolling mill 14 is carried out by the proportional-integral 25 control, and when only the feedback control of the plate thickness control on the exit side of the fourth stand rolling mill 14 is carried out. In the meantime, FIG. 8 includes graphs showing an example of a result of simulation obtained when the feedback control of the plate thickness control on the exit side of the third stand 30 rolling mill 13 that precedes the fourth stand rolling mill 14 is carried out in addition to the conditions in the case of FIG. 7. [0042] 18 Here, regarding the "plate thickness variation" in FIGs. 6 to 8, the variation in entry side plate thickness H (the entry side plate thickness deviation ΔH) is indicated with a solid line while the variation in exit side plate thickness h (the exit side plate 5 thickness deviation Δh) is indicated with a dashed line. Likewise, regarding the "tension variation" therein, the variation in entry side tension is indicated with a solid line while the variation in exit side tension is indicated with a dashed line. Regarding the "load variation" therein, the variation in rolling load is indicated 10 with a solid line while the variation in deformation resistance is indicated with a dashed line. Note that the time elapses in a direction from the left side to the right side in each of the graphs. In this context, the left end represents the current state while the right end represents the 15 most distant past state. [0043] During the simulation in the case of FIG. 6, the unevenness of hardness directly manifests as the plate thickness variation. For this reason, peak positions of waveforms of the variation in 20 deformation resistance, and the variation in entry side plate thickness H as well as the variation in exit side plate thickness h of the fourth stand rolling mill 14 coincide with one another, and there are no shifts in terms of phase relations thereof (see positions along a vertical solid line, for example). 25 [0044] Meanwhile, the simulation in the case of FIG. 7 causes a leading phase in which the phase of the variation in exit side plate thickness h of the fourth stand rolling mill 14 advances ahead of the phase of the variation in entry side plate thickness. This 30 occurs because the integral control is carried out by the plate thickness control device 64 of the fourth stand rolling mill 14, and the phase of the control output is delayed by 90 degrees. Hence, 19 the phase shift amount Δ turns out to be negative due to Formulae (1), (2-1), and (2-2) as well as the relations shown in FIGs. 24A and 24B. As a consequence, a phase difference δ of the variation in exit side plate thickness h of the fourth stand rolling mill 14 5 representing the result of the plate thickness control turns out to be positive. [0045] In the meantime, the simulation in the case of FIG. 8 causes a leading phase in which the phase of the variation in entry side 10 plate thickness H of the fourth stand rolling mill 14 advances ahead of that in deformation resistance because the feedback control is also carried out in the plate thickness control by the third stand rolling mill 13 that precedes the fourth stand rolling mill 14. [0046] 15 As described above, when prescribed control is carried out on a variation factor such as unevenness of hardness intrinsic to a control target, another variation factor with a different phase may occur whereby a phase relation between control state quantities of the control target may vary as a consequence. 20 [0047] Usually, in the tandem rolling mill, the plate thickness control is carried out in the first stand rolling mill 11 to begin with, and then in the respective stand rolling mills 12 to 14 as well. Accordingly, the phase of the variation in deformation 25 resistance is shifted from the phase of the variation in exit side plate thickness h (the exit side plate thickness deviation Δh) which emerges as a result thereof. For this reason, when the feedforward control is carried out by using the entry side plate thickness deviation ΔH of the stand rolling mill, a sufficient control effect 30 is not available due to the effect of the phase shift between the variation in deformation resistance and the variation in entry side plate thickness deviation ΔH. 20 [0048] As a conventional method of adjusting control parameters of the feedforward control, a control output timing shift amount ΔTFF for the feedforward control in FIG. 4 used to be set in consideration 5 of wasted time and a response from a control output to a control operation end, and a control gain G used to be changed depending on the exit side plate thickness deviation Δh that represents a control result. However, when this method is used, a sufficient control effect is often unavailable due to the existence of a phase 10 difference between the entry side plate thickness deviation ΔH representing the control state quantity of the target and the variation in deformation resistance which is the unevenness of hardness. [0049] 15 As shown in the aforementioned formulae (1), (2-1), and (2-2) and in FIG. 24, it is necessary in the feedforward control to appropriately set the control gain G and the control output timing shift amount ΔTFF that corresponds to the phase shift amount Δ. Moreover, this setting needs to be determined while considering a 20 rolling velocity and as to what other control is carried out, and therefore involves complicated adjustment. Regarding the rolling velocity, the response from the control output to an action of the control operation end varies due to a change in frequency of the plate thickness variation. On the other hand, in the case of the 25 tandem rolling mill, the response varies depending on what kind of the plate thickness control and/or the tension control is carried out in which stand of the rolling mill. [0050] While it is important to appropriately set the control output 30 timing shift amount ΔTFF (the phase shift amount Δ) and the control gain G in the feedforward control, these two values are also linked to each other by the relations described by using Formulae (1), (2-1), 21 and (2-2). For example, if the control gain G is changed, then the phase difference δ between the control state quantities before and after the control varies as well. On the other hand, if the control output timing shift amount ΔTFF is changed, then an amplitude X of 5 each control state quantity is changed as well. Accordingly, it is practically difficult to conduct the adjustment in such a way as to appropriately set both of the values. [0051] As shown in Formula (2-2) mentioned above, the phase 10 difference δ between the control state quantities before and after the control is an arctangent function, and therefore has a domain from -90 degrees to +90 degrees with respect to a range from -oo to +oo. Moreover, as apparent from Formula (2-2), when the variable exceeds +oo and turns into negative, the phase difference becomes 15 larger than 90 degrees. Accordingly, as shown in FIG. 24, the phase difference δ is assumed to exceed 90 degrees for the sake of convenience. Furthermore, according to Formula (2-2), the phase difference δ between the control state quantities does not exceed 90 degrees when the control gain G is not larger than 1. As a 20 consequence, it is possible to predict that the control gain G is too large when the phase difference δ between the control state quantities exceeds 90 degrees. [0052] Meanwhile, the phase shift amount Δ and the phase difference 25 δ between the control state quantities before and after the control have mutually opposite directions. Accordingly, when the phase difference δ between the control state quantities before and after the control is given, then it is possible to predict how to change the phase shift amount Δ, i.e., the control output timing shift 30 amount ΔTFF. For example, when the phase difference δ between the control state quantities is in the positive direction, the phase shift amount Δ may be changed in an increasing direction, i.e., in 22 a direction from the negative side to the positive side. In an opposite case, the phase shift amount Δ may be changed in a decreasing direction, i.e., from the positive side to the negative side. [0053] 5 As described above, in the case of the feedforward control in the plate thickness control, the phase relation between the entry side plate thickness deviation ΔH detected with the plate thickness gauge 43 on the entry side and the exit side plate thickness deviation Δh detected with the plate thickness gauge 44 on the exit side can 10 be regarded as the phase difference δ between the control state quantities. Likewise, the control output timing shift amount ΔTFF from the entry side plate thickness deviation ΔH to the control output can be regarded as the phase shift amount Δ. Accordingly, the control output timing shift amount ΔTFF and the control gain GFF 15 in the feedforward control can be adjusted by using the control state quantities. In this context, a configuration obtained by adding functions to adjust the control output timing shift amount ΔTFF and the control gain GFF to the basic control configuration of the plate thickness control device 64 shown in FIG. 4 will be hereinafter 20 referred to as an extended control configuration of the plate thickness control device 64. [0054] «3. Extended Control Configuration» <3.1 Feedforward Control Adjustment Device> 25 FIG. 9 is a diagram showing an example of an extended control configuration including the plate thickness control device 64 and a feedforward control adjustment device 101 according to the embodiment of the present invention. Here, the feedforward control adjustment device 101 is a device which obtains the control output 30 timing shift amount ΔTFF and the control gain GFF for the feedforward control to be carried out by the plate thickness control device 64. Specifically, the feedforward control adjustment device 101 is a 23 device which realizes the extended control configuration of the plate thickness control device 64, and constitutes a significant feature of this embodiment. [0055] 5 As shown in FIG. 9, in the feedforward control adjustment device 101, the entry side plate thickness deviation ΔH detected with the plate thickness gauge 43 on the entry side of the fourth stand rolling mill 14 is subjected to transfer processing to such a timing to pass immediately below the plate thickness gauge 44 on 10 the exit side of the fourth stand rolling mill 14. Then, a value obtained in the transfer processing is defined as an entry side plate thickness deviation ΔHTRK. Here, time TX3D-4 indicated in FIG. 9 represents transfer time from a position immediately below the plate thickness gauge 43 on the entry side of the fourth stand rolling 15 mill 14 to the fourth stand rolling mill 14, and time T4-X4D indicated therein represents transfer time from the fourth stand rolling mill 14 to a position immediately below the plate thickness gauge 44 on the exit side of the fourth stand rolling mill 14. [0056] 20 Meanwhile, in the feedforward control adjustment device 101, the rolling load P detected with a rolling load gauge 46 for measuring the rolling load of the fourth stand rolling mill 14 is subjected to transfer processing from a position immediately below the fourth stand rolling mill 14 to the position immediately below the plate 25 thickness gauge 44 on the exit side thereof. Then, a value obtained in the transfer processing is defined as a rolling load PTRK. Here, the variation in rolling load P represents a variation in control state quantity that occurs in response to the unevenness of hardness of the rolling object material 3. 30 [0057] A control gain and timing shift amount setting device 102 obtains the control gain GFF and the control output timing shift 24 amount ΔTFF by using the entry side plate thickness deviation ΔHTRK and the rolling load PTRK obtained in the above-mentioned transfer processing, as well as the exit side plate thickness deviation Δh detected with the plate thickness gauge 44 collectively as the input. 5 Note that methods of obtaining the control gain GFF and the control output timing shift amount ΔTFF will be described later in detail by using the drawings starting from FIG. 10. [0058] The control gain GFF and the control output timing shift amount 10 ΔTFF obtained by the control gain and timing shift amount setting device 102 are inputted to the plate thickness control device 64. The plate thickness control device 64 carries out the feedforward control by using the entry side plate thickness deviation ΔH. Here, the plate thickness control device 64 uses the control gain GFF 15 obtained by the control gain and timing shift amount setting device 102 as the control gain G of the feedforward control. In addition, the plate thickness control device 64 modifies the transfer time TFF of the entry side plate thickness deviation ΔH, which represents the timing of the control output, in such a way as to satisfy TFF 20 = TX3D-3-ΔTFF by using the control output timing shift amount ΔTFF obtained by the control gain and timing shift amount setting device 102. [0059] Although the feedforward control adjustment device 101 is 25 illustrated as a separate device provided outside the plate thickness control device 64 in FIG. 9, the feedforward control adjustment device 101 may be such a device incorporated in the plate thickness control device 64. [0060] 30 <3.2 Control Gain and Timing Shift Amount Setting Device> FIG. 10 is a diagram showing an example of a detailed configuration of the control gain and timing shift amount setting 25 device 102. As shown in FIG. 10, the control gain and timing shift amount setting device 102 includes a frequency response measurement device 201, three membership functions 105, 106 and 107, a fuzzy inference device 108, a parameter changing device 109, and the like. 5 [0061] As described previously, the control gain and timing shift amount setting device 102 calculates the control gain GFF and the control output timing shift amount ΔTFF for the feedforward control by using the entry side plate thickness deviation ΔHTRK, the exit 10 the rolling load PTRK, and the exit side plate thickness deviation Δh collectively as the input. The control gain GFF and the control output timing shift amount ΔTFF thus calculated are outputted to the plate thickness control device 64. The plate thickness control device 64 carries out the feedforward control after adjusting the 15 transfer time TFF by using the control gain GFF. In other words, the control gain and timing shift amount setting device 102 plays a role in setting and adjusting the control parameters at the time of feedforward control by the plate thickness control device 64. This is one of significant features of this embodiment which are not found 20 in the related art. [0062] An objective of the feedforward control by the plate thickness control device 64 is to render the exit side plate thickness deviation Δh smaller than the entry side plate thickness deviation 25 ΔH. For this reason, the exit side plate thickness deviation Δh becomes smaller when the feedforward control is favorably operated. However, when the exit side plate thickness deviation Δh becomes smaller, determination of the relation between the entry side plate thickness deviation ΔH and the exit side plate thickness deviation 30 Δh may be complicated. In this case, it may be difficult to obtain the control gain GFF and the control output timing shift amount ΔTFF for the feedforward control. Accordingly, the control gain and 26 timing shift amount setting device 102 of this embodiment obtains the control gain GFF and the control output timing shift amount ΔTFF for the feedforward control by using a phase relation between the rolling load PTRK to be affected by the unevenness of hardness and 5 the entry side plate thickness deviation ΔHTRK as well. This is also one of the significant features of this embodiment. [0063] In this regard, the control gain and timing shift amount setting device 102 needs to obtain signal attenuation values and 10 phase relations between time-series signals, such as between the entry side plate thickness deviation ΔHTRK and the exit side plate thickness deviation Δh, and between the entry side plate thickness deviation ΔHTRK and the rolling load PTRK. [0064] 15 According to the invention disclosed in Patent Literature 2, "square errors for one period" are calculated while shifting phases of two time-series signals, and a phase at which the square error becomes the smallest is defined as a phase difference between the two time-series signals. This method requires recognition of one 20 period of a reference signal. In addition, the application of this method may be complicated when amplitudes between the reference signal and a comparative signal differ largely due to a control effect, when multiple frequency components overlap each other, and so on. This embodiment therefore employs a frequency response 25 method which can obtain the signal attenuation and the phase relation between the two time-series signals relatively easily. [0065] (Reference 1: Regarding Frequency Response Method) FIGs. 11A and 11B are diagrams for explaining an outline of 30 the frequency response method, in which FIG. 11A is a diagram showing a time response model and FIG. 11B is a diagram showing a frequency response model. In the rolling control, the rolling object material 27 3 enters from the entry side of the fourth stand rolling mill 14, for example, reduces its plate thickness by the rolling phenomenon, and then exits from the exit side of the fourth stand rolling mill 14. In other words, the entry side plate thickness deviation ΔH 5 of the rolling object material 3 is changed into the exit side plate thickness deviation Δh due to the rolling phenomenon. [0066] Here, as shown in FIG. 11A, if a chronological change in the entry side plate thickness deviation ΔH is expressed as x(t) while 10 a chronological change in exit side plate thickness deviation Δh is expressed as y(t), then the rolling phenomenon can be expressed as a time response function g(t) that satisfies y(t) = g(t)∙x(t). In other words, the entry side plate thickness deviation x(t) being a time-domain signal (the time-series signal) is transformed into 15 the exit side plate thickness deviation y(t) being another time-domain signal by the time response function g(t) of the rolling phenomenon. [0067] The above-mentioned rolling phenomenon represented by using 20 the time response function g(t) can be expressed by using a frequency response function G(ω) shown in FIG. 11B. Specifically, if the entry side plate thickness deviation ΔH and the exit side plate thickness deviation Δh are expressed as an entry side plate thickness deviation X(ω) and an exit side plate thickness deviation Y(ω), which 25 represent frequency-domain signals (values of frequency components), then a relation between these deviations can be expresses as Y(ω) = G(ω)∙X(ω). In other words, the entry side plate thickness deviation X(ω) being the frequency-domain signal is transformed into the exit side plate thickness deviation Y(ω) being 30 the other frequency-domain signal by the frequency response function G(ω) of the rolling phenomenon. [0068] 28 The entry side plate thickness deviation x(t) and the exit side plate thickness deviation y(t) in the time domain can be obtained as time-series signals to be detected with the plate thickness gauge 43 on the entry side and the plate thickness gauge 5 44 on the exit side of the fourth stand rolling mill 14, for example. In the meantime, the entry side plate thickness deviation X(ω) and the exit side plate thickness deviation Y(ω) in the frequency domain can be obtained by subjecting the x(t) and y(t) obtained in the time domain to the Fourier transform, respectively. 10 [0069] An advantage of expressing the rolling phenomenon by using the input signal X(ω) and the output signal Y(ω) in the frequency domain and the frequency response function G(ω) is that this idea makes it easier to compare the amplitude and the phase relation 15 between the input signal and the output signal on the frequency basis. In other words, in the frequency domain, the attenuation and the phase difference between the plate thickness deviation signals due to the rolling phenomenon can be obtained easily. [0070] 20 Specifically, in this embodiment, the entry side plate thickness deviation x(t) and the exit side plate thickness deviation y(t) can be obtained as detection values with the plate thickness gauges 43 and 44. Meanwhile, the entry side plate thickness deviation X(ω) and the exit side plate thickness deviation Y(ω) in 25 the frequency domain can be obtained by subjecting the entry side plate thickness deviation x(t) and the exit side plate thickness deviation y(t) to the Fourier transform, respectively. Hence, the frequency response function G(ω) can be obtained by Formula (3) below: 30 [Formula 3] , , Y(a) Y(p)-Y(p) G(a>) = ^ = y J y J (3), X(a>) X(a>)-X(a>) 29 where X(co) is a complex conjugate of X(ω). [0071] Further, by using the frequency response function G(ω), an attenuation (gain) and a phase difference (phase) at a frequency 5 ω can be obtained by Formulae (4-1) and (4-2) below: [Formulae 4] gain = 20 • log(\G(co)\)[dB] (4-1); and phase = — arg( G(»)[deg] (4 - 2), 2;r where arg(c) represents an argument of a complex number c. 10 [0072] (Reference 2: Regarding Discrete Fourier Transform and FFT) Here, a description will be given of the Fourier transform of discrete data (the discrete Fourier transform) used for obtaining the entry side plate thickness deviation X(ω) and the exit side plate 15 thickness deviation Y(ω) in the frequency domain. Generally, when a time-series signal f(t) that includes N pieces of sampling data for one period is expressed by using N pieces of independent sinusoidal signals at a frequency of k, the time-series signal f(t) is expressed in Formula (5) below: 20 [Formula 5] N-1 f(t) = YCk-ejkt (5), where j is an imaginary unit. [0073] Here, when the numbers n = 0, 1, …, N indicating the order 25 of the sampling data for one period are associated with time t representing phases in a range from 0 to 2π, then the time t can be expressed as t = 2π∙n/N. As a consequence, Formula (5) can be expressed in Formula (6) below: [Formula 6] 30 N-1 _ j 2ak n k=0 [0074] Then, Formula (7) below is obtained by subjecting Formula (6) to the discrete Fourier transform: 5 [Formula 7] cm= — Yjf(n)-e N (7), N n=0 where m = 0, 1, 2, …, and N-1. [0075] Here, the coefficient cm is a complex number. Moreover, in 10 Formula (7), the value 2πm/N corresponds to the frequency. In other words, the coefficient cm represents the frequency component of the time-series signal f(t) expressed in Formula (5) at the frequency of 2πm/N. Accordingly, the absolute value and the argument of the coefficient cm represent the amplitude and the phase of the frequency 15 component of the time-series signal f(t) at the frequency of 2πm/N, respectively. [0076] Moreover, the fast Fourier transform (hereinafter abbreviated as the FFT) is usually used in the case of calculating the discrete 20 Fourier transform with a computer. Though the FFT has a precondition that a data quantity to be transformed needs to be a power of 2, the FFT still has a major advantage that computational complexity is significantly reduced as compared to that of the ordinary discrete Fourier transform. 25 [0077] Generally, when N pieces of data are subjected to the Fourier transform, the ordinary discrete Fourier transform requires complexity that is proportional to N2. On the other hand, it is known that the FFT only requires complexity proportional to N∙log2N. For 30 example, when 1024 pieces of data are subjected to the Fourier 31 transform, a ratio of the complexity of the FFT with respect to that of the ordinary discrete Fourier transform is calculated as log21024/1024 = 10/1024. In other words, the FFT only requires the complexity of about 1/100 as large as that of the ordinary discrete 5 Fourier transform. [0078] <3.3 Frequency Resolution and Data Collection Time in FFT> As described above, since the FFT requires the data quantity which is a power of 2, the FFT also causes a restriction of a time 10 interval for sampling the data (a sampling interval). Here, if a sampling frequency being the inverse of the sampling interval is defined as fs and the number of samples (the data quantity) is defined as N, then frequency resolution Δf can be calculated by Δf = fs/N, and data collection time MT can be calculated by MT = N/fs = 1/Δf. 15 [0079] Here, the data collection time MT means a time period from a start to an end of sampling of the data constituting the input to the FFT, and the frequency resolution Δf means resolution in a direction of a frequency axis when the FFT is carried out. Meanwhile, 20 a theoretical maximum frequency fr with which it is possible to resolve two frequency components in the data sampled at the sampling frequency fs is given by fr = fs/2. In other words, it is not possible to resolve the two frequency components unless the frequencies thereof are away from each other by at least twice as large as the 25 frequency resolution Δf. [0080] Smaller is better for both the frequency resolution Δf and the data collection time MT. Nevertheless, it is not possible to render both of the values small at the same time because of the 30 relation MT = 1/Δf as mentioned above. As a consequence, it is important to set each of the frequency resolution Δf and the data collection time MT to a practically appropriate value in utilizing 32 the FFT. [0081] In the meantime, the objective of the feedforward control of the plate thickness control according to the embodiment of the 5 present invention is to improve the control effect by adjusting the control gain G and the phase shift amount Δ. To this end, it is necessary to carry out the calculations at the shortest time interval as possible and to reduce the data collection time MT as short as possible. 10 [0082] On the other hand, when multiple frequency components are included in a disturbance to the entry side plate thickness, i.e., the entry side plate thickness deviation ΔH, the attenuation (gain) and the phase difference (phase) at each of the frequency components 15 of the disturbances cannot be calculated unless these frequencies of the disturbances are successfully separated from each other. Therefore, it is necessary to select the data collection time that satisfies the aforementioned conditions. [0083] 20 FIGs. 12A to 13B are diagrams showing results of frequency response simulation using the FFT. A simulation result in FIG. 12A represents a case where the data collection time MT is 10.24 seconds and the frequency resolution Δf is about 0.1 Hz. In this simulation, sinusoidal waves at frequencies of 0.5 Hz, 1.0 Hz, 2.0 Hz, and 3.0 25 Hz are mixed and formed into an input signal as the entry side plate thickness deviation ΔH that represents the disturbance to the plate thickness. In this case, the inputted sinusoidal waves are set such that the respective frequencies satisfy values of the attenuation (gain) and of phase difference (phase) of the exit side plate 30 thickness deviation Δh being the output signal: (frequency) (attenuation (gain)) (phase difference (phase)) 0.5 Hz -6.0 dB 60 degrees 33 1.0 Hz -4.4 dB -45 degrees 2.0 Hz -3.1 dB -30 degrees 3.0 Hz -1.9 dB 30 degrees. [0084] 5 Note that in FIG. 12A, an upper graph is a graph showing chronological changes in entry side plate thickness deviation ΔH and exit side plate thickness deviation Δh in the time domain, while a lower graph is a graph showing frequency characteristics of the entry side plate thickness deviation ΔH and the exit side plate 10 thickness deviation Δh (the output signal) in the frequency domain after carrying out the FFT. Moreover, the values of the attenuation (gain) and of phase difference (phase) are also indicated in the lower graph. [0085] 15 As can be seen from the lower frequency-domain graph in FIG. 12A, the four frequency components are clearly separated from one another in the exit side plate thickness deviation Δh as well, and the values of the attenuation (gain) and of phase difference (phase) are also obtained accurately. However, judging by the fact that 20 the smallest value of the frequencies of the disturbance, i.e., the entry side plate thickness deviation ΔH is 0.5 Hz, the data collection time of 10 seconds is indeed too long. Specifically, in this case, a time period of five times or more of a plate thickness variation period (2 seconds) is required for carrying out 25 feedforward AGC (automatic gain control) adjustment. [0086] A simulation result in FIG. 12B represents a case where the data collection time MT is 5.12 seconds and the frequency resolution Δf is about 0.2 Hz. The entry side plate thickness deviation ΔH 30 inputted in the simulation of this case is the same as that in the case of FIG. 12A, and display formats of graphs showing the simulation result are also equivalent to those in FIG. 12A. 34 [0087] As can be seen from a lower frequency-domain graph in FIG. 12B, the four frequency components included in the input signal are almost clearly separated from one another in the exit side plate 5 thickness deviation Δh as well, and the values of both the attenuation (gain) and phase difference (phase) are obtained almost accurately. Note that FIG. 12B also represents an actual example showing that the attenuation (gain) and the phase difference (phase) can be obtained almost accurately when a minimum separation width 10 (0.5 Hz) of two arbitrary frequencies included in the input signal is at least twice as long as the frequency resolution Δf (0.2 Hz). [0088] A simulation result in FIG. 13A represents a case where the data collection time MT is 2.56 seconds and the frequency resolution 15 Δf is about 0.4 Hz. The entry side plate thickness deviation ΔH inputted in the simulation of this case is the same as that in the case of FIG. 12A, and display formats of graphs showing the simulation result are also equivalent to those in FIG. 12A. [0089] 20 As can be seen from a lower frequency-domain graph in FIG. 13A, the frequencies 0.5 Hz and 1.0 Hz included in the entry side plate thickness deviation ΔH being the input signal are not adequately separated from each other both in the entry side plate thickness deviation ΔH and in the exit side plate thickness deviation 25 Δh. For this reason, the obtained values of both the attenuation (gain) and phase difference (phase) are inaccurate. [0090] A simulation result in FIG. 13B represents a case where the data collection time MT is 2.56 seconds and the frequency resolution 30 Δf is about 0.2 Hz. Here, a sinusoidal wave at a single frequency of 0.5 Hz is inputted as the entry side plate thickness deviation ΔH. In this case, as shown in lower graph in FIG. 13B, the frequency 35 of 0.5 Hz is properly separated in the exit side plate thickness deviation Δh as well, and the values of both the attenuation (gain) and phase difference (phase) are obtained almost accurately. [0091] 5 Note that the data collection time MT in this case is 2.56 seconds, which is a 2+α-second period that enables reproduction of the frequency 0.5 Hz of the exit side plate thickness deviation Δh. Thus, it is clear that the feedforward AGC adjustment can be achieved almost in the shortest time period. 10 [0092] In the meantime, the FFT significantly reduces the calculation time by limiting the data quantity to be processed to a power of 2. For this reason, it is not possible to set to the data quantity to be inputted to the FFT to an arbitrary number. Accordingly, the 15 data collection time MT is significantly changed by a combination of a sampling period and the data quantity. [0093] A case of the frequency resolution Δf at 0.1 Hz (a 10-second period) will be examined, for example. In this case, if the sampling 20 period is 10 ms and the data quantity is 1024 pieces, then the data collection time MT is 10.24 seconds. This data collection time MT is almost equal to the period of 10 seconds which derives from the frequency resolution Δf = 0.1 Hz. On the other hand, if the sampling period is 8 ms and the data quantity is 2048 pieces, then the data 25 collection time MT turns out to be 16.384 seconds, which is significantly larger than the period of 10 seconds. [0094] Next, a case of the frequency resolution Δf at 0.5 Hz (a 2-second period) will be examined. In this case, if the sampling 30 period is 10 ms and the data quantity is 256 pieces, then the data collection time MT is 2.56 seconds, which is longer than the aforementioned period of 2 seconds. Meanwhile, if the sampling 36 period is 8 ms and the data quantity is 254 pieces, then the data collection time MT turns out to be 2.048 seconds, which is substantially equal to the period of 2 seconds. [0095] 5 FIG. 14 is a diagram showing an example of a sampling period and data quantity search table. As shown in FIG. 14, the sampling period and data quantity search table is a table which stores a sampling period and a data quantity which can achieve an actual data collection time being most suitable for the minimum collection time 10 depending on the frequency resolution Δf. Here, the term "being most suitable" means "being larger than the minimum data collection time and closest to the minimum data collection time". [0096] In order to carry out the adjustment of the control parameters 15 of the feedforward control in accordance with the frequency response method using the FFT, the attenuation (gain) and the phase difference (phase) of the exit side plate thickness deviation Δh with respect to the entry side plate thickness deviation ΔH at the corresponding frequency as fast as possible (in the shortest time possible). To 20 this end, it is important to set the smallest data collection time MT to be used in the FFT. Regarding the data collection time MT, the minimum data collection time is determined based on the minimum resolution required corresponding to the plate thickness deviation (the entry side plate thickness deviation ΔH) actually coming into 25 being, and the number of samples as well as the data quantity required therefor are set accordingly. [0097] <3.4 Frequency Response Measurement Device> Next, a description will be given of a method of obtaining 30 the control gain GFF and the control output timing shift amount ΔTFF in the feedforward control of the plate thickness control by using the frequency response method. According to the frequency response 37 method, by setting the data collection time MT and subjecting the entry side plate thickness deviation ΔH and the exit side plate thickness deviation Δh to FFT processing, it is possible to obtain an amplitude in the data collection time MT. As suggested 5 previously, the unevenness of hardness is a variation in hardness in a longitudinal direction of the rolling object material 3, which occurs every time the rolling takes place. Therefore, the entry side plate thickness deviation ΔH, which is the plate thickness variation having occurred in the previous rolling process due to 10 the unevenness of hardness, and the exit side plate thickness deviation Δh representing the plate thickness variation after the rolling have substantially the same frequency. In addition, unlike the ordinary entry side plate thickness deviation ΔH, the exit side plate thickness deviation Δh due to the unevenness of hardness is 15 predicted to have a smaller value of attenuation. [0098] Accordingly, the adjustment of the control parameters of the feedforward control can be efficiently carried out by obtaining the control gain GFF and the control output timing shift amount ΔTFF of 20 the feedforward control in accordance with the following procedures. Note that the entry side plate thickness deviation ΔH in the following description often represents the entry side plate thickness deviation ΔHTRK after the transfer processing. Nonetheless, the description will be stated simply with the entry 25 side plate thickness deviation ΔH even in that case. [0099] (Procedure 1) The entry side plate thickness deviation ΔHTRK and the exit side plate thickness deviation Δh are subjected to the FFT processing. Note that the FFT processing is carried out at a 30 cycle corresponding to the frequency resolution Δf required for detection of a given plate thickness disturbance. (Procedure 2) Based on a result of the FFT processing, a 38 frequency, with which the exit side plate thickness deviation Δh becomes a predetermined value or above and the attenuation (gain) becomes the smallest is obtained as a frequency with a disturbance due to the unevenness of hardness (hereinafter referred to as an 5 adjustment target frequency), and disturbance discrimination frequency resolution Δfc, which can discriminate the frequency with the disturbance due to the unevenness of hardness and a frequency with a different disturbance, is obtained thereafter. (Procedure 3) The minimum data collection time is obtained 10 based on the disturbance discrimination frequency resolution Δfc. Furthermore, the number of samples and the sampling period in consideration of the FFT are set. (Procedure 4) The FFT is carried out by using the number of samples and the sampling period thus set, and the values of the 15 attenuation and the phase relation at the adjustment target frequency are obtained in terms of the entry side plate thickness deviation ΔHTRK and the exit side plate thickness deviation Δh. (Procedure 5) Based on the phase relation, the control gain GFF and the control output timing shift amount ΔTFF for the feedforward 20 control are obtained, and these values are outputted to the plate thickness control device 64. [0100] The above-described Procedures 1 to 5 are carried out by the control gain and timing shift amount setting device 102 shown in 25 FIG. 10. Specifically, a plate thickness disturbance measurement device 202 constituting the frequency response measurement device 201 carries out Procedure 1, while a plate thickness disturbance estimation device 203 constituting the same carries out Procedure 2, and a frequency response estimation device 204 constituting the 30 same carries out Procedures 3 and 4. In the meantime, Procedure 5 is carried out by the membership functions 105 to 107, the fuzzy inference device 108, and the parameter changing device 109. Then, 39 the adjustment of the control parameters (namely, the control gain GFF and the transfer time TFF of the entry side plate thickness deviation ΔH) in the feedforward control by the plate thickness control device 64 is executed by repeating the above-described 5 Procedures 4 and 5. [0101] Now, a detailed configuration of and contents of control by the frequency response measurement device 201 constituting the control gain and timing shift amount setting device 102 will be 10 described below. As shown in FIG. 10, the frequency response measurement device 201 includes the plate thickness disturbance measurement device 202, the plate thickness disturbance estimation device 203, and the frequency response estimation device 204. [0102] 15 FIG. 15 is a diagram showing an example of a configuration of the plate thickness disturbance measurement device 202. As shown in FIG. 15, the plate thickness disturbance measurement device 202 includes an entry side plate thickness deviation table 2021, an exit side plate thickness deviation table 2022, an entry side plate 20 thickness deviation FFT device 2023, an exit side plate thickness deviation FFT device 2024, and the like. [0103] In general, the frequency of a disturbance causing the plate thickness deviation (hereinafter referred to as a plate thickness 25 disturbance) varies depending not only on the rolling velocity but also on the type of the plate thickness disturbance, and the like. Here, the frequency resolution Δf is set to 0.1 Hz in the light of eliminating the plate thickness disturbances with frequencies equal to or above 0.5 Hz, for example. Note that these values may be 30 appropriately set and modified by a user based on actual situations of the plate thickness disturbances and on an actual state of operation. 40 [0104] When the frequency resolution Δf is set to 0.1 Hz, the minimum data collection time is equal to 10 seconds. Accordingly, the sampling period = 0.01 second and the data quantity = 1024 are 5 obtained by referring to the sampling period and data quantity search table. The subsequent FFT and other processing by the plate thickness disturbance measurement device 202 are carried out by using these numerical values. [0105] 10 The entry side plate thickness deviation table 2021 and the exit side plate thickness deviation table 2022, each of which is capable of storing 1024 pieces of data, are prepared in a storage device (not shown) of the plate thickness disturbance measurement device 202. Moreover, the values of ΔHTRK equivalent to the entry 15 side plate thickness deviation ΔH after the transfer processing and the values of the exit side plate thickness deviation Δh are inputted to the plate thickness disturbance measurement device 202 every 0.1 second of the sampling period, and are sequentially written into an address 0 to an address 1023 in the corresponding tables, 20 respectively. [0106] When the data writing into the entry side plate thickness deviation table 2021 and the exit side plate thickness deviation table 2022 is completed, the entry side plate thickness deviation 25 FFT device 2023 executes the FFT processing by using the data written into the entry side plate thickness deviation table 2021 as the input data. Likewise, the exit side plate thickness deviation FFT device 2024 executes the FFT processing by using the data written into the exit side plate thickness deviation table 2022 as the input data. 30 Then, an entry side plate thickness deviation frequency component H(f) and an exit side plate thickness deviation frequency component h(f) are obtained as a consequence of the FFT processing. 41 [0107] Here, values of the entry side plate thickness deviation frequency component H(f) and the exit side plate thickness deviation frequency component h(f) when the frequency is defined as f = m∙Δf 5 (where Δf is the frequency resolution) are obtained by calculating the value cm defined in Formula (7) described previously. In this case, the data of time-series signals f(n) included in Formula (7) are given by the entry side plate thickness deviation table 2021 and the exit side plate thickness deviation table 2022, 10 respectively. [0108] Accordingly, the time-series signals f(n) when the frequency is f = m∙Δf, namely, the amplitude attenuation and the phase difference between the entry side plate thickness deviation 15 frequency component H(f) and the exit side plate thickness deviation frequency component h(f) can be expressed in Formulae 8 based on Formulae (4-1) and (4-2) mentioned above: [Formulae 8] amplitude attenuation gain = \cm\[mm] (8-1); and 20 phase difference phase = — arg(c )[deg] (8-2). 2x m [0109] As a result of the above-described processing, in the plate thickness disturbance measurement device 202, an entry side plate thickness deviation amplitude Hg(m) and an entry side plate 25 thickness deviation phase Hp(m) are outputted from the entry side plate thickness deviation FFT device 2023. Likewise, an exit side plate thickness deviation amplitude hg(m) and an exit side plate thickness deviation phase hp(m) are outputted from the exit side plate thickness deviation FFT device 2024. 30 [0110] FIG. 16 is a graph showing an example of frequency-dependent 42 characteristics of the entry side plate thickness deviation amplitude Hg(m) and the exit side plate thickness deviation amplitude hg(m). Specifically, FIG. 16 is an example of the graph in which the horizontal axis indicates the frequency, and the values 5 of the entry side plate thickness deviation amplitude Hg(m) and the exit side plate thickness deviation amplitude hg(m) with respect to the frequencies are indicated along the vertical axis with a dashed line and a solid line, respectively. Now, contents of processing to be executed by the plate thickness disturbance 10 estimation device 203 will be described with reference to FIG. 16. [0111] In the example shown in FIG. 16, the dashed-line graph representing the entry side plate thickness deviation amplitude Hg(m) marks large values at frequency positions (A), (B), and (C), 15 namely, where the values of the frequencies are defined as mA∙Δf, mB∙Δf, and mC∙Δf, respectively. This phenomenon indicates that the entry side plate thickness deviation ΔH varies with the plate thickness disturbances having the aforementioned frequencies. [0112] 20 The entry side plate thickness deviation amplitude Hg(m) is changed into the exit side plate thickness deviation amplitude hg(m) due to the rolling phenomenon. Here, when the exit side plate thickness deviation amplitude hg(m) is sufficiently small as compared to the entry side plate thickness deviation amplitude Hg(m) 25 as observed at the position (A) in FIG. 16, this phenomenon represents large natural attenuation (a plate thickness variation suppression effect solely attributed to the rolling phenomenon) because the ordinary plate thickness control is operated effectively. Accordingly, the control parameters of the feedforward control do 30 not have to be adjusted for the frequency mA∙Δf at the position (A). [0113] On the other hand, when the exit side plate thickness deviation 43 amplitude hg(m) is hardly attenuated as compared to the entry side plate thickness deviation amplitude Hg(m) as observed at the position (B), the unevenness of hardness is presumably the cause of this phenomenon. It is therefore necessary to adjust the control 5 parameters of the feedforward control for the frequency mB∙Δf at the position (B). [0114] Meanwhile, even if the exit side plate thickness deviation amplitude hg(m) is not very large as observed at the position (C), 10 the adjustment of the control parameters of the feedforward control is determined to be necessary when the exit side plate thickness deviation amplitude hg(m) is not attenuated very much as compared to the entry side plate thickness deviation amplitude Hg(m). [0115] 15 Usually, noise components (actual noise or portions that can be regarded as noise) are included in the entry side plate thickness deviation ΔH and the exit side plate thickness deviation Δh. As a consequence, such noise components are also included in the entry side plate thickness deviation amplitude Hg(m) and the exit side 20 plate thickness deviation amplitude hg(m). Accordingly, a noise level Ln is set in advance to the entry side plate thickness deviation amplitude Hg(m) and the exit side plate thickness deviation amplitude hg(m) in this case. Then, the adjustment of the control parameters of the feedforward control is determined to be necessary 25 for such a frequency at which the exit side plate thickness deviation amplitude hg(m) exceeds the noise level Ln. [0116] In consideration of the above, in this embodiment, the plate thickness disturbance estimation device 203 first seeks the 30 frequency, at which the exit side plate thickness deviation amplitude hg(m) exceeds the preset noise level Ln, as a disturbance frequency fci. In the example in FIG. 16, the frequencies mA∙Δf, 44 mB∙Δf, and mC∙Δf are sought as the disturbance frequencies fci. Here, a parameter i may take a value of any of 1, 2, and so on to serve as an identification number when there are multiple disturbance frequencies fci. 5 [0117] Subsequently, the plate thickness disturbance estimation device 203 obtains a ratio of the exit side plate thickness deviation amplitude hg(m) with respect to the entry side plate thickness deviation amplitude Hg(m) for each of the sought disturbance 10 frequencies fci, and determines the disturbance frequency fci having the largest ratio as an adjustment target frequency fc. In the example in FIG. 16, the frequency mB∙Δf is sought as the adjustment target frequency fc. [0118] 15 As can be seen from the above description, the adjustment target frequency fc means the frequency at which the plate thickness variation caused by the unevenness of hardness becomes the largest. Accordingly, at the adjustment target frequency fc thus sought, it is necessary to appropriately adjust the control gain GFF and the 20 control output timing shift amount ΔTFF, which are the control parameters of the feedforward control. Note that while the disturbance frequencies fci are candidates for the adjustment target frequency fc, the disturbance frequencies fci also include frequencies of plate thickness variations that occur due to causes 25 other than the unevenness of hardness. [0119] Furthermore, the plate thickness disturbance estimation device 203 obtains the smallest value out of differences between the adjustment target frequency fc and the disturbance frequencies 30 fci except the adjustment target frequency fc. Then, a value obtained by multiplying the smallest value by 1/2 as the disturbance discrimination frequency resolution Δfc necessary for the 45 adjustment of the control parameters of the feedforward control. Specifically, the plate thickness disturbance estimation device 203 calculates: Δfc = (1/2)∙min{|fci-fc|:fci≠fc}. 5 Note that this calculation corresponds to processing to obtain a half value of a frequency difference between the adjustment target frequency fc and the disturbance frequency fci closest to the adjustment target frequency fc. [0120] 10 Incidentally, in the example in FIG. 16, the frequency mA∙Δf is closer to the frequency mB∙Δf being the adjustment target frequency fc than the frequency mC∙Δf is. Accordingly, the disturbance discrimination frequency resolution Δfc necessary for the adjustment of the control parameters of the feedforward control 15 is obtained as: Δfc = (mB-mA)∙Δf/2. [0121] The plate thickness disturbance estimation device 203 estimates that the adjustment target frequency fc obtained in the 20 above-described processing is the frequency of the plate thickness deviation caused by the disturbance due to the unevenness of hardness. Then, the plate thickness disturbance estimation device 203 outputs the adjustment target frequency fc and the disturbance discrimination frequency resolution Δfc thus obtained to the 25 frequency response estimation device 204. [0122] In this case, the adjustment target frequency fc is obtained based on the comparison between the entry side plate thickness deviation amplitude Hg(m) and the exit side plate thickness 30 deviation amplitude hg(m). Instead, the adjustment target frequency fc may be obtained based on a comparison between the entry side plate thickness deviation phase Hp(m) and the exit side plate 46 thickness deviation phase hp(m). [0123] FIG. 17 is a diagram showing an example of a configuration of the frequency response estimation device 204. The frequency 5 response estimation device 204 carries out the FFT based on the disturbance discrimination frequency resolution Δfc obtained by the plate thickness disturbance estimation device 203. To this end, the frequency response estimation device 204 first determines a sampling period (1/Δfs) and a data quantity Nc based on the 10 disturbance discrimination frequency resolution Δfc. [0124] As shown in FIG. 17, an entry side plate thickness deviation table 2041, an exit side plate thickness deviation table 2042, and a rolling load table 2043, each being capable of storing Nc pieces 15 of data, are prepared in the frequency response estimation device 204. Moreover, the values of the entry side plate thickness deviation ΔHTRK, the exit side plate thickness deviation Δh, and the rolling load PTRK, which are inputted to the frequency response estimation device 204 every sampling period (1/Δfs), are 20 sequentially written into an address 0 to an address Nc-1 in the corresponding tables, respectively. [0125] When the data writing into the address Nc-1 in the respective tables is completed, an entry side plate thickness deviation FFT 25 device 2044 executes the FFT processing on the data written into the entry side plate thickness deviation table 2041. Likewise, an exit side plate thickness deviation FFT device 2045 executes the FFT processing on the data written into the exit side plate thickness deviation table 2042, and a rolling load FFT device 2046 executes 30 the FFT processing on the data written into the rolling load table 2043. [0126] 47 Here, the data quantity Nc of the data used in the FFT processing by the frequency response estimation device 204 is usually set to a substantially smaller value than, such as about 1/10 as many as, the data quantity N of the data used in the FFT 5 processing by the plate thickness disturbance estimation device 203. For this reason, the FFT processing by the frequency response estimation device 204 is completed in a shorter period of time. [0127] As a consequence of the FFT processing, the entry side plate 10 thickness deviation FFT device 2044, the exit side plate thickness deviation FFT device 2045, and the rolling load FFT device 2046 obtain an entry side plate thickness deviation frequency component Hc(f), an exit side plate thickness deviation frequency component hc(f), and a rolling load frequency component Pc(f), respectively. 15 [0128] An entry side plate thickness to exit side plate thickness response measurement device 2047 calculates an entry side plate thickness to exit side plate thickness response Gh(f) based on the entry side plate thickness deviation frequency component Hc(f) and 20 the exit side plate thickness deviation frequency component hc(f) obtained as described above. Likewise, an entry side plate thickness to rolling load response measurement device 2048 calculates an entry side plate thickness to rolling load response GP(f) based on the entry side plate thickness deviation frequency 25 component Hc(f) and the rolling load deviation frequency component Pc(f). [0129] Here, the entry side plate thickness to exit side plate thickness response Gh(f) and the entry side plate thickness to 30 rolling load response GP(f) are calculated, respectively, in accordance with Formulae (9-1) and (9-2) below: [Formulae 9] 48 Gh(f) = c( f ) (9-1); and Hc(f) GP(f) = Pc( f ) (9-2). Hc(f) [0130] Next, the entry side plate thickness to exit side plate 5 thickness response measurement device 2047 assigns the adjustment target frequency fc obtained by the plate thickness disturbance estimation device 203 to the frequency f in Formula (9-1) and obtains its argument, thus calculating an entry side plate thickness and exit side plate thickness phase difference ΔTED. Likewise, the entry 10 side plate thickness to rolling load response measurement device 2048 assigns the adjustment target frequency fc to the frequency f in Formula (9-2) and obtains its argument, thus calculating an entry side plate thickness and rolling load phase difference ΔTEP. In the meantime, the exit side plate thickness deviation FFT device 15 2045 assigns the adjustment target frequency fc to the frequency f in the exit side plate thickness deviation frequency component hc(f) and obtains its absolute value, thus calculating an exit side plate thickness deviation PP value ΔhPP. [0131] 20 Specifically, the entry side plate thickness and exit side plate thickness phase difference ΔTED, the entry side plate thickness and rolling load phase difference ΔTEP, and the exit side plate thickness deviation PP value ΔhPP are calculated by Formulae (10-1) to (10-3) below: 25 [Formulae 10] AT = — arg( Gh(fc)) (10-1); 2x ATEP= — arg( GP(fc)) (10-2); and 2x 49 AhPP =\hc(fc)\ (10-3). [0132] In the above description, the plate thickness disturbance measurement device 202 and the frequency response estimation device 5 204 carry out the FFT after acquisition of result data for one period, such as N pieces of the data. Instead, it is also possible to carry out the FFT every time a piece of the result data is acquired. To this end, when a new piece of result data is written into the address 0 in the table for storing the result data such as the entry side 10 plate thickness deviation tables 2021 and 2024, the new piece of data is to be written after the data in the addresses 0 to N-1 are shifted to the addresses 1 to N. In this way, the newest result data is always written into the table such as entry side plate thickness deviation tables 2021 and 2024. As a consequence, the 15 entry side plate thickness deviation FFT devices 2023, 2044, and the like can carry out the FFT within a period of acquisition of the result data at the shortest. [0133] <3.5 Membership Functions and Fuzzy Inference Device> 20 Meanwhile, the feedforward control aims to reduce the exit side plate thickness deviation Δh by using the entry side plate thickness deviation ΔH. Therefore, the exit side plate thickness deviation Δh is the control target. An effect of the variation in deformation resistance representing the unevenness of hardness 25 emerges as the entry side plate thickness deviation ΔH on the entry side of the fourth stand. Accordingly, the plate thickness control device 64 of the fourth stand carries out the feedforward control by using the entry side plate thickness deviation ΔH. Then, the feedforward control adjustment device 101 adjusts the feedforward 30 control based on the phase relation between the entry side plate thickness deviation ΔH and the exit side plate thickness deviation Δh. 50 [0134] Consequently, as the feedforward control takes effect favorably, the detection value of the exit side plate thickness deviation Δh is decreased down to zero in an ideal case. It is 5 difficult to obtain the phase relation between the entry side plate thickness deviation ΔH and the exit side plate thickness deviation Δh in this case. On the other hand, the rolling load P varies significantly as a consequence of elimination of the exit side plate thickness deviation Δh attributed to the unevenness of hardness, 10 so that the rolling load P can be used instead of the exit side plate thickness deviation Δh. Accordingly, in this embodiment, the feedforward control adjustment device 101 has a function to adjust the control output timing shift amount ΔTFF of the feedforward control based on a phase relation between the entry side plate 15 thickness deviation ΔH and the control output timing shift amount ΔTFF. [0135] As shown in FIG. 10, the control gain and timing shift amount setting device 102 includes the membership functions 105, 106 and 20 107, and the fuzzy inference device 108. These constituents deliver the above-mentioned adjustment function. [0136] First, the membership function 105 obtains values SHS and SHB by using the exit side plate thickness deviation PP value ΔhPP as 25 the input. Here, the value SHS is a value indicating degrees of a case where the exit side plate thickness deviation Δh is small, and the value SHB is a value indicating degrees of a case where the exit side plate thickness deviation Δh is large. [0137] 30 Likewise, the membership function 106 obtains values TEDB, TEDM, TEDZ, TEDP, and TEDT by using the entry side plate thickness and exit side plate thickness phase difference ΔTED as the input. 51 Here, the value TEDB is a value indicating degrees of a case where the entry side plate thickness and exit side plate thickness phase difference ΔTED has a large negative value, while the value TEDM is a value indicating degrees of a case where the entry side 5 plate thickness and exit side plate thickness phase difference ΔTED is on the negative side. Meanwhile, the value TEDZ is a value indicating degrees of a case where the entry side plate thickness and exit side plate thickness phase difference ΔTED is zero. In the meantime, the value TEDP is a value indicating degrees of a case 10 where the entry side plate thickness and exit side plate thickness phase difference ΔTED is on the positive side, while the value TEDT is a value indicating degrees of a case where the entry side plate thickness and exit side plate thickness phase difference ΔTED has a large positive value. 15 [0138] In the meantime, the membership function 107 obtains values TEPM, TEPZ, and TEPP by using the entry side plate thickness and rolling load phase difference ΔTEP as the input. Here, the value TEPM is a value indicating degrees of a case 20 where the entry side plate thickness and rolling load phase difference ΔTEP is on the negative side. The value TEPZ is a value indicating degrees of a case where the entry side plate thickness and rolling load phase difference ΔTEP is zero. The value TEPP is a value indicating degrees of a case where the entry side plate 25 thickness and rolling load phase difference ΔTEP is on the positive side. [0139] Note that respective thresholds provided on the horizontal axes in the membership functions 105, 106, and 106 apply 30 predetermined values. A value SB applicable to the membership function 105 is a threshold used to determine whether or not it is appropriate to carry out the adjustment of the feedforward control 52 by using the exit side plate thickness deviation Δh. For example, if the exit side plate thickness deviation Δh is not used for the adjustment of the feedforward control when the exit side plate thickness deviation Δh is equal to or below 1 μm, then the threshold 5 SB is equal to 1 μm. As described above, when a range of variation of the exit side plate thickness deviation Δh fall within a predetermined range, the control gain and timing shift amount setting device 102 of this embodiment refers to the phase of the variation in rolling load P instead of the exit side plate thickness 10 deviation Δh. [0140] Values DB and DT applicable to the membership function 106 are thresholds used to determine that the control gain is too high. For example, the control gain is determined to be too high if the 15 entry side plate thickness and exit side plate thickness phase difference ΔTED exceeds 90 degrees. In this case, the value DB is set equal to -90 degrees while the value DT is set equal to 90 degrees, and the control to reduce the control gain is carried out. [0141] 20 Values DM and DP applicable to the membership function 106 and values PM and PP applicable to the membership function 107 are thresholds used to determine that it is unnecessary to adjust the output timing shift amount. For example, the adjustment of the control output timing shift amount ΔTFF is determined to be 25 unnecessary if the entry side plate thickness and exit side plate thickness phase difference ΔTED falls within a range of ±20 degrees inclusive. In this case, the value DM is set equal to -20 degrees while the value DP is set equal to 20 degrees. Likewise, the value PM is set equal to -20 degrees while the value PP is set equal to 30 20 degrees. Here, it is to be noted that these values are mere examples and can be modified as appropriate depending on the rolling situation and characteristics of facilities. 53 [0142] Meanwhile, the entry side plate thickness and exit side plate thickness phase difference ΔTED and the entry side plate thickness and rolling load phase difference ΔTEP in the case where the effect 5 of the feedforward control is maximized by rendering the exit side plate thickness deviation Δh smallest are set to values DZ and PZ. Note that these phase difference values are assumed to be predetermined based on the result data obtained by manual adjustment during rolling simulation or by actual rolling, for example. The 10 control gain and timing shift amount setting device 102 determines the control output timing shift amount ΔTFF by comparing the inputted phase difference such as the entry side plate thickness and exit side plate thickness phase difference ΔTED with the predetermined values. 15 [0143] The fuzzy inference device 108 obtains values TFFP, TFFM, GFFP, and GFFM by using the values SHS, SHB, TEDB, TEDM, TEDZ, TEDP, TEDT, TEPM, TEPZ, and TEPP, which are obtained by the membership functions 105, 106, and 107. Here, the values TFFP and TFFM are values 20 indicating degrees to change the control output timing shift amount Δ TFF to an increasing side and degrees to change the control output timing shift amount Δ TFF to a decreasing side, respectively. Meanwhile, the values GFFP and GFFM are values indicating degrees to change the control gain GFF to an increasing side and degrees to 25 change the control gain GFF to a decreasing side, respectively. [0144] While there are various inference rules in general, the fuzzy inference device 108 of this embodiment performs processing expressed by the following conditions: 30 where IF (A and B) then C is applicable: C = min (A, B), and where IF (A or B) then C is applicable: C = max (A, B). [0145] 54 If the exit side plate thickness deviation Δh is large and the entry side plate thickness and exit side plate thickness phase difference ΔTED is zero, then the control gain GFF of the feedforward control is considered to be small. Accordingly, the following 5 inference rule is applied: IF (SHB and TEDZ) then GFFP. [0146] Meanwhile, if the exit side plate thickness deviation Δh is large and the entry side plate thickness and exit side plate 10 thickness phase difference ΔTED is present, then the control output timing shift amount ΔTFF is determined to be deviated. Accordingly, the following inference rules are applied because the exit side plate thickness deviation Δh is expected to be reduced by eliminating the deviation: 15 IF (SHB and TEDP) then TFFP, and IF (SHB and TEDM) then TFFM. [0147] In the meantime, if the exit side plate thickness deviation Δh is large and the phase difference between the entry side plate 20 thickness and the exit side plate thickness largely exceeds 90 degrees, then the control gain GFF of the feedforward control is determined to be too large. In this case, it is considered preferable to reduce the control gain GFF first and then to adjust the control output timing shift amount ΔTFF after achieving the 25 appropriate control gain. Accordingly, the following inference rules are applied in this case: IF (SHB and TEDT) then GFFM, and IF (SHB and TEDB) then GFFM. [0148] 30 Meanwhile, if the exit side plate thickness deviation Δh is small and the entry side plate thickness and rolling load phase difference ΔTEP is large, then the exit side plate thickness 55 deviation Δh is expected to be further reduced by adjusting the control output timing shift amount ΔTFF. Accordingly, the following inference rules are applied in this case: IF (SHB and TEPP) then TFFM, and 5 IF (SHB and TEPM) then TFFP. [0149] According to the simulation of the rolling phenomenon, when the entry side plate thickness and rolling load phase difference ΔTEP is on the negative side, the entry side plate thickness and 10 rolling load phase difference ΔTEP is reduced by changing the control output timing shift amount ΔTFF to the increasing side. On the other hand, when the entry side plate thickness and rolling load phase difference ΔTEP is on the positive side, the entry side plate thickness and rolling load phase difference ΔTEP is reduced by 15 changing the control output timing shift amount ΔTFF to the decreasing side. The aforementioned inference rules are determined based on these simulation results. [0150] The relations in FIGs. 24A and 24B illustrate how the control 20 state quantities such as the entry side plate thickness deviation ΔH and the exit side plate thickness deviation Δh are changed before and after the control. Since the rolling load P is determined by plate thickness variations on the entry side and the exit side and on the tensions on the entry side and the exit side, a relation 25 between the entry side plate thickness and rolling load phase difference ΔTEP and the control output timing shift amount ΔTFF is different from the one illustrated in FIGs. 24A and 24B. Nonetheless, if a changing trend of the entry side plate thickness and rolling load phase difference ΔTEP in the case of the change in 30 control output timing shift amount ΔTFF is available, then the entry side plate thickness and rolling load phase difference ΔTEP can be used for the adjustment of the control output timing shift amount 56 ΔTFF as with this embodiment. [0151] By using the inference rules mentioned above, it is possible to obtain the value TFFP being the degrees to change the control 5 output timing shift amount Δ TFF for the feedforward control to the increasing side and the value TFFM being the degrees to change the control output timing shift amount Δ TFF to the decreasing side. Moreover, it is possible to obtain the value GFFP being the degrees to change the control gain GFF for the feedforward control to the 10 increasing side and the value GFFM being the degrees to change the control gain GFF to the decreasing side. [0152] Note that the above-described inference rules are mere examples, and the present invention is not limited to these inference 15 rules. For example, any inference rules are applicable as long as such inference rules are designed to reduce the exit side plate thickness deviation Δh by changing the control gain GFF, the control output timing shift amount Δ TFF, and the like for the feedforward control. In the meantime, instead of the simulation of the rolling 20 phenomenon, the inference rules may be determined based on result data obtained by the manual adjustment in the actual rolling operation. Such data are more likely to conform to the actual rolling phenomenon. [0153] 25 Using the degrees of changes TFFP, TFFM, GFFP, and GFFM obtained as described above, the parameter changing device 109 changes the control gain GFF and the control output timing shift amount Δ TFF for the feedforward control in accordance with the Formulae (11-1) and (11-2) below: 30 [Formulae 11] GFF=GFF+GFFP-CGFFP+GFFM-CGFFM (11-1); and ATFF =ATFF+ TFFP-CTFFP+ TFFM-CTFFM (11-2). 57 [0154] Here, values CTFFP, CTFFM, CGFFP, and CGFFM are adjustment parameters. The value CTFFP is a value indicating an amount of change to the increasing side of the control output timing shift amount 5 Δ TFF for each time while the value CTFFM is a value indicating an amount of change to the decreasing side thereof. Meanwhile, the value CGFFP is a value indicating an amount of change to the increasing side of the control gain GFF for each time while the value CGFFM is a value indicating an amount of change to the decreasing side 10 thereof. [0155] As described above, the feedforward control adjustment device 101 can always adjust the control gain GFF and the control output timing shift amount Δ TFF for the feedforward control to optimum 15 conditions in the feedforward control in the plate thickness control device 64. As a consequence, the control effect of the feedforward control is substantially improved. [0156] «4. Simulation Based on Extended Configuration» 20 Subsequently, a result of verification by simulation on the effect of the feedforward control adjustment device 101 will be described with reference to FIGs. 18 to 22. [0157] FIG. 18 includes graphs showing an example of a result of 25 simulation obtained when the tension control on the entry side and the exit side of the fourth stand rolling mill 14 is carried out by proportional-integral control, and when feedback control and feedforward control of the exit side plate thickness of the fourth stand rolling mill 14 is carried out. Conditions of this simulation 30 are equivalent to the case of adding the plate thickness control of the feedforward control to the conditions of the simulation in FIG. 8. In FIG. 18, regarding the "plate thickness variation", the 58 entry side plate thickness variation (the entry side plate thickness deviation ΔH) is indicated with a solid line while the exit side plate thickness variation (the exit side plate thickness deviation Δh) is indicated with a dashed line. Likewise, regarding the 5 "tension variation", a variation in tension on the entry side is indicated with a solid line while a variation in tension on the exit side is indicated with a dashed line. Regarding the "loads variation", a variation in rolling load is indicated with a solid line while a variation in deformation resistance is indicated with 10 a dashed line. In the meantime, the phase of the entry side plate thickness is indicated with a vertical solid line, the phase of the exit side plate thickness is indicated with a vertical dashed line, and the phase of the rolling load is indicated with a vertical chain dashed line. The same definitions for the solid line, the dashed 15 line, and the chain dashed apply to FIGs. 19 to 22 as well. [0158] In FIG. 18, the entry side plate thickness and exit side plate thickness phase difference ΔTED is indicated as an interval between the vertical solid line and the vertical dashed line, and is observed 20 as a leading phase. Meanwhile, the entry side plate thickness and rolling load phase difference ΔTEP is indicated as an interval between the vertical solid line and the vertical chain dashed line, and is observed as a lagging phase. Moreover, the exit side plate thickness deviation Δh is large in this example. In this case, 25 therefore, the control output timing shift amount ΔTFF for the feedforward control is firstly changed to a leading phase side, i.e., in the increasing direction. Thus, a result of simulation turns out as shown in FIG. 19. [0159] 30 FIG. 19 includes graphs showing an example of a result of simulation obtained when the control output timing shift amount ΔTFF for the feedforward control is changed in the increasing direction 59 under the same simulation conditions as those in FIG. 18. It is apparent from FIG. 19 that both the entry side plate thickness and exit side plate thickness phase difference ΔTED indicated as the interval between the solid line and the dashed line, and the entry 5 side plate thickness and rolling load phase difference ΔTEP indicated as the interval between the solid line and the chain dashed line, become smaller than those in the example in FIG. 18. Furthermore, the amplitude of the exit side plate thickness deviation Δh also becomes slightly smaller. 10 [0160] This result indicates that the control output timing shift amount ΔTFF for the feedforward control is appropriate but the control gain GFF runs short. Accordingly, when the control output timing shift amount ΔTFF for the feedforward control is increased, 15 a result of simulation turns out as shown in FIG. 20. [0161] FIG. 20 includes graphs showing an example of a result of simulation obtained when the control gain GFF for the feedforward control is increased under the same simulation conditions as those 20 in FIG. 19. It is apparent from FIG. 20 that the exit side plate thickness deviation Δh becomes significantly smaller than that in the example in FIG. 19. However, the entry side plate thickness and exit side plate thickness phase difference ΔTED is determined as a lagging phase. Accordingly, when the control output timing 25 shift amount ΔTFF for the feedforward control is changed to a lagging phase side, a result of simulation turns out as shown in FIG. 21. [0162] FIG. 21 includes graphs showing an example of a result of simulation obtained when the control output timing shift amount ΔTFF 30 for the feedforward control is changed in the decreasing direction under the same simulation conditions as those in FIG. 20. In the result of simulation shown in FIG. 21, the exit side plate thickness 60 deviation Δh is almost zero. Accordingly, the exit side plate thickness deviation Δh seems to be almost eliminated. [0163] Note that the entry side plate thickness and rolling load phase 5 difference ΔTEP exhibits a slightly lagging phase in the example in FIG. 21. Accordingly, by setting the value of the lagging phase as the value PZ applicable to the membership function 107, it is also possible to further adjust the control output timing shift amount ΔTFF for the feedforward control while using the entry side 10 plate thickness and rolling load phase difference ΔTEP. [0164] FIG. 22 includes graphs showing an example of a result of simulation obtained when the control output timing shift amount ΔTFF for the feedforward control is shifted to the opposite side from 15 that in the example in FIG. 18. In FIG. 22, the entry side plate thickness and exit side plate thickness phase difference ΔTED emerges as a leading phase. In this simulation therefore, it turns out that when the entry side plate thickness and rolling load phase difference ΔTEP is on the positive side, the exit side plate thickness deviation 20 Δh can be reduced by adjusting the control output timing shift amount ΔTFF to the decreasing side. [0165] As described above, according to this embodiment, it is possible to improve the effect of the feedforward control by 25 modifying the control output timing shift amount ΔTFF and the control gain GFF of the feedforward control while retrieving the rolling result data during the rolling operation. Moreover, in this embodiment, the entry side plate thickness and rolling load phase difference ΔTEP and the control gain GFF are basically obtained on 30 the basis of the result of subjecting the entry side plate thickness deviation ΔH and the exit side plate thickness deviation Δh to the FFT processing. For this reason, even if a lot of the frequency 61 components are included in the entry side plate thickness deviation ΔH and the exit side plate thickness deviation Δh, it is facilitated to identify the frequency of the plate thickness variation attributed to the unevenness of hardness out of the frequency 5 components, and to obtain the phase difference δ between the entry side plate thickness deviation ΔH and the exit side plate thickness deviation Δh to be defined as the control target. As a consequence, since the control output timing shift amount ΔTFF and the control gain GFF mentioned above are obtained more appropriately, and the 10 effect of the feedforward control can be significantly improved. In other words, this embodiment enables the adjustment of the control output timing for the feedforward control efficiently and in a short time by using frequency characteristics of the variations in pre-control state quantity and post-control state quantity. 15 [0166] «5. Modified Examples of Embodiment» In the above-described embodiment, the control output timing shift amount ΔTFF and the control gain GFF for the feedforward control 20 are adjusted by using the entry side plate thickness and exit side plate thickness phase difference ΔTED and the entry side plate thickness and rolling load phase difference ΔTEP. However, the method of adjusting the control output timing shift amount ΔTFF and the control gain GFF for the feedforward control is not limited to 25 this method. [0167] As shown in FIG. 17, the entry side plate thickness to rolling load response measurement device 2048 and the entry side plate thickness to exit side plate thickness response measurement device 30 2047 calculate the entry side plate thickness to rolling load response GP(f) and the entry side plate thickness to exit side plate thickness response Gh(f). It is therefore possible to obtain 62 attenuation ratios |GP(fc)| and |Gh(fc)| at the adjustment target frequency fc. Accordingly, the control rules applicable to the fuzzy inference device 108 are increased by using data of these attenuation ratios. For example, the control rule IF (SHB and TEDP) 5 then TFFM is changed in such a way as to execute GFFP at the same time when the ratio |Gh(fc)| is large (when the attenuation is small). [0168] It is thus possible to adjust the control output timing shift 10 amount ΔTFF and the control gain GFF for the feedforward control. In this case, effects such as reduction in response time necessary for the adjustment are also expected. [0169] 15 Modified Example 2 assumes an embodiment further including a database, which stores result values of the control output timing shift amount ΔTFF and the control gain GFF for the feedforward control with which the exit side plate thickness deviation Δh satisfying prescribed production quality has been obtained in an actual rolling 20 process. The control output timing shift amount ΔTFF and the control gain GFF are stored in this database while being associated with rolling conditions including the steel type of the rolling object material 3, the rolling velocity, a target plate thickness, and the like when the exit side plate thickness deviation Δh satisfying the 25 prescribed production quality was obtained in the rolling process. [0170] In this case, the database is searched at the start of rolling. If the database stores the data obtained under similar rolling conditions, then the control output timing shift amount ΔTFF and the 30 control gain GFF obtained under the similar conditions can be retrieved and used. Accordingly, in this modified example, it is possible to conduct further modifications by using the control 63 parameters for the feedforward control resulting from the past rolling process. As a consequence, it is possible to further enhance the control effect of the feedforward control. [0171] 5 The basic concept to adjust the control gain and the phase of the feedforward control in the above-described embodiment is also applicable to plate thickness control involving roll eccentricity in a single stand rolling mill, and the like. In this case, based 10 on an entry side plate thickness deviation detected with an entry side plate thickness gauge of the single stand rolling mill, its exit side plate thickness deviation is controlled while defining a roll gap (an interval between vertical work rolls) as an operating end, for example. This type of rolling control is often called a 15 gauge meter type, and a basic formula of its rolling phenomenon is expressed in Formula (12) below: [Formula 12] Ah = — + AS (12), M where Δh is an exit side plate thickness deviation, 20 ΔP is a rolling load deviation, ΔS is a roll gap deviation, and M is a mill constant. [0172] Here, if only an entry side plate thickness deviation ΔH and 25 the exit side plate thickness deviation Δh are taken into account for the rolling load deviation ΔP, then the rolling load deviation ΔP can be expressed by Formula (13) below: AP = — AH + — Ah (13), dH dh where ΔH is the entry side plate thickness deviation. 30 [0173] 64 Taking into account Formula (12), it turns out that a relation expressed by Formula (14) below holds true between the roll gap deviation ΔS and the entry side plate thickness deviation ΔH in order to satisfy the exit side plate thickness deviation Δh = 0 in Formula 5 (13): AS = - — — AH (14) . M dH [0174] Formula (14) means that the exit side plate thickness deviation Δh can be reduced to zero by subjecting the roll gap 10 deviation ΔS to feedforward proportional control based on the entry side plate thickness deviation ΔH. Specifically, the roll gap deviation ΔS can be obtained by multiplying the entry side plate thickness deviation ΔH by a control gain expressed in Formula (15) below: 15 [Formula 15] -P (15). M dH [0175] Furthermore, if the unevenness of hardness of the rolling object material 3, namely, a deformation resistance variation Δk 20 is also taken into account for the rolling load deviation ΔP, then the rolling load deviation ΔP can be expressed by Formula (16) below: [Formula 16] AP = — AH + — Ah + — Ak (16), dH dh dk where Δk is the deformation resistance variation. 25 [0176] Taking into account Formula (12), it turns out that a relation expressed by Formula (17) below holds true between the roll gap deviation ΔS and the entry side plate thickness deviation ΔH as well as the deformation resistance variation Δk in order to satisfy the 65 exit side plate thickness deviation Δh = 0 in Formula (16): [Formula 17] AS = —1|iPAH + — Ml (17). M {dH dk J [0177] 5 Here, if the entry side plate thickness deviation ΔH and the deformation resistance variation Δk have the same frequency component, then Formula (17) can be expressed as Formula (18) below: [Formula 18] AS = - — — AHsin(