TITLE OF THE INVENTION
DIAGNOSTIC APPARATUS AND DIAGNOSTIC METHOD FOR POWER PLANTS
FIELD OF THE INVENTION
The present invention relates to a diagnostic apparatus and a diagnostic method for power plants.
BACKGROUND OF THE INVENTION
A diagnostic apparatus for a power plant detects occurrence of an abnormal event or a trouble based on a measurement signal from the plant, when an abnormal transient phenomenon or the trouble has occurred in the plant.
A diagnostic apparatus for plants using an Adaptive Resonance Theory (ART) is disclosed in Japanese Unexamined Patent Publication No. 2005-165375 as a publicly-known diagnostic apparatus for plants. Herein, the ART means a technique for sorting out multidimensional data into categories depending on the similarities.
In the technique of the diagnostic apparatus for plants disclosed in Japanese Unexamined Patent Publication No. 2005-165375, the measurement signals during a period of time when the state of the plant is considered to be normal are first extracted, as data for constructing a model, from the past measurement signals in which operation data of the plant have been recorded. The data for constructing amodel are then sorted
out into a plurality of categories (normal categories) by using the ART to create a normal model to be used for diagnosis. Subsequently, the present measurement signals of the plant are sorted out into categories by the ART. When the present measurement signals do not match the normal model, i.e., when the present measurement signals cannot be sorted out into the normal categories, a new category (novel category) is created. That is, occurrence of a novel category means that the tendency of the measurement signals has been changed and that the state of the plant has been changed. Accordingly, the technique disclosed in the aforementioned Japanese Unexamined Patent is one in which: occurrence of an abnormal event is determined by occurrence of a novel category, and when a ratio of occurrences of novel categories exceeds a threshold value, it is determined that an abnormal event has occurred.
SUMMARY OF THE INVENTION
A power plant is operated under various conditions, such as start, stop, constant load, and variable load. When an operating condition is changed, a range in which a measurement signal varies is changed.
In addition, normalization processing is used as the preprocessing of the measurement signal tobe used for diagnosis. In the normalization processing, the measurement signal is subjected to the processing such that the lower limit of the
normalization is made to be 0 and the upper limit thereof is made to be 1. It is needed to set the lower limit and upper limit of the normalization in advance. In a diagnostic apparatus according to a related art, measurement signals are subjected to processing under the same normalization condition, irrespective of operating conditions. Accordingly, it is needed to set the normalization range to be so wide that it includes the range in which the measurement signals under all of the operating conditions vary.
When a normalization range is wide in comparison with the range in which a measurement signal varies, a change in a value, after the measurement signal has been normalized, becomes small. Accordingly, a change in the tendency of the measurement signals, occurring when an abnormal event has happened, cannot be detected, and there have been sometimes the cases where a novel category does not occur even when an abnormal event has occurred. This may cause a non-detection.
An object of the present invention is to provide a diagnostic apparatus in which a non-detection is suppressed by determining a normalization range appropriately in accordance with an operating condition, thereby allowing diagnostic accuracy to be improved.
A diagnostic apparatus for power plants, in which an operating state of a power plant is diagnosed based on a measurement signal in which a state quantity of the plant has
been measured and a diagnostic result is displayed on an image display device, comprises: a model construction unit that constructs a model to be used for diagnosis by applying a measurement signal in which a state quantity of the power plant has been measured to the diagnostic apparatus for power plants; a model definition unit that defines both an operating condition diagnosed by the model and a method of normalizing the measurement signal; and a diagnostic unit that diagnoses an operating state of the power plant by using the model constructed with the model construction unit, in which the model definition unit includes both an operating condition determination unit for determining an operating condition of the power plant and a normalization condition determination unit for determining a normalization condition of data for every operating condition determined by the operating condition determination unit, and in which the diagnostic unit executes diagnosis by switching a diagnostic model in accordance with an operating condition.
By using the diagnostic apparatus for power plants according to the present invention, a non-detection, in which an abnormal event is not detected when the event has occurred, can be reduced and diagnostic accuracy can be improved. Further, a normalization range can be determined automatically, and hence a period of time for adjusting the diagnostic apparatus can be shortened.
BRIEF DESCRIPTION OF THE DRAWINGS
Fig. 1 is a control block view illustrating the configuration of a diagnostic apparatus for power plants according to an embodiment of the present invention;
Fig. 2A is a flowchart view illustrating the basic operations of the diagnostic apparatus for power plants illustrated in Fig. 1;
Figs. 2B to 2D are views illustrating the timings of the operations;
Figs . 3A to 3C are views illustrating an example of mounting a function of sorting out data in a model construction unit and a diagnostic unit in the diagnostic apparatus for power plants illustrated in Fig. 1;
Figs . 4A to 4C are views illustrating an example of sorting out measurement signals by the model construction unit in the diagnostic apparatus for power plants illustrated in Fig. 1;
Figs. 5 is a graph explaining the relationship between a startingmode and a process value in the power plant illustrated in Fig. 4A;
Figs. 6A and 6B are views explaining movements of an operating condition determination unit 500 in the diagnostic apparatus for power plants illustrated in Fig. 1;
Figs. 7A and 7B are views explaining a first embodiment of a normalization condition determination unit 600 in the
diagnostic apparatus for power plants illustrated in Fig. 1;
Figs. 8A and 8B are views explaining a second embodiment of the normalization condition determination unit 600 in the diagnostic apparatus for power plants illustrated in Fig. 1;
Fig. 9 is a graph explaining a third embodiment of the normalization condition determination unit 600 in the diagnostic apparatus for power plants illustrated in Fig. 1;
Figs. 10A and 10B are views explaining flowcharts for the operations in a model construction mode and a diagnostic mode in the diagnostic apparatus for power plants illustrated in Fig. 1;
Figs. 11A to 11C are tables explaining aspects of the data stored in a database in the diagnostic apparatus for power plants illustrated in Fig. 1;
Figs. 12 is a graph explaining an advantage of applying the diagnostic apparatus for power plants illustrated in Fig. 1; and
Fig. 13 is a view explaining a screen displayed on an image display device in the diagnostic apparatus for power plants illustrated in Fig. 1.
DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
Hereinafter, a diagnostic apparatus for power plants, which is an embodiment of the present invention, will be described with reference to the accompanying drawings.
[1. Structure of Diagnostic Apparatus]
Fig. 1 is a block view explaining a diagnostic apparatus for power plants, which is an embodiment of the present invention. In the diagnostic apparatus for power plants illustrated in Fig. 1, the state of a plant 100 is diagnosed by the diagnostic apparatus 200.
The diagnostic apparatus 200 comprises, as arithmetic devices for forming the diagnostic apparatus 200, a model definition unit 400, a model construction unit 700, and a diagnostic unit 800. This diagnostic apparatus 200 comprises, as databases, a measurement signal database 310, a model definition database 320, and a diagnostic model database 330. In Fig. 1, a database is abbreviated as DB.
Electronized information is stored in the databases of the measurement signal database 310, the model definition database 320, and the diagnostic model database 330, and the information is usually referred to as electronic files (electronic data).
The model construction unit 700 creates, from a measurement signal in which an operating state of the plant 100 has been measured, a diagnostic model in which a normal state of the plant has been learned, based on accumulated data in which measurement signals of the past operating states of the plant 100 have been accumulated.
The model definition unit 400 defines both an operating
condition diagnosed by the diagnostic model and a method of normalizing the measurement signal occurring when the model is constructed.
The diagnostic unit 800 compares the data of the value of the diagnostic model created by the model construction unit 700 with the data of the measured measurement signal of the plant 100. When the measurement signal matches the value of the diagnostic model in which a normal state has been learned, the diagnostic unit 800 determines that the state of the plant is normal; on the other hand, when the two are not equal to each other, the diagnostic unit 800 determines that the state of the plant is abnormal.
The diagnostic apparatus 200 also comprises, as an interface with an external device, an external input interface 210 and an external output interface 220.
Measurement signals 1 in which various state quantities indicating an operating state of the plant 100 have beenmeasured, and an external input signal 2 created by an operation of an external input device 910, which are formed by a keyboard 920 and a mouse 930 installed in an operations management room 900, are loaded into the diagnostic apparatus 200 via the external input interface 210. Also, image display information 11 is outputted to an image display device 940 installed in the operations management room 900 via the external output interface 220.
In the diagnostic apparatus of plants according to the present embodiment, the model definition unit 400, the model construction unit 700, the diagnostic unit 800, the measurement signal database 310, the model definition database 320, and the diagnostic model database 330 are provided in the diagnostic apparatus 200; however, part of them may be arranged outside the diagnostic apparatus 200 such that only data are communicated between these devices.
In the diagnostic apparatus for plants according to the present embodiment, the single plant 100 is to be diagnosed; however, it is also possible to diagnose a plurality of the plants 100 by the single diagnostic apparatus 100.
Subsequently, operations of the diagnostic apparatus 200 provided in the diagnostic apparatus for plants according to the present embodiment will be described.
In the diagnostic apparatus for plants according to the present embodiment illustrated in Fig. 1, the measurement signals 1 in which various state quantities of the plant 100 have been measured are loaded via the external input interface 210 . Measurement signals 3 are stored in the measurement signal database 310 provided in the diagnostic apparatus 200. condition determination unit 500 and a normalization condition determination unit 600, respectively. With respect to an input of a measurement signal 4, the model definition unit 400 outputs model definition information 5 to the model definition database

320.
A power plant has four operation modes including a constant load operation where the power plant is operated with a constant output, a starting operation where the power plant is started, a stop operation where the power plant is stopped, and a variable load operation where the output is changed. The operating condition determination unit 500 sorts out the data of the plant 100 accumulated in the measurement signal database 310, for each of the operation modes during a constant load operation, a starting operation, a stop operation, and a variable load operation. The operating condition determination unit 500 also extracts a characteristic quantity of each of the operation modes. Details of this function will be described with reference to Figs. 6A and 6B.
Also, the model construction unit 700 executes normalization processing on the measurement signal as the preprocessing for constructing a model. The normalization condition determination unit 600 determines an appropriate normalization condition for each of the operation modes. Details of this function will be described later with reference to Figs. 7A to 9. The aforementioned model definition information 5 is formed by operating condition data and normalization condition data.
The model construction unit 700 constructs a model to be used for diagnosis by using the measurement signals 4 of
the plant 100 accumulated in the measurement signal database 310 and the model definition information 6 stored in the model definition database 320. The model information 8 created by the model construction unit 700 are stored in the diagnostic model database 330.
As a technique of mounting the model construction unit 700, there are clustering techniques, such as adaptive resonance theory and vector quantization. In addition, the model to be used for diagnosis is not limited to the aforementioned clustering techniques, but a model using a physics equation and a statistical model, such as a neural network, can also be used.
When the measurement signals 4 have been inputted, the diagnostic unit 800 installed in the diagnostic apparatus 200 diagnoses the operating state of the plant 100 by referring to the model definition information 7 in the model definition database 320 and the model information 9 in the diagnostic model database 330, and outputs a diagnostic result 10 thereof.
The diagnostic result 10 obtained by the diagnostic unit 800 diagnosing the present operating state of the plant 100 is transmitted, as the image display information 11, to the image display device 940 installed in the operations management room 900 via the external output interface 220. Thereby, an operator in the operations management room 900 is notified of the diagnostic result of the operating state of the plant 100.
Thus, the diagnostic apparatus 200 for plants according to the present embodiment notifies an operator of the fact that the state of the plant has been changed.
It is configured that diagnostic apparatus information 50 stored in the measurement signal database 310, the model definition database 320, and the diagnostic model database 330, which are installed in the diagnostic apparatus 200, can be arbitrarily displayed on the image display device 940 in the operations management room 900. In addition, these pieces of information can be corrected by the external input signal 2 created by an operation of the external input device 910, which are formed by the keyboard 920 and the mouse 930. [2. Operations of Diagnostic Apparatus]
Subsequently, operations of the diagnostic apparatus for plants according to the present embodiment will be described. The flowchart for the operations of the diagnostic apparatus 200 will be described below with reference to Fig. 2A, which is a flowchart view illustrating the basic operations of the diagnostic apparatus for plants illustrated in Fig. 1.
As illustrated in the flowchart view in Fig. 2A, the basic operations of the diagnostic apparatus 200 are executed by Steps 201, 202, and 203 being combined together.
In Step 201, it is first determined that an operation mode of the diagnostic apparatus 200 is the model construction mode or the diagnostic mode. When it is the model construction
mode, the processing proceeds to the step 202, and when it is the diagnostic mode, the processing proceeds to the step 203.
When Step 202 is operated, the model definition unit 400 and the model construction unit 700 are operated. As a result, the model definition information 5 and the model information 8 are created, which are stored in the model definition database 320 and the diagnostic model database 330, respectively. Details of the operations in the model construction mode will be described later with reference to Fig. 10A.
When Step 203 is operated, the diagnostic unit 800 diagnoses the operating state of the plant 100 and transmits the image display information 11 including the diagnostic result 10 to the image display device 940, thereby allowing the operating state of the plant 100 to be displayed on the image display device 940 . Details of the operations in the diagnostic mode will be described later with reference to Fig. 10B.
The respective timings when the model construction mode and the diagnostic mode are operated can be arbitrarily designated by an operator. Hereinafter, various embodiments of the timings when the model construction mode and the diagnostic mode are respectively operated will be described with reference to Figs. 2B to 2D.
In the embodiment illustrated in Fig. 2B, diagnosis is executed every period of sampling the measurement signal by operating both the model construction mode and the diagnostic
mode.
Diagnosis using the latest model can always be executed by updating the diagnostic model every time when the measurement signal is acquired.
However, when an amount of the date to be used for constructing a model is large, it takes a time to construct the model, and hence there is the possibility that a calculation may not be completed within the sampling period.
In such a case, diagnosis can also be executed in the following way: as in the embodiment illustrated in Fig. 2C, a normal state model construction mode is operated every predetermined preset period; and the diagnostic mode is only operated every sampling period. In the method in each of the embodiments illustrated in Figs. 2B and 2C, the diagnostic mode is executed every sampling period, and hence the state of the plant can be diagnosed online.
Alternatively, as in the embodiment illustrated in Fig. 2D, the model construction mode and the diagnostic mode can be respectively operated at arbitrarily timings by inputting, to the diagnostic apparatus 200, the external input signal 2 by which an operator executes model construction and diagnosis . That is, it becomes possible to diagnose the operating state of the plant 100 by changing various conditions. [3. Model construction unit]
Subsequently, a function of sorting out the measurement
signals 4 of the plant 100, which is provided in each of the model construction unit 700 and the diagnostic unit 800 that form the diagnostic apparatus 200 by which the diagnostic apparatus for plants according to the present embodiment is formed, will be described with reference to Figs. 3A to 4C.
In the diagnostic apparatus for plants according to the present embodiment, the case where the Adaptive Resonance Theory: ART) has been applied to the function of sorting out data will be described. Alternatively, other clustering technigues, such as vector guantization, may also be used as the function of sorting out data.
As illustrated in Fig. 3A, the function of sorting out data is formed by a data preprocessing device 710 and an ART module 720. The data preprocessing device 710 converts operation data into the input data for the ART module 720.
Hereinafter, the procedures (steps) executed by the data preprocessing apparatus 710 and the ART module 720 will be described.
The data preprocessing apparatus 710 first normalizes, for every measurement item, data by using the normalization condition information stored in the model definition database 320. Date including both data Nxi(n) in which measurement signals have been normalized and CNxi (n) (= 1- Nxi(n)) that are complements of the normalized data, are made to be input data Ii (n) . This input data Ii (n) are inputted to the ART module
720.
The ART module 720 sorts out the measurement signals 4 of the plant 100, which are input data, into a plurality of categories.
The ART module 720 includes an F0 layer 721, an F1 layer 722, an F2 layer 723, a memory 724, and an orienting subsystem 725, which are connected to each other. The Fl layer 722 and the F2 layer 723 are connected to each other via a weight. The weight represents a prototype (pattern) of the category into which the input data are sorted out. Herein, the prototype represents a typical value of the category.
Subsequently, the algorithm of the ART module 720 will be described.
When the input data has been inputted to the ART module 720, outline of the algorithm is what is described in the following processing 1 to 5.
Processing 1: An input vector is normalized by the F0 layer 721 to remove a noise.
Processing 2: A suitable category candidate is selected by comparing the input data inputted to the Fl layer 722 with the weight.
Processing 3: The validity of the category selected by the orienting subsystem 725 is evaluated by the ratio to a parameter p. When evaluated as being valid, the input data is sorted out into the category and the processing proceeds
to the processing 4. On the other hand, when not evaluated as being valid, the category is reset, so that a suitable category candidate is selected from other categories (processing 2 is repeated) . When the value of the parameter p is made larger, the sorting of the categories becomes finer, on the contrary, when the value thereof is made smaller, the sorting becomes
coarser. This parameter p is referred to as a vigilance parameter.
Processing 4: When all of the existing categories have been reset in the processing 2, it is determined that the input data belongs to a new category, so that a new weight representing the prototype of the new category is created.
Processing 5: When the input data have been sorted out into a category J, a weight WJ (new) corresponding to the category J is updated by the following Expression (1) with the use of the past weight WJ (old) and input data p (or data derived from the input data p). [Expression 1]
(Equation Removed)
Herein, Kw is a learning rate parameter (0 < Kw < 1) and is a value for determining a degree at which an input vector is reflected on the new weight.
Each of the arithmetic expressions of Expression (1) and the later-described Expressions (2) to (12) is incorporated into the ART module 720.
The characteristic of the data sorting algorithm of the ART module 720 exists in the above processing 4.
In the processing 4, when input data different from the learned pattern is inputted, a new pattern can be recorded without changing a recorded pattern. Accordingly, it becomes possible to record a new pattern while recording the pattern learned in the past.
As stated above, when operation data, which has been provided in advance, is provided as input data, the ART module 720 learns the provided pattern. Accordingly, when new input data is inputted to the ART module 720 that has learned, it can be determined by the aforementioned algorithm which past pattern the new input pattern is close to. In addition, when the new input pattern is different from any of the past patterns, the new input pattern is sorted out into a novel category.
Fig. 3B is a block view illustrating the configuration of the F0 layer 721. The F0 layer 721 normalizes again the input data Ii at each time such that a normalized input vector ui° to be inputted to the Fl layer 721 and the orienting subsystem 725 is created.
According to Expression (2) , Wi° is first calculated from the input data Ii. Herein, a is a constant. [Expression 2]
(Equation Removed)
Subsequently, xi°, which is a value of normalized wi°,
is calculated by using Expression (3) . Herein, // • // is a symbol representing a norm. [Expression 3]
(Equation Removed)
Subsequently, vi°, which is a value obtained by removing a noise from xi°, is calculated by using Expression (4) . Herein,
0 is a constant for removing the noise. A minute value becomes 0 because of the calculation of Expression (4), the noise in the input data is removed. [Expression 4]
(Equation Removed)
Finally, the normalized input vector Ui° is calculated by using Expression (5). The ui° is to be inputted to the F1 layer. [Expression 5]
(Equation Removed)
Fig. 3C is a block view illustrating the configuration of the Fl layer 722. The Fl layer 722 holds, as a short-term memory, the ui° calculated by Expression (5) such that pi to be inputted to the F2 layer 722 is calculated. Computation expressions executed by the F2 layer 722 are collectively shown in Expressions (6) to (12). Herein, a and b are constants,
f (•) is a function shown in Expression (4) , and Tj is a goodness of fit calculated by the F2 layer 722. [Expression 6]
(Equation Removed)
[Expression 7]
(Equation Removed)
[Expression 8]
(Equation Removed)
[Expression 9]
(Equation Removed)
[Expression 10]
(Equation Removed)
[Expression 11]
(Equation Removed)
Herein, [Expression 12]
(Equation Removed)
[4. Model Construction Function]
Subsequently, a function of constructing a model by the
measurement signals 4 of the plant 100, which is provided in the model construction unit 700 that forms the diagnostic apparatus 200 by which the diagnostic apparatus for plants according to the present embodiment is formed, will be described with reference to Fig. 4A.
An embodiment of the plant 100 will be first described with reference to Fig. 4A, and the information included in the measurement signal 4 will be described. Subsequently, a situation in which the measurement signal 4 is sorted out into a category will be described with reference to Figs. 4B and 4C.
Fig. 4A is ablock view illustrating a thermal power plant, which is an embodiment of the plant 100.
In Fig. 4A, the thermal power plant 100 includes a gas turbine generator 110, a controller 120, and a data transmitter 130. The gas turbine generator 110 includes a generator 111, a compressor 112, a combustor 113, and a turbine 114.
In generating electrical power, the air taken in by the compressor 112 is compressed to make compressed air, and the compressed air is sent to the combustor 113, where the compressed air is mixed with a fuel to be burned. The turbine 114 is rotated by using a high-pressure gas generated by the burning, so that electrical power is generated by the generator 111.
The controller 120 controls the output of the gas turbine generator 110 in accordance with an demand for electricity.
Also, the controller 120 receives, as input data, operation data 102 measured by a sensor (not illustrated) installed in the gas turbine generator 110. The operation data 102 are state quantities, such as inlet temperature, fuel supply amount, turbine exhaust gas temperature, turbine speed, power generation amount of the generator, and turbine rotor vibration, etc., which are measured every sampling period. Weather information, such as atmosphere temperature, is also measured.
The controller 120 calculates a control signal 101 for controlling the gas turbine generator 110 by using these operation data 102. The controller 120 also executes processing for issuing an alarm when the value of the operation data 102 departs from a preset range. An alarm signal is processed as a digital signal having a value of: "1" when the operation data 102 has departed from the preset range; and "0" when the operation data 102 is within the preset range. When the alarm signal is "1", an operator is notified of the content of the alarm with a sound or screen display.
The data transmitter 130 transmits, to the diagnostic apparatus 200, the operation data 102 measured by the controller 120, the control signal 101 calculated by the controller 120, and the measurement signal 1 including the alarm signal.
Fig. 4B is a view explaining a result of sorting out the measurement signals 1 obtained from the plant 100 into categories . The horizontal axis represents time, while the vertical axis
represents the measurement signals and the category numbers. Fig. 4C is a view illustrating an example of a result of sorting out the measurement signals 1 of the plant 100 into categories.
Fig. 4C illustrates an example in which two items of the measurement signals are displayed in a two-dimensional graph. The vertical axis and horizontal axis represent the normalized measurement signals of the respective items.
The measurement signals are divided into a plurality of categories 1000 (circles illustrated in Fig. 4C) by the ART module 720 in Fig. 3A. One circle corresponds to one category.
The measurement signals are divided into four categories in the present embodiment. Category number 1 is a group having a large value of an item A and a small value of an item B; category number 2 is a group having small values of the item A and item B; category number 3 is a group having a small value of the item A and a large value of the item B; and category number 4 is a group having large values of the item A and item B.
As illustrated in Fig. 4B, the data, occurring during a normal period before the start of diagnosis, have been sorted out into categories 1 to 3. The data, occurring in an early time after the start of diagnosis, have been sorted out into the category 2 that is the same category as in the normal period. In this case, the tendency of the data is the same as in the normal period, and hence it is diagnosed as being normal. On the other hand, the data, occurring in the late time after the
start of diagnosis, have been sorted out into category 4 different from that in the normal period. Because the tendency of the data is different from that in the normal period, there is the possibility that the state of the plant may have been changed, causing an abnormal event. In this case, the diagnostic apparatus 200 according to the present embodiment notifies an operator of the possibility that an abnormal event may be caused, by displaying the possibility on the image display device 940.
Although an example in which measurement signals each having two items are sorted out into categories has been described in the present embodiment, measurement signals each having three or more items can also be sorted out into categories by using a multi-dimensional coordinate.
Figs. 5 is a graph explaining the relationship between a starting mode and a process value in the power plant illustrated in Fig. 4A, which shows changes in both an output command value (A) and the turbine equipment temperature over time (B).
Examples of a typical starting mode include hot start and cold start. The start, occurring when the turbine or compressor is restarted in a hot state, is referred to as hot start. On the other hand, the start, occurring when the turbine or compressor is restated in a cool state after a relatively longperiodof stop, is referred to as cold start. As illustrated in (B) of Fig. 5, when starting modes are different from each
other, the turbine equipment temperatures at the start of operation and ranges of change in the temperature during a variable load operation are different from each other.
In a conventional diagnostic apparatus, a single diagnostic model is constructed irrespective of operating conditions, and hence it is needed to determine a normalization range so as to include data. That is, for example, the upper limit of a normalization range has been made to be 1010 and the lower limit thereof to be 1011.
For example, in the case of a low load operation in hot start, the normalization range 1022 becomes larger than the range 1020 in which a process value is changed.
If a normalization range is large in comparison with a range in which data is changed, a change in the value becomes small after normalization. Accordingly, a change in data, occurring when an abnormal event has happened, cannot be detected, and there are sometimes the cases where a new category does not occur even when an abnormal event has occurred. This may cause a non-detection.
A normalization range is appropriately determined in accordance with operating conditions by using the model definition unit 400 provided in the present invention. With this function, a non-detection can be suppressed and diagnosis accuracy can be improved. Hereinafter, a specific method thereof will be described.

[5. Operating Condition Determination Unit]
Fig. 6A is a flowchart view explaining movements of the operating condition determination unit 500 that is a component of the model definition unit 400. As illustrated in Fig. 6A, this algorithm is executed by Steps 510, 520, 530, 540, and 550 being combined together.
In Step 510, the data accumulated in the measurement signals 4 are first divided for each of the periods of time during a constant load operation, a starting operation, a stop operation, and a variable load operation.
When an output is not changed, i.e. , when a rate of change in an output measurement signal is small, the operation is assumed as a constant load operation. In addition, the data are divided for each of the periods of time during a starting operation, a stop operation, and a variable operation, based on the signals for distinguishing the above three operation modes from each other, the signals being included in the measurement signals of the power plant.
When the operation mode is a constant load operation, the processing proceeds to Step 520; when the operation mode is a starting operation, proceeds to Step 530; when the operation mode is a stop operation, proceeds to Step 540; and when the operation mode is a variable load operation, proceeds to Step 550, respectively.
In Step 520, the information with respect to load bands
is extracted. The load bands are sorted out in accordance with outputs, for example, an output of 0 to 50% of the rated output is assumed as a low output, and an output of 50 to 100% thereof is assumed as a high output, etc. In Step 530, the information with respect to the type of a starting mode and a rate of change in a load during a starting operation is extracted. Examples of the starting mode include a hot start mode and a cold start mode, etc. In Step 540, the information with respect to a stop mode and an output at the start of a stop operation are extracted. Examples of the stop mode include a mode in which the plant is stopped in a usual operation and a mode in which the plant is urgently blocked when an abnormal event has occurred, etc. In Step 550, the information with respect to a rate of change in a load, an output at the start of a variable load operation, and an output at the end of the variable load operation are extracted.
Although, in the present embodiment, the aforementioned information is extracted in Steps 520, 530, 540, and 550 in order to distinguish, from each other, the states during a constant load operation, a starting operation, a stop operation, and a variable load operation, an amount of the information can also be increased. For example, the information with respect to turbine speed, a rate of increase in speed, the type of the fuel supplied to the plant, and atmosphere temperature, etc., may be added to the information extracted in Steps 520,
530, 540, and 550, as far as the information is obtained by processing the measurement signals of the plant 100.
As illustrated in Fig. 6B, a diagnostic model 560 using the same measurement item is constructed by a set of models, which have been sorted out into sub diagnostic models 570 based on the information extracted in Fig. 6A. Diagnosis is executed by defining normalization conditions different for every sub diagnostic model.
[6. Normalization Condition Determination Unit]
Hereinafter, an embodiment of the normalization condition determination unit 600 for determining a normalization condition will be described with reference to Figs. 7A to 9. The data preprocessing apparatus 710 normalizes measurement signals by using the information with respect to the normalization condition determined by the normalization condition determination unit 600 . It is assumed that the number of the data items of measurement signals xi is N and the n-th measurement signal is indicated by x(n) . The normalized data Nxi (n) is represented by the following Expression (13). Herein, Nmin(n) is the lower limit of the normalization and Nmax(n) is the upper limit thereof.
[Expression 13]
Nxi(n)=(xi(n)- Nmin(n))/(Nmax(n)- Nmin(n))
The normalization condition determination unit 600 determines Nmin (n) and Nmax (n) shown in Expression (13).
Figs. 7A and 7B are views explaining a first embodiment of the normalization condition determination unit 600.
Fig. 7A is a flowchart view illustrating the first embodiment of the normalization condition determination unit 600. As illustrated in Fig. 7A, this algorithm is executed by Steps 611, 612, and 613 being combined together.
In Step 611, the data for constructing a model are first divided for every operating condition. In Step 612, the information with respect toawidthofa change in the measurement signal is extracted for every operating condition.
In Step 613, the upper limit Nmaxl (n) of a normalization range and the lower limit Nminl (n) thereof are determined.
In the case of a constant load operation, Expressions (14) and (15) are used. Herein, Dmaxl (n) is the maximum of the measurement signals, Dminl (n) is the minimum thereof, and a, P are constants. [Expression 14] Nmaxl (n)=Dmax1 (n)x(T + a) [Expression 15] Nmin1(n)=Dmin1(n)x(1 - ß)
In the case of a variable load operation, a starting operation, or a stop operation, the normalization range is determined in accordance with an output command value, as shown in Expressions (16) and (17). [Expression 16]
Nmax2=MW*a4-b
[Expression 17] Nmin2=MW*c+d
Herein, Mw is an output command value, and a and c are the slope, occurring when measurement signals during a variable load operation are subjected to linear approximation. And, b and d are values calculated in Expressions (18) and (19). [Expression 18] b=f+Dmax2x1. 2 [Expression 19] d=f—Dmin2x1. 2
Herein, f is the intercept, occurring when the measurement signals during a variable load operation are subjected to linear approximation, and Dmax2 is the maximum of the deviations between the straight line, occurring when the measurement signals are subjected to linear approximation, and the measurement signals, and Dmin2 is the minimum thereof.
Fig. 7B are graphs explaining changes in an output and a measurement signal over time, occurring when a load has been changed from a low output to a high output. An output is: low before the time TOa; changing between the time TOa and the time Tub; and high after the time Tub.
By operating the flowchart illustrated in Fig. 7A, the upper limit of the normalization range during a low output operation is determined as 1030, the lower limit thereof is
determined as 1040, the upper limit of the normalization range during a variable load operation is determined as 1050, the lower limit thereof is determined as 1060, the upper limit of the normalization range during a high output operation is determined as 1070, and the lower limit thereof is determined as 1080 . Thus, the normalization range according to the present invention is changed in accordance with an operating condition.
Figs. 8A and 8B are views explaining a second embodiment of the normalization condition determination unit 600. As illustrated in Fig. 8A, this algorithm is executed by Steps 631, 632, and 633 being combined together.
In Step 631, the data in a variable loadmode are extracted. In Step 632, dead time is estimated for every data item. The dead time is estimated by an impulse response using a process signal as input/output data. A specific example of the calculation method includes the method described in a reference document: "Upper System Identification for Control" (Tokyo Denki University Press). In Step 633, the data in the dead time estimated in Step 632 are shifted, followedby determination of a normalization range for every data item.
This situation will be described with reference to Fig. 8B. The time when a measurement signal begins to vary becomes late from the time T1a when an output command value begins to increase. This period of time is the dead time. The dead time for the data item A is (T2a - Tla) and that for the data item
B is (T3a - T1a).
After a measurement signal has been shifted by the dead time with the use of the flowchart in Fig. 8A, a normalization range is determined with the use of the flowchart in Fig. 7A. Thereby, the normalization range of a measurement signal can be determined in accordance with an output command value.
Fig. 9 is a graph explaining a third embodiment of the normalization condition determination unit 600 in the diagnostic apparatus for power plants illustrated in the Fig. 1.
The normalization ranges near to the maximum and the minimum of a measurement signal are made to be enlarged such that changes in the measurement signal near to the maximum and the minimum can be easily detected.
A normalization method executed by the normalization condition determination unit 600 is not limited to the aforementioned contents, but any of the methods of determining a normalization condition for every operating condition may be adopted. For example, a range in which data values vary, estimated from the information for designing a plant or the specification of a measuring instrument, may be adopted as a normalization range. [7. Operation Modes of Diagnostic Apparatus]
Figs . 10A and 10B are flowchart views explaining operation modes of the diagnostic apparatus 200 for power plants
illustrated in Fig. 1. Fig. 10A illustrates a flowchart for the operations in the model construction mode in Fig. 2, while Fig. 10B illustrates a flowchart for the operations in the diagnostic mode.
As illustrated in Fig. 10A, the model construction mode is executed by Steps 1200, 1210, 1220, and 1230 being combined together. In Step 1200, data in a period of time used for constructing a model are extracted from the measurement signal database 310. This period of time can be arbitrarily set by an operator of the plant 100. Subsequently, the operating condition determination unit 500 is operated in Step 1210. The data in the period of time used for constructing a model are divided for each of the operation modes during a constant load operation, a starting operation, a stop operation, anda variable load operation, by operating the flowchart illustrated in Fig. 6A; and the data in each of the modes are further divided for each of the characteristics by operating Steps 520, 530, 540, and 550. Subsequently, the normalization condition determination unit 600 is operated in Step 1220. A normalization condition is determined for each of the groups divided in Step 1210 by operating the flowcharts described in Figs. 7A and 8A. The model definition information 5 obtained by operating Steps 1210 and 122 0 is stored in the model definition database 320. Finally, the model construction unit 700 is operated in Step 1230. The each of the groups divided in Step
1210 is defined as the sub diagnostic model 570 (see Fig. 6B) , and measurement signals are processedby using the normalization condition defined for every sub diagnostic model to construct a diagnostic model by using the ART described in Fig. 3.
As illustrated in Fig. 10B, the diagnosticmode is executed by Steps 1300, 1310, and 1320 being combined together. In Step 1300, data in a period of time to be diagnosed are extracted from the measurement signal database 310. Subsequently, the data in the period of time to be diagnosed are processed by the operating condition determination unit 500 in Step 1310. The sub diagnostic models whose operating conditions are the same as each other are extracted from a plurality of the sub diagnostic models constructed in the model construction mode. Finally, the diagnostic unit 800 is operated in Step 1320. The diagnostic unit 800 extracts, from the model definition database 320, the sub diagnostic model information extracted in Step 1330. The diagnostic unit compares the category number of the sub diagnosticmodel with the category number obtainedby sorting out the data in the period of time to be diagnosed with the ART. The diagnostic unit diagnoses the information as being normal when the two categories are the same as each other, while diagnoses the information as being abnormal when different category number has occurred. The diagnostic result 10 is outputted to the external output interface 220. [8. Aspects of Data]
Figs. 11A to 11C are tables explaining aspects of the data stored in the database according to the present invention.
As illustrated in Fig. 11A, values of the measurement signals 1 (data items A, B, and C are described in the table) , which are operation data measured for the plant 100, are stored in the measurement signal database 310 every sampling period (time on the vertical axis).
Data varying in a large range can be scroll displayed by using, on a display screen 311, scroll boxes 312 and 313 with which a table can be moved vertically and horizontally.
As illustrated in Fig. 11B, the information with respect to an operating condition and a normalization range are stored in the model definition database 320 by associating them with each other.
As illustrated in Fig. 11C, the relationship between the category number and a weight is stored in the diagnostic model database. Herein, the weight means the central coordinates of a category.
Figs. 12A to 12D are graphs explaining an advantage of applying the diagnostic apparatus for power plants illustrated in Fig. 1.
A change in a load is started at the time T4 and is ended at the time T5. An abnormal event occurred after the change in a load had been ended.
Fig. 12 illustrates the relationship between an output
command value (A) and the temperature of the turbine (B) and the relationship between the category number and a measurement signal for temperature (C) , (D) . With occurrence of an abnormal event, the temperature has been increased; however, the increased temperature has been within a range of the category 4, and hence the abnormal event has been sorted out into the same category as that in a normal state. In a conventional method, the normalization range is large, and accordingly an increase in the temperature, with occurrence of an abnormal event, was not able to be detected, thereby not allowing the abnormal event to be detected.
In the present invention, a diagnostic model is switched in accordance with an operating condition. In the present embodiment, the diagnosticmodel is switched such that diagnosis is performed: in an operating condition 1 diagnostic model (during a constant load operation, a low output model) before the time T4; in an operating condition 2 diagnostic model (during a variable load operation) between the time T4 and the time T5; and in an operating condition 3 diagnostic model (during a constant load operation, a high output model) after the time T5.
In the operating condition 3 diagnostic model, normal states are sorted out into the category numbers 1 to 5. Although the data, before occurrence of an abnormal event, have been sorted out into the same category as that in a normal state,
the data, after the occurrence of the abnormal event, have been sorted out into a new category with the category number 6, thereby allowing the abnormal event to be detected. That is, in the present embodiment, the state can be sorted out more finely than a conventional method, and hence a non-detection in which an abnormal event cannot be detected has been suppressed.
Fig. 13 is a view explaining a screen displayed on the image display device 940 in the diagnostic apparatus 200 for power plants illustrated in Fig. 1.
The upper limits (954, 956, and 958) of a normalization range, the lower limits (955, 957, and 959) thereof, and the boundary times (952 and 953) of an operating condition can be arbitrarily adjusted by operating the mouse 930 with a cursor 951. An execution button 960 is clicked on the screen in Fig. 13 in order to reflect an adjusted result on a model. With this operation, the information with respect to the operating condition and the normalization range in the model definition database illustrated in Fig. 10B are changed, and hence the adjusted result can be reflected on the construction of a model and the diagnostic operation.
The present invention should not be limited to the aforementioned embodiment, but various variations can be included. For example, the aforementioned embodiments have been described in detail for easy description of the invention, and accordingly the invention should not always be limited to
embodiments including all of the described configurations.
Alternatively, the aforementioned configurations, functions, and processing units may be achieved by hardware with part or all of them being designed by integrated circuits, etc. Alternatively, the aforementioned configurations and functions, etc., may be achieved by software in which processors interpret and execute programs that achieve the respective functions. Programs, tables, files, measurement signals, and information with respect to calculated information, etc., can be stored in memory device, such as a memory and hard disk, or in memory media, such as an IC card, SD card, and DVD. Accordingly, each of the processing and configurations can be achieved as a processing unit or a program module.
The information lines considered to be necessary for description have been described above, but all of the control lines and information lines in terms of products have not always been described. It may be considered that, in fact, almost all of the configurations are connected to each other.
According to the present embodiment, a diagnostic apparatus and a diagnostic method for power plants can be obtained, in which an abnormal event occurring in a power plant is detected with high accuracy.
The present invention can be widely applied to various plants, etc., as a diagnostic apparatus and a diagnostic method for plants.

WHAT IS CLAIMED IS:
1. A diagnostic apparatus for power plants, in which an operating state of a power plant is diagnosed based on a measurement signal in which a state quantity of the plant has been measured and a diagnostic result is displayed on an image display device, the diagnostic apparatus for power plants comprising:
a model construction unit that constructs a model to be used for diagnosis by applying a measurement signal in which a state quantity of the power plant has been measured to the diagnostic apparatus for power plants;
a model definition unit that defines both an operating condition diagnosed by the model and a method of normalizing the measurement signal; and
a diagnostic unit that diagnoses an operating state of the power plant by using the model constructed with the model construction unit, wherein
the model definition unit includes both an operating condition determination unit for determining an operating condition of the power plant and a normalization condition determination unit for determining a normalization condition of the measurement signal for every operating condition determined by the operating condition determination unit, and wherein
the diagnostic unit executes diagnosis by switching a diagnostic model in accordance with an operating condition.
2. In the diagnostic apparatus for power plants according to claim 1, wherein
the operating condition determination unit includes: an arithmetic device for sorting out the operating condition of the power plant into one of the operation modes during a constant load operation, a starting operation, a stop operation, and a variable load operation; an arithmetic device for extracting a load band when the operating condition is the constant load operation; an arithmetic device for extracting the type of a starting mode and a rate of change in a load when the operating condition is the starting operation; an arithmetic device for extracting the type of a stop mode and an output at the start of a stop operation when the operating condition is the stop operation; and an arithmetic device for extracting a rate of change in a load, an output at the start of a variable load operation, and an output at the end of the variable load operation when the operating condition is the variable load operation.
3 . The diagnostic apparatus for power plants according to claim 1, wherein
the normalization condition determination unit includes : an arithmetic device for dividing data for constructing a model,
for every operating condition determined by the operating condition determination unit; an arithmetic device for extracting information with respect to a width of a change in the data for every operating condition; and an arithmetic device for determining a normalization range based on the information with respect to the width of a change in the data.
4 . The diagnostic apparatus for power plants according
to claim 3 comprising:
an arithmetic device for making a normalization range of the measurement signal during the variable load operation to be a function of an output command value.
5 . The diagnostic apparatus for power plants according
to claim 3, wherein
in the normalization condition determination unit, after both an arithmetic device for estimating dead time with respect to the measurement signal during the variable load operation and an arithmetic device for shifting the measurement signal by the estimated dead time, have been operated, the arithmetic devices of claim 3 are operated.
6. The diagnostic apparatus for power plants according to any one of claims 3 to 5, wherein
the normalization condition determination unit includes
an arithmetic device for enlarging the normalization range near to the maximum and the minimum of the measurement signal.
7 . The diagnostic apparatus for power plants according to claim 1 comprising:
a screen display device that is configured to display in a superimposed manner: a trend graph showing a change in the measurement signal over time; the time when the operating condition determined by the operating condition determination unit is switched; and a normalization range determined by the normalization condition determination unit, and that is used for arbitrarily changing the time when the operating condition is switched and the normalization range.
8. A diagnostic method for power plants, in which an operating state of a power plant is diagnosed based on a measurement signal in which a state quantity of the plant has been measured and a diagnostic result is displayed on an image display device, the diagnostic method for power plants comprising the steps of:
determining an operating condition of the power plant;
determining a normalization condition of the measurement signal for every operating condition determined by an operating condition determination unit;
constructing a model to be used for diagnosis by applying
the measurement signal in which a state quantity of the power plant has been measured to a diagnostic apparatus for power plants;
defining both an operating condition diagnosed by the model and a method of normalizing the measurement signal; and
executing diagnosis by switching a diagnostic model in accordance with an operating condition.
9. The diagnostic method for power plants according to claim 8, wherein
the step of determining an operating condition of the power plants includes the steps of: sorting out the operating condition of the power plant into one of operation modes during a constant load operation, a starting operation, a stop operation, and a variable load operation; extracting a load band when the operating condition is the constant load operation; extracting the type of a starting mode and a rate of change in a load when the operating condition is the starting operation; extracting the type of a stop mode and an output at the start of a stop operation when the operating condition is the stop operation; and extracting a rate of change in a load, an output at the start of a variable load operation, and an output at the end of the variable load operation when the operating condition is the variable load operation.
10. The diagnostic method for power plants according
to claim 8, wherein
the step of determining a normalization condition includes the steps of: dividing data for constructing a model, for every operating condition determined by the operating condition determination unit; extracting information with respect to a width of a change in the data for every operating condition; and determining a normalization range based on the information with respect to the width of a change in the data.
11. The diagnostic method for power plants according
to claim 10 comprising the step of:
making a normalization range of the measurement signal during the variable load operation to be a function of an output command value.
12. The diagnostic method for power plants according
to claim 10, wherein
in the step of determining a normalization condition, after both steps of estimating dead time with respect to the measurement signal during the variable load operation and shifting the measurement signal by the estimated dead time, have been executed, the steps of claim 10 are executed.
13. The diagnostic method for power plants according
to any one of claims 10 to 12, wherein
the determining of a normalization condition includes a step of enlarging the normalization range near to the maximum and the minimum of the measurement signal.
14. The diagnostic method for power plants according
to claim 8, wherein
a trend graph showing a change in the measurement signal over time, the time when an operating condition determined by the operating condition determination unit is switched, and a normalization range determinedby the normalization condition determination unit, are displayed in a superimposed manner, and wherein
the time when the operating condition is switched and the normalization range can be arbitrarily changed.
15. A diagnostic apparatus for power plants,
substantially as herein described with reference to
accompanying drawings.
16. A diagnostic method for power plants, substantially as herein described with reference to accompanying drawings.