Description:
The present invention relates to fight against disturbances resulting from the random variation of wind speed. By attenuation or complete rejection, the wind turbine would operate without disturbance, which would extend useful life and reduce the number of failures. This work presented the LPV control strategy for attenuating torque harmonics.
[02] BACKGROUND OF THE INVENTION

The reduction or cancellation of torque ripples has been the subject of much research since the installation of wind turbines became widespread. From Battista et al. [DBM98] addressed the problem of attenuating torque ripple early on while applying slip mode control (SMC). Damping was performed on the basis of a speed feedback while producing a reference torque to follow. Besides SMC, traditional control methods have been used to solve the problem of adaptive control, robust control (RC) and internal model control (IMC). Nevertheless, these methods are known to be slow in dynamics and they are however recognized as being methods of active control (AC) vis-à-vis the disturbance. These methods meet the requirements of stability, monitoring, regulation, performance, robustness and rejection of disturbances. Modern systems require adaptive and responsive line control to disturbances. Thus, it is necessary to find more advanced control techniques. In the following paragraphs, some of the most used advanced methods for disturbance mitigation are presented.
[03] SUMMARY OF THE PRESENT INVENTION
This work presented the LPV control strategy for attenuating torque harmonics. The LPV control made it possible to reduce the angular speed harmonics on the high speed side by an electrical control applied to the generator. The major concern is how much current the generator can withstand without damage it. Three approaches to estimating harmonics and insulation were applied. The Unknown Input Observer (UIO) showed better results compared to the extended Kalman filter (EKF) and the modified Sliding Window Empirical Mode Decomposition (SWEMD) work. This is related to the ability of the LPV control to attenuate more harmonics for less current injected. Either way, modified SWEMD remains a very efficient and accurate technique for signal analysis and system diagnostics. This is due to the possibility of applying it to any signal without having the signal model. This approach proves that the operation of the wind turbine can be made more flexible while protecting various mechanical components, including the driveline. This can impact the life of these components, while reducing the stresses and fatigue imposed by the wind profile and, possibly, by torque harmonics. Another note concerns the harmonics added to the electromagnetic torque: additional control and filtering is necessary with regard to the fluctuation of the output currents and voltages. This could be done by DC bonding and the proper design of converters before sending the generated electrical power to the grid.
[04] BRIEF DESCRIPTION OF THE DRAWINGS
ESO studies process dynamics and external disturbances, the whole process comes down to estimating and cancelling. From the ESO produced, the whole domain of disturbance rejection was developed to allow all the uncertainties and disturbances of the system to be dealt with in what was later defined as the active control of the system disturbance rejection (ADRC) (see FIG 1).
The original Active Control Methodology (ADRC) is a combination of tracking differential and ESO with a non-linear shape. However, the basic ADRC framework, later called Disturbance Observer Control (DOBC), combines a disturbance observer, feedforward control block, and feedback control. A major part of this design lies in the principle of separation on which the adjustment of the feedback and direct-acting control blocks is carried out. In this diagram, the predictive controller rejects the disturbance, while the feedback controller monitors the performance of the control (see FIG 2).
By examining FIG 3, one can note the similarities existing with the DOBC diagram of FIG 2. The main difference is that there are multiple loops for the CHADC to ensure cancellation of multiple types of disturbances.
The estimated harmonics will be used by the proposed LPV control strategy in order to produce an electrical control capable of reducing its effects. FIG 4 shows the block diagram of the control strategy applied for the simulation and experimental part.
FIG 5 shows the angular velocity of the high-speed side before and after the activation of the LPV control.
FIG 6 shows how the harmonics of Te are amplified.
The last observation for constant wind speed is the current produced by the LPV controller. FIG 7 shows the injected current.
FIG 8 proves that even with varying wind speeds; the LPV controller is able to generate an appropriate current command. In this case, the FFT analysis is obviously unnecessary.
Using the EKF, the harmonics of the aerodynamic torque are isolated and fed into the LPV control. The block diagram of the approach is given in FIG 9.
FIG 10 shows the effect of harmonics isolated by the EKF and used by the LPV control.
By observing FIG 11, it can be noted that, compared to the UIO, the injected current resulted in stronger oscillations on the torque Te.
The greatest amplitudes during LPV control are due to the injection of iqref2 observed in FIG 12.
FIG 13 shows good results with the EKF but with a greater amount of current generated.
The principle diagram is given in FIG 14 for harmonic isolation and LPV control.
FIG 15 shows the effect of LPV control when using the modified SWEMD at constant wind.
FIG 16 shows the effect of LPV control on the electromagnetic torque Te.
The greatest amplitudes during LPV control are due to the injection of iqref2 observed in FIG 17.
FIG 18 shows good results as for the EKF. The results show steady state disturbance attenuation for the signal processing and estimation methods used.
DETAILED DESCRIPTION OF THE INVENTION

[05] LPV control and application to wind turbines.
When designing controllers for nonlinear systems, the most widely used multivariable controllers are designed under specified operating conditions. In some cases it is possible to use a linearized model of the system dynamics for each condition. These linearized models can be dependent on one or more parameters defining the operating points at which the linearized system exists. This is equivalent to designing a specified linear controller into a linear system at each operating condition. System settings determine the operating region of each controller. Correcting the linear controllers together results in a nonlinear controller for the nonlinear system. The study of the gains, although simplifying the design of the controller, has certain drawbacks such as:
• Stability may not be guaranteed at certain operating points where no controller is defined. Since the parameters may not cover all regions of operation, stability may not exist under some unselected operating conditions.
• When it comes to complex systems, the design of multivariable controllers becomes tedious and requires a computational effort.
• Stability, robustness and overall performance may not be guaranteed, as the system parameters do not cover all operating points.
In order to reduce the effects of these drawbacks on synthesis Linear parameter-varying (LPV) controllers are used as a specific systematic design procedure to obtain controllers. LPV techniques integrate the performance, robustness and bandwidth requirements of the structure. The LPV system models are written in the same way as the linear invariant form with the inclusion of the time varying parameter. Two representations are possible: polytopic and affine.
[06] State of the art of LPV for the mitigation and rejection of disturbances in wind power applications
Sinusoidal or periodic disturbance rejection is a hot topic in all fields of engineering, whether it is electric vehicles, helicopters, airplanes, wind turbines, etc. Asymptotic regulation is well known in the rejection of disturbances of known autonomous and unstable exogenous systems. However, with unsteady, uncertain frequencies, asymptotic regulation of changing frequencies cannot be applied. In this case, robust and adaptive control methods are mainly used. In addition, in one of the cases where frequencies are either estimable or measurable, or in-line, as is the case with rotating machines, LPV control can also be applied for robust and adaptive disturbance attenuation for the non-stationary case.
Disturbances in wind turbines can be rejected using LPV control methods as it is a rotating system. Several applications of the LPV control could be used for the different electrical systems and sub-techniques existing in wind turbines.
LPV control is widely used in wind turbines with promising and interesting results in terms of disturbance rejection or mitigation. That hasn't stopped researchers from developing other strategies, including Active Disturbance Rejection Control (ADRC).
[07] Active control for interference rejection
Proportional integral and derivative control (PID) has dominated the controller community over the past century with 98% use of PI in control loops and 95% of PID for process control. However, PID has been overwhelmed by the increasing demands for efficiency, robustness and stability of real-time systems operating in complex environments. The motivation for work is linked to two main objectives: to make the PID linear non-linear, while using the derivative part of the PID which will improve the transient performance of the control systems. It is known as the Extended State Observer (ESO). ESO studies process dynamics and external disturbances, the whole process comes down to estimating and cancelling. From the ESO produced, the whole domain of disturbance rejection was developed to allow all the uncertainties and disturbances of the system to be dealt with in what was later defined as the active control of the Active disturbance rejection control (ADRC) (see FIG 1).
The original ADRC is a combination of tracking differential and ESO with a non-linear shape. However, the basic ADRC framework, later called Disturbance Observer Control (DOBC), combines a disturbance observer, feedforward control block, and feedback control. A major part of this design lies in the principle of separation on which the adjustment of the feedback and direct-acting control blocks is carried out. In this diagram, the predictive controller rejects the disturbance, while the feedback controller monitors the performance of the control (see FIG 2).
Based on these characteristics, DOBC has shown superior performance compared to traditional active control methods. Compared to traditional methods, the DOBC has a faster dynamic response while handling disturbances. This is due to the direct compensation which introduces a signal which directly neutralizes the disturbances. Traditional methods rely solely on feedback regulation for the rejection of disturbances. This strategy of the DOBC methods uses the classic feedback regulation technique used in gain computation, model predictive control and PID control, while adding disturbance-based compensation that improves robustness and disturbance rejection capabilities. . The control strategy developed is then complementary to the classical theorems and methods. One of the advantages of DOBC over AC is that it is not designed on a worst-case scenario basis. It is an online estimator of disturbances, while the design of the feed-forward and feed-forward controllers adapt accordingly. With this strategy, the DOBC ensures better nominal dynamic performance.
[08] Composite hierarchical anti-disturbance control (CHADC)
A common approach found in the literature in all mechanical systems considers a single equivalent perturbation to deal with multiple perturbations. Since disturbances can be of several types (stochastic, Gaussian, constants, harmonics, periodic processes, etc.), it has become essential to develop advanced techniques for rejecting disturbances in order to deal with them separately. In line with the requirements of current applications, an upgrade to the ADRC has recently been proposed, known as the Composite Hierarchical Anti-Disturbance Control (CHADC).
By examining FIG 3, one can note the similarities existing with the DOBC diagram of FIG 2. The main difference is that there are multiple loops for the CHADC to ensure cancellation of multiple types of disturbances. The baseline controller can be introduced for tracking or mitigation purposes (H1, stochastic control theory, etc.). While in itself, stability and performance for disturbance mitigation are no longer satisfied. Applications of CHADC include the introduction of PID (Proportional, Integral, and Derivative), PD control, Model Predictive Control (MPC), Sliding Mode - Sliding Mode Control (SMC), Adaptive Control, H1, Fuzzy Control - Fuzzy Control.
The combination of DOBC and ADRC techniques and control theory makes CHADC strategies more complex and highly coupled. this results in difficulties related to the analysis, modeling of disturbance characteristics, restrictions on modeling disturbances in accordance with the DOBC model, and dealing with uncertain and unknown parameters existing in the CHADC process.
[09] Reduction of torque harmonics by LPV control
The estimated harmonics will be used by the proposed LPV control strategy in order to produce an electrical control capable of reducing its effects. FIG 4 shows the block diagram of the control strategy applied for the simulation and experimental part. The control leads to the calculation of an appropriate reference current iqref2 (torque control in vector control). The LPV controller is with appropriate adaptations and modifications to meet the wind power system and operating requirements.
The harmonics estimated by the UIO are considered to be an exogenous disturbance. The design of the LPV control ensures its ability to act on one or more harmonic components, so it takes into account the sum of the harmonics and generates the iqref2 signal accordingly. The design of the controller is based on the internal model principle (IMP) and generalized asymptotic regulation. The quadratic stability of the closed loop system is an objective of the approach.
Simulation results
[010] UIO-based control
A constant wind speed is applied to visualize the effect of the LPV control. The control produces a signal grouping together components in opposition to the disturbance harmonics in order to attenuate them. In theory, it should not generate components that did not originally exist in the disturbance signal. Before applying the command generated by the LPV controller, one must keep in mind the ability of the generator to withstand the amount of energy injected. In this case, we are only interested in the harmonic attenuation and not the total rejection. As the latter require large amount of energy injected which can compromise the entire system, we will check the consequences of the control on the electromagnetic torque of the generator.
FIG 5 shows the angular velocity of the high-speed side before and after the activation of the LPV control.
The FFT analysis shows that no additional harmonics have been added. It can be observed that after the introduction of LPV control, the amplitude of each harmonic is considerably reduced especially from 4 rad/ s.
The main concern is electromagnetic torque. The quantity of current injected must respect the level at which the generator can make face.
FIG 6 shows how the harmonics of Te are amplified. However, at this level, the machine is operating in a safe region. Although harmonics are aggravated in the machine and output voltages and currents are more fluctuating, the strategy helps reduce vibrations on mechanical components, especially the gearbox. Fluctuating currents and voltages will be filtered by a continuous link before reaching the network. The harmonic amplification is approximately 20% of Te.
The last observation for constant wind speed is the current produced by the LPV controller. FIG 7 shows the injected current. The value of the current is negative since it opposes the pre-existing harmonics. Additionally, the shape is similar to that found in Taero. This is shown in the FFT analysis where the frequency components are similar. The LPV controller was successful in preserving the frequency signature information found throughout the system. A simulation is performed using the original wind speed profile in order to study the efficiency of the controller in the non-stationary case. FIG 8 proves that even with varying wind speeds; the LPV controller is able to generate an appropriate current command. In this case, the FFT analysis is obviously unnecessary.
This section has proven the effectiveness of the controller in attenuating harmonics through current control while using the UIO as the harmonic estimator.
[011] Comparison with an EKF-based test
In this section, the EKF function is applied in order to prove the usefulness of the UIO in relation to the EKF function. Using the EKF, the harmonics of the aerodynamic torque are isolated and fed into the LPV control. The block diagram of the approach is given in FIG 9.
FIG 10 shows the effect of harmonics isolated by the EKF and used by the LPV control. The FFT analysis of the angular velocity at high speed after the start of the LPV control shows good results in harmonic attenuation. However, compared to the results in the previous paragraph, the unknown input observer gives better results. The performance of the UIO is further demonstrated in the electromagnetic torque Te. By observing FIG 11, it can be noted that, compared to the UIO, the injected current resulted in stronger oscillations on the torque Te.
The greatest amplitudes during LPV control are due to the injection of iqref2 observed in FIG 12. Larger amplitudes of the harmonics are remarkable, linked to a higher current generated by the LPV control.
The same test is carried out for a variable wind. FIG 13 shows good results with the EKF but with a greater amount of current generated. The EKF associated with the active control requires a higher current injection which risks bringing the machine to the limit of its normal operating zone.
[012] SWEMD-based control
In this section, the modified SWEMD is applied in order to prove the effectiveness of the proposed method while comparing its synthesis with the EKF and UIO methods. Using the modified SWEMD, the aerodynamic torque harmonics are isolated and fed into the LPV control. The principle diagram is given in FIG 14 for harmonic isolation and LPV control.
FIG 15 shows the effect of LPV control when using the modified SWEMD at constant wind. The FFT analysis of the angular velocity at high speed after the start of the LPV control shows good results in harmonic attenuation. A peak is expected in the first window of the SWEMD because the final effect cannot be eliminated in the first data window. Compared to the UIO, it should be noted that the UIO has better attenuation results. FIG 16 shows the effect of LPV control on the electromagnetic torque Te. It can be noted that the injected current resulted in an amplitude greater than that observed for the UIO case. This shows that when using UIO, the attenuation of harmonics may be greater than when using EKF or modified SWEMD to isolate harmonics. As indicated in the EKF case, the greatest amplitudes during LPV control are due to the injection of iqref2 observed in FIG 17. The amplitudes of the harmonics are greater and are remarkable, they are linked to a higher current generated by the LPV control.
FIG 18 shows good results as for the EKF. The results show steady state disturbance attenuation for the signal processing and estimation methods used. On the other hand, the operating mode of wind turbines is non-stationary, hence the drawbacks for the modified SWEMD and especially the EKF concerning the estimation of individual harmonics and total signal.

Claims:

1. The LPV control made it possible to reduce the angular speed harmonics on the high speed side by an electrical control applied to the generator.
2. Three approaches to estimating harmonics and insulation were applied. The Unknown Input Observer (UIO) showed better results compared to the extended Kalman filter (EKF) and the modified Sliding Window Empirical Mode Decomposition (SWEMD) work. This is related to the ability of the LPV control to attenuate more harmonics for less current injected. Either way, modified SWEMD remains a very efficient and accurate technique for signal analysis and system diagnostics. This is due to the possibility of applying it to any signal without having the signal model. This approach proves that the operation of the wind turbine can be made more flexible while protecting various mechanical components, including the driveline. This can impact the life of these components, while reducing the stresses and fatigue imposed by the wind profile and, possibly, by torque harmonics.
3. Another note concerns the harmonics added to the electromagnetic torque: additional control and filtering is necessary with regard to the fluctuation of the output currents and voltages. This could be done by DC bonding and the proper design of converters before sending the generated electrical power to the grid.