WO2008046942A1

WO2008046942A1 - Closed-loop identification of wind turbines

Info

Publication number: WO2008046942A1
Application number: PCT/ES2007/000584
Authority: WO
Inventors: Javier SANZ RODRÍGUEZ; Mikel Lasa Morán; Miguel Iribas Latour; Ioan-Doré LANDAU
Original assignee: Fundación Cener-Ciemat
Priority date: 2006-10-17
Filing date: 2007-10-17
Publication date: 2008-04-24
Also published as: ES2299375B1; ES2299375A1

Abstract

The invention relates to the closed-loop identification of wind turbines, involving a process including an experimentation protocol (1), data collection (2), identification routines (3), acquisition (4) of linear input/output models and application (5) of said linear models to the functions to be controlled. The experimentation protocol (1) is performed with the wind turbine in normal operation, by introducing a reference signal with the algorithm of the function to be controlled and returning the output from said function to the input while an excitation signal is being applied with the control algorithm.

Description

IDENTIFICATION OF AEROGENERATORS IN CLOSED LOOP

Technical sector

The present invention is related to the functional control of wind turbines, proposing a closed loop identification system for said application.

State of the art

Numerous and diverse theoretical studies exist regarding the physics that involves a wind turbine.

These studies have led to the development of theoretical nonlinear models of wind turbine behavior with different degrees of success.

The development of these non-linear models and their implementation in simulators, together with the increasing complexity of wind turbines has made the growth in complexity of their control necessary. To be able to design controllers, it is necessary to have linear models that describe in the best possible way the behavior of the wind turbine, for different wind conditions.

The development of simulators based on complex non-linear models, together with the needs of control designers, have led to the use of non-linear simulators in order to obtain linear models, using different linearization techniques, obtaining results whose suitability is In some cases moot.

A true fact in all areas of the science, and its description, is the need to have experimental data that somehow allow numerical values to be given to the different constants involved in a model. This is a complex problem and in many areas not fully resolved.

A wind turbine is a very complex mechanism, which includes a large number of mechanical elements, actuators, aerodynamic elements, etc. In addition, a wind turbine is an extremely complex mechanism to analyze independently, which makes it extremely complex to correctly adjust the entire number of parameters that, together with the dynamics equations, describe the behavior of the wind turbine.

These problems related to the quantification of certain parameters of the non-linear model, together with the poorly correct numerical linearization techniques, sometimes lead to the obtaining of linear models that are not very successful, or that are not adjusted to reality. These models are difficult to assess until the control that is designed is not tested on the real machine, and it is analyzed whether the simulated behavior resembles the real one. It has been proven with some frequency that a control design based on linear models obtained from said non-linear simulators does not work in the real system, which requires manual and little precise adjustments to the designed control.

This shows that the models obtained in this way are not entirely correct and that experimentation on the real system (or on its theoretical nonlinear model) is advisable for obtaining of suitable models.

The most common way to obtain values of certain physical properties of a real machine to identify, is to operate experimentally with the machine in a concrete way. Experimentation for identification in the field of control is carried out on the real system, traditionally operating in open loop. That is, applying an exciting signal to the system and reading the output variables. This process is possible when you have access to the source of energy that governs the behavior of the mechanism.

This method basically has two limitations or contraindications. The first of the limitations is that it is not advisable to make this type of identification in open loop when the integrity of the machine can be compromised by operating in said mode. The second of the limitations comes from the fact of not having access to the power source that governs the behavior of the machine. In the case of a wind turbine, we are therefore faced with two limitations that open loop identification techniques present.

For all these reasons, it is essential to use closed loop identification techniques in the field of wind turbine control.

Open loop identification is a technique known and widely used in the field of control and other scientific or financial areas. The concept of identification in the field of control refers to a method for experimentally obtaining input-output models of a mechanism or machine. The concept of open loop refers in turn to the fact that the system is developed without the control of the application mechanism or machine being active, which makes the system operate freely.

This open loop identification technique is developed experimentally, based on a specific configuration in relation to the system for which the application is intended.

In the case of wind turbines, the conventional controller design process is developed using linear models obtained from non-linear simulators. The protocols and linearization algorithms are diverse, but they have in common that they are numerical methods and that they are based exclusively on the theoretical equations that describe the behavior of a wind turbine. This protocol consists in applying a differential impulse to the different inputs to the non-linear models. These can be wind speed, pitch angle and generator torque, etc. After these small disturbances, the wind turbine is allowed to evolve freely. Thus, the derivatives of all the states defined by the non-linear simulator are estimated and from these values that are obtained, linear models of the machine to be controlled are determined, with which the controllers are designed and then implemented in real machines .

Due to the variability of the factors that They influence the behavior of wind turbines, the controllers that are obtained in the indicated way, based on models obtained after the linearization of non-linear simulators, sometimes they do not work as expected on the real machine. This can be critical, since the design of the components of a wind turbine is based on the controllers designed against the linear models.

It is known, on the other hand, the concept of closed loop, which refers to the fact that the control loops that automatically govern the behavior of a mechanism, are active during the normal operation of the machine.

The concept of closed-loop identification, however, was initially rejected by the researchers, considering that it was not applicable for technical reasons. However, recent studies show that this technology is possible, making a correct formulation of the problem and the algorithms used in the process.

Object of the invention

In accordance with the invention, a system of identification of closed-loop wind turbines is proposed, by means of which linear models that accurately represent the dynamic behavior of the application wind turbine can be obtained, such models being used for various purposes, such as controller design, use as part of a control algorithm, parameter identification, component analysis, etc. The system is based on the development of a process that is carried out through software, through which a series of steps are performed, obtaining and storing data, which are subsequently analyzed and treated in order to obtain models of input and output that can be subsequently used for any of the functions indicated above.

As mentioned, the objective of this experimentation protocol is to obtain input / output models for the wind turbine. When reference is made throughout the text, it should be assumed that it can be any of the following input variables:

Wind-related: Wind speed, at any height from the ground or at any distance from the wind turbine, whether measured or inferred.

Related to the generator: Generator torque, real or demanded, torque in the generator air gap, generator power, real or demanded, active power, real or demanded, reactive power, real or demanded, currents and voltages of each of the lines of the three-phase system, real or demanded, voltages and currents dyq, real or demanded, of the generator control system, etc.

Related to the pitch actuator: Pitch angle, real or demanded, pitch speed, real or demanded, pitch acceleration, real or demanded, powers, voltages and currents, _ • η

actual or demanded of the pitch motor. Pressures, flows, positions, speeds and accelerations of the actuator, in case of being hydraulic, etc.

- Related to the yaw actuator: Yaw angle, real or demanded, yaw speed, real or demanded, yaw acceleration, real or demanded, powers, voltages and currents, real or demanded of the yaw motor. Pressures, flows, positions, speeds and accelerations of the actuator, in case of being hydraulic, etc.

Related to the geometry of the blade: Variation in geometry of the section of the blade and the length of the blade, position, speed, acceleration of elements or parts of elements, used for aerodynamic control, mobile in the blades of the wind turbine, as well as voltages, currents, pressures, flows, etc., of the actuators that modified the geometric properties of the blades

On the other hand, when the text refers to outputs within the identification process, it will be referring to any of the following variables, sensible or inferred, of a wind turbine:

Speed of the generator, speed of the rotor on the high side, speed of the rotor on the low side, real and demanded active power to the generator, real and demanded reactive power to the generator, actual and demanded currents and voltages of each of the lines of the three-phase system, actual and demanded voltages and currents dyq of the generator control system, deformations, positions, speeds, accelerations, linear and angular, forces and moments in the bushing, along the power train, along each of the blades, along the tower, along the nacelle, etc.

All variables cited as input variables can also be considered as output variables.

The process includes an experimentation protocol, data collection, identification routines, and obtaining linear models for the desired application.

The experimentation protocol is developed during the normal operation of the wind turbine, with the control of the same asset. This control consists in the introduction of a reference signal as the desired value for the variable to be controlled. The signal to be controlled, the output, is fed back to compare it with the reference signal. This difference, called error, feeds the controller that will be responsible for applying a signal that will cause the output signal to approximate its reference value by means of the actuator, thus trying to reject external and uncontrollable disturbances to the system such as bursts of wind etc.

For its part, the closed loop experimentation protocol for the identification of wind turbines is carried out by introducing a signal that can be added to the reference signal of the control loop or can be added to the value that the controller have calculated. Data collection and storage should be done by software, and then analyzed using identification routines.

The identification routines must consist of a copy of the controller and an adjustable model of the closed loop input / output model, of any combination of the input and output variables mentioned above. The corresponding output of the real wind turbine is compared with the closed loop output predicted by the identification algorithm. The error is processed by another algorithm, which will vary the values of the parameters in the appropriate way, in order to minimize the difference between the actual measurement of the wind turbine output variable and the output variable predicted by the identification algorithm.

Once the model is obtained, it can be validated by statistical, frequency and temporal methods. If these validation conditions are met, the models are given as correct and can then be used for all the exposed activities.

A system is thus obtained with which the following advantages are achieved:

1. Obtaining reliable models. Experience in working with linear models obtained through linearization techniques from non-linear simulators, demonstrates that these sometimes provide unreliable models; so that you can find models of the same machine that have very different properties and even contradictory. This makes the design of controllers against these models raise reasonable doubts about their validity. On the contrary, the models obtained in closed loop are closer to reality and therefore more reliable, obtaining better results from them.

2. Reduction of numerical errors in the models. One of the possible consequences of the techniques currently used for estimating models is the obtaining of poorly conditioned models. This results in limitations in their treatment and in their correct transformation of a system of representation of linear models (state spaces), to other methods of representation of linear input-output models (transfer functions). On the contrary, the proposed system eliminates the problem of obtaining poorly conditioned models.

3. Theoretical-practical models.

The fact of using an experimental technique allows the use of the system both in the field of simulation, and in the field of experimentation on real machines.

4. "Ad-hoc" models for each machine and location. The fact of being able to obtain a model for each machine, makes it possible to design an appropriate controller for each of them, in relation to its own specificities due to construction, assembly, location, etc. 5. Obtaining temporary models of the same machine.

The fact that one can experiment on real machines, makes it possible to obtain models that describe their behavior when deemed appropriate, so that degradation between the theoretical model and the real machine can be studied over time.

6. Maintenance applied to control algorithms.

Since real machine models can be obtained at any time, the modifications deemed appropriate within the control algorithms can be made, taking into account the changes that the machine undergoes over time and They detect with the identifier model.

7. Use of the models identified within the control routines.

There are different control techniques and strategies that explicitly use the plant model and the model of the disturbances that can be obtained through identification, so that when the identified models mentioned are available, these techniques are likely to be used.

Description of the figures

Figure 1 shows a block diagram of the development process of the system of the invention.

Figure 2 is a diagram of the application of the excitation signal in the system experimentation protocol, according to an embodiment. Figure 3 is a diagram of the application of the excitation signal in the experimentation protocol, according to another embodiment.

Detailed description of the invention

The object of the invention relates to a closed loop wind turbine identification system, which aims to obtain linear models that accurately represent the dynamic behavior of the application wind turbine, to design the algorithms that govern the behavior of the wind turbine in the actual operation of it.

The system consists of the realization of a process of generation, storage and analysis of data, by means of a sequence like the one represented in figure 1, which includes an experimentation protocol (1), a data collection (2), routines of identification (3), the obtaining (4) of linear input-output models, and the application (5) of the linear models identified for the design and implementation of wind turbine controllers.

In accordance with the proposed system, the experimentation protocol (1) is developed as follows:

During normal operation of the wind turbine (6), according to the representation of figures 2 and 3, its control is always active, applying a reference signal (7) in relation to the signal to be controlled, with feedback (8 ) of the output of said signal to the input, so that at an operation time of the wind turbine (6) an excitation signal (9) is introduced into the control algorithm (10) that governs the system.

Said excitation signal (9) must have a high frequency content, particularly between 0 and 1000 Hz, as well as a determined and limited length depending on the application wind turbine (β), which excitation signal (9) can be applied at the input of the reference signal (7), as shown in figure 2, or in an intermediate part with respect to one or more of the actuators that govern the behavior of the wind turbine (6), as shown in figure 3.

That is to say, in relation to a wind turbine (6), it is possible to mark, for example, a speed of rotation of the blades of said wind turbine (6), whereby it tries to reach said speed of rotation of the blades adapting to each wind speed, for which the output signal of the speed of the blades is introduced back into the machine after comparing it with the input signal, which is indicating to the wind turbine (6) a greater acceleration or lower in the speed of rotation, preventing the operation of the wind turbine (6) from leaving the logical operating ranges.

On the other hand, the introduction of an excitation signal (9), which is coupled to the input reference signal (7), allows, by studying the output and the effect of the disturbances, to design the algorithms that govern the wind turbine operation (6) in a real situation. The mentioned experimentation protocol (1) can also be carried out, under the same conditions, on a simulator instead of the real wind turbine (6), achieving an identical result.

By collecting data (2) the experiment can be repeated as many times as desired, and the procedure can be applied to all control loops that are active in the wind turbine or simulator (6), with the possibility of measuring all the appropriate variables for Get the models you want.

During the experiment, the data corresponding to all the sensors that the wind turbine or simulator (6) has, as well as the sequences of all the signals involved in the wind turbine control loop, as well as the excitation signal, whether it is saved entered as a sum to the reference, as if it is introduced as a sum to the action calculated by the controller.

The identification routines (3) of the process are developed by submitting the data obtained to conventional parameter identification algorithms, so that these identification routines (3) are intended to find the relationship between temporary collections of data, the results being in these identification routines (3), linear empirical models are obtained.

The result of applying the experimentation protocol (1), as well as the data collection (2) and their subsequent treatment with the identification routines (3), are linear models that They describe the behavior of the system, so that once (4) these models are obtained, they can be applied (5) for activities such as: component design, start-up of wind turbines, certification, adaptive control, predictive control, maintenance of control, control tuning, dynamic operation monitoring, etc.

Claims

1.- Identification of wind turbines in a closed loop, to obtain linear models that accurately represent the dynamic behavior of the application wind turbine, characterized in that a process is developed that includes an experimentation protocol (1), a data collection ( 2), identification routines (3), obtaining (4) of linear input-output models and the application (5) of said linear models for the design and synthesis of controllers, where the experimentation protocol (1) includes the supply of a reference signal (7) in the algorithm (10) of the function to be controlled, with the application wind turbine (6) in normal operation, the output signal of said function being fed back (8) to the input, same time that an excitation signal (9) is applied in the control algorithm (10) of the system.

2.- Identification of wind turbines in closed loop, according to the first claim, characterized in that the excitation signal (9) is applied at the same input input of the reference signal (7).

3.- Identification of wind turbines in closed loop, according to the first claim, characterized in that the excitation signal (9) is applied in an intermediate part on the actuators that govern the operation of the application wind turbine (6).

4.- Identification of wind turbines in closed loop, according to claims first to third, characterized in that the experimentation protocol can be performed under the same conditions on a real wind turbine (6), or on a ^' simulator.

5.- Identification of wind turbines in closed loop, according to the first claim, characterized in that the input models in relation to which the system is applied refer to variables related to wind, with the generator, with the pitch actuator , with the yaw actuator, or with the geometry of the wind turbine blades.

6.- Identification of wind turbines in closed loop, according to the first claim, characterized in that the output models in relation to which the system is applied refer to variables related to the generator, with the rotor, with the system lines three-phase, with the control system, or with the geometry of any of the structural parts of the wind turbine.