A method for suppressing subsynchronous oscillation of a doubly-fed wind turbine based on data-driven predictive control

Abstract

The present invention discloses a method for suppressing subsynchronous oscillation of a doubly-fed wind turbine based on data-driven predictive control, including: collecting historical measurement sequences of the doubly-fed wind turbine under continuous excitation to establish a historical data matrix; collecting the measurement sequence of the doubly-fed wind turbine in real time to establish a real-time data matrix; constructing a data-driven predictive controller suitable for subsynchronous oscillation suppression, solving the controller to obtain the optimal control law to achieve wind turbine oscillation suppression; updating the historical data matrix according to two update strategies of timed update and abnormal update, and updating the data-driven predictive controller. The present invention only requires the input and output data of the doubly-fed wind turbine, does not require accurate modeling of the internal mechanism of the unit, reduces the dependence on modeling accuracy, and effectively improves the practicality of the oscillation suppressor. The present invention takes into account the uncertainty of the operating conditions of the doubly-fed wind turbine, realizes stable control of the time-varying wind turbine system, and ensures the robustness and effectiveness of the subsynchronous oscillation suppression method.

Description

Technical Field

[0001] The present invention relates to the technical field of power system subsynchronous oscillation suppression, and in particular to a method for suppressing subsynchronous oscillation of a doubly-fed wind turbine based on data-driven predictive control. Background Art

[0002] Wind power and photovoltaic power generation are important directions for the development of green energy and have become one of the most important clean energy sources in the world. However, with the continuous increase in wind power installed capacity, there is a risk of sub-synchronous oscillation (SSO) in large-scale wind power grid-connected systems, which will threaten the safety and stability of wind farms and grid-connected systems, and seriously restrict the new power system's ability to absorb new energy.

[0003] At present, wind farm synchronous oscillation suppression strategies can be divided into two categories according to different suppression methods, namely grid-side suppression measures and unit-side suppression measures. Among them, grid-side suppression measures will add additional power electronic equipment, resulting in higher system complexity and operating costs, while unit-side suppression measures are simple and easy to implement, with faster response speed and greater practical significance.

[0004] There are various existing unit-side suppression methods, such as additional damping control, converter parameter optimization, and filter introduction into the control loop. However, the existing unit-side suppression methods are based on the aggregation model, which does not consider the difference between the actual unit and the theoretical aggregation model. When the system parameters change greatly, the subsynchronous oscillation cannot be accurately suppressed; and the robustness is weak, making it difficult to cope with the situation where multiple oscillation frequencies occur at the same time. Therefore, a robust method for suppressing subsynchronous oscillations of a doubly fed wind turbine is needed. Summary of the invention

[0005] The purpose of the present invention is to overcome the defects of the prior art and to propose a method for suppressing subsynchronous oscillation of a doubly-fed wind turbine based on data-driven predictive control.

[0006] The present invention is achieved through the following technical solutions:

[0007] The present invention first provides a method for suppressing subsynchronous oscillation of a doubly-fed wind turbine based on data-driven predictive control, and the implementation process is as follows:

[0008] 1) Collect the historical measurement sequence of the doubly fed wind turbine under continuous excitation and establish a historical data matrix;

[0009] 2) Collect the measurement sequence of the doubly-fed wind turbine in real time and establish a real-time data matrix;

[0010] 3) Based on the historical data matrix and the real-time data matrix, a data-driven predictive controller suitable for subsynchronous oscillation suppression is constructed, the reference value of the control target is set and the controller is solved to obtain the optimal control law to achieve wind turbine oscillation suppression;

[0011] 4) According to the two updating strategies of scheduled update and abnormal update, the historical data matrix is ​​updated, and the data-driven predictive controller is updated. Thus, the purpose of suppressing subsynchronous oscillation of doubly-fed wind turbine based on data-driven predictive control is achieved.

[0012] The present invention also provides a doubly-fed wind turbine subsynchronous oscillation suppression system based on data-driven predictive control for implementing the method, which comprises:

[0013] A historical data matrix construction module is used to collect historical measurement sequences of the doubly-fed wind turbine generator set under continuous excitation and construct a historical data matrix;

[0014] Real-time data matrix building module, used to collect the measurement sequence of the double-fed wind turbine in real time and build a real-time data matrix;

[0015] The subsynchronous oscillation suppression module is used to build a data-driven predictive controller suitable for subsynchronous oscillation suppression based on the historical data matrix and the real-time data matrix, and solve the optimization problem to obtain the optimal control law to achieve wind turbine oscillation suppression;

[0016] The historical data matrix update module is used to update the historical data matrix according to scheduled updates and abnormal updates.

[0017] Compared with the prior art, the present invention targets the subsynchronous oscillation phenomenon of doubly fed wind turbines and proposes a method for suppressing subsynchronous oscillation of doubly fed wind turbines based on data-driven predictive control. The method can suppress subsynchronous oscillation of doubly fed wind turbines by taking into account the differences among units, and effectively improves the stability of the wind power grid-connected system.

[0018] The controller involved in the present invention only requires input and output data of the doubly-fed wind turbine and does not require accurate modeling of the internal mechanism of the unit, thereby reducing dependence on modeling accuracy and effectively improving the practicality of the oscillation suppressor.

[0019] The control target of the present invention can be specified as needed. For example, the control target of the present invention can be that the active power increment and reactive power increment of the doubly fed wind turbine are both 0. When the power fluctuations are caused by wind speed changes, reference power changes, etc., the oscillation problem in the system can be more effectively suppressed to achieve effectiveness for multiple operating points.

[0020] The present invention takes into account the uncertainty of the operating conditions of the doubly fed wind turbine, updates the controller in a manner combining the two strategies of scheduled update and abnormal update, achieves stable control of the time-varying wind turbine system, and ensures the robustness and effectiveness of the subsynchronous oscillation suppression method. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1The flowchart of the method for suppressing subsynchronous oscillation of a doubly-fed wind turbine based on data-driven predictive control is shown in FIG.

[0022] Figure 2 The subsynchronous oscillation suppression method of a doubly-fed wind turbine based on data-driven predictive control and the control structure diagram of the doubly-fed wind turbine;

[0023] Figure 3 It is the update strategy diagram of the subsynchronous oscillation suppression method of the doubly-fed wind turbine based on data-driven predictive control;

[0024] Figure 4 It is a topological diagram of the simulation system in the embodiment;

[0025] Figure 5 This is the effect diagram of the subsynchronous oscillation suppression method of the doubly fed wind turbine based on data-driven predictive control. DETAILED DESCRIPTION

[0026] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0027] The present invention designs a method for suppressing subsynchronous oscillation of a doubly-fed wind turbine based on data-driven predictive control, and the steps are as follows:

[0028] Step 1: Collect the historical measurement sequence of the doubly-fed wind turbine under continuous excitation and establish a historical data matrix.

[0029] According to some embodiments of the present invention, step one specifically includes the following steps:

[0030] Step 11: Collect historical measurement sequences of the doubly-fed wind turbine under continuous excitation. Figure 2 The control block diagram of a typical doubly-fed wind turbine connected to the grid with a series compensation line is shown in Figure 1, where DFIG represents a doubly-fed wind turbine generator set, RSC and GSC are the rotor-side and grid-side converters, respectively. is the DC side capacitance, is the inductance on the unit side, , are the transmission line reactance and resistance, is the series compensation capacitance of the transmission line, , They are respectively the actual value of the RSC power outer loop power, , They are the reference values ​​of the RSC power outer loop power, , are the dq axis currents of RSC respectively, , are the RSC control loop voltage command values, , They are the DC voltage and reactive power actual values ​​of the GSC control outer loop, , They are the DC voltage and reactive power reference values ​​of the GSC control outer loop, , are the dq axis currents of GSC, , are the voltage command values ​​of the GSC control loop, and PI represents the proportional integral controller. The noise power injected into the control loops of the rotor side and the grid side is 10 -4 The white noise signal is used to trigger the continuous excitation of the controlled system. The specific location is the dq axis current reference value signal of the two converters of the doubly fed unit. , , , ,Right now Figure 2 The red part in the diagram; records the active power of the unit’s grid connection point and reactive power , and the corresponding increment and , active power fluctuation at the collection point and wind speed , record the data Figure 2 The green part of the

[0031] Step 12: Construct the Hankel matrix.

[0032] Define an input sequence of length T , the output sequence , output increment sequence , the perturbation sequence ,in, Represents the input data at time i, which also includes the d-axis current reference value signals of the two converters of the doubly fed wind turbine , and q-axis current reference value signal , ; Represents the output data at time i, and also includes the active power at time i and reactive power , Indicates the output increment data at time i, and also includes the active power increment at time i and reactive power increment , Represents the disturbance data at time i, and also includes the active power fluctuation at time i and wind speed ;

[0033] According to the defined sequences, construct input Hankel matrices with L rows and T-L+1 columns respectively. , output Hankel matrix , output incremental Hankel matrix and the perturbed Hankel matrix , the jth column of the Hankel matrix is ​​the data from the k0+j-1th moment to the k0+L+j-2th moment of the corresponding sequence, where L≤T, j≤T-L+1, to input the Hankel matrix For example, its form is as follows:

[0034]

[0035] Step 13: Construct the historical data matrix. Divide each Hankel matrix horizontally into two parts. ini The first N rows are used to implicitly estimate the initial conditions of the system, denoted by the subscript P, and the next N rows are used to predict the "future" trajectory of the system, denoted by the subscript F. ini is the time length of the preset real-time data matrix, and T ini +N=L:

[0036]

[0037] , , , , , , , is the submatrix after the corresponding Hankel matrix is ​​divided, and the superscript k0 represents the initial recording time of the data sequence;

[0038] Construct the historical data matrix at time k0 for:

[0039]

[0040] Step 2: Collect the measurement sequence of the doubly-fed wind turbine in real time and establish a real-time data matrix.

[0041] According to some embodiments of the present invention, the second step is: recording the unit operation data starting at time k, and the length of time for recording the data is T ini , the data includes the dq axis current reference value signal of the converter , , , , active power of the unit grid connection point and reactive power , and the corresponding increment and , active power fluctuation at the collection point and wind speed . The real-time data matrix at time k is constructed as:

[0042]

[0043] in, , , , They start at time k and have a length of T. ini The input sequence, perturbation sequence, output sequence and output increment sequence of ; , , , is a data sequence of length N to be predicted.

[0044] Step three, based on the historical data matrix and the real-time data matrix, construct a data-driven predictive controller suitable for subsynchronous oscillation suppression (i.e., establish an optimization problem with the goal of suppressing power oscillations), set the reference value of the control target and solve the controller to obtain the optimal control law to achieve wind turbine oscillation suppression.

[0045] According to some embodiments of the present invention, step three specifically includes the following steps:

[0046] Step 31, establish the optimization problem description. The main goal of the distributed oscillation suppressor proposed in this embodiment is to suppress power oscillation, that is, the output signal fluctuation is 0, and the input signal approaches 0. For the convenience of subsequent proof, the control goal is changed to guide y and Δy to the origin, that is, the control object is the difference between the active power and the reference signal, and the difference between the reactive power and the reference signal. The optimization problem considering the above goals can be expressed as:

[0047] ;

[0048] Among them, k is the initial time of the control problem, is the control cost matrix; is the output cost matrix; and are auxiliary slack variables to ensure the feasibility of the initial condition equality constraints; is the regularization parameter; is the output reference trajectory; N is the prediction length; Represents a quadratic form ; g is the solution to the above optimization problem; is the quadratic square penalty of g; Δ represents the incremental form of the signal;

[0049] Step 32, simplify the optimization problem and define F and ΔY F The historical matrix , not including and The running data sequence h1, , solve have:

[0050]

[0051] in, is any vector. So we have:

[0052]

[0053] Substituting the above formula into the objective function of the optimization problem, the decision variables g and y can be eliminated, thereby eliminating the equality constraint. At the same time, assuming that g and y are adjustable decision variables at the algorithm deployment time t+k stage, the optimization problem is obtained:

[0054]

[0055] For the disturbance signal d in the formula, the wind speed disturbance sequence It has low-frequency characteristics. We can use downsampling to reduce the data dimension and use the difference method to fit its trajectory, that is:

[0056]

[0057] Where M is the downsampling factor, Represents the lower rounding of x.

[0058] For another disturbance sequence, the power fluctuation at the convergence point is usually not a random bounded signal. It has a strong correlation with the wind turbine power and has a certain smoothness at low frequency. The correlation in the disturbance can effectively reduce uncertainty. Based on the optimization problem, the following disturbance dimension reduction strategy is considered:

[0059]

[0060] in, is the power disturbance signal in the algorithm deployed at the jth DFIG; is the rated power of the jth unit; is the operating status of the jth unit, which is 1 when it is operating normally and 0 when it is faulty; is the fluctuation of the convergence point, is the active power of the grid-connected point.

[0061] Step 33, solve the optimization problem, you can use the YALMIP toolbox, or you can use programming to solve it. Get the optimal control law And inject the converter dq axis current reference value , , , , to achieve wind turbine oscillation suppression.

[0062] Step 34, collect the output data of the doubly-fed wind turbine, repeat step 2 to update the real-time data matrix, and repeat step 33 to obtain the real-time optimal control law to achieve wind turbine oscillation suppression.

[0063] Step 4: Update the historical data matrix and the data-driven prediction controller according to the two update strategies of scheduled update and abnormal update. Figure 3 shown.

[0064] According to some embodiments of the present invention, the timing update strategy is as follows: define a time-dependent historical data matrix , where the superscript t represents the initial recording time of the data sequence in the data matrix; the time threshold T is set r , when the historical data matrix Yes r When the time is not updated, repeat step 1 to update the historical data matrix to obtain the historical data matrix after regular update. , t0 is a matrix The time of the last update.

[0065]

[0066] According to some embodiments of the present invention, the abnormal update strategy is as follows: when the active power increment of the unit grid connection point is detected The RMS value within the preset cycle threshold Greater than threshold , repeat step 1 to update the historical data matrix:

[0067]

[0068] ;

[0069] , ;

[0070] Among them, t0 is The time of the last update; T a For distance The time of the last update; is the transformation matrix of the history matrix; To add a new history matrix; , , , The size is 4T respectively a , 2T a , T-1, the identity matrix of n; , , , It is the newly added row of the Hankel matrix after the data is updated; , , , Respectively Time to Input data, disturbance data, output data and incremental output data at each moment; is the local transformation matrix; Representation Matrix Move up m rows and set the last m rows to 0; is the Kronecker product.

[0071] According to some embodiments of the present invention, based on the above method, the present invention designs a doubly-fed wind turbine subsynchronous oscillation suppression system based on data-driven predictive control, which includes:

[0072] A historical data matrix construction module is used to collect historical measurement sequences of the doubly-fed wind turbine generator set under continuous excitation and construct a historical data matrix;

[0073] Real-time data matrix building module, used to collect the measurement sequence of the double-fed wind turbine in real time and build a real-time data matrix;

[0074] The subsynchronous oscillation suppression module is used to build a data-driven predictive controller suitable for subsynchronous oscillation suppression based on the historical data matrix and the real-time data matrix, and solve the optimization problem to obtain the optimal control law to achieve wind turbine oscillation suppression;

[0075] The historical data matrix update module is used to update the historical data matrix according to scheduled updates and abnormal updates.

[0076] The method of the present invention is verified by using an actual wind farm equivalent simulation model on MATLAB software. The test system is an equivalent model built based on a real wind farm, such as Figure 4 As shown, the power reference value is 100MVA, the voltage reference value is 345kV, and DFIG in the figure represents a double-fed unit. is the series compensation capacitor of the No. 1 wind farm transmission line, , are the line inductance and resistance from area A to area B, , are the line inductance and resistance from area B to area C, , , are the line inductance, resistance and capacitance from area C to area D, , are the line inductance and resistance from area D to area E, , is the line inductance and resistance from area E to area F, , is the line reactance and resistance from region F to grid No. 2. The test system includes three wind farms located in different regions. Wind farm No. 1 is connected to region A by a series compensation line. Wind farms No. 1, No. 2 and No. 3 respectively operate 267, 300 and 167 doubly fed wind turbines with a rated power of 1.5MW. The rated wind speed of the doubly fed wind turbine is 11m / s, the proportional coefficient and integral coefficient of the rotor-side converter are 0.12 and 1.2 respectively, and the proportional coefficient and integral coefficient of the grid-side converter are 0.83 and 5 respectively.

[0077] At the initial moment, the wind speed of the three wind farms remained at 7.5 m / s. The series line compensation level of wind farm No. 1 was 10%. At 1 second, the connection line between area E and area F was interrupted, resulting in a radial connection between wind farm No. 3 and area E. At 1 second, wind farms No. 2 and 3 experienced subsynchronous oscillations of 21 Hz, and wind farm No. 1 experienced subsynchronous oscillations of 21 Hz and 34 Hz ​​at the same time. The active power curve of the grid connection point of wind farm No. 1 is shown in Figure 1. Figure 5 As shown in Figure 1, subsynchronous oscillation occurs in the wind power system, and there are two oscillation frequencies.

[0078] By comparing the active power curves of the system under different controls, for the original system without any control, the oscillation frequency continues to increase, eventually leading to system instability; when additional damping control is added to the system, the oscillation of 21Hz frequency is effectively suppressed, but the oscillation of 34Hz frequency still exists and gradually increases, and the system becomes unstable; by adopting the data-driven predictive control technology proposed in the present invention, with the assistance of the update strategy, the multi-modal subsynchronous oscillation is effectively suppressed.

[0079] The above-mentioned embodiments only express several implementation methods of the present invention, and the description is relatively specific and detailed, but it cannot be understood as limiting the scope of the present invention. For ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention.

Claims

1. A method for suppressing subsynchronous oscillation of a doubly-fed wind turbine based on data-driven predictive control, characterized in that: The steps include: 1) Collect the historical measurement sequence of the doubly fed wind turbine under continuous excitation and establish a historical data matrix; The step 1) is: 11) Inject a white noise signal into the control loop of the doubly fed wind turbine converter to trigger the continuous excitation of the doubly fed wind turbine. The specific injection position of the white noise signal is the dq axis current reference value signal of the two converters of the doubly fed wind turbine; record the active power P of the doubly fed wind turbine grid connection point out,i , reactive power Q out,i , active power increment ΔP out,i , reactive power increment ΔQ out,i , and the active power fluctuation d at the collection point cs,i and unit wind speed v i ; 12) Construct the Hankel matrix, output Hankel matrix, output increment Hankel matrix and perturbation Hankel matrix respectively according to the obtained data; 13) Divide each Hankel matrix and construct a historical data matrix based on the divided sub-matrices; 2) Collect the measurement sequence of the doubly-fed wind turbine in real time and establish a real-time data matrix; The step 2) is specifically as follows: record the operation data of the double-fed wind turbine starting at time k, and construct a real-time data matrix at time k. for: ； in, , , , They start at time k and have a length of T. ini The input sequence, perturbation sequence, output sequence and output increment sequence of ; , , , is a data sequence of length N to be predicted; 3) Based on the historical data matrix and the real-time data matrix, a data-driven predictive controller suitable for subsynchronous oscillation suppression is constructed, the reference value of the control target is set and the controller is solved to obtain the optimal control law to achieve wind turbine oscillation suppression; The step 3) comprises the following steps: 31) With the goal of suppressing power oscillation, the optimization problem is established as follows: ； ； Among them, k is the initial time of the control problem; is the control cost matrix; is the output cost matrix; and are auxiliary slack variables to ensure the feasibility of the initial condition equality constraints; is the regularization parameter; is the output reference trajectory; N is the prediction length; Represents a quadratic form ; g is the solution to the above optimization problem; is the quadratic square penalty of g; Δ represents the incremental form of the signal; 32) Solve the optimization problem and obtain the optimal control law And inject into the two converter dq axis current reference value , , , , realize wind turbine oscillation suppression; 33) Collect the output data of the doubly-fed wind turbine generator set, repeat step 2) to update the real-time data matrix, and repeat step 32) to obtain the real-time optimal control law to achieve wind turbine generator set oscillation suppression; 4) Update the historical data matrix and the data-driven prediction controller according to the two update strategies of scheduled update and abnormal update; 2. The method according to claim 1, characterized in that The step 12) is specifically as follows: Define an input sequence of length T , the output sequence , output increment sequence , the perturbation sequence ,in, Represents the input data at time i, which also includes the d-axis current reference value signals of the two converters of the doubly fed wind turbine , and q-axis current reference value signal , ; Represents the output data at time i, and also includes the active power at time i and reactive power , Indicates the output increment data at time i, and also includes the active power increment at time i and reactive power increase , Represents the disturbance data at time i, and also includes the active power fluctuation at time i and wind speed ; According to the defined sequences, construct input Hankel matrices with L rows and T-L+1 columns respectively. , output Hankel matrix , output incremental Hankel matrix and the perturbed Hankel matrix , the j-th column of the Hankel matrix is ​​the data from the k0+j-1th moment to the k0+L+j-2th moment of the corresponding sequence, where L≤T, j≤T-L+1.

3. The method according to claim 2, characterized in that The step 13) is specifically as follows: Each Hankel matrix is ​​divided horizontally into two parts. ini The first N rows are used to implicitly estimate the initial conditions of the system, denoted by the subscript P, and the next N rows are used to predict the "future" trajectory of the system, denoted by the subscript F. ini is the time length of the preset real-time data matrix, and T ini +N=L: ; , , , , , , , is the submatrix after the corresponding Hankel matrix is ​​divided, and the superscript k0 represents the initial recording time of the data sequence; Construct the historical data matrix at time k0 for: 。 4. The method according to claim 1, characterized in that: The timing update strategy in step 4) is: Defining a time-dependent historical data matrix , where the superscript t represents the initial recording time of the data sequence in the data matrix; the time threshold T is set r , when the historical data matrix Yes r When the time is not updated, repeat step 1) to update the historical data matrix to obtain the historical data matrix after the time is updated. , t0 is a matrix The time of the last update.

5. The method according to claim 1, characterized in that The abnormal update strategy in step 4) is: When the active power increment of the unit grid connection point is detected The RMS value within the preset cycle threshold Greater than threshold When , repeat step 1) to update the historical data matrix: ； ； ， ； Among them, t0 is The time of the last update; T a For distance The time of the last update; is the transformation matrix of the history matrix; To add a new history matrix; , , , The size is 4T respectively a , 2T a , T-1, the identity matrix of n; , , , It is the newly added row of the Hankel matrix after the data is updated; , , , Respectively Time to Input data, disturbance data, output data and incremental output data at each moment; is the local transformation matrix; Representation Matrix Move up m rows and set the last m rows to 0; is the Kronecker product.

6. A doubly-fed wind turbine subsynchronous oscillation suppression system based on data-driven predictive control implementing the method according to any one of claims 1 to 5, characterized in that: include: A historical data matrix construction module is used to collect historical measurement sequences of the doubly-fed wind turbine generator set under continuous excitation and construct a historical data matrix; Real-time data matrix building module, used to collect the measurement sequence of the doubly-fed wind turbine in real time and build a real-time data matrix; The subsynchronous oscillation suppression module is used to build a data-driven predictive controller suitable for subsynchronous oscillation suppression based on the historical data matrix and the real-time data matrix, and solve the optimization problem to obtain the optimal control law to achieve wind turbine oscillation suppression; The historical data matrix update module is used to update the historical data matrix according to scheduled updates and abnormal updates.

Images