Virtual inertia identification method suitable for doubly-fed wind power plant

Abstract

The invention discloses a virtual inertia identification method suitable for a doubly-fed wind power plant. The method comprises the following steps: constructing a second-order transfer function and a virtual rotational inertia expression of grid-connected power and power grid frequency of the doubly-fed wind power plant; collecting data, selecting a stable active power sequence and a power grid angular frequency sequence as identification data and denoising, and then selecting quasi-steady-state data to calculate a droop control coefficient; and discretizing the transfer function, identifying a coefficient of the discretized transfer function based on a regularized least square method, restoring a second-order transfer function based on the identified parameter, and calculating the virtual rotational inertia by using the coefficient of the second-order transfer function and a virtual rotational inertia expression. According to the method, no phase delay exists in the process of identifying the virtual inertia of the doubly-fed wind power plant, and the peak characteristic of inertia response is reserved; the inertia identification error caused by parameter coupling in the traditional method is solved; and by introducing a regularization factor, a severe oscillation phenomenon of fitting parameters is prevented, and the identification precision of the virtual inertia is remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of power system control and parameter identification technology, specifically to a method for virtual inertia identification of doubly-fed wind farms. Background Technology

[0002] As the penetration rate of renewable energy sources such as wind power increases, the rotational inertia of power systems gradually decreases, posing challenges to frequency stability. Doubly fed induction generators (DFIGs) typically support the grid frequency through additional frequency control, and DFIG wind farms are power generation bases composed of numerous DFIG wind turbines. To assess the frequency support capability of the system, accurately identifying the equivalent inertia constant of a DFIG wind farm is crucial.

[0003] Existing technologies for inertia identification mainly suffer from the following problems:

[0004] 1) Noise sensitivity: Inertia identification depends on the rate of change of frequency. The power grid frequency signal usually contains measurement noise. Direct differential will amplify high-frequency noise, causing the identification results to oscillate violently.

[0005] 2) Filtering Delay: To suppress noise, existing technologies often employ low-pass filters such as Butterworth filters, but this introduces a significant phase delay. This affects the power change at time t in the identification equation. The corresponding one is Moment This phase delay disrupts the physical causal relationship of the inertial response, causing a large identification error.

[0006] 3) Parameter Coupling: Traditional least squares methods tend to minimize global error. Since inertia response exists only in the transient process, while droop control dominates the steady-state process, if the overall fit is performed indiscriminately, the algorithm often sacrifices the accuracy of transient inertia parameters to achieve steady-state accuracy, resulting in low accuracy of the identified inertia values. Furthermore, traditional least squares methods are even unable to perform calculations and fitting when the data matrix is ​​non-invertible. Summary of the Invention

[0007] In view of this, the present invention provides a virtual inertia identification method suitable for doubly-fed wind farms, in order to solve the problems of large differential signal noise, high filtering delay and inaccurate identification caused by steady-state parameter coupling in existing doubly-fed wind farm inertia identification.

[0008] The virtual inertia identification method applicable to doubly-fed wind farms of this invention includes the following measures:

[0009] 1) For doubly-fed induction generator (DFIG) wind farms in power systems, construct the second-order transfer function of the grid-connected power and grid frequency of the DFIG wind farm:

[0010] (1)

[0011] (2)

[0012] In the formula: It is the change in grid-connected power. It is the change in the angular frequency of the power grid. The moment of inertia of the doubly-fed asynchronous wind turbine generator. The equivalent time constant of the active power control loop of a doubly-fed asynchronous wind turbine generator. is the damping coefficient of the doubly-fed asynchronous wind turbine. is the aerodynamic damping coefficient of a doubly-fed asynchronous wind turbine. For virtual inertia coefficients, This is the droop control coefficient. represents the number of pole pairs of a doubly-fed asynchronous wind turbine. For steady-state mechanical angular velocity, For steady-state stator active power, The steady-state angular frequency at the grid connection point. is the maximum power tracking coefficient; s is the Laplace operator;

[0013] b) Approximate the doubly-fed wind farm as a virtual doubly-fed asynchronous wind turbine, and construct a virtual moment of inertia expression for the doubly-fed wind farm:

[0014] (3)

[0015] In the formula, This represents the number of pole pairs for generators in the power system other than doubly-fed asynchronous wind turbines.

[0016] 2) Apply load disturbance to the power system, and then collect the active power sequence and grid angular frequency sequence at the grid connection point of the doubly-fed wind farm; collect real-time wind speed data at each doubly-fed asynchronous wind turbine in the doubly-fed wind farm, and calculate the average wind speed of the whole field at each sampling time based on the collected real-time wind speed data to obtain the average wind speed sequence of the whole field of the doubly-fed wind farm.

[0017] 3) Determine whether the collected average wind speed sequence is a stationary sequence, and select the active power sequence and power grid angular frequency sequence corresponding to the collection time of the stationary sequence as identification data;

[0018] 4) Denoise the active power sequence and grid angular frequency sequence used as identification data;

[0019] 5) Calculate the droop control coefficient by selecting quasi-steady-state data from the later stages of frequency disturbances in the denoised active power sequence and the grid angular frequency sequence:

[0020] (4)

[0021] in, This represents the droop control coefficient. The average value of the quasi-steady-state active power change. The average value of the quasi-steady-state power grid angular frequency change is given by t, which represents the sampling time.

[0022] (5)

[0023] (6)

[0024] in, It is the active power sampled at time t from the selected quasi-steady-state active power data sequence. It is a sample value in the original denoised active power data sequence;

[0025] (7)

[0026] (8)

[0027] in, It is the grid angular frequency sampled at time t from the selected quasi-steady-state data sequence of grid angular frequency. It is the first sampled value in the original denoised power grid angular frequency data sequence.

[0028] in, These are the lengths of the active power quasi-steady-state data sequence and the grid angular frequency quasi-steady-state data sequence, respectively.

[0029] Then, the active power sequence data obtained by denoising using measure 4) is subtracted from the steady-state component to obtain the dynamic power sequence:

[0030] (9)

[0031] in, Represents a dynamic power sequence. This represents the original denoised active power sequence. This is a sequence of changes in the angular frequency of the power grid. For steady-state components;

[0032] 6) Transfer function Discretization yields:

[0033] (10)

[0034] In the formula, It is a discrete transformation operator;

[0035] The least-squares form is constructed using the power grid angular frequency sequence data obtained through noise reduction processing (measure 4) and the dynamic power sequence obtained through measure 5):

[0036] (11)

[0037] In the formula: This is the k-th sample value in the dynamic power sequence. T is the sampling period; It is the first sampled value in the dynamic power sequence;

[0038] The parameters to be identified , This refers to measure 4) the power grid angular frequency sequence data obtained through noise reduction processing;

[0039] The objective function for least squares is constructed as follows:

[0040] (12)

[0041] in, Must meet:

[0042] (13)

[0043] In the formula: N is the sequence length; ;

[0044] Solving :

[0045] (14)

[0046] Introducing regularization factors Construct the cost function and apply the ridge regression algorithm to equation (13) to obtain:

[0047] (15)

[0048] In the formula, I is the identity matrix, and the parameters are obtained by solving equation (15). ;

[0049] 7) The parameters identified in measure 6) Substitute the transfer function Using the following formula:

[0050] (16)

[0051] Will Reverting to the continuous-domain transfer function form shown in equation (1), we can obtain the coefficients in equation (1). Then, the virtual moment of inertia contributed by the doubly fed wind farm to the power system is calculated using equation (3).

[0052] Furthermore, in measure 3), the methods for determining whether the collected average wind speed sequence is a stationary sequence include:

[0053] Calculate the average of all samples in the overall wind speed average sequence. Subtract the average of all samples in the overall wind speed average sequence from each sample value in the wind speed average sequence. Then compare the absolute value of the difference with the set screening threshold. If the absolute value of each difference is less than or equal to the screening threshold, the wind speed average sequence is a stationary sequence; otherwise, the wind speed average sequence is a non-stationary sequence.

[0054] Furthermore, in measure 4), a Savitzky-Golay filter is used to denoise the active power sequence and the grid angular frequency sequence, which are used as identification data.

[0055] The beneficial effects of this invention are:

[0056] This invention eliminates phase delay in identifying the virtual inertia of a doubly-fed wind farm, preserving the peak characteristics of the inertia response; it solves the inertia identification error caused by parameter coupling in traditional methods; and by introducing a regularization factor, it prevents severe oscillations in the fitted parameters, significantly improving the identification accuracy of the virtual inertia. Attached Figure Description

[0057] Figure 1 Flowchart for estimating the virtual moment of inertia of a doubly-fed wind farm;

[0058] Figure 2 A comparison chart of power prediction and measured values ​​from the virtual inertia identification algorithm for a doubly-fed wind farm under frequency disturbance;

[0059] Figure 3 This is a comparison of the frequency responses of the original system and the inertial background system under frequency disturbance. Detailed Implementation

[0060] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0061] As shown in the figure, the virtual inertia identification method applicable to doubly-fed wind farms in this embodiment includes the following measures:

[0062] 1) For doubly-fed induction generator (DFIG) wind farms in power systems, construct the second-order transfer function of the grid-connected power and grid frequency of the DFIG wind farm:

[0063] (1)

[0064] (2)

[0065] In the formula: It is the change in grid-connected power. It is the change in the angular frequency of the power grid. The moment of inertia of the doubly-fed asynchronous wind turbine generator. The equivalent time constant of the active power control loop of a doubly-fed asynchronous wind turbine generator. is the damping coefficient of the doubly-fed asynchronous wind turbine. is the aerodynamic damping coefficient of a doubly-fed asynchronous wind turbine. For virtual inertia coefficients, This is the droop control coefficient. represents the number of pole pairs of a doubly-fed asynchronous wind turbine. For steady-state mechanical angular velocity, For steady-state stator active power, The steady-state angular frequency at the grid connection point. is the maximum power tracking coefficient; s is the Laplace operator.

[0066] b) Approximate the doubly-fed wind farm as a virtual doubly-fed asynchronous wind turbine, and construct a virtual moment of inertia expression for the doubly-fed wind farm:

[0067] (3)

[0068] In the formula, This represents the number of pole pairs for generators in a power system other than doubly-fed asynchronous wind turbines.

[0069] The expression for the virtual moment of inertia described in this embodiment is obtained through the following process:

[0070] The dynamic characteristics of the grid angular frequency after a doubly-fed asynchronous wind turbine is connected to the grid are related to the power variation as follows:

[0071] (4)

[0072] In the formula, It is the moment of inertia of generators in the power system other than doubly-fed asynchronous wind turbines. It is the number of pole pairs of generators in a power system other than doubly-fed asynchronous wind turbines. It is the damping coefficient of generators in the power system other than doubly-fed asynchronous wind turbines. This represents the change in mechanical power. This represents the change in load disturbance power.

[0073] Combining equations (2) and (4), the transfer function between the grid angular frequency and active power It can be represented as:

[0074] (5)

[0075] (6)

[0076] For the second-order transfer function (5), the Routh approximation theory is used for order reduction to ensure the dynamic characteristics in the low-frequency band. The order reduction results are as follows:

[0077] (7)

[0078] Combining formulas (5), (6), and (7), the overall rotational inertia transfer function of the power system is derived:

[0079] (8)

[0080] In the transfer function Represents the overall rotational inertia of the power system , It consists of three parts: The moment of inertia of other generators in the power system; This is the damping coefficient related to other generators and doubly-fed asynchronous wind turbines in the power system. This term is very small and can be ignored. The virtual moment of inertia contributed to the power system by the doubly-fed asynchronous wind turbine with additional frequency control. For a doubly-fed wind farm, if the control parameters of the doubly-fed asynchronous wind turbines inside are similar, the doubly-fed wind farm can be equivalent to a virtual doubly-fed asynchronous wind turbine. The virtual moment of inertia of this equivalent unit is determined by all units in the farm, thus deriving formula (3).

[0081] 2) Apply load disturbance to the power system, and then collect the active power sequence and grid angular frequency sequence at the grid connection point of the doubly-fed wind farm; collect real-time wind speed data at each doubly-fed asynchronous wind turbine in the doubly-fed wind farm, and calculate the average wind speed of the whole field at each sampling time based on the collected real-time wind speed data to obtain the average wind speed sequence of the whole field of the doubly-fed wind farm.

[0082] 3) Determine whether the collected average wind speed sequence is a stationary sequence. Select the active power sequence and grid angular frequency sequence corresponding to the collection time of the stationary sequence as identification data. In this step, the methods for determining whether the collected average wind speed sequence is a stationary sequence include:

[0083] Calculate the average of all samples in the overall wind speed average sequence. Using each sample value in the wind speed average series Subtract the average of all samples from each sample, and then compare the absolute value of the difference with the set screening threshold, as shown in the following expression:

[0084]

[0085] If the absolute value of each difference is less than or equal to the screening threshold If the wind speed average sequence is positive, then the wind speed average sequence is a stationary sequence; otherwise, the wind speed average sequence is a non-stationary sequence.

[0086] 4) The active power sequence and the grid angular frequency sequence, which are used as identification data, are denoised using a Savitzky-Golay filter.

[0087] 5) Select quasi-steady-state data from the later stages of frequency disturbances in the denoised active power sequence and grid angular frequency sequence (in this embodiment, specifically select the last 20% of the active power sequence and grid angular frequency sequence data) to calculate the droop control coefficient:

[0088] (9)

[0089] in, This represents the droop control coefficient. The average value of the quasi-steady-state active power change. The average value of the quasi-steady-state power grid angular frequency change is given by t, which represents the sampling time.

[0090] (10)

[0091] (11)

[0092] in, It is the active power sampled at time t from the selected quasi-steady-state active power data sequence. It is the first sampled value in the original denoised active power data sequence.

[0093] (12)

[0094] (13)

[0095] in, It is the grid angular frequency sampled at time t from the selected quasi-steady-state data sequence of grid angular frequency. It is the first sampled value in the original denoised power grid angular frequency data sequence.

[0096] in, These are the lengths of the active power quasi-steady-state data sequence and the grid angular frequency quasi-steady-state data sequence, respectively.

[0097] Then, the active power sequence data obtained by denoising using measure 4) is subtracted from the steady-state component to obtain the dynamic power sequence:

[0098] (14)

[0099] in, Represents a dynamic power sequence. This represents the denoised active power sequence. This is a sequence of changes in the angular frequency of the power grid. This is the steady-state component.

[0100] 6) Transfer function Discretization yields:

[0101] (15)

[0102] In the formula, It is a discrete transformation operator.

[0103] The least-squares form is constructed using the power grid angular frequency sequence data obtained through noise reduction processing (measure 4) and the dynamic power sequence obtained through measure 5):

[0104] (16)

[0105] In the formula: This is the k-th sample value in the dynamic power sequence. T is the sampling period; It is the first sampled value in the dynamic power sequence.

[0106] The parameters to be identified , The power grid angular frequency sequence data obtained by measure 4) noise reduction processing.

[0107] The objective function for least squares is constructed as follows:

[0108] (17)

[0109] in, Must meet:

[0110] (18)

[0111] In the formula: N is the sequence length; ;

[0112] Solving :

[0113] (19)

[0114] Introducing regularization factors Construct the cost function and apply the ridge regression algorithm to equation (19) to obtain:

[0115] (20)

[0116] In the formula, I is the identity matrix, and the parameters are obtained by solving equation (20). .

[0117] 7) The parameters identified in measure 6) Substitute the transfer function Using the following formula:

[0118] (twenty one)

[0119] Will Reverting to the continuous-domain transfer function form shown in equation (1), we can obtain the coefficients in equation (1). Then, the virtual moment of inertia contributed by the doubly fed wind farm to the power system is calculated using equation (3).

[0120] The effectiveness of the invention will be verified through simulation experiments below:

[0121] The relevant parameters of a doubly-fed wind farm are shown in Table 1:

[0122] Table 1 Parameters of Doubly Fed Wind Farm

[0123] parameter numerical values Number of DFIG units 3 units DFIG rated power 2 MVA Stator voltage 575 V <![CDATA[DFIG moment of inertia J d > <![CDATA[580 kg·m 2 ]]> <![CDATA[Resistance R on the stator and rotor windings s , R r > 23, 16 mΩ <![CDATA[Leakage inductance L on the stator and rotor windings ls , L lr > 0.18, 0.16 mH <![CDATA[Mutual inductance L between the stator winding and the rotor winding m > 2.9 mH Current inner loop PI parameters [1.6, 80] Power outer loop PI parameters [0.2, 20] <![CDATA[Virtual inertia coefficient k df > 0.2 <![CDATA[Droop control coefficient k pf > 0.2

[0124] Simulation verification: The load disturbance is set to 4MW, and the sampling time interval is... Sampling duration When using the Savitzky-Golay filter to denoise the original active power sequence and the grid angular frequency sequence, a p=3rd order polynomial is used to fit the local window data, the sliding window size is set to 11, and the ridge regression regularization factor is used. Set to 1e-5. Figure 2 This is a comparison chart of the power prediction and measured values ​​from the virtual inertia identification algorithm for a doubly-fed inertial wind farm under frequency disturbance. The inertia identification results are as follows: =633.15 kg·m 2 .

[0125] To verify the correctness of the derived virtual moment of inertia, The moment of inertia is added to the synchronous generator, while the additional frequency control of each doubly-fed induction generator (DFIG) in a doubly-fed wind farm is removed. This creates a new test system called the inertial background system. If the derived virtual moment of inertia is accurate enough, the inertial background system should have the same moment of inertia as the original system, and the same frequency response characteristics under load disturbances. Figure 3The comparison diagram shows the frequency response of the original system and the inertial background system under frequency disturbance. The root mean square error of the identification results is only 0.0386, and the average absolute percentage error is only 1.84%. Simulation results show that the inertia identification method of the present invention can identify the virtual inertia coefficient of the doubly-fed wind farm well, with high identification accuracy, and is suitable for evaluating the frequency support capability of power systems with high renewable energy penetration.

[0126] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A virtual inertia identification method suitable for doubly-fed wind farms, characterized in that: Including the following measures: 1) For doubly-fed induction generator (DFIG) wind farms in power systems, construct the second-order transfer function of the grid-connected power and grid frequency of the DFIG wind farm: （1） （2） In the formula: It is the change in grid-connected power. It is the change in the angular frequency of the power grid. The moment of inertia of the doubly-fed asynchronous wind turbine generator. The equivalent time constant of the active power control loop of a doubly-fed asynchronous wind turbine generator. is the damping coefficient of a doubly-fed asynchronous wind turbine. is the aerodynamic damping coefficient of a doubly-fed asynchronous wind turbine. This is the virtual inertia coefficient. This is the droop control coefficient. represents the number of pole pairs of a doubly-fed asynchronous wind turbine. For steady-state mechanical angular velocity, For steady-state stator active power, The steady-state angular frequency at the grid connection point. This is the maximum power tracking coefficient; s is the Laplace operator; b) Approximate the doubly-fed wind farm as a virtual doubly-fed asynchronous wind turbine, and construct a virtual moment of inertia expression for the doubly-fed wind farm: （3） In the formula, This represents the number of pole pairs for generators in the power system other than doubly-fed asynchronous wind turbines. 2) Apply load disturbance to the power system, and then collect the active power sequence and grid angular frequency sequence at the grid connection point of the doubly-fed wind farm; collect the real-time wind speed data at each doubly-fed asynchronous wind turbine in the doubly-fed wind farm, and calculate the average wind speed of the whole field at each sampling time based on the collected real-time wind speed data to obtain the average wind speed sequence of the whole field of the doubly-fed wind farm. 3) Determine whether the collected average wind speed sequence is a stationary sequence, and select the active power sequence and power grid angular frequency sequence corresponding to the collection time of the stationary sequence as identification data; 4) Denoise the active power sequence and grid angular frequency sequence used as identification data; 5) Calculate the droop control coefficient by selecting quasi-steady-state data from the later stages of frequency disturbances in the denoised active power sequence and the grid angular frequency sequence: （4） in, This represents the droop control coefficient. The average value of the quasi-steady-state active power change. The average value of the quasi-steady-state power grid angular frequency change is given by t, which represents the sampling time. （5） （6） in, It is the active power sampled at time t from the selected quasi-steady-state active power data sequence. It is the first sampled value in the original denoised active power data sequence; （7） （8） in, It is the grid angular frequency sampled at time t from the selected quasi-steady-state data sequence of grid angular frequency. It is the first sampled value in the original denoised power grid angular frequency data sequence; in, These are the lengths of the active power quasi-steady-state data sequence and the grid angular frequency quasi-steady-state data sequence, respectively. Then, the active power sequence data obtained by denoising using measure 4) is subtracted from the steady-state component to obtain the dynamic power sequence: （9） in, Represents a dynamic power sequence. This represents the original denoised active power sequence. This is a sequence of changes in the angular frequency of the power grid. For steady-state components; 6) Transfer function Discretization yields: （10） In the formula, It is a discrete transformation operator; The least-squares form is constructed using the power grid angular frequency sequence data obtained through noise reduction processing (measure 4) and the dynamic power sequence obtained through measure 5): （11） In the formula: This is the k-th sample value in the dynamic power sequence. T is the sampling period; It is the first sampled value in the dynamic power sequence; The parameters to be identified , This refers to measure 4) the power grid angular frequency sequence data obtained through noise reduction processing; The objective function for least squares is constructed as follows: （12） in, Must meet: （13） In the formula: N is the sequence length; ; Solving : （14） Introducing regularization factors Construct the cost function and apply the ridge regression algorithm to equation (13) to obtain: （15） In the formula, I is the identity matrix, and the parameters are obtained by solving equation (15). ; 7) The parameters identified in measure 6) Substitute the transfer function Using the following formula: （16） Will Reverting to the continuous-domain transfer function form shown in equation (1), we can obtain the coefficients in equation (1). Then, the virtual rotational inertia contributed by the doubly fed wind farm to the power system is calculated using equation (3).

2. The virtual inertia identification method for doubly-fed wind farms according to claim 1, characterized in that: In measure 3), the methods for determining whether the collected average wind speed sequence is a stationary sequence include: Calculate the average of all samples in the overall wind speed average sequence. Subtract the average of all samples in the overall wind speed average sequence from each sample value in the wind speed average sequence. Then compare the absolute value of the difference with the set screening threshold. If the absolute value of each difference is less than or equal to the screening threshold, the wind speed average sequence is a stationary sequence; otherwise, the wind speed average sequence is a non-stationary sequence.

3. The virtual inertia identification method for doubly-fed wind farms according to claim 1, characterized in that: In measure 4), the Savitzky-Golay filter is used to denoise the active power sequence and the grid angular frequency sequence, which are used as identification data.

Images