A method for identifying control parameters of a doubly-fed wind generator converter considering the whole wind speed range

Abstract

The present application relates to a method for identifying control parameters of a doubly-fed wind turbine converter considering the whole wind speed range, comprising: determining the control parameters of the doubly-fed wind turbine converter to be identified, collecting each observation output response data set under different control parameters, calculating the correlation between the to-be-identified parameters and different observations, and selecting the observation; using the gating cycle unit to obtain the preliminary identification value of each to-be-identified control parameter; based on the preliminary identification value, using the improved particle swarm algorithm to accurately identify the control parameters of the doubly-fed wind turbine under different operating states; calculating the response error between the measured curve and the identification curve under three-phase fault disturbance, and verifying the accuracy of the identification result. The present application divides three different operating states according to the wind speed on the basis of the preliminary identification value of the GRU model, and uses the improved particle swarm algorithm to accurately identify the control parameters of the doubly-fed wind turbine converter under different operating states, effectively solving the problem of the influence of the random output characteristics of wind power on the identification accuracy.

Description

Technical Field

[0001] This invention relates to the field of wind power generation control, and in particular to a method for identifying control parameters of a doubly fed wind turbine converter that considers the entire wind speed range. Background Technology

[0002] The rapid growth of new energy sources such as wind power and photovoltaics, coupled with the intermittent and random nature of wind power, introduces significant challenges to its safe and stable operation through the integration of numerous power electronic devices into the power grid. To analyze the impact of large-scale grid-connected wind power on the power grid, it is necessary to establish accurate wind turbine generator models. The accuracy of these models depends on the accuracy of their parameters, which are often difficult to obtain directly from nameplates or manuals. Therefore, parameter identification becomes an effective means of acquiring these model parameters.

[0003] Due to the stochastic power output characteristics of double-fed induction generators (DFIGs), the operating states of DFIGs are divided into Maximum Power Point Tracking (MPPT), constant speed, and constant power operating states based on wind speed intervals. The control structure is the same under these operating states, but the parameters may differ. Therefore, the differences between these operating states must be considered when identifying the control parameters of the DFIG converter. Especially under constant power operating state, the DFIG pitch angle controller will activate, but this controller only triggers under specific wind speed conditions and does not increase the complexity of converter control parameter identification. Therefore, if the stochastic power output characteristics of the DFIG are not considered, the identification results will not accurately reflect the dynamic characteristics of the DFIG.

[0004] In view of this, this invention studies the random output characteristics of doubly fed wind turbines and, in combination with the operating characteristic curves, divides the doubly fed wind turbines into three different operating states based on wind speed across the entire wind speed range. Based on the gated recurrent unit (GRU) neural network model and the improved particle swarm optimization (PSO) algorithm, the control parameters of the doubly fed wind turbine converter under different operating states across the entire wind speed range are identified. Summary of the Invention

[0005] The purpose of this invention is to address the problem of the influence of the random characteristics of wind power on the accuracy of parameter identification. It proposes a method for identifying control parameters of doubly-fed induction generator (DFIG) converters considering the entire wind speed range. The method employs a two-step identification approach: first, a gated cyclic unit is used to perform preliminary identification of the DFIG converter control parameters; then, based on the preliminary identification results, an improved particle swarm optimization (IPSO) algorithm is used for further precise identification, resulting in high-precision wind turbine converter control parameters. The parameters to be identified include the control parameters of the rotor-side converter (RSC) PI controller and the grid-side converter (GSC) PI controller.

[0006] The technical solution of this invention is a method for identifying control parameters of a doubly-fed induction generator (DFIG) considering the entire wind speed range, comprising the following steps:

[0007] Step 1: Determine the control parameters of the doubly fed wind turbine converter to be identified, collect the output response datasets of various observations under different control parameters, calculate the correlation between the parameters to be identified and different observations using the distance correlation coefficient method, and select the observations.

[0008] Step 2: Organize and prepare multiple sets of observation datasets, and use the processed datasets to train the gated recurrent unit neural network model to obtain the preliminary identification values ​​of each control parameter to be identified;

[0009] Step 3: Based on the operating characteristic curve of the doubly fed fan, the doubly fed fan is divided into different operating states according to the wind speed in the whole wind speed range. Based on the preliminary identification value of the gated cyclic unit neural network, the improved particle swarm algorithm is used to accurately identify the control parameters of the doubly fed fan under different operating states.

[0010] Step 4: Extract the optimal control parameter identification value from multiple identification results under different operating conditions, and input the optimal control parameter identification value into the doubly fed wind turbine control model. Calculate the response error between the measured curve and the identification curve under three-phase short-circuit fault disturbance to verify the accuracy of the identification results.

[0011] Preferably, in step 1, the observed quantity includes the rotor current dq-axis component i. dr and i qr Stator current dq axis component i ds and i qs Rotor voltage dq axis component u dr and u qr Stator output active power P s and reactive power Q s dq-axis component of grid-side converter output current i dg and iqg dq-axis component of grid-side converter output voltage u dg and u qg The active power output P of the grid-side converter g and reactive power Q g and DC bus voltage U dc .

[0012] Furthermore, step 2 specifically includes the following sub-steps:

[0013] Step 201: Collect observation data under different control parameters, and use the distance correlation coefficient method to screen out observations that are strongly correlated with the control parameters;

[0014] Step 202: Normalize the selected observations and divide the processed observation dataset into training and test sets according to the proportions.

[0015] Step 203: Set the uniform error as the loss function, use Adam as the optimizer, and identify the control parameters for the training set sequence obtained from the simulation.

[0016] Step 204: Calculate the error between the output of the gated recurrent unit network model and the training set, and continuously train the gated recurrent unit network model based on the calculated error until the training error is less than the training requirement;

[0017] Step 205: Use the actual dataset of the doubly fed wind turbine as the input of the gated cyclic unit network model, and use the output of the gated cyclic unit network model as the preliminary identification value of the control parameters based on the gated cyclic unit network model.

[0018] Furthermore, step 3 specifically includes the following sub-steps:

[0019] Step 301: Based on the operating characteristic curve of the doubly fed wind turbine, the operating states are divided into three different states: maximum power point tracking, constant speed, and constant power, according to the wind speed.

[0020] Step 302: Expand the preliminary identification values ​​of the control parameters obtained in Step 2 into a range of control parameter values, and use the range of control parameter values ​​as the parameter search range of the improved particle swarm algorithm under different operating states;

[0021] Step 303: Use the current values ​​of the particles in the population as control parameters, run the doubly fed wind turbine control model, and use the weighted method to calculate the objective function value of the wind turbine converter control parameters, and pass it to the improved particle swarm algorithm as the particle fitness value.

[0022] Step 304: Update the velocity, position, individual optimal and global optimal values ​​of particles in the population, and determine whether the identification results meet the accuracy or the number of iterations has reached the maximum value. If yes, output the optimal identification value of the control parameters; otherwise, return to step 303 to continue the iteration.

[0023] Preferably, the improved particle swarm optimization algorithm nonlinearly adjusts the inertia weight ω and learning factors c1 and c2 to enhance the algorithm's ability to handle nonlinear problems. In the early stage of iteration, c1 is initially large and then small, biased towards individual optima, while a larger ω facilitates global search. In the later stage of iteration, c2 is initially small and then large, biased towards global optima, while a smaller ω facilitates local search and algorithm convergence.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] 1. This invention takes into account the stochastic characteristics of wind power and proposes a method for identifying control parameters of doubly-fed induction generator (DFIG) converters that considers the entire wind speed range. Based on the preliminary identification values ​​of the GRU network model, this method combines the DFIG operating characteristic curves and divides the operating state into three different states based on wind speed. An improved particle swarm optimization algorithm is used to accurately identify the control parameters of the DFIG converter under different operating states, effectively solving the problem that the stochastic output characteristics of wind power affect the identification accuracy.

[0026] 2. In order to improve the ability of the PSO algorithm in handling nonlinear parameter identification research, this invention makes nonlinear improvements to the inertia weight ω and learning factors c1 and c2, which effectively avoids the problem that traditional PSO is prone to getting trapped in local optima, and improves the global search capability and convergence speed of the PSO algorithm.

[0027] 3. This invention uses the distance correlation coefficient method to analyze the correlation between each control parameter and different observations, selects the observations with high correlation with each control parameter as the input features of the GRU neural network model and the construction index of the objective function of the improved particle swarm algorithm weighted method, effectively improving the parameter identification accuracy.

[0028] 4. This invention uses weighted average absolute deviation to measure response curve error. Under three-phase short-circuit fault conditions, the response error between the identified curve and the actual curve is calculated, further verifying the accuracy of the identification results. Attached Figure Description

[0029] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0030] Figure 1 This is a flowchart illustrating the identification of control parameters for a doubly fed wind turbine converter considering the entire wind speed range, as described in an embodiment of the present invention.

[0031] Figure 2 This is a typical grid-connected structure and control block diagram for a doubly fed wind turbine.

[0032] Figure 3 These are the power-wind speed and angular velocity-wind speed operating characteristic curves of a doubly fed wind turbine.

[0033] Figure 4 The graph shows the calculated distance correlation coefficients between different observations and RSC control parameters.

[0034] Figure 5 The graph shows the calculated distance correlation coefficients between different observations and GSC control parameters.

[0035] Figure 6 This is a comparison chart of the actual active power curve and the identified curve response under MPPT operating conditions.

[0036] Figure 7 This is a comparison chart of the actual and identified reactive power response curves under MPPT operation conditions.

[0037] Figure 8 This is a comparison chart of the actual curve and the identified curve response of active power under constant speed operation.

[0038] Figure 9 This is a comparison chart of the actual and identified reactive power response curves under constant speed operation.

[0039] Figure 10 This is a comparison chart of the actual curve and the identified curve response of active power under constant power operation.

[0040] Figure 11 This is a comparison chart of the actual and identified reactive power response curves under constant power operation conditions. Detailed Implementation

[0041] like Figure 1 As shown, the method for identifying control parameters of a doubly-fed induction generator (DFIG) considering the entire wind speed range includes the following steps:

[0042] Step 1: Determine the parameters to be identified based on the typical control block diagram of the doubly fed wind turbine converter, collect simulation datasets of the output response of each observation under different control parameters, and use the distance correlation coefficient method to calculate the correlation between the parameters to be identified and different observations, thereby determining the selection of observations;

[0043] Step 2: Process multiple sets of observation simulation datasets, and use the processed datasets to train a gated recurrent unit neural network model to obtain preliminary identification values ​​for each parameter to be identified;

[0044] Step 3: Based on the operating characteristic curve of the doubly fed wind turbine, the doubly fed wind turbine is divided into three different operating states according to the wind speed in the whole wind speed range. Based on the preliminary identification value of the gated recurrent unit neural network, the improved particle swarm algorithm is used to accurately identify the parameters of different operating states respectively.

[0045] Step 4: Extract the optimal identification value from multiple sets of identification results under different operating conditions, and input the optimal identification value into the model. Calculate the response error between the measured curve and the identification curve under three-phase short-circuit fault disturbance to verify the accuracy of the identification results.

[0046] In step 1, the parameters to be identified are the control parameters of the rotor-side converter PI controller and the control parameters of the grid-side converter PI controller. The formula for calculating the distance correlation coefficient describing the two variables x and y is as follows:

[0047]

[0048] Where dcor(x,y) represents the distance correlation coefficient; dcov() represents the distance covariance;

[0049]

[0050] In the formula, S1, S2, and S3 are all intermediate variables; the calculation formulas for S1, S2, and S3 are as follows:

[0051]

[0052]

[0053]

[0054] In the formula x i x j Represents the i-th and j-th data points of variable x; y i y j represents the i-th and j-th data points of variable y; n represents the total length of the sample data. This represents the correlation distance between each data point of variables x and y.

[0055] Similarly, dcov(x,x) and dcov(y,y) can be calculated.

[0056] Figure 2 The diagram shows a typical grid-connected structure and control block diagram for a doubly fed wind turbine. Based on this, the parameters to be identified are the RSC controller parameters and the GSC controller parameters.

[0057] The distance correlation coefficient calculation results are as follows Figure 4 and Figure 5 As shown, the observed quantity P s and Q s ids and i qs With rotor current dq axis component i dr and i qr The correlation coefficients are very close because the active power P output by the stator of the doubly-fed wind turbine is... s and reactive power Q s The rotor current d-axis component i dr and q-axis component i qr The stator current d-axis and q-axis components i are determined. ds and i qs The rotor current d and q-axis components i dr and i qr Therefore, i is selected. dr i qr P s and Q s As an observation, identify each control parameter of RSC; observation i dg i qg P g and Q g The correlation between i and GSC control parameters is relatively large, therefore i is selected. dg i qg P g and Q g As an observation, identify the various control parameters of GSC.

[0058] In step 2, the gated recurrent unit network model is an improved model of the long short-term memory (LSTM) network. It integrates the forget gate and input gate of the LSTM network into a single update gate and adds a reset gate. Its network structure is as follows:

[0059] The current state memory variable h of the GRU t h is the state memory variable from the previous moment. t-1 and the current candidate set state linear combination, h t It can be represented as

[0060]

[0061] In the formula, z t To update the state of the gate; I represents the identity matrix; ⊙ is the Hadamard product symbol.

[0062] GRU current candidate set status It can be represented as

[0063]

[0064] In the formula, x t r is the input vector;t To reset the door's state; For candidate set and x t and h t-1 The weight parameters are multiplied by the connection matrix; φ is the tanh activation function.

[0065] Update Gate Z t and reset door r t They can be represented as follows:

[0066] z t =σ(W z ·[h t-1 ,x t ])(7)

[0067] r t =σ(W r ·[h t-1 ,x t ])(8)

[0068] In the formula, σ() is the sigmoid function; W z W r These represent the weight parameters for updating the gate and resetting the gate, respectively.

[0069] The current output vector y of the GRU t It can be represented as

[0070] y t =σ(W o ·h t (9)

[0071] In the formula W o This represents the output weight parameters.

[0072] Step 2, obtaining preliminary identification values ​​based on the GRU network model, includes the following steps:

[0073] Step 201: Collect observation data under different control parameters and use the distance correlation coefficient method to screen out observations that are strongly correlated with the control parameters;

[0074] Step 202: Normalize the selected observations and divide the processed observation simulation dataset into training and test sets in an 8:2 ratio;

[0075] Step 203: Set the mean squared error (MSE) as the loss function, use Adam as the optimizer, and identify the training set sequence obtained from the simulation.

[0076] Step 204: Set the last set of features in the training set to the actual output data, calculate the error between the output of the GRU network model and the training set, output the final training result based on the actual dataset, and use this result as the preliminary parameter identification value based on the GRU network model.

[0077] The preliminary identification results of the control parameters of the rotor-side converter and grid-side converter based on the GRU neural network model are shown in Tables 1 and 2, where K p4 =K p2 K i4 =K i2 K p7 =K p6 K i7 =K i6 K p1 K i1 These are the proportional and integral coefficients of the RSC active power controller, respectively; K p3 K i3 These are the proportional and integral coefficients of the RSC reactive power controller, respectively; K p2 K p4 K i2 K i4 These are the proportional and integral coefficients of the RSC inner loop current controller, respectively; K p5 K i5 These are the proportional and integral coefficients of the GSC DC voltage controller; K p6 K p7 K i6 K i7 These are the proportional and integral coefficients of the GSC inner loop current controller, respectively.

[0078] Table 1. Preliminary identification results of RSC control parameters based on the GRU network model.

[0079]

[0080] Table 2. Preliminary identification results of GSC control parameters based on the GRU network model.

[0081]

[0082] In step 3, each particle in the particle swarm optimization algorithm has a fitness value determined by the objective function, and iterative updates are performed using a velocity-position search model to obtain the optimal solution. The velocity and position update formulas are as follows:

[0083]

[0084]

[0085] In the formula, Let be the velocity of the i-th particle after the k-th iteration; Let be the velocity of the i-th particle after the k-th iteration; ω be the inertia weight; r1 and r2 are both random numbers between (0,1); c1 and c2 are individual and social learning factors, respectively. This represents the optimal value of the i-th particle after the k-th iteration. This is the globally optimal value after the k-th iteration.

[0086] A nonlinear dynamic variation coefficient is introduced to adjust the inertia weight ω and learning factors c1 and c2 of the PSO algorithm, thereby improving the PSO algorithm's ability to handle nonlinear problems. In the early stage of iteration, c1 is initially made larger and then smaller, biased towards individual optima, while a larger ω is beneficial for global search. In the later stage of iteration, c2 is initially made smaller and then larger, biased towards global optima, while a smaller ω is beneficial for local search and algorithm convergence. The improved expression is as follows:

[0087]

[0088]

[0089]

[0090] In the formula, ω max and ω min These represent the maximum and minimum inertia weights, respectively; T is the maximum number of iterations; and t is the current number of iterations.

[0091] Step 3, which involves accurately identifying control parameters under different operating conditions based on the IPSO algorithm, specifically includes the following steps:

[0092] Step 301: Combining Figure 3 The operating characteristic curves of the doubly fed wind turbine shown are divided into three different operating states based on wind speed: MPPT, constant speed, and constant power.

[0093] Step 302: Use 0.8 to 1.2 times the preliminary identification result as the parameter search range for the IPSO algorithm under different operating conditions;

[0094] Step 303: Assign each particle in the population to each control parameter, run the model to be identified, calculate the weighted objective function value corresponding to the set of parameters, and pass it to the IPSO algorithm as the fitness value of the particle.

[0095] Step 304: Update the particle's velocity, position, and individual and global optimal values, and determine whether the identification result meets the accuracy requirements or whether the number of iterations has reached the maximum value. If yes, output the optimal identification value of the parameters; otherwise, return to step 303 to continue iterating.

[0096] Step 3, the construction of the weighted objective function, specifically includes the following steps:

[0097] 1) Based on the correlation coefficients between each observation and the control parameters obtained from the distance correlation coefficient, the correlation coefficients of each individual observation with respect to all control parameters are summed to obtain I. d I q The coefficients corresponding to the P and Q observations and d1, d2, d3 and d4;

[0098] 2) Calculate the weighting coefficient k for each observation according to the following formula. i ;

[0099]

[0100] 3) Construct the objective function based on the root mean square error formula below;

[0101]

[0102] In the formula, J represents the objective function; N is the number of data sets, and I... d,e (), I q,e (), P e (), Q e () represent the observation I d I q The deviation between the measured data and the identified data corresponding to P and Q;

[0103] 4) Substitute the calculated weighted coefficients of each observation into equation (16) to obtain the objective function of the weighted method as follows:

[0104]

[0105] Tables 3 and 4 present the accurate identification results of control parameters for rotor-side converters and grid-side converters under different operating conditions based on the IPSO algorithm.

[0106] Table 3. Accurate identification results of RSC control parameters based on IPSO algorithm

[0107]

[0108] Table 4. Precise identification results of GSC control parameters based on IPSO algorithm

[0109]

[0110] In step 4, the error between the two response curves is measured by the weighted average deviation of the intervals. Taking active power as an example, the absolute deviation of active power in each error interval is calculated.

[0111] The absolute deviation of active power F in the pre-fault stage P,A The calculation formula is as follows:

[0112]

[0113] In the formula K Start,A K End,A These are the first and last data sequence numbers within the error interval before the fault, respectively; P M (k) and P S (k) represent the measured and identified active power data of the kth interval, respectively;

[0114] Absolute deviation of active power F during the fault phase P,B The calculation formula is as follows:

[0115]

[0116] In the formula K Start,B K End,B These are the first and last data sequence numbers within the error interval of the fault stage, respectively.

[0117] The absolute deviation of active power F during the fault recovery phase P,C The calculation formula is as follows:

[0118]

[0119] In the formula K Start,C K End,C These are the first and last data sequence numbers within the error interval of the fault recovery phase, respectively.

[0120] The interval-weighted average absolute deviation of the three error intervals—pre-fault, during-fault, and fault recovery phases—is defined as follows:

[0121]

[0122] In the formula F P,A F P,B and F P,C These represent the active power deviations between the measured system and the identification system during the pre-fault, fault, and fault recovery phases, respectively; F Q,A F Q,B and F Q,C These represent the reactive power deviations between the measured system and the identification system during the pre-fault, fault, and fault recovery phases, respectively; F P F Q F and F represent the active power, reactive power, and absolute deviation of the total system, respectively.

[0123] Table 5 presents the active power, reactive power, and weighted average absolute deviation of the entire system under three-phase short-circuit fault conditions, operating in different states. As shown in Table 5, the initial identification of the converter control parameters using the GRU neural network model, and the simulation results under different operating states, resulted in an average error of approximately 2% in the response curve. Further simulation using the optimal identification results obtained under different operating states via the IPSO algorithm resulted in a weighted average total deviation of approximately 0.2% between the identified response curve and the actual response curve, representing a reduction of nearly 10 times in error.

[0124] Table 5 Weighted average absolute deviation of response curves under different operating conditions

[0125]

Claims

1. A method for identifying control parameters of a doubly-fed induction generator (DFIG) considering the entire wind speed range, characterized in that, Includes the following steps: Step 1: Determine the control parameters of the doubly fed wind turbine converter to be identified, collect the output response datasets of various observations under different control parameters, calculate the correlation between the parameters to be identified and different observations using the distance correlation coefficient method, and select the observations. Step 2: Organize and prepare multiple sets of observation datasets, and use the processed datasets to train the gated recurrent unit neural network model to obtain the preliminary identification values ​​of each control parameter to be identified; Step 201: Collect observation data under different control parameters, and use the distance correlation coefficient method to screen out observations that are strongly correlated with the control parameters; Step 202: Normalize the selected observations and divide the processed observation dataset into training and test sets according to the proportions. Step 203: Set the loss function, select the optimizer, and identify the control parameters for the training set sequence obtained from the simulation; Step 204: Calculate the error between the output of the gated recurrent unit network model and the training set, and continuously train the gated recurrent unit network model based on the calculated error until the training error is less than the training requirement; Step 205: Use the actual dataset of the doubly fed wind turbine as the input of the gated cyclic unit network model, and use the output of the gated cyclic unit network model as the preliminary identification value of the control parameters based on the gated cyclic unit network model; Step 3: Divide the doubly fed wind turbine into different operating states based on wind speed within the full wind speed range. Based on the preliminary identification value of the gated cyclic unit neural network, use the improved particle swarm algorithm to accurately identify the control parameters of the doubly fed wind turbine under different operating states. Step 301: Based on the operating characteristic curve of the doubly fed wind turbine, classify the wind turbine operating state according to the wind speed. The wind turbine operating state includes maximum power point tracking, constant speed and constant power. Step 302: Expand the preliminary identification values ​​of the control parameters obtained in Step 2 into a range of control parameter values, and use the range of control parameter values ​​as the parameter search range of the improved particle swarm algorithm under different operating states; Step 303: Use the current values ​​of the particles in the population as control parameters, run the doubly fed wind turbine control model, and use the weighted method to calculate the objective function value of the wind turbine converter control parameters, and pass it to the improved particle swarm algorithm as the particle fitness value. Step 304: Update the velocity, position, individual optimal and global optimal values ​​of particles in the population, and determine whether the identification results meet the accuracy or the number of iterations has reached the maximum value. If yes, output the optimal identification value of the control parameters; otherwise, return to step 303 to continue iterating. Step 4: Extract the optimal control parameter identification value from multiple identification results under different operating conditions, and input the optimal control parameter identification value into the doubly fed wind turbine control model. Calculate the response error between the measured curve and the identification curve under three-phase fault disturbance to verify the accuracy of the identification results.

2. The method for identifying control parameters of a doubly fed wind turbine converter according to claim 1, characterized in that, The improved particle swarm optimization algorithm nonlinearly adjusts the inertia weight ω and learning factors c1 and c2 to enhance its ability to handle nonlinear problems. In the early stage of iteration, c1 is initially large and then small to favor individual optima, while a larger ω facilitates global search. In the later stage of iteration, c2 is initially small and then large to favor global optima, while a smaller ω facilitates local search and algorithm convergence. The formula for calculating the inertia weight ω is as follows: In the formula ω max ω min These represent the maximum and minimum inertia weights, respectively; T is the maximum number of iterations; and t is the current iteration number. The formulas for calculating learning factors c1 and c2 are as follows:

3. The method for identifying control parameters of a doubly-fed wind turbine converter according to claim 2, characterized in that, In step 1, the observations include the rotor current dq-axis component i dr and i qr Stator current dq axis component i ds and i qs Rotor voltage dq axis component u dr and u qr Stator output active power P s and reactive power Q s dq-axis component of grid-side converter output current i dg and i qg dq-axis component of grid-side converter output voltage u dg and u qg The active power output P of the grid-side converter g and reactive power Q g and DC bus voltage U dc .

4. The method for identifying control parameters of a doubly-fed wind turbine converter according to claim 3, characterized in that, In step 1, the formula for calculating the distance correlation coefficient between the two variables x and y is: In the formula, dcor(x,y) represents the distance correlation coefficient; dcov() represents the distance covariance. In the formula, S1, S2 and S3 are all intermediate variables; The formulas for calculating S1, S2, and S3 are: In the formula x i x j Represents the i-th and j-th data points of variable x; y i y j represents the i-th and j-th data points of variable y; n represents the total length of the sample data. This represents the correlation distance between each data point of variables x and y; Similarly, calculate dcov(x,x) and dcov(y,y).

5. The method for identifying control parameters of a doubly fed wind turbine converter according to claim 4, characterized in that, In step 3, the objective function value of the wind turbine converter control parameters is calculated using a weighted method. The specific process is as follows: 1) Based on the correlation coefficients between each observation and the control parameters obtained from the distance correlation coefficient, the correlation coefficients of each individual observation with respect to all control parameters are summed to obtain the d-axis component I of the current. d Current q-axis component I q The correlation coefficients and d1, d2, d3, and d4 corresponding to the observed active power P and reactive power Q; 2) Calculate the weighting coefficient k for each observation. i ; In the formula k i i = 1, 2, 3, 4 represent the correlation coefficient and d, respectively. i Weighting coefficients; 3) Construct the objective function In the formula, J represents the objective function; N is the number of data sets, and I... d,e (), I q,e (), P e (), Q e () represent the observation I d I q The deviation between the measured data and the identified data corresponding to P and Q; 4) Substitute the calculated weighted coefficients of each observation into the objective function constructed in step 3), and the resulting objective function is as follows:

6. The method for identifying control parameters of a doubly-fed wind turbine converter according to claim 5, characterized in that, In step 4, the interval weighted average absolute deviation is used to measure the response error between the measured curve and the identification curve. Taking active power as an example, the active power absolute deviation of each error interval is calculated. The absolute deviation of active power F in the pre-fault stage P,A The calculation formula is as follows: In the formula K Start,A K End,A These are the first and last data sequence numbers within the error interval before the fault, respectively; P M (k) and P S (k) represent the measured and identified active power data of the kth interval, respectively; Absolute deviation of active power F during the fault phase P,B The calculation formula is as follows: In the formula K Start,B K End,B These are the first and last data sequence numbers within the error interval of the fault stage, respectively. The absolute deviation of active power F during the fault recovery phase P,C The calculation formula is as follows: In the formula K Start,C K End,C These are the first and last data sequence numbers within the error interval of the fault recovery phase, respectively. The interval-weighted average absolute deviation of the three error intervals—pre-fault, during-fault, and fault recovery phases—is defined as follows: In the formula F P,A F P,B and F P,C These represent the active power deviations between the measured system and the identification system during the pre-fault, fault, and fault recovery phases, respectively; F Q,A F Q,B and F Q,C These represent the reactive power deviations between the measured system and the identification system during the pre-fault, fault, and fault recovery phases, respectively; F P F Q F and F represent the active power, reactive power, and absolute deviation of the total system, respectively.

Images