Data and knowledge driven SVG electromagnetic transient model parameter identification method

Abstract

The invention discloses a data and knowledge-driven SVG electromagnetic transient model parameter identification method, which comprises the following steps of S1, analyzing SVG fault evolution characteristics, and establishing and utilizing SVG actual control strategy knowledge to drive an electromagnetic transient identification model; s2, hardware-in-loop testing is carried out on an RT-LAB platform, data sets of an SVG actual controller under different fault working conditions are obtained, and an HIL data drive is adopted to improve a GOOSE algorithm; s3, identifying parameters in the control strategy knowledge by using an improved GOOSE algorithm driven by HIL data; according to the invention, the adaptability and accuracy of identification are improved, and accurate identification of SVG control parameters is realized.

Description

Technical Field

[0001] The invention relates to the field of SVG parameter identification, and in particular to a data and knowledge driven SVG electromagnetic transient model parameter identification method. Background Art

[0002] The transformation of traditional energy to clean, low-carbon new energy is the primary feature of future energy transformation. The 14th Five-Year Plan clearly points out: accelerate the development of non-fossil energy, adhere to centralized and distributed development, and vigorously increase the scale of wind power and photovoltaic power generation. However, the large-scale access of new energy units to the power grid through nonlinear power electronic converters will pose a huge threat to the safe and stable operation of the power system. The configuration of reactive power compensation device SVG is a common method to improve the reactive power regulation capability of new energy stations. In the electromagnetic transient simulation of power systems, the accuracy of SVG control parameters is closely related to various links such as electromagnetic modeling, planning and design of new energy stations, and power grid safety and stability analysis. Unfortunately, in the actual engineering modeling process, due to realistic factors such as commercial secrets and patent protection, the models usually provided by manufacturers are black box models, which brings major challenges to the analysis of SVG electromagnetic transient characteristics. Traditional identification algorithms are known for their intuitive principles and convenient implementation, but their identification accuracy is often unsatisfactory when faced with a variety of different working conditions. In contrast, the application of artificial intelligence methods driven by data and knowledge in parameter identification can better cope with complex problems. Data-driven methods usually use a large amount of actual data as learning samples to train intelligent algorithms, showing strong versatility and the ability to continuously learn and evolve, but they do not pay much attention to the internal mechanism of the research object, lack theoretical support, and have weak interpretability. Knowledge-driven methods conduct rigorous principle analysis and mathematical modeling of the research object based on systematic theoretical knowledge and rule experience. They have strong interpretability, but there are problems such as high cost of knowledge acquisition, difficulty in balancing calculation accuracy and efficiency, and inability to continuously learn and evolve. Summary of the invention

[0003] The purpose of the present invention is to overcome the above-mentioned shortcomings and provide a data and knowledge driven SVG electromagnetic transient model parameter identification method to improve the adaptability and accuracy of the identification so as to realize the accurate identification of SVG control parameters.

[0004] In order to solve the above technical problems, the technical solution adopted by the present invention is: a data and knowledge driven SVG electromagnetic transient model parameter identification method, comprising the following steps:

[0005] Step S1, analyzing the SVG fault evolution characteristics, establishing and utilizing the SVG actual control strategy knowledge, and driving the electromagnetic transient identification model;

[0006] Step S2, performing hardware-in-the-loop testing on the RT-LAB platform to obtain data sets of the actual SVG controller under different fault conditions, and using HIL data to drive the improvement of the GOOSE algorithm;

[0007] Step S3, using the HIL data-driven improved GOOSE algorithm to identify parameters in the control strategy knowledge.

[0008] Preferably, the step S1 comprises:

[0009] Step S11, dividing the SVG fault process into five stages: pre-fault steady state A, fault occurrence transient state B, fault ride-through period C, fault recovery transient state D and post-fault steady state E;

[0010] Among them, the SVG controller adopts the fault ride-through control strategy after the voltage in the fault stages B, C, and D reaches the ride-through threshold; when the voltage in the fault stages A, B, C, D, and E does not reach the ride-through threshold, the steady-state control strategy is adopted;

[0011] Step S12, based on the fault stages divided in S11, the SVG steady-state control strategy knowledge model is as follows:

[0012]

[0013] Where: K pv , K iv are the proportional and integral coefficients of the DC side voltage outer loop control respectively; K pq , K iq They are the proportional and integral coefficients of the reactive power outer loop control on the AC side; K pi , K ii are the proportional and integral coefficients of the current inner loop control respectively; t is the time; U dc is the DC side voltage value; U dcref is the DC side voltage reference value; U d , U q are the d-axis and q-axis voltage control signals respectively; U gd , U gq are the d-axis and q-axis components of the grid-connected point voltage respectively; i d 、i q are d-axis and q-axis currents respectively; i dref 、i qref is the reference value of d and q axis current; Q is the actual value of reactive power at the grid connection point; Q ref is the reactive power given value of the grid connection point; ω is the angular frequency; L is the filter inductance value;

[0014] The parameters that need to be identified in the SVG steady-state control strategy model are: K pv , K iv , K pq , Kiq , K pi , K ii ;

[0015] Step S13, the SVG fault ride-through control strategy knowledge model is as follows:

[0016]

[0017] Where: I Lsq ,I Hsq are the steady-state currents (pu) of LVRT and HVRT fault phase C respectively; k Ls1 , k Ls2 k is two different LVRT reactive current calculation coefficients; Hs1 , k Hs2 For two different HVRT reactive current calculation coefficients; I N is the per unit value (pu) of the SVG rated current; I q0 is the steady-state value of reactive current in fault stage A; I qmin ,I qmax They are the upper and lower limits of reactive current (pu); U Lref , U Href They are LVRT voltage reference value and HVRT voltage reference value (pu) respectively; U is the grid connection point voltage per unit value (pu);

[0018] The parameters that need to be identified in the SVG fault crossing model are: k Ls1 , k Ls2 , k Hs1 , k Hs2 ,I qmin ,I qmax , U Lref , U Href ;

[0019] Step S14, using the actual control strategy knowledge in S12 and S13 to drive the electromagnetic transient identification model.

[0020] Preferably, step S2 comprises:

[0021] Step S21, HIL data acquisition: based on the RT-LAB simulation system, connect the RT-LAB with the actual SVG controller through optical fiber, test the fault response characteristics of the SVG under various fault conditions, and generate corresponding simulation data sets;

[0022] Step S22, define the objective function: The objective function is defined as:

[0023]

[0024] In the formula: U, I qare the grid-connected voltage and reactive current output by the actual SVG model, obtained through the RT-LAB platform test; U * , are the grid-connected voltage and reactive current output by the SVG electromagnetic transient identification model; n is the total length of the data; m is the current data sequence;

[0025] Step S23, algorithm driven: using the HIL data in S21 in combination with the objective function in S22 to drive the improvement of the GOOSE algorithm.

[0026] Preferably, the step S3 comprises:

[0027] Step S31, initializing the parameters of the improved GOOSE algorithm according to the HIL data, and calculating the initial fitness using the objective function in S22;

[0028] Step S32, update the position, the formula is as follows:

[0029] When rnd≥0.5, pro>0.2 and When , the position update formula is:

[0030]

[0031] When rnd≥0.5, pro≤0.2 or When , the position update formula becomes:

[0032]

[0033] When rnd<0.5, it enters the exploration phase, and the position update formula is:

[0034]

[0035] Where: dim represents the dimension of the problem, represents the weight of the stone picked up by the i-th goose during the t-th iteration, which is a random number between 5 and 25; are two random time constants from 0 to 1 corresponding to the i-th goose in each dimension during the t-th iteration; is the average value of the time constant corresponding to the i-th goose in one dimension during the t-th iteration; Ttolmin is the minimum value of the average time constant; represents the random coefficient corresponding to the i-th goose in the t-th iteration, and its value is between 0 and 0.17; T is the maximum number of iterations; α is the time weight factor, which changes linearly from 2 to 0 as the number of iterations decreases; represents the position of the i-th goose at the t+1 iteration; represents the best position found after t iterations; V Sis the speed of sound, which is 343.2; b is the damping factor, which takes values ​​in [0,1]; rnd and pro are random numbers in the interval [0,1] respectively;

[0036] Step S33, randomly use the Levy flight strategy to update the position, the formula is:

[0037]

[0038] Where: L step (dim) is the flight step function, fy is the flight probability constant, and the flight probability is set to 10%. step The calculation formula of (dim) flight function is:

[0039]

[0040] In the formula: ν~N(0,1), μ~N(0,σ 2 ), β is a constant;

[0041] Step S34, performing boundary check on each goose and adjusting the position;

[0042] Step S35, check whether the conditions are met, if the maximum number of iterations is reached or the requirements are met, output the global optimal parameters;

[0043] Step S36, bringing the global optimal parameters of S35 into the identification model of S13, and comparing the identification results with the HIL data through the knowledge-driven identification model;

[0044] Step S37, calculate the error by the following formula, and if the error is satisfied, output the optimal parameter;

[0045]

[0046] Where: A, B, C, D, E are the serial numbers of the five fault stages, F Gi is the average identification error of fault stage i, X R is the measured voltage or reactive current data, X S The voltage or current data output by the identification model; N start and N end are the first and last simulation data numbers of the corresponding stage respectively, and k is the current data number.

[0047] Beneficial effects of the invention: The invention takes SVG as the research object, and aims at the problems of poor accuracy and adaptability of the electromagnetic transient model parameter identification method of the traditional SVG controller. The invention uses the relevant knowledge of the actual control strategy of SVG to drive the electromagnetic transient identification model, thereby improving the adaptability of the identification. At the same time, HIL data is used to drive the improved GOOSE algorithm, thereby improving the accuracy of the identification and realizing the accurate identification of the SVG control parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a flow chart of the parameter identification method driven by data and knowledge;

[0049] Figure 2 It is a schematic diagram of the fault stage;

[0050] Figure 3 It is the SVG steady-state control strategy model diagram;

[0051] Figure 4 It is the SVG fault ride-through control strategy model diagram;

[0052] Figure 5 This is the schematic diagram of the HIL simulation system of the SVG controller;

[0053] Figure 6 It is a comparison chart of iteration curves of different algorithms;

[0054] Figure 7 This is the comparison chart of simulation results when high voltage ride-through Q = 0.2pu;

[0055] Figure 8 This is a comparison chart of simulation results when low voltage ride-through Q=0.2pu. DETAILED DESCRIPTION

[0056] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0057] Example 1: Figure 1 , a data and knowledge driven SVG electromagnetic transient model parameter identification method, the steps include:

[0058] Step S1, analyzing the SVG fault evolution characteristics, establishing and utilizing the SVG actual control strategy knowledge, and driving the electromagnetic transient identification model;

[0059] Step S2, performing hardware-in-loop (HIL) testing on the RT-LAB platform to obtain data sets of the actual SVG controller under different fault conditions, and using HIL data to drive the improvement of the GOOSE algorithm;

[0060] Step S3, using the HIL data-driven improved GOOSE algorithm to identify parameters in the control strategy knowledge;

[0061] Furthermore, the specific steps of S1 are as follows:

[0062] Step S11, referring to NB / T 31066-2015 "Guidelines for Modeling Electrical Simulation Models of Wind Turbines", the SVG fault process is divided into five stages: pre-fault steady state A, fault transient state B, fault ride-through period C, fault recovery transient state D and post-fault steady state E. Figure 2 shown.

[0063] Among them, the SVG controller adopts the fault ride-through control strategy after the voltage in the fault stages B, C, and D reaches the ride-through threshold; when the voltage in the fault stages A, B, C, D, and E does not reach the ride-through threshold, the steady-state control strategy is adopted.

[0064] Step S12, based on the fault stage divided in S11, refer to Figure 3 , the knowledge model of SVG steady-state control strategy is as follows:

[0065]

[0066] Where: K pv , K iv are the proportional and integral coefficients of the DC side voltage outer loop control respectively; K pq , K iq They are the proportional and integral coefficients of the reactive power outer loop control on the AC side; K pi , K ii are the proportional and integral coefficients of the current inner loop control respectively; t is the time; U dc is the DC side voltage value; U dcref is the DC side voltage reference value; U d , U q are the d-axis and q-axis voltage control signals respectively; U gd , U gq are the d-axis and q-axis components of the grid-connected point voltage respectively; i d 、i q are d-axis and q-axis currents respectively; i dref 、i qref is the reference value of d and q axis current; Q is the actual value of reactive power at the grid connection point; Q ref is the given value of reactive power at the grid connection point; ω is the angular frequency; L is the filter inductance value.

[0067] The parameters that need to be identified in the SVG steady-state control strategy model are: K pv , K iv , K pq , K iq , K pi , Kii .

[0068] Step S13, refer to Figure 4 ,The knowledge model of SVG fault ride-through control strategy is as follows:

[0069]

[0070] Where: I Lsq ,I Hsq are the steady-state currents (pu) of LVRT and HVRT fault phase C respectively; k Ls1 , k Ls2 k is two different LVRT reactive current calculation coefficients; Hs1 , k Hs2 For two different HVRT reactive current calculation coefficients; I N is the per unit value (pu) of the SVG rated current; I q0 is the steady-state value of reactive current in fault stage A; I qmin ,I qmax They are the upper and lower limits of reactive current (pu); U Lref , U Href are LVRT voltage reference value and HVRT voltage reference value (pu) respectively; U is the per unit value of the grid connection point voltage (pu).

[0071] The parameters that need to be identified in the SVG fault crossing model are: k Ls1 , k Ls2 , k Hs1 , k Hs2 ,I qmin ,I qmax , U Lref , U Href .

[0072] Step S14, using the actual control strategy knowledge in S12 and S13 to drive the electromagnetic transient identification model;

[0073] Furthermore, the specific steps of S2 are as follows:

[0074] Step S21, HIL data acquisition: refer to Figure 5 ,Based on the RT-LAB simulation system, the RT-LAB is connected to the actual SVG controller through optical fiber, the fault response characteristics of SVG under various fault conditions are tested, and the corresponding simulation data sets are generated.

[0075] Step S22, define the objective function: The objective function is defined as:

[0076]

[0077] In the formula: U, I qare the grid-connected voltage and reactive current output by the actual SVG model, obtained through the RT-LAB platform test; U * , They are the grid-connected voltage and reactive current output by the SVG electromagnetic transient identification model; n is the total length of the data; and m is the current data sequence.

[0078] Step S23, algorithm driving: using the HIL data in S21 combined with the objective function in S22 to drive the intelligent algorithm;

[0079] Furthermore, the specific steps of S3 include:

[0080] Step S31, initializing the parameters of the improved GOOSE algorithm according to the HIL data, and calculating the initial fitness using the objective function in S22.

[0081] Step S32, update the position, the formula is as follows:

[0082] When rnd≥0.5, pro>0.2 and When , the position update formula is:

[0083]

[0084] When rnd≥0.5, pro≤0.2 or When , the position update formula becomes:

[0085]

[0086] When rnd<0.5, it enters the exploration phase, and the position update formula is:

[0087]

[0088] Where: dim represents the dimension of the problem, represents the weight of the stone picked up by the i-th goose during the t-th iteration, which is a random number between 5 and 25; are two random time constants from 0 to 1 corresponding to the i-th goose in each dimension during the t-th iteration; is the average value of the time constant corresponding to the i-th goose in one dimension during the t-th iteration; Ttolmin is the minimum value of the average time constant; represents the random coefficient corresponding to the i-th goose in the t-th iteration, and its value is between 0 and 0.17; T is the maximum number of iterations; α is the time weight factor, which changes linearly from 2 to 0 as the number of iterations decreases; represents the position of the i-th goose at the t+1 iteration; represents the best position found after t iterations; V Sis the speed of sound, which is 343.2; b is the damping factor, which takes a value in [0,1]; rnd and pro are random numbers in the interval [0,1] respectively.

[0089] Step S33, randomly use the Levy flight strategy to update the position, the formula is:

[0090]

[0091] Where: L step (dim) is the flight step function, fy is the flight probability constant, and the flight probability is set to 10%. step The calculation formula of (dim) flight function is:

[0092]

[0093] In the formula: ν~N(0,1), μ~N(0,σ 2 ), β is a constant, usually 1.

[0094] Step S34, perform boundary check on each goose and adjust the position.

[0095] Step S35, check whether the conditions are met. If the maximum number of iterations is reached or the requirements are met, output the global optimal parameters.

[0096] Step S36, bringing the global optimal parameters of S35 into the identification model of S13, and comparing the identification results and HIL data through the knowledge-driven identification model.

[0097] Step S37, calculate the error using the following formula, and if the error is satisfied, output the optimal parameters.

[0098]

[0099] Where: A, B, C, D, E are the serial numbers of the five fault stages, F Gi is the average identification error of fault stage i, X R is the measured voltage or reactive current data, X S The voltage or current data output by the identification model; N start and N end are the first and last simulation data numbers of the corresponding stage respectively, and k is the current data number.

[0100] Example 2: To verify the feasibility and accuracy of the data and knowledge driven SVG electromagnetic transient model parameter identification method, a certain model of SVG actual controller of a domestic manufacturer is taken as an example, and the RT-LAB platform is used to perform HIL testing to obtain its fault condition data set. Figure 5The test results are shown in Table 1. The improved GOOSE algorithm driven by HIL data is used to identify the control parameters in the actual control strategy knowledge of SVG, and then the identification results are brought into the knowledge-driven identification model. Finally, the simulation results are compared with the HIL test results to verify the effectiveness of the proposed identification method.

[0101] Table 1 Steady-state U and I before, during and after faults under different working conditions q Data / pu

[0102]

[0103] In order to verify the effectiveness and superiority of the proposed data-driven method, GA, PSO, GOOSE and improved GOOSE algorithms are selected and driven by the same HIL data set. The iteration curves corresponding to each algorithm are shown in Figure 2. Figure 6 The population size of all algorithms is 50, and the maximum number of iterations is set to 50.

[0104] Depend on Figure 6 It can be seen that the PSO particle swarm algorithm is prone to fall into the global optimum, resulting in low algorithm accuracy; although the GA genetic algorithm has acceptable convergence accuracy, its convergence speed is slow; the traditional GOOSE algorithm also has the problem of slow convergence speed.

[0105] In summary, compared with other intelligent algorithms, the improved GOOSE algorithm driven by HIL data performs better in terms of convergence speed, convergence accuracy and global search capability, which verifies the effectiveness of the proposed data-driven method.

[0106] The improved GOOSE algorithm driven by HIL data is used to identify the parameters in the control strategy knowledge, and the identification results are brought into the knowledge-driven identification model. The grid-connected point voltage U and reactive current I output by the identification model are converted into q The response curve of the HIL is compared with the actual measurement results. Figure 7 and Figure 8 After that, the average identification error of each fault stage is calculated using the formula of S37. The error results are shown in Tables 2 and 3, which verify the feasibility and effectiveness of the proposed method.

[0107] Table 2 Fault stages I q Average identification error / %

[0108]

[0109] Table 3 Average identification error of U at each fault stage (%)

[0110]

[0111] It can be seen that the deviations under various working conditions are less than 5%. It can be concluded that the proposed data and knowledge-driven SVG electromagnetic transient model parameter identification method has good identification accuracy and strong adaptability under various working conditions, and can meet the requirements of electromagnetic transient simulation of power systems.

[0112] The above embodiments are only preferred technical solutions of the present invention and should not be regarded as limiting the present invention. The protection scope of the present invention shall be the technical solutions recorded in the claims, including equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.

Claims

1. A data and knowledge driven SVG electromagnetic transient model parameter identification method, characterized by: The following steps are involved: Step S1, analyzing the SVG fault evolution characteristics, establishing and utilizing the SVG actual control strategy knowledge, and driving the electromagnetic transient identification model; Step S2, performing hardware-in-the-loop testing on the RT-LAB platform to obtain data sets of the actual SVG controller under different fault conditions, and using HIL data to drive the improvement of the GOOSE algorithm; Step S3, using the HIL data-driven improved GOOSE algorithm to identify parameters in the control strategy knowledge.

2. The data and knowledge driven SVG electromagnetic transient model parameter identification method according to claim 1, characterized in that: The step S1 comprises: Step S11, dividing the SVG fault process into five stages: pre-fault steady state A, fault occurrence transient state B, fault ride-through period C, fault recovery transient state D and post-fault steady state E; Among them, the SVG controller adopts the fault ride-through control strategy after the voltage in the fault stages B, C, and D reaches the ride-through threshold; when the voltage in the fault stages A, B, C, D, and E does not reach the ride-through threshold, the steady-state control strategy is adopted; Step S12, based on the fault stages divided in S11, the SVG steady-state control strategy knowledge model is as follows: Where: K pv , K iv are the proportional and integral coefficients of the DC side voltage outer loop control respectively; K pq , K iq They are the proportional and integral coefficients of the reactive power outer loop control on the AC side; K pi , K ii are the proportional and integral coefficients of the current inner loop control respectively; t is the time; U dc is the DC side voltage value; U dcref is the DC side voltage reference value; U d , U q are the d-axis and q-axis voltage control signals respectively; U gd , U gq are the d-axis and q-axis components of the grid-connected point voltage respectively; i d 、i q are d-axis and q-axis currents respectively; i dref 、i qref is the reference value of d and q axis current; Q is the actual value of reactive power at the grid connection point; Q ref is the reactive power given value at the grid connection point; ω is the angular frequency; L is the filter inductance value; The parameters that need to be identified in the SVG steady-state control strategy model are: K pv , K iv , K pq , K iq , K pi , K ii ; Step S13, the SVG fault ride-through control strategy knowledge model is as follows: Where: I Lsq ,I Hsq are the steady-state currents (pu) of LVRT and HVRT fault phase C respectively; k Ls1 , k Ls2 k is two different LVRT reactive current calculation coefficients; Hs1 , k Hs2 For two different HVRT reactive current calculation coefficients; I N is the per unit value (pu) of the SVG rated current; I q0 is the steady-state value of reactive current in fault stage A; I qmin ,I qmax They are the upper and lower limits of reactive current (pu); U Lref , U Href They are LVRT voltage reference value and HVRT voltage reference value (pu) respectively; U is the grid connection point voltage per unit value (pu); The parameters that need to be identified in the SVG fault crossing model are: k Ls1 , k Ls2 , k Hs1 , k Hs2 ,I qmin ,I qmax , U Lref , U Href ; Step S14, using the actual control strategy knowledge in S12 and S13 to drive the electromagnetic transient identification model.

3. The data and knowledge driven SVG electromagnetic transient model parameter identification method according to claim 2 is characterized by: The step S2 comprises: Step S21, HIL data acquisition: based on the RT-LAB simulation system, connect the RT-LAB with the actual SVG controller through optical fiber, test the fault response characteristics of the SVG under various fault conditions, and generate corresponding simulation data sets; Step S22, define the objective function: The objective function is defined as: In the formula: U, I q are the grid-connected voltage and reactive current output by the actual SVG model, obtained through the RT-LAB platform test; U * , are the grid-connected voltage and reactive current output by the SVG electromagnetic transient identification model; n is the total length of the data; m is the current data sequence; Step S23, algorithm driven: using the HIL data in S21 in combination with the objective function in S22 to drive the improvement of the GOOSE algorithm.

4. The data and knowledge driven SVG electromagnetic transient model parameter identification method according to claim 3 is characterized by: The step S3 comprises: Step S31, initializing the parameters of the improved GOOSE algorithm according to the HIL data, and calculating the initial fitness using the objective function in S22; Step S32, update the position, the formula is as follows: When rnd≥0.5, pro>0.2 and When , the position update formula is: When rnd≥0.5, pro≤0.2 or When , the position update formula becomes: When rnd<0.5, it enters the exploration phase, and the position update formula is: Where: dim represents the dimension of the problem, represents the weight of the stone picked up by the i-th goose during the t-th iteration, which is a random number between 5 and 25; are two random time constants from 0 to 1 corresponding to the i-th goose in each dimension during the t-th iteration; is the average value of the time constant corresponding to the i-th goose in one dimension during the t-th iteration; Ttolmin is the minimum value of the average time constant; represents the random coefficient corresponding to the i-th goose in the t-th iteration, and its value is between 0 and 0.17; T is the maximum number of iterations; α is the time weight factor, which changes linearly from 2 to 0 as the number of iterations decreases; represents the position of the i-th goose at the t+1 iteration; represents the best position found after t iterations; V S is the speed of sound, which is 343.2; b is the damping factor, which takes values ​​in [0,1]; rnd and pro are random numbers in the interval [0,1] respectively; Step S33, randomly use the Levy flight strategy to update the position, the formula is: Where: L step (dim) is the flight step function, fy is the flight probability constant, and the flight probability is set to 10%. step The calculation formula of (dim) flight function is: In the formula: ν~N(0,1), μ~N(0,σ 2 ), β is a constant; Step S34, performing boundary check on each goose and adjusting the position; Step S35, check whether the conditions are met, if the maximum number of iterations is reached or the requirements are met, output the global optimal parameters; Step S36, bringing the global optimal parameters of S35 into the identification model of S13, and comparing the identification results with the HIL data through the knowledge-driven identification model; Step S37, calculate the error by the following formula, and if the error is satisfied, output the optimal parameter; Where: A, B, C, D, E are the serial numbers of the five fault stages, F Gi is the average identification error of fault stage i, X R is the measured voltage or reactive current data, X S The voltage or current data output by the identification model; N start and N end are the first and last simulation data numbers of the corresponding stage respectively, and k is the current data number.

Images