New energy GCI dq impedance acquisition method based on BKA-BiLSTM model

Abstract

The invention relates to a new energy GCI dq impedance acquisition method based on a BKA-BiLSTM model, and the method comprises the following steps: firstly, carrying out the mechanism modeling of a new energy grid-connected inverter based on the Norton's theorem, and analyzing the nonlinear mapping relation between the equivalent dq impedance of the new energy GCI and the port characteristics; secondly, establishing a new energy unit grid-connected system simulation model by utilizing MATLAB / Simulink software, and acquiring a data set required by BiLSTM neural network training under various working conditions; secondly, training a BiLSTM neural network by utilizing simulation data, and optimizing hyper-parameters of the BiLSTM by adopting a black-wing optimization algorithm (BKA); and finally, an RT-LAB hardware-in-loop test technology is combined with a small perturbation method to scan impedance characteristics of the GCI, and an actual impedance value required by model verification is obtained. Through example simulation analysis, the correctness of the GCI dq impedance identification method provided by the invention is verified.

Description

Technical Field

[0001] The present invention relates to the field of photovoltaic power electronics technology, and in particular to a new energy GCI dq impedance acquisition method based on a BKA-BiLSTM model. Background Art

[0002] In recent years, the penetration rate of new energy sources such as wind power and photovoltaic power in the power grid has continued to increase, causing a series of broadband oscillation problems, resulting in a large number of new energy units being disconnected from the grid and equipment being damaged, which has seriously affected the reliable power supply of new energy and the safe and stable operation of the power system. Among them, the accurate acquisition of the impedance of the grid-connected inverter is one of the key prerequisites for system stability analysis. The commonly used impedance acquisition methods can be divided into two methods: white box mechanism modeling and black box measurement. White box mechanism modeling methods include state space averaging method, harmonic linearization method, harmonic state space analysis method, etc., but the methods used require that the internal information of the new energy grid-connected inverter (GCI) such as structure, parameters, and control mode is known. In actual engineering, due to the restrictions of commercial technology confidentiality, these key information is often unknown, so the above-mentioned white box modeling method is no longer applicable. In order to effectively deal with the above problems, the black-box modeling method can overcome the problem that the white-box modeling method needs to obtain the internal parameters of the GCI. The basic principle of this type of method is: connect a voltage disturbance source in series or a current disturbance source in parallel at the GCI grid-connected point, and the voltage / current disturbance generates a current / voltage response through the inverter circuit, collect the voltage and current data of the grid-connected inverter port, and calculate the corresponding impedance value. However, the calculation results of the black-box modeling method are relatively random and usually lack practical physical meaning. The effective combination of the two methods can avoid obtaining the internal parameter information of the GCI, and can also ensure that the results have practical physical meaning. At the same time, combined with the advantage of the strong nonlinear ability of the artificial intelligence algorithm to fit the input and output variables, the white-box modeling results are used as the output label, and the big data obtained by the black-box frequency sweep is used as the input feature. By training the artificial intelligence algorithm model, the actual impedance characteristics of the GCI are obtained.

[0003] According to the above background, a high-precision identification method of GCI dq impedance based on artificial intelligence algorithm is proposed, which has important practical significance for in-depth research and ensuring the stable operation of high-proportion new energy power systems. Summary of the invention

[0004] The purpose of the present invention is to overcome the shortcomings of the prior art and to effectively identify the GCI dq impedance. A new energy GCI dq impedance acquisition method based on the BKA-BiLSTM model is proposed. BiLSTM - bidirectional long short-term memory (bi-directional long short-term memory).

[0005] The technical solution adopted by the present invention is: a new energy GCI dq impedance acquisition method based on the BKA-BiLSTM model, comprising the following steps:

[0006] S1: Model the mechanism of new energy GCI based on Norton's theorem and analyze the equivalent dq impedance Z of new energy GCI dq Nonlinear mapping relationship with port characteristics;

[0007] S2: Establish the dq impedance identification model of the grid-connected inverter based on the BiLSTM neural network;

[0008] S3: Use BKA to optimize the hyperparameters of the BiLSTM neural network;

[0009] S4: Evaluate the recognition effect of the model.

[0010] Specifically, according to the Norton theorem, the S1 converts the GCI of the new energy generator set into a DC current source I o and impedance Z o In the parallel form, the voltage and current at the common coupling point of the grid connection are represented by U PCC and I2, the impedance of the power grid is represented by Z g It means that according to KVL, the frequency domain expressions of the port voltage and current in the dq coordinate system can be written as: I 2dq (s) represents the s-domain port current of GCI in the dq coordinate system, I odq (s) represents the magnitude of the s-domain DC current in the dq coordinate system, represents the s-domain equivalent admittance of GCI, U PCCdq (s) represents the voltage magnitude of the common coupling point s domain in the dq coordinate system;

[0011] According to the above formula, the output impedance Z o (s) and the d-axis grid connection point voltage U in the dq coordinate system PCCd , q-axis grid-connected point voltage U PCCq , d-axis grid-connected point current I 2d , q-axis grid-connected point current I 2q There is a complex nonlinear mapping relationship between the frequency and f, which can be expressed by the general function formula: Z o (s)=F(f,U PCCd ,U PCCq ,I 2d ,I 2q ), where F(·) represents the input variable f, U PCCd , U PCCq ,I 2d ,I 2q With the output variable Z o(s), given five input variables, the output variable Z can be obtained through F(·) o (s) Impedance elements on the four channels.

[0012] When the control structure and parameters of the GCI of the new energy unit are clearly given, the "white box" modeling can be directly used to obtain the accurate analytical expression of the nonlinear mapping relationship F. However, due to the inherent technical confidentiality of commercial new energy units and the unknown nature of the GCI control structure and parameters, it is impossible to directly establish the analytical expression of the mapping relationship F. To address this problem, a data-driven artificial intelligence algorithm model can be used to analyze the data features between the input and output data in the training data set, so that the nonlinear mapping relationship F can be established more accurately.

[0013] Specifically, the specific process of step S2 is as follows:

[0014] Step S21: Data acquisition. In order to accurately establish the BiLSTM identification model, the GCI dq impedance data is needed as the basis. Therefore, it is necessary to first build a simulation model in MATLAB / Simulink, change the grid connection point voltage and current by setting the active and reactive output of the new energy unit at different frequencies, and use the impedance scanning module to obtain the GCI dq impedance data set under various operating conditions.

[0015] Step S22: Data processing: In order to improve the training efficiency of the BiLSTM model, the acquired data is first normalized, usually scaling the data to the range of (0,1), and then divided into a training set and a test set at a ratio of 8:2 for the recognition model to be trained and then denormalized to make it have actual physical meaning. The normalization and denormalization calculation formulas are shown below:

[0016]

[0017] x i (k) = x i* (k)(x i,max -x i,min )+x i,min

[0018] In the formula, x i (k) is the original data of the kth sample of feature i; x i,max 、x i,min are the maximum and minimum values ​​in the sequence respectively; x i* (k) is the normalized value.

[0019] Step S23: Model construction: transform f, U PCCd , U PCCq ,I2d ,I 2q As the input variable of model training, GCI dq impedance is used as the output variable to construct the BiLSTM model for impedance identification.

[0020] Step S24: Identification verification: A semi-physical simulation platform is built based on RT-LAB software, and the dq impedance of GCI under different working conditions is tested by the small perturbation method, thereby forming a verification data set for model identification verification. The GCI impedance obtained by the small perturbation method can be expressed as:

[0021]

[0022] In the formula, subscripts 1 and 2 represent two disturbance injections during the test, U d1 , U d2 It represents the voltage of two disturbances on the d-axis in the dq coordinate system, U q1 , U q2 It represents the voltage of two disturbances on the q axis in the dq coordinate system, I d1 ,I d2 It represents the current of two disturbances on the d-axis in the dq coordinate system, I q1 ,I q2 Represents the current of two disturbances on the q axis in the dq coordinate system.

[0023] The small signal interference source injected during the experiment is a voltage signal. The disturbance voltage source is connected in series. Considering the dynamic influence of PLL, a single frequency signal is used as the injection amount of each disturbance, and the amplitude of the disturbance signal is ensured to be moderate, otherwise it will interfere with the steady-state test results of the system, or the impedance measurement accuracy may be low due to the unclear disturbance signal. Taking all factors into consideration, the voltage amplitude of the small signal interference source injected in the present invention is selected as 3% of the grid connection point voltage.

[0024] Specifically, in step S3, the hyperparameters of the BiLSTM neural network are optimized using BKA. BKA has unique biological inspiration features, which not only captures the flight and predation behavior of black kites in nature, but also deeply simulates their high adaptability to environmental changes and target locations. The imitation of this biological mechanism provides the algorithm with a robust dynamic search capability, enabling it to effectively cope with the ever-changing optimization environment. The specific process of the algorithm is as follows:

[0025] (1) Population initialization

[0026] Assume that the population of black-winged kites is n, and the dimension of the problem to be optimized is D, then all individuals can be represented by the following matrix:

[0027] n is a natural number greater than 1,

[0028] Randomly initialize the i-th individual:

[0029] X i =X lb +rand(X ub -X lb )

[0030] Where, X ub , X lb are the upper and lower boundaries of the solution space respectively; rand(·) is a random number generation function in the interval [0,1].

[0031] (2) Aggressive behavior

[0032] The black-winged kite is a predator of small mammals and insects in the grassland. During flight, it adjusts the angle of its wings and tail according to the wind speed, hovers quietly to observe its prey, and then quickly dives to attack. This strategy includes different attack behaviors for global exploration and search. The attack behavior of the black-winged kite can be expressed as:

[0033]

[0034] In the formula, They represent the position of the i-th black kite in the j-dimensional space at the t-th and t+1-th iterations respectively; r is a randomly generated parameter ranging from 0 to 1; p represents the probability of a black kite successfully attacking its prey, which is usually taken as 0.9; and T is the total number of iterations.

[0035] (3) Migration behavior

[0036] Migration is usually led by a leader, based on the assumption of bird migration: if the fitness value of the current population is less than the fitness value of the random population, then the leader will give up leadership and join the migrating population, indicating that it is not suitable to lead the population forward. On the contrary, if the fitness value of the current population is greater than the fitness value of the random population, it will guide the population to the destination. This strategy can dynamically select excellent leaders to ensure the success of migration. The mathematical model of black-winged kite migration behavior is:

[0037]

[0038] In the formula, represents the score of the black-winged kite leader in the j-dimensional space at the t-th iteration; F i represents the position of any black-winged kite in the j-dimensional space at the tth iteration; F ri represents the fitness value of any black kite at a random position in the j-dimensional solution space at t iterations; C(0,1) represents the Cauchy mutation operator. The Cauchy mutation strategy is introduced in BKA, which helps the algorithm to jump out of the local optimal solution and increase the probability of finding a better solution in the global search space. The one-dimensional Cauchy distribution is a continuous probability distribution with two parameters, and its probability density function can be expressed as:

[0039]

[0040] In the formula, μ and σ are the mean and variance respectively, and x represents the independent variable of the probability density function.

[0041] Then, the best hyperparameters of the BiLSTM neural network are optimized through BKA to avoid the impact of poor hyperparameter settings on the experiment and further improve the accuracy of the experiment and the generalization ability of the model. The specific steps are:

[0042] Step S31, setting BKA algorithm parameters, including population number n, maximum number of iterations T, and problem dimension D to be optimized;

[0043] Step S32, determine the hyperparameters to be optimized for the BiLSTM model: the number of nodes in the two hidden layers, the number of training times, and the learning rate;

[0044] Step S33, setting the range of hyperparameter optimization: the optimization range of the number of nodes H1 and H2 in the two hidden layers is [50, 200]; the optimization range of the number of training times N is [50, 500]; the optimization range of the learning rate l is [0.001, 0.1];

[0045] Step S34, initializing the population distribution of the algorithm;

[0046] Step S35, using the mean square error between the actual value and the identified value as the fitness function to calculate the fitness function value of the BKA population individual;

[0047] Step S36, determine the initial population position and fitness value, and determine whether it is the global optimum;

[0048] Step S37, randomly generate a parameter r, and compare it with the probability p of a black-winged kite attacking its prey successfully to update the position and fitness value of the population;

[0049] Step S38, determine whether the fitness value of the current population is less than the fitness value of the random population. If so, the leader will give up leadership and join the migrating population, which indicates that it is not suitable to lead the population forward. On the contrary, if the fitness value of the current population is greater than the fitness value of the random population, the leader will continue to guide the population to the destination;

[0050] Step S39, determine whether the algorithm has reached the maximum number of iterations. If so, stop the iteration and output the optimal parameters of the two hidden layer nodes, training times and learning rate of BiLSTM to obtain the BKA-BiLSTM recognition model; otherwise, return to step (7) until the maximum number of iterations is met.

[0051] Specifically, in step S4, the mean absolute percentage error (MAPE ), root mean square error (R MSE ) is used to evaluate the recognition effect of the model. The specific calculation formula is:

[0052]

[0053] Where y(k), Y(k) and are the identified value and true value of the sample respectively; m is the number of data samples.

[0054] Compared with the prior art, the present invention has the following beneficial effects:

[0055] 1. The present invention adopts Norton's theorem to model the mechanism of new energy GCI, and can establish a nonlinear mapping relationship between GCI dq impedance and port characteristics, and then can use artificial intelligence algorithm to effectively fit this relationship;

[0056] 2. The present invention adopts BiLSTM neural network to identify GCI dq impedance, which can fit the nonlinear mapping relationship between input and output variables, and then adopts BKA to optimize the hyperparameters of BiLSTM, which can avoid the problems of strong subjectivity, time-consuming and labor-intensive manual tuning. After parameter optimization, the accuracy of impedance identification is significantly improved. Compared with the traditional swarm intelligence algorithm, the Cauchy mutation strategy is introduced in BKA, which helps the algorithm to jump out of the local optimal solution and increase the probability of finding a better solution in the global search space. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 This is a new energy GCI dq impedance acquisition method process based on the BKA-BiLSTM model in the present invention;

[0058] Figure 2 The main circuit structure of the LCL type GCI in the present invention;

[0059] Figure 3 is the Norton equivalent circuit of the GCI in the present invention;

[0060] Figure 4 is the iteration curve of BKA in the present invention;

[0061] Figure 5 is the GCI dq impedance identification error under 33 working conditions in the embodiment of the present invention;

[0062] Figure 6 1 is the identification of the amplitude and phase angle curve of the GCI dq impedance Bode diagram in the embodiment of the present invention. DETAILED DESCRIPTION

[0063] The effectiveness of the proposed method will be verified by using an embodiment below, and the technical solution in the embodiment of the present invention will be clearly and completely described in conjunction with the accompanying drawings. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0064] The present invention provides a new energy GCI dq impedance acquisition method based on the BKA-BiLSTM model, comprising the following steps:

[0065] S1: Model the mechanism of new energy GCI based on Norton's theorem and analyze the equivalent dq impedance Z of new energy GCI dq Nonlinear mapping relationship with port characteristics;

[0066] S2: Establish the dq impedance identification model of the grid-connected inverter based on the BiLSTM neural network;

[0067] S3: Use BKA to optimize the hyperparameters of the BiLSTM neural network;

[0068] S4: Evaluate the recognition effect of the model.

[0069] The new energy GCI dq identification process based on the BKA-BiLSTM model is shown in the attached figure. Figure 1 shown.

[0070] Attached Figure 2 is the main circuit topology of the new energy GCI. In the figure, the grid side voltage is recorded as u ga 、u gb 、u gc ;i 1a 、i 1b and i 1c is the filter inductor current; i dc is the DC side current; i 2a 、i 2b 、i 2c is the three-phase current on the grid side; C dc is the DC bus capacitance; u dc is the DC bus capacitor voltage; L f , C f and R f They are filter inductor, filter capacitor and damping resistor respectively; R g , L g are the resistance and inductance on the grid side; Q1~Q6 are switching devices; PCC is the common coupling point, that is, the connection point between the grid and the filter output port.

[0071] In order to obtain the sample data required for training the BKA-BiLSTM model, the present invention takes the doubly fed wind turbine grid-connected system as an example, uses MATLAB / Simulink software to build its typical structure and control simulation model, and obtains its GCI dq impedance data through the impedance scanning module, and then normalizes the acquired data to better serve the model for impedance identification. The active output step size of the doubly fed wind turbine is set to 10%, the output variation range is [0, 100%], and the reactive output includes three working conditions: maximum capacitive reactive output, reactive power of 0, and maximum inductive reactive output. Combined with different active and reactive output conditions, a total of 33 working conditions can be obtained. Under each working condition, the dq axis voltage and current data at the grid-connected point are collected and recorded, and the dq impedance in the range of 1Hz to 1000Hz is measured.

[0072] The BKA-BiLSTM model was trained and tested using the above sample data. The training set and test set were divided into 8:2, and the population and iteration times of the BKA algorithm were set to 50 and 100 respectively, and the dimension of the problem was 4. The optimal hyperparameters of BiLSTM obtained by BKA optimization are: H1=65, H2=50, l=0.04, N=221. The BKA optimization iteration curve is shown in the attached figure. Figure 3 As shown in the figure, when the number of iterations reaches about 40 generations, the fitness drops to the lowest point. As the number of iterations increases, the iteration curve does not show obvious changes, which means that the algorithm has reached the global optimum. The output particle position is the optimal hyperparameter of BiLSTM. It also shows the rationality of the BKA parameter setting. By optimizing the hyperparameters of BiLSTM through BKA, it can achieve better recognition effect.

[0073] In order to verify the effectiveness of the BKA-BiLSTM model, a semi-physical simulation platform for the doubly fed wind turbine grid-connected system was constructed using the hardware-in-the-loop test technology through the RT-LAB software. The GCI dq impedance was then scanned and obtained based on the small perturbation method to obtain the actual impedance value required for model effectiveness verification. The trained BKA-BiLSTM model was used to fit the GCI dq impedance under 33 test conditions, and the error between the measured value and the identified value under 33 conditions was calculated. The calculation results are shown in the attached figure. Figure 4 As shown, the Bode diagram corresponding to the impedance is then drawn. Due to the large number of working conditions, the present invention only selects a group of working conditions for impedance acquisition results to display. The working condition corresponds to an active output of 50% and a reactive power of the maximum capacitive reactive power. The comparison between the measured value and the identified value is shown in the attached figure. Figure 5 As shown in the figure, the amplitude error R MSE is 0.125dB, the phase angle error R MSE is 3.433°, and the amplitude error is M APE is 3.2%, the phase angle error MAPE It is 6.4%.

[0074] By the attached Figure 4 It can be seen that under 33 groups of working conditions, the GCI impedance values ​​obtained by the proposed method are basically consistent with the impedance values ​​actually measured by the small perturbation method, and the amplitude error does not exceed 0.6dB, and the phase angle error does not exceed 4°. These errors are generally small, which effectively verifies the accuracy and effectiveness of the proposed method. It also shows that the sample data required for model training under various working conditions obtained by MATLAB / Simulink simulation software is reliable, and the proposed method can provide more accurate results for the impedance characteristic identification of GCI under various operating conditions. Figure 5 It can be seen intuitively that the curve fitting between the identification value and the measured value under this group of working conditions is good, and only a certain identification deviation occurs at the position with a large fluctuation amplitude, but the overall error is small. At the same time, by comparing the BiLSTM identification effect before and after hyperparameter optimization, the BKA-BiLSTM has a better identification effect, which verifies the effectiveness of the proposed method.

[0075] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the present invention can be implemented in other specific forms without departing from the spirit or essential features of the present invention. Therefore, the embodiments should be considered exemplary and non-restrictive in all respects, and the scope of the present invention is defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims be included in the present invention.

Claims

1. A new energy GCI dq impedance acquisition method based on BKA-BiLSTM model, characterized in that: The following steps are involved: S1: Based on the Norton theorem, the mechanism modeling of new energy GCI is carried out, and the nonlinear mapping relationship between the equivalent dq impedance Zdq of new energy GCI and the port characteristics is analyzed; S2: Establish the dq impedance identification model of the grid-connected inverter based on the BiLSTM neural network; S3: Use BKA to optimize the hyperparameters of the BiLSTM neural network; S4: Evaluate the recognition effect of the model.

2. According to a method for acquiring new energy GCI dq impedance based on BKA-BiLSTM model according to claim 1, it is characterized in that: According to step S1, according to Norton's theorem, the GCI of the new energy unit is equivalent to a DC current source I o and impedance Z o In the parallel form, the voltage and current at the common coupling point of the grid connection are represented by U PCC and I2, the impedance of the power grid is represented by Z g According to KVL, the frequency domain expressions of the port voltage and current in the dq coordinate system are written as: I 2dq (s) represents the s-domain port current of GCI in the dq coordinate system, I odq (s) represents the magnitude of the s-domain DC current in the dq coordinate system, represents the s-domain equivalent admittance of GCI, U PCCdq (s) represents the voltage magnitude of the common coupling point s domain in the dq coordinate system; According to the above formula, the output impedance Z o (s) and the d-axis grid connection point voltage U in the dq coordinate system PCCd , q-axis grid-connected point voltage U PCCq , d-axis grid-connected point current I 2d , q-axis grid-connected point current I 2q There is a complex nonlinear mapping relationship between the frequency and f, which can be expressed as: o (s)=F(f,U PCCd ,U PCCq ,I 2d ,I 2q ), where F(·) represents the input variable f, U PCCd , U PCCq ,I 2d ,I 2q With the output variable Z o (s) is a nonlinear mapping function between five input variables. After given five input variables, the output variable Z is obtained through F(·) o (s) Impedance elements on the four channels.

3. According to a method for acquiring new energy GCI dq impedance based on BKA-BiLSTM model according to claim 1, it is characterized in that: The specific process of step S2 is as follows: Step S21: data acquisition, building a simulation model in MATLAB / Simulink, changing the voltage and current at the grid connection point by setting the active and reactive output of the new energy unit at different frequencies, and using the impedance scanning module to obtain the dq impedance data set of the GCI under various operating conditions; Step S22: Data processing, normalizing the acquired data, usually scaling the data to the range of (0,1), and then dividing the data into a training set and a test set at a ratio of 8:2 for the identification model to be trained and then denormalizing the data to make it have actual physical meaning. The normalization and denormalization calculation formulas are shown as follows: x i (k) = x i* (k)(x i,max -x i,min )+x i,min , where x i (k) is the original data of the kth sample of feature i; x i,max 、x i,min are the maximum and minimum values ​​in the sequence respectively; x i* (k) is the normalized value; Step S23: Model construction, f, U PCCd , U PCCq ,I 2d ,I 2q As the input variable for model training, the GCI dq impedance is used as the output variable to construct the BiLSTM model for impedance identification; Step S24: Identification verification: a semi-physical simulation platform is built based on RT-LAB software, and the dq impedance of GCI under different working conditions is tested by the small perturbation method, thereby forming a verification data set for model identification verification. The GCI impedance obtained by the small perturbation method is expressed as: In the formula, subscripts 1 and 2 represent two disturbance injections during the test, U d1 , U d2 It represents the voltage of two disturbances on the d-axis in the dq coordinate system, U q1 , U q2 It represents the voltage of two disturbances on the q axis in the dq coordinate system, I d1 ,I d2 It represents the current of two disturbances on the d-axis in the dq coordinate system, I q1 ,I q2 It represents the current of two disturbances of q axis in dq coordinate system; The small signal interference source injected during the experiment is a voltage signal, and the disturbance voltage source is connected in series.

4. According to a method for acquiring new energy GCI dq impedance based on BKA-BiLSTM model according to claim 3, it is characterized in that: The voltage amplitude of the injected small signal interference source is selected to be 3% of the grid connection point voltage.

5. According to a method for acquiring new energy GCI dq impedance based on BKA-BiLSTM model according to claim 1, it is characterized in that: The step S3 comprises the following specific steps: Step S31, setting BKA algorithm parameters, including population number n, maximum number of iterations T, and problem dimension D to be optimized; Step S32, determine the hyperparameters to be optimized for the BiLSTM model: the number of nodes in the two hidden layers, the number of training times, and the learning rate; Step S33, setting the range of hyperparameter optimization: the optimization range of the number of nodes H1 and H2 in the two hidden layers is [50, 200]; the optimization range of the number of training times N is [50, 500]; the optimization range of the learning rate l is [0.001, 0.1]; Step S34, initializing the population distribution of the algorithm; Step S35, using the mean square error between the actual value and the identified value as the fitness function to calculate the fitness function value of the BKA population individual; Step S36, determine the initial population position and fitness value, and determine whether it is the global optimum; Step S37, randomly generate a parameter r, and compare it with the probability p of a black-winged kite attacking its prey successfully to update the position and fitness value of the population; Step S38, determine whether the fitness value of the current population is less than the fitness value of the random population. If so, the leader will give up leadership and join the migrating population, which indicates that it is not suitable to lead the population forward. On the contrary, if the fitness value of the current population is greater than the fitness value of the random population, the leader will continue to guide the population to the destination; Step S39, determine whether the algorithm has reached the maximum number of iterations, if yes, then stop the iteration, output the optimal parameters of the two hidden layer nodes, training times and learning rate of BiLSTM, and obtain the BKA-BiLSTM identification model; Otherwise, return to step S37 until the maximum number of iterations is met.

6. The method for acquiring the dq impedance of a new energy GCI based on the BKA-BiLSTM model according to claim 1, characterized in that: In step S4, the mean absolute percentage error M is used. APE , Root mean square error R MSE The identification effect of the model is evaluated, and the specific calculation formula is: Where y(k), Y(k) and are the identified value and true value of the sample respectively; m is the number of data samples.

Images