Optimization method of grid-connected short-circuit current parameters of grid-connected photovoltaic energy storage system based on Kriging surrogate model and improved particle swarm algorithm

Abstract

A grid-connected short-circuit current parameter optimization method for a grid-connected photovoltaic energy storage system based on a Kriging proxy model and an improved particle swarm algorithm is used to construct a short-circuit current simulation model for a grid-connected photovoltaic energy storage system; the control parameters are screened through local sensitivity analysis to select the control parameters that have a greater impact on short-circuit current suppression; the relationship between the fitted control parameters and the short-circuit current response is predicted using the Kriging proxy model; the short-circuit current control parameters fitted by the Kriging proxy model are globally optimized through the improved particle swarm algorithm to obtain the optimized optimal short-circuit current control parameter combination. After the optimization is completed, a global sensitivity analysis is performed to further select the key control parameters that have the most significant effect on short-circuit current suppression. The present invention identifies the most critical control parameters for short-circuit current suppression by analyzing a variety of control parameters and VSG parameters of power electronic equipment, and optimizes these parameters to ensure stable operation of the system during power grid failures.

Description

Technical Field

[0001] The present invention relates to the technical field of grid-connected control of a grid-connected photovoltaic energy storage system, and in particular to a method for optimizing grid-connected short-circuit current parameters of a grid-connected photovoltaic energy storage system based on a Kriging proxy model and an improved particle swarm algorithm. Background Art

[0002] New energy will become an important part of my country's power structure, and the proportion of renewable energy such as photovoltaics, wind power, and energy storage will increase significantly. With the large-scale grid connection of photovoltaic and energy storage systems, how to ensure the stability and security of the power system has become one of the urgent issues to be solved. In the context of a large number of new energy sources connected to the power grid, existing research still has the following deficiencies in short-circuit current control and parameter optimization during the grid connection of photovoltaic energy storage systems:

[0003] 1) Insufficient coordinated optimization of multiple control parameters: Most current research on grid-connected photovoltaic energy storage systems focuses on the optimization of single control parameters, such as reactive current limiting, virtual damping coefficient, virtual moment of inertia, etc. However, in complex grid fault scenarios, multiple control parameters often need to be coordinated and optimized to ensure that the system can respond quickly and stably when a fault occurs. However, existing research on the interaction of multiple parameters is relatively weak, making it difficult to effectively improve the robustness and dynamic stability of the system in different fault scenarios.

[0004] 2) Lack of accurate methods for identifying control parameters: The control parameters of the photovoltaic energy storage system, such as the inertia parameters and damping coefficient of the virtual synchronous generator, have an important impact on the response of the grid short-circuit current. However, there is currently a lack of a method that can accurately identify and optimize these key control parameters, especially in complex scenarios where multiple parameters are coupled with each other. Existing methods are difficult to provide accurate parameter adjustment mechanisms, resulting in the system being unable to effectively suppress the rise or suppression of short-circuit current in the event of a sudden fault. Summary of the invention

[0005] The present invention provides a method for optimizing the grid-connected short-circuit current parameters of a grid-connected photovoltaic energy storage system based on a Kriging proxy model and an improved particle swarm algorithm. The method analyzes multiple control parameters of power electronic equipment and VSG parameters, identifies key control parameters for short-circuit current, and optimizes these parameters to ensure stable operation of the system during a grid fault.

[0006] The technical solution adopted by the present invention is:

[0007] The method for optimizing the short-circuit current parameters of a grid-connected photovoltaic energy storage system based on the Kriging agent model and the improved particle swarm algorithm includes the following steps:

[0008] Step 1: Construct a grid-connected photovoltaic energy storage system;

[0009] Step 2: Screen the control parameters through local sensitivity analysis to select the control parameters that have a greater impact on short-circuit current suppression;

[0010] Step 3: Use the Kriging surrogate model to predict the relationship between the fitted control parameters and the short-circuit current response;

[0011] Step 4: Globally optimize the short-circuit current control parameters fitted by the Kriging proxy model through the improved particle swarm algorithm to obtain the optimized optimal short-circuit current control parameter combination.

[0012] In the step 1, a grid-type photovoltaic energy storage system is constructed in the Matlab / Simulink platform, which includes a photovoltaic module, an energy storage module, a main circuit, a virtual synchronous generator (VSG) control module, an observation module, an SPWM modulation module, and a dq axis conversion module; Figure 1 As shown:

[0013] The photovoltaic module, after the voltage is increased by a DC-DC boost circuit, is connected to the virtual synchronous generator; the electricity generated by the photovoltaic module first passes through the battery part of the energy storage module, part of the electricity is used for storage, and the rest is converted into AC power by the inverter and then connected to the grid;

[0014] The output of the energy storage module is directly connected to the DC input of the inverter; the energy storage module stores electrical energy and releases it when needed. The energy storage module is usually composed of a battery system, which is used to regulate the load of the power grid and help balance the supply and demand of electricity, especially in the case of unstable photovoltaic power generation, to stabilize the DC bus voltage in the system;

[0015] The main circuit is used for the conversion and transmission of electric energy. The main circuit consists of an inverter, a filter, a line resistor, and a line connected to the grid. After the inverter converts the DC power from the photovoltaic module and the energy storage module into AC power, it first passes through a filter to remove harmonics to ensure the quality of the output current and voltage waveforms; then, the current passes through the line impedance to simulate the transmission loss in the actual system; finally, the current enters the grid to achieve grid connection;

[0016] The virtual synchronous generator (VSG) control module is used to measure the voltage, current and frequency of the system and adjust the system in real time. The virtual synchronous generator (VSG) control module is connected to the inverter and is responsible for controlling the output frequency and voltage of the inverter to synchronize it with the power grid. At the same time, the virtual synchronous generator (VSG) control module also provides inertia and damping characteristics for the system. When a fault or disturbance occurs, the VSG can provide reactive and active support similar to that of a traditional generator. In Matlab / Simulink, the virtual synchronous generator (VSG) control module is connected to the inverter through a signal line to adjust its output.

[0017] The observation module is used to monitor the operating status of the system, including signals such as voltage, current and frequency. The observation module is directly connected to the main circuit and inverter, collects the operating data of the system in real time, and displays it through charts or numerical values. In Simulink, the observation module is usually connected to each subsystem through the sensor module, and records and displays key variables;

[0018] The SPWM modulation module is used to generate pulse width modulation signals to directly control the power switch tube of the inverter, thereby adjusting its output voltage and frequency; the SPWM modulation module receives signals from the virtual synchronous generator (VSG) control module, such as voltage and frequency reference values, and converts them into corresponding modulation signals. In Matlab / Simulink, the SPWM module output signal is connected to the switch module of the inverter to control its dynamic response;

[0019] The dq-axis transformation module is used to convert the three-phase AC power of the power grid into a DC equivalent dq-axis representation to simplify control. The dq-axis transformation module receives the three-phase current and voltage signals of the main circuit and transforms them into DC signals under dq coordinates, which are provided to the virtual synchronous generator (VSG) control module. At the same time, it also converts the dq-axis control signal output by the virtual synchronous generator (VSG) control module back into a three-phase AC signal and transmits it to the inverter. In Matlab / Simulink, the dq-axis transformation module is connected to the VSG and the main circuit through signal lines.

[0020] In step 2, the control parameters that have a greater impact on the short-circuit current response are preliminarily screened out through local sensitivity analysis to reduce the number of parameters that need to be considered during optimization, quickly locate key parameters, and improve efficiency. Specifically, the following steps are included:

[0021] S2.1. Define initial conditions:

[0022] First, determine the initial values ​​of all control parameters, including:

[0023] a. Reactive current limit value: ;

[0024] b. Virtual damping coefficient (VSG parameter): ;

[0025] c. Virtual moment of inertia (VSG parameter): ;

[0026] d. Current limiting setting: ;

[0027] The system output is the short-circuit current response, such as the peak value or recovery time of the short-circuit current. Indicates that is the vector of control parameters.

[0028] S2.2. Change a single control parameter:

[0029] Keep other control parameters unchanged and make only a small change to one control parameter. For example, suppose the reactive current limit value is currently adjusted. , so that it increases by a small amount , becomes ;

[0030] S2.3. Calculate the changes in system output:

[0031] Calculate the adjusted system , , Indicates other control parameters in the grid-connected photovoltaic energy storage system. For example: virtual damping coefficient (VSG parameter), virtual moment of inertia (VSG parameter), etc.

[0032] and with the original output Make a comparison. This comparison is done using the following formula:

[0033] (1);

[0034] S2.4. Calculate the local sensitivity coefficient:

[0035] Local sensitivity coefficient Quantization control parameters The impact on system output can be calculated by the following formula:

[0036] (2);

[0037] in, Indicates a small change in the system output; Indicates control parameters Small changes in Indicates that the system is changing the reactive current limit The output value after Indicates that the system is reactive current limiting before change The output value of is the response of the system in the initial state;

[0038] S2.5. Repeat the sensitivity analysis for other control parameters:

[0039] Repeat the above steps to adjust other control parameters one by one, such as adjusting the virtual moment of inertia etc., calculate each parameter to get the sensitivity;

[0040] S2.6, Analysis and Sorting:

[0041] The calculated sensitivity coefficient The control parameters with larger sensitivity coefficients have a greater impact on the system output. In subsequent optimization, priority is given to adjusting and optimizing these parameters with greater sensitivity.

[0042] In step 3, the Kriging proxy model is used to fit the nonlinear relationship between the control parameters and the short-circuit current response. It is a statistical method based on sample data, which can make efficient predictions based on a small number of sample points. The Kriging proxy model provides an approximate model for the optimization algorithm by establishing the relationship between input parameters, including reactive current limiting, current limiting constant, virtual moment of inertia, etc., and output response: short-circuit current peak or recovery time, thereby reducing the burden of simulation calculations. The Kriging proxy model is as follows:

[0043] According to the required sample size, within the range of the optimized variable, samples are collected evenly as the sample data required for training samples.

[0044] 3.1、Sample data collection:

[0045] Before building the Kriging proxy model, it is necessary to collect a small amount of sample data through simulation or experiment. sample points, each sample contains a control parameter vector , Respectively expressed in The control parameter vector of sample points; Indicates The first control parameter of the sample point, Indicates The second control parameter of the sample point; and the corresponding short-circuit current response .

[0046] Sample points: , Represents the combination of different control parameters, each sample point Represents the specific values ​​of a set of control parameters in the system, such as reactive current limit, virtual damping coefficient, etc.;

[0047] Response point: , Respectively represent the control parameters in each group The output response of the system under these parameter combinations. Specifically, it is the short-circuit current peak value or recovery time of the system under these parameter combinations;

[0048] These data points will be used in the Kriging surrogate model.

[0049] 3.2 Basic formula of Kriging proxy model:

[0050] The Kriging surrogate model is constructed by parameter Predicting Response , its basic formula is:

[0051] (3);

[0052] Where: is the predicted value; is a constant or trend function, used to represent the global trend; It is a local deviation based on the covariance and is used to capture the impact of local changes in the control parameters on the short-circuit current response.

[0053] 3.3. Definition of covariance function:

[0054] In order to describe the correlation between control parameters, the Kriging surrogate model uses a covariance function. The common form of the covariance function is the Gaussian kernel function:

[0055] (4);

[0056] Where: Represents sample points , Covariance in Kriging models; Indicates that the control parameters The response value of the system below; Indicates that the control parameters The response value of the system below; is the variance of the model, representing the noise between data points; is the correlation length size parameter, indicating the control parameter the extent to which changes in the response affect the response; is the sample point and In the parameters The distance on Indicates Among the sample points The control parameter value, i.e. Quantity; Indicates Among the sample points The control parameter value, i.e. Quantity; Indicates the total dimension of the control parameters, that is, how many control parameters are there in the system; Represents the power exponent, which indicates the nonlinearity of the distance metric. In the covariance function, Determines the degree of influence of changes in control parameters on covariance;

[0057] The covariance function represents the correlation between adjacent sample points, that is, if two sample points are close in the control parameter space, their short-circuit current responses should also be similar.

[0058] 3.4. Training of Kriging surrogate model:

[0059] The training process of the Kriging surrogate model is the fitting process, which requires estimating the parameters in the Kriging surrogate model. , , Etc. The commonly used estimation method is the maximum likelihood estimation (MLE), whose goal is to find the optimal model parameters by maximizing the likelihood function.

[0060] Likelihood function is defined as:

[0061] (5);

[0062] Where: is the actual observed response value vector; is the covariance matrix composed of covariance functions; Indicates transposing the matrix; is the global trend term; and are the model parameters to be estimated ;

[0063] 3.5. Predict the short-circuit current response of the new parameter combination system:

[0064] After training is completed, the Kriging agent model can be used according to the new control parameters Make a prediction. The predicted output response is:

[0065] (6);

[0066] Where: For new samples The predicted response value of It's new The covariance vector with the training dataset;

[0067] Represents the global trend term; Indicates that the matrix is ​​transposed; is the covariance matrix of the training dataset; is the response value of the training data set.

[0068] The Kriging proxy model is used to predict the relationship between the fitting control parameters and the short-circuit current response, and a short-circuit current response prediction model is constructed. The short-circuit current response prediction model includes the Kriging proxy model. The model is trained by historical data to predict the response of the short-circuit current under different control parameters. The formula involved includes formula (6).

[0069] In step 4, the improved particle swarm algorithm is specifically a particle swarm optimization algorithm based on the credibility of the three-black hole system particle capture strategy, which is mainly used to optimize the control parameter combination predicted by the Kriging agent model to find the optimal short-circuit current suppression solution. Specifically, the following steps are included:

[0070] S4.1: Random Black Hole Particle Swarm Optimization (RBHPSO):

[0071] The particle space vector under RBHPSO can be expressed as:

[0072] (7);

[0073] Where: represents the probability of a particle being captured by a black hole. When the particle escapes from the black hole, When , the particle is captured by the black hole; is the current iteration number; and Represents particles No. Vidi The speed and position of the generation; and They represent the individual optimal solution of the particle and the global optimal solution respectively; is the inertia weight; and is the learning coefficient; , , and is a random number uniformly distributed between [0,1]; Indicates The particle in The first The location of the dimension; express The particle in The first The speed of dimension.

[0074] S4.2, introduce credibility factor:

[0075] Since the global optimal particle of RBHPSO in the early stage of iteration is less likely to be the true optimal particle, and as the algorithm iterates, the particles will move towards the true optimal solution, and the possibility of the global optimal particle being generated in it will continue to increase, so the credibility factor q is introduced:

[0076] (8);

[0077] Where: Gen is the total number of iterations; is a random number uniformly distributed between [0,1].

[0078] S4.3. Introduce the Three Black Hole System Capture Strategy (TCS):

[0079] In the early stages of the iteration, particles will be captured by random black holes, and the mass of the black holes will increase, increasing their gravitational force and the radius of the black holes. As the iterations proceed, the mass of the black holes close to the global optimal solution continues to increase until the remaining black holes are destroyed. During the iteration process, the black hole will not only attract the remaining particles, but its own position will also change between the midpoint between the global optimal particle and the upper bound, the midpoint between the global optimal particle and the lower bound, and a point near the global optimal.

[0080] (9);

[0081] Where: is the probability of updating the black hole position; Represents the probability value, which is used to determine the change in the black hole position; Indicates the upper limit of the black hole's position; Represents the lower bound of the black hole's position. The black hole's own position will vary between the midpoint between the global optimal particle and the upper bound, and between the midpoint between the global optimal particle and the lower bound. ∈[0,1]; is a random number uniformly distributed between [0,1].

[0082] The present invention provides a method for optimizing key parameters of short-circuit current of a grid-connected photovoltaic energy storage system based on a Kriging proxy model and an improved particle swarm algorithm, and the technical effects are as follows:

[0083] 1) The present invention uses the Kriging proxy model to fit the nonlinear relationship between the control parameters and the short-circuit current response, and can make high-precision predictions based on a small amount of simulation data. This greatly reduces the reliance on large-scale simulations and solves the problem of inaccurate prediction of short-circuit current response in the prior art.

[0084] 2) The present invention introduces an improved particle swarm algorithm to dynamically adjust the particle positions, thereby ensuring that the global optimal solution is found in the high-dimensional control parameter space, thereby improving the efficiency and reliability of short-circuit current control parameter optimization.

[0085] 3) The present invention significantly improves the dynamic response capability of the photovoltaic energy storage system when it is connected to the grid by globally optimizing the control parameters, especially effectively suppressing the short-circuit current under fault conditions. When a grid fault occurs, the system can recover quickly and stably, ensuring the overall stability of the grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0086] The present invention will be further described below in conjunction with the accompanying drawings and embodiments:

[0087] Figure 1 This is a schematic diagram of the grid-connected structure of the grid-connected photovoltaic energy storage system of the present invention;

[0088] Figure 2 It is a relationship diagram between the Kriging model prediction control parameters and the short-circuit current response of the present invention;

[0089] Figure 3 It is a schematic diagram of updating the three-black hole particle swarm algorithm based on credibility of the present invention;

[0090] Figure 4 It is a flow chart of the Kriging proxy model and improved particle swarm optimization algorithm according to the present invention;

[0091] Figure 5 It is a Pareto frontier solution set diagram based on the Kriging proxy model and the improved particle swarm optimization algorithm of the present invention;

[0092] Figure 6 It is a comparison chart of short-circuit current response curves before and after optimization. DETAILED DESCRIPTION

[0093] A method for optimizing key parameters of short-circuit current of grid-connected photovoltaic energy storage systems based on Kriging proxy model and improved particle swarm algorithm. First, the control parameters and short-circuit current response data of the system are collected through the simulation model, and the parameters with greater influence are screened out using local sensitivity analysis to reduce the complexity of the model; then, the Kriging model is used to efficiently fit the nonlinear relationship between the control parameters and the short-circuit current response to reduce the dependence on large-scale simulation; then, the Kriging model is globally optimized using the credibility-based three-black hole particle swarm optimization algorithm to ensure that the globally optimal control parameter combination is found; finally, the optimization effect is verified through simulation, which improves the short-circuit current suppression capability and response speed of the photovoltaic energy storage system when it is connected to the grid. Specifically, the following steps are included:

[0094] Step 1: Construct a short-circuit current simulation model of the grid-connected photovoltaic energy storage system based on the Matlab / Simulink platform;

[0095] Step 2: Screen the control parameters through local sensitivity analysis to select the control parameters that have a greater impact on short-circuit current suppression;

[0096] Step 3: Use the Kriging proxy model to predict the relationship between the fitting control parameters, including reactive current limit value, current limit setting value, virtual synchronous generator parameters, and short-circuit current response, and build a short-circuit current response prediction model;

[0097] Step 4: Globally optimize the short-circuit current control parameters fitted by the Kriging model through the improved particle swarm algorithm to obtain the optimized optimal short-circuit current control parameter combination;

[0098] Step 5: After the optimization is completed, analyze the short-circuit current level before and after the optimization.

[0099] The present invention analyzes various control parameters of power electronic equipment and VSG parameters, identifies the most critical control parameters for short-circuit current suppression, and optimizes these parameters to ensure stable operation of the system during power grid faults.

[0100] (1): Construct a short-circuit current simulation model for a grid-type photovoltaic energy storage system:

[0101] The present invention builds a grid-connected photovoltaic energy storage system in the Matlab / Simulink platform. Figure 1 As shown, it includes a photovoltaic module, an energy storage module, a main circuit part, a virtual synchronous generator (VSG) control module, an observation module, an SPWM modulation module and a dq axis conversion module.

[0102] (2): Local sensitivity analysis:

[0103] The present invention preliminarily screens out the control parameters that have a greater impact on the short-circuit current response through local sensitivity analysis, so as to reduce the number of parameters that need to be considered during optimization, quickly locate key parameters, and improve efficiency. The specific steps are as follows:

[0104] 2.1. Define initial conditions:

[0105] First, determine the initial values ​​of all control parameters. For example:

[0106] a. Reactive current limit value: ;

[0107] b. Virtual damping coefficient (VSG parameter): ;

[0108] c. Virtual moment of inertia (VSG parameter): ;

[0109] d. Current limiting setting: ;

[0110] The system output is the short-circuit current response, such as the peak value or recovery time of the short-circuit current. Indicates that is the vector of control parameters.

[0111] 2.2. Change a single control parameter:

[0112] Keep other control parameters unchanged and make only a small change to one control parameter. For example, suppose the reactive current limit value is currently adjusted. , so that it increases by a small amount , becomes

[0113] 2.3. Changes in calculation system output:

[0114] Calculate the adjusted system , and with the original output Make a comparison. This comparison is done using the following formula:

[0115] (1);

[0116] 2.4. Calculate the local sensitivity coefficient:

[0117] Local sensitivity coefficient Quantization control parameters The impact on system output can be calculated by the following formula:

[0118] (2);

[0119] 2.5. Repeat the sensitivity analysis of other control parameters:

[0120] Repeat the above steps to adjust other control parameters one by one, such as adjusting the virtual damping coefficient Etc., calculate each parameter to get the sensitivity.

[0121] 2.6. Analysis and sorting:

[0122] The calculated sensitivity coefficient The control parameters with larger sensitivity coefficients have a greater impact on the system output. In subsequent optimization, priority is given to adjusting and optimizing these parameters with greater sensitivity.

[0123] (3): Krging agent model construction:

[0124] The Kriging proxy model is used to fit the nonlinear relationship between the control parameters and the short-circuit current response. It is a statistical method based on sample data that can make efficient predictions based on a small number of sample points. The Kriging model provides an approximate model for the optimization algorithm by establishing the relationship between input parameters such as reactive current limiting, current limiting constant, virtual moment of inertia, etc. and output responses such as short-circuit current peak or recovery time, thereby reducing the burden of simulation calculations. The steps of the Kriging proxy model are as follows:

[0125] According to the required sample size, samples are collected evenly within the range of the optimized variable as the training samples.

[0126] 3.1: Collection of sample data:

[0127] Before building the Kriging model, a small amount of sample data needs to be collected through simulation or experiment. sample points, each sample contains a control parameter vector and the corresponding short-circuit current response These data points will be used in the Kriging surrogate model.

[0128] Sample points:

[0129] Response point:

[0130] 3.2: Basic formula of Kriging surrogate model:

[0131] The Kriging surrogate model is constructed by parameter Predicting Response , its basic formula is:

[0132] (3);

[0133] Where: is the predicted value, is a constant or trend function, used to represent the global trend; It is a local deviation based on the covariance and is used to capture the impact of local changes in the control parameters on the short-circuit current response.

[0134] 3.3: Definition of covariance function:

[0135] In order to describe the correlation between control parameters, the Kriging model uses a covariance function. The common form of the covariance function is the Gaussian kernel function:

[0136] (4);

[0137] Where: is the variance of the model, representing the noise between data points; is the correlation length size parameter, indicating the control parameter the extent to which changes in the response affect the response; is the sample point and In the parameters The distance on

[0138] The covariance function represents the correlation between adjacent sample points, that is, if two sample points are close in the control parameter space, their short-circuit current responses should also be similar.

[0139] 3.4: Kriging model training

[0140] The training process of the Kriging model is the fitting process, which requires estimating the parameters in the model. , , Etc. The commonly used estimation method is the maximum likelihood estimation (MLE), whose goal is to find the optimal model parameters by maximizing the likelihood function.

[0141] Likelihood function is defined as:

[0142] (5);

[0143] Where: is the actual observed response value vector; is the covariance matrix composed of covariance functions; is the global trend term; and are the model parameters to be estimated

[0144] 3.5: Predict the short-circuit current response of the new parameter combination system:

[0145] After training, the Kriging model can be used according to the new control parameters Make a prediction. The predicted output response is:

[0146] (6);

[0147] Where: It's new The covariance vector with the training dataset; is the covariance matrix of the training dataset; is the response value of the training data set.

[0148] Figure 2This is the relationship diagram between the Kriging model prediction control parameters and the short-circuit current response. The results show that the prediction results of the Kriging proxy model are very close to the true values, and after calculating the error between the proxy model prediction value and the true value, it can be seen that the error under the sample parameters is within 10%. At the same time, the trend of gradually decreasing error is reflected in the comparison between the true value and the predicted value, especially in the later sampling points. With the increase of sample points, the training data of the model becomes more sufficient, the error is also within 5%, and the fitting effect is also significantly improved.

[0149] (4). Improved particle swarm optimization algorithm:

[0150] The improved particle swarm optimization algorithm is specifically a particle swarm optimization algorithm based on the credibility of the three-black hole system particle capture strategy, which is mainly used to optimize the control parameter combination predicted by the Kriging agent model to find the optimal short-circuit current suppression solution. Figure 3 The update diagram is shown in the figure, and the specific steps are as follows:

[0151] 4.1: Random Black Hole Particle Swarm Optimization (RBHPSO):

[0152] The particle space vector under RBHPSO can be expressed as:

[0153] (7);

[0154] Where: l represents the probability of a particle being captured by a black hole. When the particle escapes from the black hole, When , the particle is captured by the black hole; k is the current iteration number, and They represent the velocity and position of particle i in the dth dimension at the k+1th generation respectively; and They represent the individual optimal solution of the particle and the global optimal solution respectively; is the inertia weight; and is the learning coefficient; l, , , and is a random number uniformly distributed between [0,1].

[0155] 4.2: Introducing credibility factor:

[0156] Since the global optimal particle of RBHPSO in the early stage of iteration is less likely to be the true optimal particle, and as the algorithm iterates, the particles will move towards the true optimal solution, and the possibility of the global optimal particle being generated in it will continue to increase, so the credibility factor q is introduced.

[0157] (8);

[0158] Where: Gen is the total number of iterations; is a random number uniformly distributed between [0,1].

[0159] 4.3: Introducing the Three Black Hole System Capture Strategy (TCS)

[0160] In the early stages of the iteration, particles will be captured by random black holes, and the mass of the black holes will increase, increasing their gravitational force and the radius of the black holes. As the iterations proceed, the mass of the black holes close to the global optimal solution continues to increase until the remaining black holes are destroyed. During the iteration process, the black hole will not only attract the remaining particles, but its own position will also change between the midpoint between the global optimal particle and the upper bound, the midpoint between the global optimal particle and the lower bound, and a point near the global optimal.

[0161] (9);

[0162] Where: is the probability of updating the black hole position, ∈[0,1]; is a random number uniformly distributed between [0,1].

[0163] (5). Analysis of solution results:

[0164] The flow chart of the method proposed by the present invention is as follows Figure 4 As shown in the figure, a short-circuit current control parameter identification and optimization method based on the Kriging agent model and the three-black hole particle swarm algorithm (BHPSO) is proposed based on the grid-connected scenario of the grid-connected photovoltaic and energy storage system. By analyzing various control parameters of power electronic equipment and VSG parameters (such as moment of inertia, damping coefficient, etc.), the key control parameters for short-circuit current suppression are identified and the parameters are optimized to ensure the stable operation of the grid-connected photovoltaic storage system.

[0165] Figure 5The Pareto front solution set diagram based on the Kriging agent model and the improved particle swarm optimization algorithm. In this simulation, the short-circuit current peak and recovery time of the grid-connected photovoltaic energy storage system were multi-objective optimized based on the Kriging agent model and the improved particle swarm optimization algorithm, and the Pareto front solution set was obtained. This solution set shows the balance state of the system between these two objectives. The simulation results show this balance relationship, and the optimal solution set under different control parameter combinations is intuitively displayed through the Pareto front solution set. Generally, it is hoped that the short-circuit current peak can be reduced by controlling and optimizing parameters, thereby improving the safety and stability of the system. The recovery time represents the time required for the system to return to normal operation after a fault occurs. The shorter the recovery time, the stronger the dynamic response capability of the system, and the faster it can recover from the fault.

[0166] In the simulation diagram, 40 solution points are shown, representing the balance between the short-circuit current peak and the recovery time under different control parameter combinations. In the low peak current area: In the left area of ​​the graph, the short-circuit current peak is close to 0.2 to 0.4. It can be seen that the recovery time has increased significantly. This shows that in order to reduce the short-circuit current peak, the system must sacrifice the dynamic response speed, resulting in a significant increase in the recovery time, up to 1.5 seconds. This is consistent with the phenomenon in the actual system. The control strategy to reduce the short-circuit current peak usually introduces more protection or control delays. In the high peak current area: In the right area of ​​the graph, the short-circuit current peak is close to 0.7 to 0.9; the recovery time is short, about 0.7 to 0.9 seconds. This shows that the system's dynamic response speed is fast, but at the cost of a higher short-circuit current peak. This is in line with the actual situation, that is, when improving the dynamic response capability of the system, the short-circuit current peak is difficult to be effectively suppressed.

[0167] In summary: In the optimized solution set (red dots), it can be seen that the three-black hole particle swarm algorithm has significantly improved the performance of the system under multiple control parameter combinations through optimization. For example, in the area where the short-circuit current peak is about 0.5 to 0.6, the optimized solution set shows a shorter recovery time of about 0.8 to 1.0 seconds; at the same time, the short-circuit current peak is also controlled. This shows that through optimization, the system can find a better balance between the short-circuit current peak and the recovery time. The optimized solution set is closer to the Pareto frontier than the unoptimized solution set, indicating that a better control parameter combination can be found through the global search of the three-black hole particle swarm algorithm. Especially in the short-circuit current peak range of 0.4 to 0.7, the performance of short-circuit current and recovery time in the optimized solution set is relatively ideal.

[0168] Figure 6The short-circuit current response curve comparison diagram before and after optimization is shown. The short-circuit current level before optimization is displayed in a gradient from blue to yellow, representing the short-circuit current response of the system under different control parameter settings. The short-circuit current level after optimization is displayed as a red surface, showing the significant improvement of the short-circuit current response after optimization. From the simulation diagram, it can be seen that with the increase of reactive current limit and current limit constant, the short-circuit current level gradually decreases before optimization, which shows that the appropriate increase of these parameters can effectively reduce the impact of short-circuit current on the system. It is displayed by the color bar on the right. Before optimization, it was observed that a larger virtual moment of inertia would increase the short-circuit current peak. This is because a larger virtual moment of inertia will make the system response sluggish, resulting in the system being unable to adjust quickly under a short-circuit fault, resulting in a higher short-circuit current peak. It is displayed by the color bar on the left. It can be observed from the figure that a lower virtual damping coefficient will significantly increase the short-circuit current level, indicating that a smaller virtual damping coefficient will weaken the damping capacity of the system, resulting in a longer short-circuit current recovery time.

[0169] When control parameter 1: reactive current limit and control parameter 2: current limit setting are at lower values ​​(i.e., near the lower left corner in the figure), the short-circuit current level is high, close to 2.0. This shows that when the reactive current limit is small and the current limit setting is low, the system is difficult to provide effective reactive power and current support under short-circuit faults, resulting in the short-circuit current being unable to be suppressed in time, forming a high peak current. When the virtual moment of inertia (control parameter 3) is large, the short-circuit current level is high. Large moment of inertia will cause the system to respond slowly and be unable to respond quickly to short-circuit current shocks, resulting in an increase in the system short-circuit current level. This can be reflected by observing the change in the color of the surface from blue to yellow.

[0170] After the optimization of the three-black-hole particle swarm algorithm based on credibility mentioned in the present invention, the optimized red surface is significantly lower than the blue surface before optimization, especially under the combination of high reactive current limiting and high current limiting constant, the short-circuit current level is greatly reduced. The virtual moment of inertia (control parameter 3) and the virtual damping coefficient (control parameter 4) are adjusted to a suitable range in the optimized combination. Under the reasonable combination of these parameters, the system response is accelerated, the short-circuit current recovery time is significantly shortened, and the short-circuit current peak is significantly reduced to about 0.8. In particular, the improvement of control parameters 1 and 2 significantly improves the reactive support capacity and current limiting capacity of the system, thereby effectively reducing the peak value of the short-circuit current and accelerating the recovery process of the system.

[0171] In summary: The Kriging proxy model effectively fits the nonlinear relationship between the short-circuit current control parameters and the short-circuit current response, and globally optimizes the key control parameters through the improved particle swarm optimization algorithm. In the optimized figure, through the optimal combination of multiple groups of control parameters, the short-circuit current level has been greatly improved in the entire parameter space, especially under the conditions of high reactive current limiting and high current limiting constant, the short-circuit current peak is reduced to below 1.0. The joint optimization of the virtual moment of inertia and the virtual damping coefficient ensures that the system can balance the short-circuit current peak and recovery time under fault conditions. The optimized result is a smoother and more uniform short-circuit current response on the surface.

Claims

1. A method for optimizing the short-circuit current parameters of a grid-connected photovoltaic energy storage system based on the Kriging agent model and improved particle swarm algorithm, characterized in that The following steps are involved: Step 1: Construct a short-circuit current simulation model of a grid-connected photovoltaic energy storage system; Step 2: Screen the control parameters through local sensitivity analysis to find out the control parameters that have an impact on short-circuit current suppression; Step 3: Use the Kriging surrogate model to predict the relationship between the fitted control parameters and the short-circuit current response; Step 4: Globally optimize the short-circuit current control parameters fitted by the Kriging proxy model by using an improved particle swarm algorithm to obtain the optimized optimal short-circuit current control parameter combination; In step 4, the improved particle swarm algorithm is used to optimize the control parameter combination predicted by the Kriging agent model to find the optimal short-circuit current suppression solution; the following steps are included: S4.1: The particle space vector under the random black hole particle swarm algorithm RBHPSO can be expressed as: Where: represents the probability of a particle being captured by a black hole. When the particle escapes from the black hole, When , the particle is captured by the black hole; is the current iteration number; and Represents particles No. Vidi The speed and position of the generation; and They represent the individual optimal solution of the particle and the global optimal solution respectively; is the inertia weight; and is the learning coefficient; , , and is a random number uniformly distributed between [0,1]; Indicates The particle in The first The location of the dimension; express The particle in The first The speed of the dimension; S4.2, introduce credibility factor: Since the global optimal particle of RBHPSO in the early stage of iteration is less likely to be the true optimal particle, and as the algorithm iterates, the particles will move towards the true optimal solution, and the possibility of the global optimal particle being generated in it will continue to increase, so the credibility factor q is introduced: Where: Gen is the total number of iterations; is a random number uniformly distributed between [0,1]; S4.

3. Introducing the three-black hole system particle capture strategy TCS: At the beginning of the iteration, particles will be captured by random black holes and increase the mass of the black holes, increasing their gravitational force and the radius of the black holes. As the iteration proceeds, the mass of the black holes close to the global optimal solution continues to increase until the remaining black holes are destroyed. During the iteration process, the black hole will not only attract the remaining particles, but its own position will also change between the midpoint between the global optimal particle and the upper bound, the midpoint between the global optimal particle and the lower bound, and a point near the global optimal. Where: is the probability of updating the black hole position; Represents the probability value, which is used to determine the change in the black hole position; Indicates the upper limit of the black hole's position; Represents the lower bound of the location of the black hole; ∈[0,1]; is a random number uniformly distributed between [0,1].

2. The method for optimizing the grid-connected short-circuit current parameters of a grid-connected photovoltaic energy storage system based on the Kriging proxy model and the improved particle swarm algorithm according to claim 1 is characterized in that: In the step 1, a grid-based photovoltaic energy storage system is constructed in the Matlab / Simulink platform, which includes a photovoltaic module, an energy storage module, a main circuit, a virtual synchronous generator VSG control module, an observation module, an SPWM modulation module, and a dq axis conversion module; The photovoltaic module, after the voltage is increased by a DC-DC boost circuit, is connected to the virtual synchronous generator; the electricity generated by the photovoltaic module first passes through the battery part of the energy storage module, part of the electricity is used for storage, and the rest is converted into AC power by the inverter and then connected to the grid; The output end of the energy storage module is directly connected to the DC input end of the inverter; the energy storage module stores electrical energy and releases it when needed; the energy storage module is used to adjust the load of the power grid and balance the supply and demand of electricity; The main circuit is used for the conversion and transmission of electric energy. After the inverter converts the DC power from the photovoltaic module and the energy storage module into AC power, it first passes through the filter to remove harmonics. Then, the current passes through the line impedance, simulating the transmission loss in the actual system. Finally, the current enters the grid to achieve grid connection. The virtual synchronous generator VSG control module is used to measure the voltage, current and frequency of the system and adjust the system in real time. The virtual synchronous generator VSG control module is connected to the inverter and is responsible for controlling the output frequency and voltage of the inverter to synchronize it with the power grid. At the same time, the virtual synchronous generator VSG control module also provides inertia and damping characteristics for the system. Observation module, used to monitor the operating status of the system, including signals such as voltage, current and frequency; the observation module is directly connected to the main circuit and inverter, collects the operating data of the system in real time, and displays it in charts or numerical values; The SPWM modulation module is used to generate a pulse width modulation signal to directly control the power switch tube of the inverter, thereby adjusting its output voltage and frequency; the SPWM modulation module receives the signal from the virtual synchronous generator VSG control module and converts it into a corresponding modulation signal; The dq-axis conversion module is used to convert the three-phase AC power of the power grid into a DC equivalent dq-axis representation to simplify the control; the dq-axis conversion module receives the three-phase current and voltage signals of the main circuit, and converts them into DC signals under dq coordinates, and provides them to the virtual synchronous generator VSG control module; at the same time, it also converts the dq-axis control signal output by the virtual synchronous generator VSG control module back into a three-phase AC signal and transmits it to the inverter.

3. The method for optimizing the grid-connected short-circuit current parameters of a grid-connected photovoltaic energy storage system based on the Kriging proxy model and the improved particle swarm algorithm according to claim 1 is characterized in that: The step 2 comprises the following steps: S2.

1. Define initial conditions: First, determine the initial values ​​of all control parameters, including: a. Reactive current limit value: ; b. Virtual damping coefficient: ; c. Virtual moment of inertia: ; d. Current limiting setting: ; The system output is the short-circuit current response, the peak value or the recovery time of the short-circuit current. Indicates that is the vector of control parameters; S2.

2. Change a single control parameter: Keep other control parameters unchanged and only make a small change to one control parameter; suppose the current adjustment is the reactive current limit value , so that it increases by a small amount , becomes ; S2.

3. Calculate the changes in system output: Calculate the adjusted system , , Indicates other control parameters in the grid-connected photovoltaic energy storage system; and with the original output Make a comparison; this comparison is performed using the following formula: （1）； S2.

4. Calculate the local sensitivity coefficient: Local sensitivity coefficient Quantization control parameters The impact on system output is calculated by the following formula: （2）； in, Indicates a small change in the system output; Indicates control parameters Small changes in Indicates that the system is changing the reactive current limit The output value after Indicates that the system is reactive current limiting before change The output value of is the response of the system in the initial state; S2.

5. Repeat the sensitivity analysis for other control parameters: Repeat the above steps to adjust other control parameters one by one and adjust the virtual moment of inertia , calculate each parameter to get the sensitivity; S2.6, Analysis and Sorting: The calculated sensitivity coefficient The control parameters with larger sensitivity coefficients have greater impact on the system output.

4. The method for optimizing the grid-connected short-circuit current parameters of a grid-connected photovoltaic energy storage system based on the Kriging proxy model and the improved particle swarm algorithm according to claim 1 is characterized in that: In step 3, the Kriging proxy model is used to fit the nonlinear relationship between the control parameters and the short-circuit current response; the Kriging proxy model provides an approximate model for the optimization algorithm by establishing the relationship between the input parameters, including reactive current limiting, current limiting constant, virtual moment of inertia, and the output response: short-circuit current peak value or recovery time; specifically as follows: 3.

1. Before constructing the Kriging proxy model, a small amount of sample data is collected through simulation or experiment; sample points, each sample contains a control parameter vector , Respectively expressed in The control parameter vector of sample points; Indicates The first control parameter of the sample point, Indicates The second control parameter of the sample point; Sample points: , Represents the combination of different control parameters, each sample point Represents a specific value of a set of control parameters in the system; Response point: , Respectively represent the control parameters in each group The output response of the system is: 3.2 Kriging surrogate model through parameters Predicting Response , its basic formula is: （3）； Where: is the predicted value; is a constant or trend function, used to represent the global trend; It is a local deviation based on covariance, which is used to capture the impact of local changes in control parameters on the short-circuit current response; S3.

3. In order to describe the correlation between control parameters, the Kriging proxy model uses a covariance function; the covariance function is in the form of a Gaussian kernel function: （4）； Where: Represents sample points , Covariance in Kriging models; Indicates that the control parameters The response value of the system below; Indicates that the control parameters The response value of the system below; is the variance of the model, representing the noise between data points; is the correlation length size parameter, indicating the control parameter the extent to which changes in the response affect the response; is the sample point and In the parameters The distance on Indicates Among the sample points The control parameter value, i.e. Quantity; Indicates Among the sample points The control parameter value, i.e. Quantity; Indicates the total dimension of the control parameters, that is, how many control parameters are there in the system; Represents the power exponent, which indicates the degree of nonlinearity of the distance metric; In the covariance function, Determines the degree of influence of changes in control parameters on covariance; 3.

4. The training process of the Kriging surrogate model is the fitting process, estimating the parameters in the Kriging surrogate model , , ; The estimation method adopts the maximum likelihood estimation method, whose goal is to find the optimal model parameters by maximizing the likelihood function; Likelihood function is defined as: （5）； Where: is the actual observed response value vector; is the covariance matrix composed of covariance functions; Indicates transposing the matrix; is the global trend term; and are the model parameters to be estimated ; 3.

5. After training, the Kriging agent model can be used according to the new control parameters Make a prediction; the predicted output response is: （6）； Where: For new samples The predicted response value of It's new The covariance vector between the training dataset and Represents the global trend term; Indicates that the matrix is ​​transposed; is the covariance matrix of the training dataset; is the response value of the training data set.

Images