Multi-level photovoltaic grid-connected converter model prediction control method, system and equipment based on SO optimization neural network
Through the multi-level photovoltaic grid-connected converter model prediction control method based on SO-optimized neural network, the initial weight and threshold are optimized by using the snake group optimization algorithm to achieve accurate prediction of the number of MMC submodules, and the prediction in the existing technology is solved The problem of low efficiency improves the efficiency of photovoltaic power generation systems.
Patent Information
- Application Number
- CN202411043295.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-31
- Publication Date
- 2024-11-08
AI Technical Summary
The prior art cannot quickly predict the number of sub-modules installed in multi-level photovoltaic grid-connected converters, resulting in troublesome operation and low MMC operation efficiency.
The multi-level photovoltaic grid-connected converter model prediction control method based on SO optimization neural network is adopted. The MMC model is built through the MATLAB/Simulink simulation platform, data is collected for preprocessing, and the initial weight of the neural network is optimized by using the snake group optimization algorithm. Threshold, build SO-neural network-MPC controller to achieve accurate prediction of the number of submodules.
It improves the prediction accuracy of the number of submodules, improves the startup efficiency of MMC, reduces energy waste, solves the problem of operational troubles, and improves the power generation efficiency of the photovoltaic power generation system.
Smart Images

Figure CN118917199A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of flexible direct current transmission of electric power systems, and in particular to a model predictive control method, system and device for a multi-level photovoltaic grid-connected converter based on a SO optimized neural network. Background Art
[0002] Modular Multilevel Converter (MMC) is an important technology to support modern power transmission equipment. This technology is widely used in AC transmission system (FACTS) equipment and high voltage direct current (HVDC) systems, which can greatly improve the stability of power grid operation and long-distance power transmission capacity. A typical three-phase MMC consists of six bridge arms. A large number of submodules (SM) are connected in series in each bridge arm, and a bridge arm inductor is connected in series. MMC essentially controls the output voltage of each bridge arm by controlling the output level of each submodule, thereby controlling the bridge arm current, and finally controlling the AC, DC and circulating current. Due to the advantages of high efficiency, reliability, flexibility, reduced electromagnetic interference, and reduced line loss, MMC is widely used in the field of photovoltaic grid-connected power generation. In actual practice, the number of submodules has a significant impact on the energy storage of photovoltaic power generation systems. Therefore, a reasonable selection of the number of submodules can maximize the benefits of photovoltaic power generation systems.
[0003] In the prior art, the number of sub-modules put into use in the bridge arm can be obtained by the algebraic sum of all switch signals in the bridge arm. The prior art determines the number of sub-modules through three methods: carrier stacking, carrier phase shift modulation, and nearest level modulation. However, carrier stacking, carrier phase shift modulation, and nearest level modulation are all based on the intersection of the sinusoidal modulation wave and the triangular carrier to control the on and off of the sub-modules. The results of multiple carriers are superimposed together to obtain the number of sub-modules put into use. In actual operation, due to the uncertainty of the actual current demand, the period of the carrier needs to be continuously adjusted according to the actual situation. Therefore, these methods cannot quickly predict the number of sub-modules to be installed. Continuous debugging is required during the installation process to obtain the accurate number of sub-modules. The operation is cumbersome, resulting in low MMC operating efficiency. Summary of the invention
[0004] The present invention aims to provide a model predictive control method for a multi-level photovoltaic grid-connected converter based on a SO optimized neural network, which is mainly used to solve the technical problems that the prior art cannot quickly predict the number of sub-modules to be installed, and the number of sub-modules can only be determined after multiple tests and operations, which is cumbersome to operate and leads to low MMC operating efficiency.
[0005] In order to solve the above technical problems, the present invention provides the following technical solutions:
[0006] A model predictive control method for a multi-level photovoltaic grid-connected converter based on a SO optimized neural network comprises the following steps:
[0007] S1: Use MATLAB / Simulink simulation platform to build a simulation model with MMC as the main circuit and MPC as the control circuit;
[0008] S2: Collecting data and preprocessing the data through the MATLAB / Simulink simulation platform, the data are several electrical variables output by the MMC;
[0009] S3: Input the processed data into the neural network in the simulation model for training and testing to obtain a neural network-MPC controller, and obtain the initial weights and thresholds of the neural network during the training process;
[0010] S4: Using the SO algorithm to optimize the initial weights and thresholds of the neural network, a SO-neural network-MPC controller is obtained;
[0011] S5: Simulate the multi-level photovoltaic grid-connected converter model predictive control through the SO-neural network-MPC controller and output reasonable sub-module quantity prediction value data.
[0012] Preferably, in step S2, relevant data is sampled from the simulation model, the data is electrical variables, and the correlation coefficient between the feature and the target variable is calculated through feature selection, and the feature with a very strong correlation with the target variable is selected. The relevant expression of feature selection is:
[0013]
[0014] Where X i is a column of data in the data set, Y i is a column of target variable data, where the target variable is the number of submodules, Corr(X i ,Y i ) is the correlation coefficient between the feature and the target variable, Cov(X i ,Y i ) is the covariance between the feature and the target variable, σ X and σ Y Represent the standard deviation of feature X and target variable Y respectively;
[0015] The collected data includes the AC side current.
[0016] Preferably, in step S2, the correlation coefficient between the feature and the target variable ranges from 0.8 to 1.0.
[0017] Preferably, in step S2, by using variational mode decomposition VMD, the AC side current i ac Perform filtering operation, VMD will i ac Decomposed into K modal functions, after decomposition i ac The expression is as follows:
[0018]
[0019] In the formula, is the fundamental frequency component, u kn (t) represents the nth mode function.
[0020] Preferably, in step S2, the AC side current variational model is as follows:
[0021]
[0022]
[0023] In the formula {u k} represents the set of modal components, {ω k} is the set of center frequencies corresponding to each mode; δ(t) is the Dirac function, e ―jωkt Indicates that each mode u k The analytical signal spectrum of (t) is shifted to the baseband, Represents the entire expression δ Taking the partial derivative with respect to time t, It means that each u k (t) is converted into an analytical signal form to optimize the decomposition result of VMD by utilizing its instantaneous frequency and instantaneous phase information.
[0024] Preferably, in step S3, the specific structure of the neural network is as follows: the input layer has 11 electrical variables; the hidden layer has 11 nodes; the output layer has 2 nodes; ij and θ ij is the initial weight and threshold between the input layer and the hidden layer, ω jk and θ jk is the initial weight and threshold between the hidden layer and the output layer.
[0025] Preferably, in step S4, the SO algorithm includes the following steps:
[0026] S31: Initialize the snake group. The initialization formula is as follows:
[0027] x i =x min +r(x max ―x min )
[0028] In the formula, xi is the solution for the ith position; r is a random number in [0,1]; x min and x max It is the upper and lower boundaries of problem solving;
[0029] Assess each group and determine temperature and food amount:
[0030]
[0031] Where t represents the current iteration; temp represents the calculated temperature value; Q represents the amount of food; T represents the maximum number of iterations, and c1 = 0.5.
[0032] Determine the exploration phase and development phase:
[0033]
[0034] In the formula, Indicates the position of male snakes; represents the random snake position; c2 is a constant, c2=0.05; rand∈[0,1] is a random number, x min and x max is the upper and lower bounds of the problem, A m It represents the ability of male snakes to search for food, and its formula is as follows:
[0035]
[0036] In the formula, yes applicability; yes applicability;
[0037] S32: Close to food, when Q>0.25, temp>0.6:
[0038] x i,j (t+1)=x food ±c3×temp×rand×(x food ―x i,j (t))
[0039] In the formula, x i,j represents the entire snake population, x food Indicates the global optimal position, c3 = 2;
[0040] S33: Combat, the formula is as follows:
[0041]
[0042] In the formula, It is the optimal position for female snakes; It is the female snake position; This is the optimal position for male snakes; It is the male snake position; Indicates the updated position of the male snake. Indicates the updated position of the female snake; FF is the combat value of the female snake, and FM is the combat value of the male snake;
[0043] The FM and FF formulas are as follows:
[0044]
[0045] In the formula, For the best female snake position, is the position of the best male snake, f i It is the proxy position of the snake;
[0046] S34: Mating, the formula is as follows:
[0047]
[0048] Where M m and M f are the mating abilities of male and female snakes, respectively, and the formula is as follows:
[0049]
[0050] In the formula, Represents a male snake search agent, Represents a search agent for female snakes;
[0051]
[0052] In the formula, Indicates the worst position in the male snake group. Indicates the worst position in the female snake group, represents the maximum position boundary value in the male snake group, represents the minimum position boundary value in the male snake group, represents the maximum position boundary value in the female snake group, represents the minimum value of the position boundary in the female snake group;
[0053] S35: Terminate and determine whether the maximum number of iterations has been reached; if so, output the optimal snake individual, that is, the optimal initial weight and threshold; otherwise, return to S33 to continue iterating.
[0054] To solve the above technical problems, the second solution proposed by this patent is as follows:
[0055] A multi-level photovoltaic grid-connected converter model predictive control system based on SO optimized neural network, comprising:
[0056] The simulation module is used to build a simulation model with MMC as the main circuit and MPC as the control circuit;
[0057] The preprocessing module is used to collect data through the simulation platform and perform data preprocessing; the training optimization module is used to input the data processed by the preprocessing module into the neural network in the simulation model for training and testing to obtain the neural network-MPC controller, and obtain the initial weights and thresholds of the neural network through training;
[0058] An optimization module is used to optimize the initial weights and thresholds of the neural network using the SO algorithm to obtain a SO-neural network-MPC controller;
[0059] The output module is used to simulate the model predictive control of the multi-level photovoltaic grid-connected converter through the SO-neural network-MPC controller and output reasonable sub-module quantity prediction value data.
[0060] To solve the above technical problems, the third solution proposed by this patent is as follows:
[0061] A computer-readable storage medium includes a stored program, wherein when the program is run, the device where the computer-readable storage medium is located is controlled to execute the model predictive control method of a multi-level photovoltaic grid-connected converter based on a SO optimized neural network.
[0062] To solve the above technical problems, the fourth solution proposed by this patent is as follows:
[0063] A processor is used to run a program, wherein when the program is run, a model predictive control method for a multi-level photovoltaic grid-connected converter based on a SO optimized neural network is executed.
[0064] The beneficial effects of this program are as follows:
[0065] The Snake Optimizer (SO) algorithm was proposed in 2022. The algorithm simulates the foraging and reproductive behavior of snakes. It is a heuristic optimization algorithm based on snake behavior and is used to solve optimization problems. The SO algorithm has the advantages of easy implementation, strong global search capability and strong robustness. The SO algorithm simulates the fighting and mating modes of snakes to achieve global optimization. In the fighting mode, males compete for the best females, and each female selects the best male. In the mating mode, mating between males and females is affected by food availability. If mating occurs during the search phase, female snakes may lay eggs and hatch new offspring. The basic idea of the present invention is to use the snake optimization algorithm to train and optimize the initial weights and thresholds of the neural network to improve the prediction performance of the neural network. The algorithm has strong optimization performance and wide applicability, which improves the convergence speed of the algorithm, can find the optimal solution quickly, and is widely used in the engineering field. The SO algorithm has the advantages of strong robustness, which can improve the convergence speed and quickly find the optimal solution.
[0066] The multi-level photovoltaic grid-connected converter model predictive control method, system and device based on SO optimized neural network proposed by the present invention optimize a large number of initial values and corresponding thresholds and weights through SO optimized neural network, greatly improve the training accuracy of the network, realize the accurate prediction of the model predictive control method, and accurately predict the number of submodules in the MMC bridge arm by constructing an algorithm model, thereby improving the benefits of MMC in engineering applications. The multi-level photovoltaic grid-connected converter model predictive control method based on SO optimized neural network can predict the reasonable number of submodules in advance, improve the bridge arm startup efficiency, thereby improving the power generation efficiency of the photovoltaic power generation system, reducing energy waste, and solving the technical problems that the existing technology cannot quickly predict the number of submodules to be installed, and needs to go through multiple tests and operations to determine the number of submodules, which is cumbersome to operate and leads to low MMC operation efficiency. BRIEF DESCRIPTION OF THE DRAWINGS
[0067] Figure 1 The three-phase modular multi-level converter topology structure of the present invention's patent;
[0068] Figure 2 It is the submodule topology of the patent of this invention;
[0069] Figure 3 The neural network structure of the patent of the present invention;
[0070] Figure 4 It is the SO algorithm step of the present invention patent;
[0071] Figure 5 This is the fitting regression diagram of the SO-neural network-MPC controller of the patent of this invention. DETAILED DESCRIPTION
[0072] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0073] A model predictive control method for a multi-level photovoltaic grid-connected converter based on a SO optimized neural network mainly comprises the following steps.
[0074] S1: Use MATLAB / Simulink simulation platform to build a simulation model with MMC as the main circuit and MPC as the control circuit;
[0075] S2: Collect data and perform data preprocessing through MATLAB / Simulink simulation platform (the data are several electrical variables collected by MMC);
[0076] The first step is feature selection. The correlation coefficient between the feature and the target variable is calculated, and the feature with a high correlation with the target variable is selected. The relevant expression for feature selection is:
[0077]
[0078] Where X i is a column of data in the data set, Y i is a column of target variable data, where the target variable is the number of submodules, Corr(X i ,Y i ) is the correlation coefficient between the feature and the target variable, Cov(X i ,Y i ) is the covariance between the feature and the target variable, σ X and σ Y Represent the standard deviation of feature X and target variable Y respectively.
[0079] When defining the correlation between features and target variables, it is usually necessary to refer to the following logic and standards. The judgment criteria of the correlation coefficient are generally divided into the following intervals:
[0080] 0.8~1.0: Very strong correlation
[0081] 0.6~0.8: Strong correlation
[0082] 0.4~0.6: Moderate correlation
[0083] 0.2~0.4: Weak correlation
[0084] 0.0~0.2: very weak or no correlation;
[0085] The correlation coefficient between the feature and the target variable in step S2 ranges from 0.8 to 1.0.
[0086] Then, during the experiment, the preferred characteristics include the AC side current i ac , because the AC side current i ac In the presence of harmonic interference, filtering is performed by using the AC side current variational mode decomposition VMD, which has no effect on its amplitude and phase.
[0087] VMD will i ac Decomposed into K modal functions, so i ac The expression is as follows:
[0088]
[0089] In the formula, is the fundamental frequency component, u kn (t) represents the nth mode function;
[0090] The variational model of the AC side current is as follows:
[0091]
[0092] In the formula {u k} represents the set of modal components, {ω k} is the set of center frequencies corresponding to each mode; δ(t) is the Dirac function, e ―jωkt Indicates that each mode u k The analytical signal spectrum of (t) is shifted to the baseband, Represents the entire expression Take the partial derivatives with respect to time t, that is, we need to minimize the sum of the squares of the time derivatives of each mode function, It means that each u k (t) is converted into an analytical signal form to optimize the decomposition result of VMD by utilizing its instantaneous frequency and instantaneous phase information.
[0093] S3: Input the processed data into the neural network for training and testing to obtain a neural network-MPC controller (NN-MPC). Use the SO algorithm to optimize the initial weights and thresholds to obtain a SO-neural network-MPC controller.
[0094] First, the neural network is constructed. The input layer has 11 electrical variables; the hidden layer has 11 nodes; and the output layer has 2 nodes. ij and θij is the weight and threshold between the input layer and the hidden layer, ω jk and θ jk are the weights and thresholds between the hidden layer and the output layer. The above thresholds and weights are obtained through neural network training in MATLAB.
[0095] like Figure 4 As shown, the SO algorithm consists of five steps:
[0096] S31: Initialize the snake group. The initialization formula is as follows:
[0097] x i =x min +r(x max ―x min )
[0098] In the formula, x i is the solution for the ith position; r = a random number in [0,1]; x min and x max These are the upper and lower boundaries for solving the problem.
[0099] Assess each group and determine temperature and food amount:
[0100]
[0101] Where t represents the current iteration; temp represents the calculated temperature value; Q represents the amount of food; T represents the maximum number of iterations, and c1 = 0.5.
[0102] Determine the exploration phase and development phase:
[0103]
[0104] In the formula, Indicates the position of male snakes; represents the random snake position; c2 is a constant, c2=0.05; rand∈[0,1] is a random number, x min and x max is the upper and lower bounds of the problem, A m It represents the ability of male snakes to search for food, and its formula is as follows:
[0105]
[0106] In the formula, yes applicability; yes applicability.
[0107] S32: Close to food, when Q>0.25, temp>0.6:
[0108] x i,j (t+1)=x food ±c3×temp×rand×(x food ―x i,j (t))
[0109] In the formula, x i,j represents the entire snake population, x food Indicates the global optimal position, c3=2.
[0110] S33: Combat, the formula is as follows:
[0111]
[0112] In the formula, It is the optimal position for female snakes; It is the female snake position; This is the optimal position for male snakes; It is the male snake position; Indicates the updated position of the male snake. Indicates the updated position of the female snake; FF is the combat value of the female snake, and FM is the combat value of the male snake;
[0113] The FM and FF formulas are as follows:
[0114]
[0115] In the formula, For the best female snake position, is the position of the best male snake, f i is the snake's proxy position;,f i It is the proxy position of the snake.
[0116] S34: Mating, the formula is as follows:
[0117]
[0118] Where M m and M f are the mating abilities of male and female snakes, respectively, and the formula is as follows:
[0119]
[0120] In the formula, Represents a male snake search agent, Represents a search agent for female snakes;
[0121]
[0122] In the formula, Indicates the worst position in the male snake group. Indicates the worst position in the female snake group, represents the maximum position boundary value in the male snake group, represents the minimum position boundary value in the male snake group, represents the maximum position boundary value in the female snake group, Represents the minimum position boundary in the female snake group.
[0123] S35: Terminate and determine whether the maximum number of iterations has been reached; if so, output the optimal snake individual (i.e., the initial weights and thresholds of the optimal neural network); otherwise, return to S33 to continue iterating.
[0124] S4: Simulate the model predictive control of the multi-level photovoltaic grid-connected converter through the SO-neural network-MPC controller and output the predicted value data.
[0125] Finally, the root mean square error (RMSEP) and R (representing the correlation between the predicted output and the target output) are used to evaluate the prediction accuracy of the neural network model. The value range of R is [0,1]. The closer R is to 1, the closer the predicted value is to the true value, and the better the model fitting effect; conversely, the worse the model prediction accuracy is. If the accuracy meets the requirements, the data can be output directly. If it does not meet the requirements, the initial weights and thresholds are optimized again, and iterated again. Repeat the above operations until the requirements are met. The calculation formula is as follows:
[0126]
[0127] Where: n is the total number of predicted samples, i is the total number of predicted samples, It is the predicted value minus the true value on the total predicted sample. is the sample mean.
[0128] At the same time, the present invention also proposes the following three solutions:
[0129] A multi-level photovoltaic grid-connected converter model predictive control system based on SO optimized neural network, comprising:
[0130] The simulation module is used to build a simulation model with MMC as the main circuit and MPC as the control circuit;
[0131] The preprocessing module is used to collect data through the simulation platform and perform data preprocessing; the training optimization module is used to input the data processed by the preprocessing module into the neural network in the simulation model for training and testing to obtain the neural network-MPC controller, and obtain the initial weights and thresholds of the neural network through training;
[0132] An optimization module is used to optimize the initial weights and thresholds of the neural network using the SO algorithm to obtain a SO-neural network-MPC controller;
[0133] The output module is used to simulate the model predictive control of the multi-level photovoltaic grid-connected converter through the SO-neural network-MPC controller and output reasonable sub-module quantity prediction value data.
[0134] A computer-readable storage medium includes a stored program, wherein when the program is run, the device where the computer-readable storage medium is located is controlled to execute the model predictive control method of a multi-level photovoltaic grid-connected converter based on a SO optimized neural network.
[0135] A processor is used to run a program, wherein when the program is run, a model predictive control method for a multi-level photovoltaic grid-connected converter based on a SO optimized neural network is executed.
[0136] The following are specific cases, mainly including:
[0137] Step 1: First determine the structure of the neural network controller and collect the corresponding data samples in the MPC-MMC control system. The structure of the neural network consists of an input layer, a hidden layer, and an output layer. The 11 input nodes of the input layer correspond to the 11 electrical variables output by the MMC, which are the AC side voltage u ac , AC side current i ac , DC side voltage u dc , AC side current reference value DC side current i dc , upper and lower bridge arm current i p and i n , the average voltage of the upper and lower bridge arms u p and u n , the total voltage of the upper and lower bridge arms u zp and u zn ; The number of hidden layer nodes is 11, which is selected from experience; the number of output layer nodes is 2, which is the number of upper and lower bridge arm submodules n pj and n nj .
[0138] Step 2: Preprocess the original sample set, select the data features on the MATLAB experimental platform, and then import it into the Python3.13 simulation platform to build a virtual environment. ac There are harmonic interferences of different frequencies, which interfere with the neural network training. Therefore, the AC side current i ac VMD decomposition is performed. After VMD decomposition filtering, the harmonic content is greatly reduced, and a better sample data set is obtained.
[0139] Step 3:
[0140] 1) The optimal initial weights and thresholds between the input layer and the hidden layer, and between the hidden layer and the output layer of the SO algorithm are set. As the initial weights and thresholds between neurons in each layer of the neural network;
[0141] 2) Forward propagate the samples sequentially and calculate the weights and thresholds between neurons in each layer.
[0142] 3) Calculate the output error. If it does not meet the requirements, back propagate the error correction ω ij ,θ ij and ω jk ,θ jk ; If the requirements are met, continue to the next group.
[0143] 4) If the data set samples are not exhausted, return to 2); otherwise, calculate the root mean square error (RMSEP) and R of the neural network. If both meet the requirements, the neural network training is completed; if not, return to 1) and retrain the network.
[0144] Step 4: Apply the trained SO-neural network MPC controller to the MMC system.
[0145] In order to verify the prediction accuracy of the proposed SO-neural network-MPC controller, the correlation coefficient curve between the measured value and the predicted value of the verification data is used as shown in Figure 5 As shown. RMSEP (Root Mean Square Error of Prediction) is the root mean square error on the prediction set. It indicates the error level when using the model to predict new samples. The smaller the RMSEP, the better the prediction ability of the SO-neural network-MPC controller for new samples. The R coefficient represents the correlation between the predicted output and the target output. The closer the R value is to 1, the closer the relationship between the predicted and output data is. Figure 5 It can be seen that the R value of the model is 0.99286, the root mean square error is 0.0124, and the error value is small, indicating that the overall fitting of the SO optimized neural network model is good and the prediction accuracy is high.
[0146] In summary, the present invention obtains the optimal bridge arm submodule quantity prediction value through the multi-level photovoltaic grid-connected converter model predictive control method based on SO optimization neural network, and improves the prediction accuracy. Compared with the traditional neural network model predictive control method, the SO algorithm has excellent global search ability, optimizes the initial weights and thresholds, and enhances the training efficiency and accuracy of the neural network model. The SO algorithm is suitable for applications in various engineering fields and has strong practicality.
[0147] The above is only an embodiment of the present invention, and the common knowledge such as the known specific structure and characteristics in the scheme is not described in detail here. It should be pointed out that for those skilled in the art, several deformations and improvements can be made without departing from the structure of the present invention, which should also be regarded as the protection scope of the present invention, and these will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.
Claims
1. A model predictive control method for multi-level photovoltaic grid-connected converter based on SO optimized neural network, characterized in that: The steps include: S1: Use MATLAB / Simulink simulation platform to build a simulation model with MMC as the main circuit and MPC as the control circuit; S2: Collecting data and preprocessing the data through the MATLAB / Simulink simulation platform, the data are several electrical variables output by the MMC; S3: Input the processed data into the neural network in the simulation model for training and testing to obtain a neural network-MPC controller, and obtain the initial weights and thresholds of the neural network during the training process; S4: Using the SO algorithm to optimize the initial weights and thresholds of the neural network, a SO-neural network-MPC controller is obtained; S5: Simulate the multi-level photovoltaic grid-connected converter model predictive control through the SO-neural network-MPC controller and output reasonable sub-module quantity prediction value data.
2. The model predictive control method for a multi-level photovoltaic grid-connected converter based on a SO optimized neural network according to claim 1, characterized in that: In step S2, relevant data is sampled from the simulation model, and the correlation coefficient between the feature and the target variable is calculated through feature selection, and the feature with a very strong correlation with the target variable is selected. The relevant expression of feature selection is: Where X i is a column of data in the data set, Y i is a column of target variable data, where the target variable is the number of submodules, Corr(X i ,Y i ) is the correlation coefficient between the feature and the target variable, Cov(X i ,Y i ) is the covariance between the feature and the target variable, σ Y and σ Y Represent the standard deviation of feature X and target variable Y respectively; The collected data include the AC side current i ac .
3. The model predictive control method for a multi-level photovoltaic grid-connected converter based on a SO optimized neural network according to claim 2, characterized in that: In step S2, the correlation coefficient between the feature and the target variable ranges from 0.8 to 1.
0.
4. The model predictive control method for multi-level photovoltaic grid-connected converter based on SO optimized neural network according to claim 3 is characterized in that: In step S2, by using variational mode decomposition VMD, the AC side current i ac Perform filtering operation, VMD will i ac Decomposed into K modal functions, after decomposition i ac The expression is as follows: In the formula, is the fundamental frequency component, u kn (t) represents the nth mode function.
5. According to the model predictive control method of a multi-level photovoltaic grid-connected converter based on SO optimized neural network according to claim 4, in step S2, the AC side current variational model is as follows: In the formula {u k } represents the set of modal components, {ω k } is the set of center frequencies corresponding to each mode; δ(t) is the Dirac function, e ―jωkt Indicates that each mode u k The analytical signal spectrum of (t) is shifted to the baseband, Represents the entire expression Taking the partial derivative with respect to time t, It means that each u k (t) is converted into an analytical signal form to optimize the decomposition result of VMD by utilizing its instantaneous frequency and instantaneous phase information.
6. The model predictive control method for multi-level photovoltaic grid-connected converter based on SO optimized neural network according to claim 1, characterized in that: In step S3, the specific structure of the neural network is as follows: the input layer has 11 electrical variables; the hidden layer has 11 nodes; the output layer has 2 nodes; ij and θ ij is the initial weight and threshold between the input layer and the hidden layer, ω jk and θ jk is the initial weight and threshold between the hidden layer and the output layer.
7. The model predictive control method for multi-level photovoltaic grid-connected converter based on SO optimized neural network according to claim 6, characterized in that: In step S4, the SO algorithm includes the following steps: S31: Initialize the snake group. The initialization formula is as follows: x i =x min +r(x max ―x min ) In the formula, x i is the solution for the ith position; r is a random number in [0,1]; x min and x max It is the upper and lower boundaries of problem solving; Assess each group and determine temperature and food amount: Where t represents the current iteration; temp represents the calculated temperature value; Q represents the amount of food; T represents the maximum number of iterations, and c1 = 0.
5. Determine the exploration phase and development phase: In the formula, Indicates the position of male snakes; represents the random snake position; c2 is a constant, c2=0.05; rand∈[0,1] is a random number, x min and x max is the upper and lower bounds of the problem, A m It represents the ability of male snakes to search for food, and its formula is as follows: In the formula, yes applicability; yes applicability; S32: Close to food, when Q>0.25, temp>0.6: x i,j (t+1)=x food ±c3×temp×rand×(x food ―x i,j (t)) In the formula, x i,j represents the entire snake population, x food Indicates the global optimal position, c3 = 2; S33: Combat, the formula is as follows: In the formula, It is the optimal position for female snakes; It is the female snake position; This is the optimal position for male snakes; It is the male snake position; Indicates the updated position of the male snake. Indicates the updated position of the female snake; FF is the combat value of the female snake, and FM is the combat value of the male snake; The FM and FF formulas are as follows: In the formula, For the best female snake position, is the position of the best male snake, f i It is the proxy position of the snake; S34: Mating, the formula is as follows: Where M m and M f are the mating abilities of male and female snakes, respectively, and the formula is as follows: In the formula, Represents a male snake search agent, Represents a search agent for female snakes; In the formula, Indicates the worst position in the male snake group. Indicates the worst position in the female snake group, represents the maximum position boundary value in the male snake group, represents the minimum position boundary value in the male snake group, represents the maximum position boundary value in the female snake group, represents the minimum value of the position boundary in the female snake group; S35: Terminate and determine whether the maximum number of iterations has been reached; if so, output the optimal snake individual, that is, the optimal initial weight and threshold; otherwise, return to S33 to continue iterating.
8. A multi-level photovoltaic grid-connected converter model predictive control system based on SO optimized neural network, characterized in that: A model predictive control method for a multi-level photovoltaic grid-connected converter based on a SO optimized neural network as described in claims 1 to 7, comprising: The simulation module is used to build a simulation model with MMC as the main circuit and MPC as the control circuit; The preprocessing module is used to collect data through the simulation platform and perform data preprocessing; the training optimization module is used to input the data processed by the preprocessing module into the neural network in the simulation model for training and testing to obtain the neural network-MPC controller, and obtain the initial weights and thresholds of the neural network through training; An optimization module is used to optimize the initial weights and thresholds of the neural network using the SO algorithm to obtain a SO-neural network-MPC controller; The output module is used to simulate the model predictive control of the multi-level photovoltaic grid-connected converter through the SO-neural network-MPC controller and output reasonable sub-module quantity prediction value data.
9. A computer-readable storage medium, the computer-readable storage medium comprising a stored program, wherein: When the program is running, the device where the computer-readable storage medium is located is controlled to execute the model predictive control method of a multi-level photovoltaic grid-connected converter based on SO optimized neural network as described in any one of claims 1 to 7.
10. A processor, characterized in that: The processor is used to run a program, wherein when the program is run, the model predictive control method for a multi-level photovoltaic grid-connected converter based on a SO optimized neural network as described in any one of claims 1 to 7 is executed.
Citation Information
Cited By
Quality data asset value quantitative evaluation method
CN120821719A
