Intelligent buck converter control method based on Kriging agent model assisted genetic algorithm

Abstract

The invention provides a buck converter intelligent control method based on a Kriging agent model assisted genetic algorithm. The buck converter intelligent control method comprises the steps of 1, collecting historical data of a power grid; step 2, constructing a Kriging agent model; 3, evaluating the global optimality of the parameter combination; step 4, using a genetic algorithm to realize directional search of a parameter space; 5, updating the Kriging agent model; step 6, substituting the updated Kriging agent model into a genetic algorithm to realize a closed-loop feedback mechanism; 7, iteratively executing the steps 4-6, and finding out the minimum extreme point; 8, controller parameters corresponding to the optimal individual and fitness function values under the controller parameters are obtained; and step 9, applying the controller parameters corresponding to the optimal fitness to an actual buck converter system. According to the invention, the gradient penalty function is added in the genetic algorithm, the generalization ability of the updated proxy model to the fluctuation of the input voltage is enhanced, and the robustness of the system is significantly improved.

Description

Technical Field

[0001] The invention belongs to the technical field of power electronic converters, and in particular relates to an intelligent control method for a buck converter based on a Kriging agent model assisted genetic algorithm. Background Art

[0002] Surrogate models, also known as approximate models, are a computational method widely used in modeling, simulation, and optimization of complex systems. They use a simplified model to approximate a more complex or difficult to calculate directly system, thereby reducing computational cost and time. The Kriging surrogate model constructs an approximate mapping relationship between controller parameters and performance indicators through Gaussian process regression, replacing time-consuming electromagnetic transient simulation, and is particularly suitable for situations where computational cost is high, parameters are numerous, or system behavior is not easy to model directly. Genetic algorithms perform parameter search based on surrogate models, and gradually approach the optimal solution through selection, crossover, and mutation operations. Buck converter, also known as step-down converter, is a single-tube non-isolated DC converter with an output voltage less than the input voltage. Traditional Buck circuit simulation usually requires a lot of computing resources and time, especially when optimizing the design, the computational cost is high. In traditional methods, the surrogate model is only used to replace the circuit simulation calculation, and the optimization algorithm is used to mechanically call the surrogate model output, and the two lack closed-loop feedback and dynamic coordination. There are several main problems with this method:

[0003] The proxy model cannot be dynamically updated based on new samples from the genetic algorithm, and the model accuracy gradually degrades with iterations. This fragmented collaboration mode makes it difficult to balance optimization efficiency and global convergence.

[0004] The genetic algorithm cannot use the prediction variance provided by the surrogate model to adjust the search strategy during the iteration process, resulting in the potential optimal solution in the high uncertainty area being ignored.

[0005] Rigid constraint processing: Traditional methods use fixed penalty coefficients to process constraints (such as kp range restrictions or non-negativity requirements for performance indicators), and are unable to dynamically adjust the penalty intensity, causing the optimization results to deviate from the actual engineering feasibility.

[0006] In order to solve the above problems, researchers began to explore the control strategy of dynamically optimizing energy storage converters by combining surrogate models and algorithms. Among them, the use of surrogate models for models that are complex, computationally expensive, or difficult to evaluate can significantly shorten the optimization time, and the combination with evolutionary algorithms has been widely studied due to its advantages in solving complex optimization problems. Evolutionary algorithms can find approximate optimal solutions in multi-objective optimization problems by simulating natural selection and genetic mechanisms. The use of surrogate models avoids the need for computational complexity of complex physical models or detailed simulation models, which significantly reduces the computational complexity of each optimization iteration while maintaining sufficient accuracy. However, there are also some challenges when combining evolutionary algorithms with surrogate models for buck converter optimization: 1) Cost of acquiring high-fidelity data: Proxy models rely on high-quality simulation or experimental data, but the high-frequency switching characteristics of the buck converter require fine-mesh simulation (such as finite element analysis), and data generation is time-consuming, which conflicts with the original intention of reducing computational costs.

[0007] 2) Trade-off in computational efficiency: When it comes to multivariable optimization such as controller parameters (such as PI gain) and circuit component parameters (inductance, capacitance), the generalization ability of the proxy model in high-dimensional space is significantly reduced, and it is necessary to balance the model complexity and training cost.

[0008] 3) Limitations of evolutionary algorithms: Traditional evolutionary algorithms may converge to local optimality too early, especially in multi-modal or high-dimensional problems.

[0009] In summary, the present invention combines the agent model and the genetic algorithm to propose a buck converter intelligent control method based on the Kriging agent model assisted genetic algorithm. This method not only shortens the optimization time and improves the calculation efficiency, but also provides an efficient and accurate control optimization method, reduces the calculation burden, and is of great significance for promoting the development of power systems. Summary of the invention

[0010] Purpose of the invention: In view of the shortcomings of the prior art, the present invention provides a buck converter intelligent control method based on Kriging agent model assisted genetic algorithm. It includes the following steps: Step 1, using the optimal Latin hypercube method to collect historical data of the power grid to form a training set of the model, and preprocessing the historical data of the power grid; Step 2: construct a Kriging proxy model and fit the mapping relationship between controller parameters and performance indicators through Gaussian process regression; Step 3, designing the genetic algorithm fitness function according to the performance index of the Kriging proxy model fitting; Step 4: embed the gradient penalty function in the genetic algorithm, dynamically adjust the crossover rate and mutation rate according to the uncertainty quantification result of the Kriging proxy model, realize the directional search of the parameter space, and obtain the improved genetic algorithm; Step 5, iteratively generate new parameters and performance indicators based on the improved genetic algorithm, add them to the training set of the model, and update the Kriging proxy model; Step 6, using the updated Kriging model combined with the genetic algorithm to implement a closed-loop feedback mechanism for the entire system; Step 7, iteratively execute steps 4 to 6 until the termination condition is met and the minimum extreme point is found; Step 8, obtaining controller parameters corresponding to the optimal individual and the fitness function value under the controller parameters; Step 9: Apply the controller parameters corresponding to the optimal fitness to the actual buck converter system.

[0011] Step 1 includes: collecting historical grid data through smart meters, sensors, phasor measurement units and grid management systems, and performing normalization processing; the historical grid data includes input parameters kp and ki of the buck converter, and output performance indicators, and the output performance indicators include tracking time st, overshoot ov and steady-state error sse.

[0012] Step 2 includes: Step 2.1, define the Kriging proxy model architecture: the parameters kp and ki preprocessed in step 1 constitute the input training set S. The input layer of the Kriging proxy model is used to receive the input training set S. The output layer is used to predict the tracking time st, overshoot ov, and steady-state error sse. The predicted tracking time st, overshoot ov, and steady-state error sse constitute the input training set Y. The Kriging proxy model of the Gaussian process regression framework is selected, and the nonlinear relationship of the parameter space is characterized by the covariance function. Step 2.2, model training: call the dacefit function of the MATLAB DACE toolbox, input the training set S and the training set Y, set the linear regression function for the indicators ov and sse, set the quadratic polynomial regression function for the indicator st, set the Gaussian exponential model for the indicators ov and sse, set the exponential model for the indicator st, and verify the accuracy of the model through the mean square error and root mean square error.

[0013] Step 3 includes: fitness function Defined as: , in is the penalty coefficient.

[0014] Step 4 includes: Step 4.1, set the initial parameters of the genetic algorithm, including population size, mutation rate, crossover rate and number of evolutions, and substitute the performance index of the Kriging proxy model into the fitness function defined in step 3; Step 4.2, design the gradient penalty function, define the constraints: the ranges of parameters kp and ki are [0.0001, 0.01], [100, 200] respectively, the performance indicators need to satisfy st>0, sse>0, ov>0, calculate the penalty value to check whether each set of predicted data meets the constraints, and calculate the penalty value according to the degree of violation of the constraints by the predicted results. The calculation formula is: , in is the penalty coefficient, which is determined through sensitivity analysis; Step 4.3, update the fitness function value: fitness=fitness+penalty; Step 4.4, use uncertainty-driven genetic algorithm for dynamic adjustment: the population enters iterative optimization, and if the following formula is satisfied, the indicator is determined to be in the high uncertainty area: ; in is the prediction variance of ov, is the prediction variance of sse, is the prediction variance of st, calculated by the Kriging proxy model, the mutation rate Pm increases from 0.8 to 0.9, the crossover probability Pc decreases from 0.8 to 0.05, and high potential individuals are retained to enhance the exploration ability; If not, the indicator is judged to be in the low uncertainty region, the mutation rate is 0.01, the crossover rate is 0.2, the population diversity is retained, the mutation rate and crossover rate are dynamically updated, and an adaptive balance between exploration and utilization is achieved; Step 4.5, the initialized population is put into selection, crossover and mutation operations to iteratively search for the best result; Step 4.6, the population enters the selection operation, using the roulette method, based on the fitness ratio selection strategy, to select individuals with higher fitness to form a new population, and the selection probability of each individual is inversely proportional to its fitness value; Step 4.7, the population enters the crossover operation, and the real number crossover method is used to perform a crossover operation on the selected individuals to generate new individuals. The crossover points are randomly selected, and the individuals after the crossover form a new population.

[0015] Step 5 includes: Step 5.1, after the improved genetic algorithm iterates for 2 generations, it regularly calls the real data collected by the buck circuit simulation, integrates the new parameters kp, ki and performance indicators tracking time, overshoot and steady-state error into the new sample buffer, and ensures that the model update is based on the actual new data; Step 5.2, normalize the new parameters kp, ki and performance indicators obtained in step 5.1, use the new data to train the Kriging proxy model, and obtain an updated Kriging proxy model to maintain the timeliness and accuracy of the model. This realizes the dynamic coordination of data generation, model optimization, and parameter search.

[0016] Step 6 includes the following steps: Step 6.1, fit the performance index through the updated Kriging proxy model, and bring it into the fitness function to obtain the fitness function value, and divide the high uncertainty area and low uncertainty area according to the prediction variance of different performance indicators provided by the updated Kriging proxy model; Step 6.2, combining the high uncertainty area and the low uncertainty area, calling the improved genetic algorithm for directional search, dynamically updating the mutation rate and crossover rate, and realizing the dynamic coordination mechanism of the Kriging agent model and the genetic algorithm.

[0017] The updated model is improved through the above steps, and the closed-loop feedback mechanism of the whole system is realized, forming a dynamic coordination of model optimization, parameter search and data generation.

[0018] Step 7 includes: Use the MATLAB min function to find the extreme point of the fitness function obtained in step 4. The specific implementation code is: , Among them, fit is the extreme value, and id is the position order of the extreme value.

[0019] The present invention also provides an energy storage converter intelligent control optimization device implemented by the method, comprising: a data acquisition module for randomly collecting historical power grid data; an agent model module for fitting the relationship between energy storage converter controller parameters and performance indicators; a weight coefficient setting module for quantitatively setting the weight coefficients between various performance indicators; a fitness function construction module for constructing the fitness function of the energy storage converter system; and an optimization algorithm module for executing an improved genetic algorithm.

[0020] The present invention also provides an electronic device, comprising a processor and a memory, wherein the memory stores program code, and when the program code is executed by the processor, the processor executes the steps of the described method.

[0021] The present invention has the following technical features: 1) Kriging agent model and genetic algorithm two-way closed-loop regulation: According to the characteristics of the proxy model and genetic algorithm, this solution realizes the dynamic interaction between the two, which drives each other, and forms a closed-loop system of model optimization and parameter search, which is different from the traditional method in which the proxy model only statically replaces the simulation. The genetic algorithm dynamically adjusts the mutation rate and crossover rate of the genetic algorithm according to the prediction results of the proxy model, and feeds back the newly generated high-value samples to the Kriging proxy model in real time for incremental training.

[0022] 2) Dynamic constraint mechanism of gradient penalty function: formula The introduction of the penalty coefficient is optimized through sensitivity analysis , and , adjust the degree of constraint violation. Force the population to evolve towards the feasible domain while retaining the potential optimal solution on the boundary, avoiding overfitting or underconstraint problems caused by fixed penalties.

[0023] 3) Dynamic feedback parameter search: The genetic algorithm dynamically divides the exploration and utilization interval according to the prediction variance provided by the Kriging proxy model: in the high variance interval, the population diversity is increased to generate new samples and improve the local accuracy of the proxy model; in the low variance interval, the local search is focused to accelerate convergence.

[0024] In summary, the present invention provides an effective intelligent control optimization strategy for buck converters by combining the agent model and the genetic algorithm. This strategy not only shortens the optimization time, but also can perform control optimization efficiently and accurately, while reducing the computational burden and avoiding the problem of the optimization process falling into the local optimum.

[0025] Beneficial effects: The present invention combines the dynamic synergy advantages of the Kriging agent model and the genetic algorithm, and provides an efficient intelligent control optimization method for a buck converter. First, based on the high-precision fitting of the nonlinear dynamic characteristics of the circuit by the Kriging agent model, the minute-level parameter evaluation time required for the traditional simulation is compressed to the second level, and the global parallel search capability of the genetic algorithm is combined to significantly improve the design iteration speed. Secondly, the agent model is equivalently fitted to the buck converter, and combined with the genetic algorithm to ensure that the algorithm accurately captures the coupling relationship between the controller parameters and the dynamic performance (overshoot, steady-state error) while reducing the consumption of computing resources. Through the closed-loop feedback mechanism, the Kriging agent model is incrementally updated with the genetic algorithm iteration, and the training set is updated after 10 iterations to avoid the accuracy degradation problem of the traditional static agent model. On the other hand, the Kriging agent model is used to provide uncertain information to guide the genetic algorithm to search, and the crossover rate and mutation rate of the genetic algorithm are dynamically updated to increase the diversity of the population, so as to avoid the traditional algorithm from missing the potential optimal solution. Finally, the gradient penalty function is added to the genetic algorithm to enhance the generalization ability of the updated agent model to the input voltage fluctuation, which significantly improves the robustness of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a flow chart of a method according to an embodiment of the present invention.

[0027] Figure 2 1 is a control system optimization diagram of a buck converter in an embodiment of the present invention.

[0028] Figure 3 It is a comparison chart of the predictions of the tracking time predicted by the proxy model.

[0029] Figure 4 It is a prediction comparison chart of the steady-state error predicted by the surrogate model.

[0030] Figure 5 It is a comparison chart of the predictions of the surrogate model predicting overshoot.

[0031] Figure 6 It is a schematic diagram comparing the simulation output voltage tracking of the method of the present invention and the traditional method.

[0032] Figure 7 It is a comparison chart of the running time of the real-time simulation model and the proxy model after 10 iterations. DETAILED DESCRIPTION

[0033] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments, and the above and / or other advantages of the present invention will become more clear.

[0034] like Figure 1 As shown, this embodiment provides a buck converter intelligent control method based on Kriging agent model assisted genetic algorithm, comprising the following steps: Step 1: Use the optimal Latin hypercube method to collect historical power grid data and perform preprocessing, which specifically includes the following steps: Step 1.1, use the optimal Latin hypercube method to collect historical data of the power grid and put the collected data into an Excel table, including fitting various combinations of energy storage converter controller parameters kp and ki, where kp is the gain coefficient of the proportional control term, which determines the instantaneous response strength of the controller to the error, kp mainly affects the dynamic response speed of the system, and ki eliminates the steady-state error through integral action, and the corresponding performance indicators time (st), overshoot (ov), and steady-state error (sse). The selection of kp and ki is random, the range of kp is [0.0001, 0.01], and the approximate range of ki is [100, 200], a total of 800 sets of data, and the data are normalized; Step 2, constructing a Kriging proxy model with two inputs and three outputs, using the data from step 1 to fit the relationship between the buck converter parameters and the performance indicators, specifically including the following steps: Step 2.1, using MATLAB software to construct a Kriging proxy model in code to fit the relationship between the buck converter parameters and performance indicators, including: kp, ki and st; kp, ki, and ov; kp, ki and sse; Step 2.2: This example uses the Kriging proxy model in the DACE toolbox that comes with MATLAB. You can call it using the relevant code. The specific formula is as follows: , Where S represents the sample data matrix of the input variables, which is the sample matrix composed of kp and ki here, Y represents the sample data matrix of the output variables, which is the matrix composed of performance indicators here, regr represents the handle of the regression polynomial function, which is used to fit the regression relationship between the input variables and the response variables, corr represents the handle of the correlation function, which is used to describe the correlation between the input variables, theta represents the parameter vector of the correlation function, which is used to adjust the shape and range of the correlation function, lob represents the lower bound vector of the correlation function parameters, upb represents the upper bound vector of the correlation function parameters, dmodel represents the fitted DACE model, which can be used for prediction and interpolation, and perf represents the performance indicator of the fitted model; This embodiment includes two parameters and three performance indicators, so a Kriging proxy model with two inputs and three outputs is designed; The fitting training sets 80% of the total data for training and 20% for testing. The predicted value of the Kriging proxy model is: , in (x) represents the mean of the random process, It is represented as a Gaussian process with a mean of zero, x is the input variable, represents the predicted value of the Kriging proxy model; Step 2.3, set the model parameters, S is the vector composed of kp and ki, Y is the vector composed of sse, st, and ov, theta = [10 10], lob = [0.0001 100], upb = [0.1 200]; Step 2.4, when performing the controller parameter and ov, sse, st fitting training, Set S to be the vector composed of kp and ki; Y is the vector composed of ov, sse, and st; ov and sse are @regpoly1: linear regression; st uses @regpoly2: quadratic polynomial regression; ov and sse are @corrgauss: Gaussian model; st is @correxp: exponential model; In step 2.5, 160 sets of independent test set data are used to verify the prediction accuracy of the proxy model.

[0035] Step 2.6, predict the output of the test set, and judge the fitting effect of the trained Kriging proxy model through specific evaluation indicators, such as mean absolute error, mean square error, root mean square error and determination coefficient. In this example, mean square error and root mean square error are used as evaluation indicators. Their calculation formulas are as follows: , , Where y1 represents the true value, It represents the predicted value, MSE represents the mean square error, and RMSE represents the root mean square error, as shown in Table 1 below, which shows the comparison of the root mean square error of the Kriging proxy model and the neural network prediction index.

[0036] Table 1 Root mean square error of the Kriging proxy model Root Mean Square Error of Neural Network Tracking time st: 0.000160 Tracking time st: 0.00004967 Overshoot sse: 0.005370 Overshoot sse: 0.0082589 Steady-state error ov: 0.025180 Steady-state error ov: 0.02259 In this example, the MSE and RMSE of the three prediction models are less than 0.001; The three performance indicators in this example are as described in step 1.1. Figure 3 It is a comparison chart of the predicted value and the true value of st. Figure 4 This is a comparison chart of sse predicted values ​​and true values. Figure 5 This is a comparison chart of ov predicted values ​​and true values. The comparison shows that the prediction of the proxy model is close enough to the actual simulation results. Step 3: Since the existing 800 sets of data cannot fully represent the coupling between the buck converter parameters and the performance indicators, more data is needed. Therefore, the relationship between the buck converter parameters and the performance indicators is fitted according to the existing proxy model, the weight coefficients between the performance indicators are quantitatively set, and the fitness function is constructed in the form of: , k1 is the weight coefficient of the performance index ov, k2 is the weight coefficient of the performance index sse, and k3 is the weight coefficient of the performance index st, where k1=50, k=50, k3=50; Step 4: embed the gradient penalty function in the genetic algorithm, predict the system parameters and prediction variance at different controllers through the Kriging proxy model, dynamically adjust the crossover rate and mutation rate, and realize the directional search of the parameter space, which specifically includes the following steps: Step 4.1, set the parameters of the genetic algorithm and initialize them, set the population size to 40, the mutation probability Pm to 0.8, the crossover probability Pc to 0.05, the number of evolutions to 10, and predict the performance indicators st, ov, and sse of the system under the proportional-integral controller parameter combination by fitting the trained Kriging agent model; Step 4.2, design the gradient penalty function and define the constraints: the ranges of parameters kp and ki are 0.0001, 0.01, 100, 200 respectively, and the performance indicators st>0, sse>0, ov>0. Through the orthogonal experimental method, the sensitivity of each constraint item is analyzed, and the penalty coefficient β=120 (corresponding to the constraint weight of the overshoot ov), γ=80 (corresponding to the constraint weight of the steady-state error sse), and δ=60 (corresponding to the constraint weight of the tracking time st) are determined. The penalty value calculation formula is: , in are penalty coefficients, which are determined through sensitivity analysis and are 120, 80, and 60 respectively.

[0037] Step 4.3, update the fitness function value: Fitness=fitness+penalty; fitness is the fitness function calculated in step 3, penalty is the penalty value calculated in step 4.2; Step 4.4, use the Kriging proxy model to fit the prediction variance provided by different indicators, divide the population into high and low uncertainty areas, and determine that the indicator is in the high uncertainty area if the following formula is satisfied: , is the predicted variance of x, calculated by the Kriging proxy model, the mutation rate Pm increases from 0.8 to 0.9, the crossover probability Pc decreases from 0.8 to 0.05, and high-potential individuals are retained to enhance exploration ability; if it is not satisfied, in the low uncertainty area, the mutation rate is 0.01, the crossover rate is 0.2, the population diversity is retained, the mutation rate and crossover rate are dynamically updated, and the mutation rate and crossover rate are dynamically updated; Step 4.5, the 40 initialized populations are subjected to selection, crossover and mutation operations to iteratively search for the best result; Step 4.6, the population enters the selection operation, using the roulette method, based on the fitness ratio selection strategy, to select individuals with higher fitness to form a new population, and the selection probability of each individual is inversely proportional to its fitness value; Step 4.7, the population enters the crossover operation, using the real number crossover method to perform a crossover operation on the selected individuals to generate new individuals. The crossover points are randomly selected, and the individuals after the crossover form a new population; Step 5: Based on the improved genetic algorithm in step 4, new parameters and performance indicators are iteratively generated, added to the training set of the model, and the Kriging proxy model is updated.

[0038] Step 5.1, first combine the improved genetic algorithm obtained in step 4 to iterate 2 generations, then regularly call the real data collected by the buck circuit simulation, integrate the new parameters and performance indicators into the new sample buffer, and ensure that the model update is based on the actual new data.

[0039] Step 5.2, normalize the new parameters kp, ki and performance indicators (time, overshoot and steady-state error) obtained in step 5.1, use the new data to update the Kriging proxy model, and obtain the updated proxy model to maintain the timeliness and accuracy of the model. This realizes the dynamic coordination of data generation, model optimization and parameter search.

[0040] Step 6: Using the updated proxy model obtained in step 5 and combining it with the genetic algorithm, the optimization results are fed back to the Kriging proxy model training set through a closed-loop feedback mechanism to form a dynamic collaboration of model optimization, parameter search, and data generation. Specifically, the steps include: Step 6.1, bring the performance index fitted by the proxy model updated in step 2.3 into the fitness function to obtain the fitness function value, and divide the high uncertainty area and the low uncertainty area according to the prediction variance of different performance indicators provided by the Kriging proxy model; Step 6.2, combining the high uncertainty area and low uncertainty area obtained in step 5.1, calls the genetic algorithm for directional search, dynamically updates the mutation rate and crossover rate, and realizes the dynamic coordination mechanism of the agent model and the genetic algorithm.

[0041] Step 7: Use the MATLAB min function to find the extreme point of the fitness function obtained in step 5. The specific implementation code is: , Among them, fit is the extreme value, id is the position of the extreme value, min is the function name, and Fitness is the predicted data set. Iterate steps 4 to 7 until the termination condition is met and the minimum extreme point is found, such as reaching the preset number of iterations or the fitness value converges. In each iteration, update the population and record the optimal individual and its fitness value; Step 8, result output, output the controller parameters corresponding to the optimal individual, and the fitness function value under the parameters, that is, the extreme point and extreme value of the system performance. In this example, a set of parameters is obtained, kp=0.03217078, ki=160.9565. The system voltage output under this set of controller parameters meets the requirements, and the next step can be performed. If it does not meet the requirements, repeat steps 4 to 7 until it meets the requirements. Step 9: Result verification and output: Apply the optimal controller parameters to the actual buck converter system. Figure 6 As shown in the figure, it represents the output voltage tracking comparison diagram. Figure 6 It can be seen from the results that the performance of this experiment is improved and according to the above steps, the method proposed by the present invention reduces the simulation time, such as Figure 7 As shown, it represents the time of substituting the real simulation and the proxy model into the genetic algorithm for 40 populations and iterating 10 times. It can be seen that the experimental time of the method proposed by the present invention is shorter than that of the traditional method.

[0042] like Figure 2 As shown, the control system optimization diagram consists of three parts: buck circuit, proportional-integral controller and genetic algorithm.

[0043] This embodiment also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the intelligent control method for a buck converter based on a Kriging agent model assisted genetic algorithm are implemented.

[0044] This embodiment also provides a computer device, including: Memory, used to store instructions.

[0045] The processor is used to execute the instructions so that the computer device executes the steps of the intelligent control method for a buck converter based on a Kriging agent model assisted genetic algorithm.

[0046] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0047] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0048] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0049] The present invention provides a buck converter intelligent control method based on a Kriging agent model assisted genetic algorithm. There are many methods and ways to implement the technical solution. The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the protection scope of the present invention. All components not specified in this embodiment can be implemented by existing technologies.

Claims

1. A buck converter intelligent control method based on Kriging agent model assisted genetic algorithm, characterized in that: The following steps are involved: Step 1, using the optimal Latin hypercube method to collect historical data of the power grid to form a training set of the model, and preprocessing the historical data of the power grid; Step 2: construct a Kriging proxy model and fit the mapping relationship between controller parameters and performance indicators through Gaussian process regression; Step 3, designing the fitness function of the genetic algorithm according to the performance index of the Kriging proxy model fitting; Step 4: embed the gradient penalty function in the genetic algorithm, dynamically adjust the crossover rate and mutation rate according to the uncertainty quantification result of the Kriging proxy model, realize the directional search of the parameter space, and obtain the improved genetic algorithm; Step 5, iteratively generate new parameters and performance indicators based on the improved genetic algorithm, add them to the training set of the model, and update the Kriging proxy model; Step 6, using the updated Kriging model combined with the genetic algorithm to implement a closed-loop feedback mechanism for the entire system; Step 7, iteratively execute steps 4 to 6 until the termination condition is met and the minimum extreme point is found; Step 8, obtaining controller parameters corresponding to the optimal individual and the fitness function value under the controller parameters; Step 9: Apply the controller parameters corresponding to the optimal fitness to the actual buck converter system.

2. The method according to claim 1, characterized in that Step 1 includes: collecting historical grid data through smart meters, sensors, phasor measurement units and grid management systems, and performing normalization processing; the historical grid data includes input parameters kp and ki of the buck converter, and output performance indicators, and the output performance indicators include tracking time st, overshoot ov and steady-state error sse.

3. The method according to claim 2, characterized in that Step 2 includes: Step 2.1, define the Kriging proxy model architecture: the parameters kp and ki preprocessed in step 1 constitute the input training set S. The input layer of the Kriging proxy model is used to receive the input training set S. The output layer is used to predict the tracking time st, overshoot ov, and steady-state error sse. The predicted tracking time st, overshoot ov, and steady-state error sse constitute the input training set Y. The Kriging proxy model of the Gaussian process regression framework is selected, and the nonlinear relationship of the parameter space is characterized by the covariance function. Step 2.2, model training: call the dacefit function of the MATLAB DACE toolbox, input the training set S and the training set Y, set the linear regression function for the indicators ov and sse, set the quadratic polynomial regression function for the indicator st, set the Gaussian exponential model for the indicators ov and sse, set the exponential model for the indicator st, and verify the accuracy of the model through the mean square error and root mean square error.

4. The method according to claim 3, characterized in that Step 3 includes: fitness function Defined as: ， in , and is the weight coefficient.

5. The method according to claim 4, characterized in that Step 4 includes: Step 4.1, set the initial parameters of the genetic algorithm, including population size, mutation rate, crossover rate and number of evolutions, and substitute the performance index of the Kriging proxy model into the fitness function defined in step 3; Step 4.2, design the gradient penalty function, define the constraints: the ranges of parameters kp and ki are [0.0001, 0.01], [100, 200] respectively, the performance indicators need to satisfy st>0, sse>0, ov>0, calculate the penalty value to check whether each set of predicted data meets the constraints, and calculate the penalty value according to the degree of violation of the constraints by the predicted results. The calculation formula is: ， in is the penalty coefficient; Step 4.3, update the fitness function value: fitness=fitness+penalty; Step 4.4, use uncertainty-driven genetic algorithm for dynamic adjustment: the population enters iterative optimization, and if the following formula is satisfied, the indicator is determined to be in the high uncertainty area: ; in is the prediction variance of ov, is the prediction variance of sse, is the prediction variance of st, calculated by the Kriging proxy model, the mutation rate Pm increases from 0.8 to 0.9, and the crossover probability Pc decreases from 0.8 to 0.05; If it is not satisfied, the judgment index is in the low uncertainty region, with a mutation rate of 0.01 and a crossover rate of 0.2; Step 4.5, the initialized population is put into selection, crossover and mutation operations to iteratively search for the best result; Step 4.6, the population enters the selection operation, using the roulette method, based on the fitness ratio selection strategy, to select individuals with high fitness to form a new population, and the selection probability of each individual is inversely proportional to the fitness value; Step 4.7, the population enters the crossover operation, and the real number crossover method is used to perform a crossover operation on the selected individuals to generate new individuals. The crossover points are randomly selected, and the individuals after the crossover form a new population.

6. The method according to claim 5, characterized in that Step 5 includes: Step 5.1, after the improved genetic algorithm iterates for 2 generations, it regularly calls the real data collected by the buck circuit simulation, and integrates the new parameters kp, ki and the performance indicators tracking time, overshoot and steady-state error into the new sample buffer; Step 5.2, normalize the new parameters kp, ki and performance indicators obtained in step 5.1, use the new data to train the Kriging proxy model, and obtain the updated Kriging proxy model.

7. The method according to claim 6, characterized in that Step 6 includes the following steps: Step 6.1, fit the performance index through the updated Kriging proxy model, and bring it into the fitness function to obtain the fitness function value, and divide the high uncertainty area and low uncertainty area according to the prediction variance of different performance indicators provided by the updated Kriging proxy model; Step 6.2, combining the high uncertainty area and the low uncertainty area, calling the improved genetic algorithm for directional search, dynamically updating the mutation rate and crossover rate, and realizing the dynamic coordination mechanism of the Kriging agent model and the genetic algorithm.

8. The method according to claim 7, characterized in that Step 7 includes: Use the MATLAB min function to find the extreme point of the fitness function obtained in step 4. The specific implementation code is: ， Among them, fit is the extreme value, and id is the position order of the extreme value.

9. An energy storage converter intelligent control optimization device implemented by the method described in any one of steps 1 to 8, characterized in that: include: Data acquisition module, used to randomly collect historical data of power grid; A proxy model module is used to fit the relationship between the controller parameters and performance indicators of the energy storage converter; The weight coefficient setting module is used to quantitatively set the weight coefficients between various performance indicators; the fitness function construction module is used to construct the fitness function of the energy storage converter system; and the optimization algorithm module is used to execute the improved genetic algorithm.

10. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory stores program codes, and when the program codes are executed by the processor, the processor executes the steps of the method according to any one of claims 1 to 8.

Images