Research On Equivalent Modeling Of Microgrid Based On Optimized Neural Network

As a new form of power supply,the micro grid has received extensive attention.The micro grid has the advantages of low pollution,high power utilization,flexible installation and so on,which can make up for the shortcomings of the large power grid.With more and more micro grid connected to large grid,the stability and reliability of power supply are improved.Micro grid modeling research can be more comprehensive and in-depth study of micro grid.Neural networks have the ability to learn and generalize,and are widely used in system modeling.In this paper,a new method of micro-grid equivalent modeling based on optimized neural network is proposed,In this method,the micro grid is regarded as an integral external system,and the neural network is used to conduct the equivalent modeling of the grid connection points of the micro grid.The working principle and mathematical model of each distributed power supply and battery energy storage system in the micro grid are introduced to prepare for the equivalent modeling research of the micro grid.Aiming at the grid-connected modeling problem of micro grid,an equivalent modeling method of micro grid based on optimized echo state network(ESN)is proposed.Taking the micro grid as an integral external system and taking the current and power data of the grid-connected access point of the micro grid as the input and output of the ESN respectively,the grid-connected equivalent model of the micro grid based on the ESN is constructed.Since the initialization parameter of ESN no longer changes,lacking adaptability,it lead to the approaching ability of ESN is not optimal.In order to improve the accuracy of the modeling,the fireworks algorithm is used to optimize the parameters of the ESN.Fireworks algorithm establishes a mathematical model by simulating the explosion of fireworks,and selects the best individual by calculating individual fitness values.The fireworks algorithm has the advantages of explosiveness,instantaneousity,parallelism and scalability.In order to verify the fitting accuracy and generalization ability of the model,compare the output of the equivalent model with the grid-connected simulation measured data of the micro grid.Aiming at the grid-connected modeling problem of dc micro-grid,a grid-connected equivalent modeling method of dc micro-grid based on extreme learning machine(ELM)is proposed.Taking the micro grid as an integral external system and taking the voltage and power data of the grid-connected access point of the dc micro grid as the input and output of the ELM respectively,the grid-connected equivalent model of the dc micro grid based on the ELM is constructed.In the initialization process of ELM,input weight and hidden layer node bias are randomly set without any change later,which leads to the lack of adaptability in modeling and affects the accuracy of model.The shark optimization algorithm is used to optimize the initialization parameters of the ELM,so as to obtain the optimal initialized parameters and improve the modeling accuracy.Shark algorithm simulates the hunting process of sharks for optimization,and the smell size of smell particles guides the updating of shark positions.In order to verify the fitting accuracy and generalization ability of the model,compare the output of the equivalent model with the grid-connected simulation measured data of the micro grid.