Short-term Load Forecasting Based On Improved Genetic Algoritlum To Optimize Elman Neural Network

In today's high-speed economic development,electricity is an indispensable part of the development of various industries,plus the characteristics of electricity,such as the out-of-the-box storage.Therefore,research on short-term load forecasting is very necessary.In addition,for the power dispatch department,the dispatch of electrical energy depends on accurate load forecasting.Accurate short-term load forecasting affects the supply and demand relationship in the power market and can also improve the reliability of the power system operation.This article first makes a systematic analysis of the existing forecasting methods and briefly summarizes the advantages and disadvantages of the existing forecasting methods.In the existing prediction methods,Elman neural network has strong learning ability,fault tolerance,associative memory,and also has a strong ability to adapt to time-varying characteristics,which is very suitable for short-term load forecasting.However,in terms of learning rules,the Elman neural network uses the momentum gradient descent method,which will make the convergence speed slower during the entire learning process,resulting in too long learning time,and the unstable convergence process leads to the failure to achieve the ideal Output.Aiming at the problems existing in Elman neural network,this paper proposes a method based on improved genetic algorithm to optimize the initial weight and threshold of Elman neural network.The genetic algorithm of a single population is easy to converge earlier,and the phenomenon of prematurity occurs.In response to this phenomenon,this paper proposes a genetic algorithm for co-evolution ofmultiple populations to optimize the initial weights and thresholds in the Elman neural network.In specific applications,real-number coding is used to encode the network parameters,and crossover,selection,and mutation operations are used to perform genetic evolution operations.Finally,the optimal solution is obtained by decoding,and the optimal initial weight of the network is determined.3.Threshold parameters.In order to verify the feasibility of the method proposed in this paper,three different prediction methods were used in this paper,and the load prediction for a certain area was done,and simulation experiments were carried out on matlab.The simulation data demonstrates that the improved genetic algorithm optimized Elman model is superior to the traditional Elman neural network in prediction accuracy and efficiency,thus proving the feasibility and scientificity of the method proposed in this paper.