Research On Power Load Forecasting Based On Lifting Wavelet And Improved Neural Network

Power system short-term load forecasting is an important foundation for power system dispatch operation and development planning.It is of great significance to the stable operation of power systems and the formulation of power generation plans.The accuracy of short-term load prediction directly affects the economics,safety and reliability of power system operations.The key issue of short-term load forecasting is how to choose a reasonable forecasting method.Therefore,in order to improve the accuracy of power load forecasting,this paper establishes a short-term load forecasting model based on lifting wavelet and improved neural network.Elman Neural Network(ENN)is an intelligent method that has achieved good application results in the field of short-term load forecasting.However,during the load forecasting process of the Elman neural network,when the excitation function Sigmoid reaches saturation,the convergence speed of the network may be reduced,and the network may fall into a local minimum,resulting in poor training and learning effects of the Elman neural network.This paper uses Mexican Hat wavelet function instead of Sigmoid function as the transfer function of the hidden layer nodes of the network,which can make the network in a real saturation state,and realize the network learning and training well.Experiments show that the improved Elman Neural Network(IENN)improves the prediction accuracy compared to the traditional ENN.IENN is a network structure improvement based on BP neural network(Back Propagation Neural Network,BPNN).There are still inherent defects of BPNN slow convergence and easy to fall into local optimization.This paper uses improved particle swarm optimization(Chaos Generalized Particle Swarm Optimization,CGPSO)to optimize the training of IENN.The CGPSO algorithm is based on ant colony algorithm improved particle swarm optimization(Generalized Particle Swarm Optimization,GPSO),which selects some particles with poor fitness values,uses chaos theory to chaotically disturb them,and corrects the position of the particles to map again.In the optimization area,the traversal of chaos theory is applied to improve the accuracy of GPSO algorithm detailed search and enhance the global search capability.The application of CGPSO algorithm to optimize IENN is to use the initial parameters of the neural network as CGPSO particles,optimize the initial distribution of thresholds and weights,and when searching for a better position,use the improved Elman neural network for local optimization to further improve the convergence speed of the Elman neural network avoids the defects such as local optimization.The experiment proves that the CGPSO-IENN method improves the load prediction accuracy and has higher prediction accuracy.For short-term and ultra-short-term load data with strong randomness and trend characteristics,the lifting wavelet algorithm is used to decompose the load data at different scales,which reduces the redundancy of the original load data and the separation characteristics of the spectrum are very obvious and convenient Carrying out prediction alone and reconstructing the results can improve the accuracy of load prediction.Combining lifting wavelet and improved Elman neural network,a short-term load forecasting model(LWT-CGPSO-IENN)based on lifting wavelet and improved Elman neural network is proposed.The experimental results show that the prediction accuracy of the LWT-CGPSO-IENN model is improved by 2.7777% compared with the traditional ENN model,which significantly improves the prediction accuracy.Compared with the single neural network prediction model,it has better adaptability to meet short-term load forecasting requirements.