Research On Short-term Load Forecasting Method Based On Neural Network

The power industry plays a vital role in the national industrial system.The smooth running of electrical power system is the lifeblood of the national economy.In power system management,the electric load forecasting is indispensable.Accurate load forecasting could provide important basis for the stable operation of power system,setting reasonable electricity price,electricity real-time scheduling.Especially in economic sphere,load forecasting plays significant role in rational allocation of resources,optimizing power generation plan,achieving the optimal social and economic benefits.However,With the rapid development of global economy,demand for electricity power is increasing quickly.Meanwhile,the load is affected by date,climate,market and policy,which greatly increases the difficulty of power load forecasting.This paper introduces the background of power load forecasting and its influence on economy,analyzes the influence factors in and introduced several basic models in load forecasting.This paper also introduces some frequently-used methods for load forecasting.A new combined method based on GRNN,Elman NN and LSSVM optimized by PSO is proposed.The main contents included in this paper are showed as follows.Introduces the necessity of data preprocessing,explains how to select valid training samples and normalize the samples.The empirical mode decomposition(EMD)method is used to eliminate part of noise components in the original data.The seasonal adjustment is used to remove the influence on load data made by temporal periodicity.All data are normalized before the prediction to improve the prediction accuracy.The basic principle of neural network is introduced,and the principle,advantages and disadvantages of three prediction methods are described in detail.In this paper,three forecasting models are built,they are the generalized regression neural network(GRNN),Elman neural network(ENN)and the least squares support vector machine(LSSVM)respectively.When predict the load date by LSSVM model,the choice ofhas great influence on the result,so the particle swarm optimization(PSO)is used to optimized the parameters in LSSVM model.Finally,the three prediction results are combined by simulated annealing algorithm,then the final result is obtained.In this paper,the forecasting model is used to predict the load data of NSW and QLD in Australia.The simulation results show that the prediction model has good prediction accuracy,and it has good ability of dealing with historical data and good generalization ability.In this paper,the prediction results of combined forecasting model and single forecasting models are compared.The results show that the prediction accuracy of the combined model is obviously less than that of the single model.