Improvement Of PSO Algorithm Based On GSA And GSO And Its Application In Power Load Forecasting

Load forecasting is an important part of power system,and it is also the basis of power system's secure and stable operation.Accurate load forecasting can bring positive effect on the reasonable dispatching of power generation capacity and effective reduction of the cost of electricity generation.It is of great significance to carry out load forecasting,especially short-term load forecasting of electric power system.This paper establishes a serious of neural network forecasting models based on swarm intelligence optimization algorithms.Before the establishment of the model,the local mean decomposition(LMD)algorithm is applied to process load data,then glowworm swarm Optimization algorithm(GSO),gravitational search algorithm(GSA)and particle swarm optimization algorithm(PSO)are used to optimize the parameters of three kinds of neural networks:Wavelet Neural Network(WNN),?-support vector machine regression(SVR)and BP neural network(BPNN),finally the model realizes the accurate forecasting of power load of Queensland,Australia.The research is carried out in five steps:First step,the raw data is decomposed into a sequence of product functions and a residual component by LMD,then the highest frequency componentPF₁ is excluded from the original data,at last the data with noise removed is used for network training.Second step,GSO,GSA and PSO algorithms are used respectively to optimize parameters of the three kinds of neural networks above,and the procedure is aimed at improving prediction performance of these networks.Followed by it the trained neural networks are used to rolling forecast one day load of Queensland.Third step,in order to further improve the forecasting precision of the models,this article sums up the advantages of GSO,GSA,PSO algorithms respectively in global and local searching,combining GSO,GSA algorithms with PSO algorithm,four hybrid algorithms are proposed innovatively on the basis of complementary advantages,namelyGSOPSO₁,GSOPSO₂,GSAPSO₁,GSAPSO₂ and,then the four algorithms re-optimize parameters of the neural networks to carry out prediction process.Fourth step,given the parameter aa inGSOPSO₂ and GSAPSO₂ has great impact on their predictive performance,this paper uses grid searching method with step 0.05 to find the optimal aa,as a result their optimization performance get improved.Fifth step,the whole forecasting models used in the paper are tested in redundancy,and then combined with grey relational grade analysis and forecasting validity analysis,we get the overall order of the prediction effectiveness in all models.The experiment results show thatGSOPSO₁,GSOPSO₂,GSAPSO₁ andGSAPSO₂ can significantly improve the prediction performance,hybrid models based on the four algorithms can achieve really high prediction accuracy.