Shor-term Load Forecasting Model Based On Least Square Support Vector Machine

The short-term load forecasting of e lectr ic syste m is one o f the important forsafety and econo my operatio n of the power system. Accurate short-term loadforecasting is adva ntageous to improving the secure and economic e ffect o f powersystem a nd improve supply qua lity. Thus, to find e ffective method has importantapplied value for maximum limit enhancing short-term forecast precision.Primarily, based on the background of electr ic load forecasting, theresearchstatus, and the s ignificance of research, the paper analyzes the characteristicsof electric load,and the non-linear relations hip between the electr ic load forecastingand the vario us affecting factors. This paper has expounded the periodic law of loadchanges and ame nded the abnor mal data in histor ical load,and norma lized the relatedele ments in load forecasting. Then Least Squares Support Vector Machine (LSSVM)which has ma ny advantages on solve nonlinear,high dime ns ion and other proble ms isproposed in this paper. Based on the research, we change the inequa lity constra ints toequa lity constraints, the training error square to slack var iables, and Least SquaresSupport Vector Machine is proposed, which greatly accelerate the solving process andneed less optimized parameters.Cons idering it’s cruc ia l of the model para meters to the prediction accuracy,weproposed the Partic le Swar m Optimization(PSO) to optimize the parameters inLSSVM,in order to improve the accuracy of the prediction. Partic le SwarmOptimization(PSO) algor ithm is easily trapped in the local optimum and appearedpremature converge nce.To solve this proble m,an improved PSO algor ithm isproposed.To avo id the local convergence appeared in the process ofoptimizatio nprediction model for short-term load forecasting,based on Improved PSOLSSVM(IPSO-LSSVM) is proposed，and the relative error and relative standarddeviation were used as its eva luatio n criteria. I mprove the LSSVM model pred ictio naccuracy.Fina lly, this paper uses the histor ica l load data of an area of guangdong provincein2010,respective ly based on LSSVM, PSO LSSVM, IPSO LSSVM model to forecastsimulation.The simulatio n results show that IPSO-LSSVM is better than LSSVM,and PSO-LSSVM model,whic h shows the effectiveness and super iority o f theIPSO-LSSVM,and also shows it has certain research va lue and socia l s ignificance.The IPSO-LSSVM model with good converge nce, a higher predictio n precis io n andfaster training speed.