The Short-term Load Forecasting Research Based On Frequency Domain Decomposition

The purpose of load forecasting is to provide reference for system analysis, orto make decisions to electric power system. The STLF (short term load forecasting)play an important role in planning the electric power production and dispatch,ensuring the power balance between supply and demand, maintaining the power gridsecurity, stability and economic.The basic task of STLF in the research is to reduce the error of forecasting, so asto improve the precision of result. The three essential questions in short-term loadforecasting are processing historical data, mining the rules of short-term load variation,and making a mathematic model to suit the local load. Around these main problems,the predecessors constantly study indepth. In this paper. In this paper author sums upthe traditional methods and intelligent methods of STLF, and carries out the researchconsidering the global cyclical and local probabilistic characteristics when the loadchanges.The periodic component of load change must be consider, in order to study theload cycle change rule. In the past, Fourier or Wavelet Transform method iscommonly used to decompose the load curve, so as to describe the periodiccomponent. To avoiding one problem the lost of the high frequency information usingFourier decomposition and the other problem of selecting wavelet basis,this paperadopts a new methods frequency domain decomposition EEMD. This paper firstlyproposes index of the accuracy evaluation, use the component characteristics todescribe the scope of normal load changes, identify the false data according to thecharacteristics of the high frequency component, and correct the bad data according tothe characteristics of the trend component.Considering the statistical regular of the same time’s load in different days, wefind it obeys the gaussian distribution. This paper introduces the generalized relevance vector machine regression (RVS) model and parameter identification algorithm. Inorder to check the model’s efficiency, this paper uses the Matlab software and theload historical data from the international EUNITE competition, trains a model forload forecasting, then use the new trained modle to forcast the next day’s load.Through analysis and evaluate to the forecasting model, we obtained a satisfactoryresults.The achievements in this paper can provide principle and practical guides todevelopment of the load forecasting intelligent system in the future.