The Research Of Load Forecasting In Huizhou Power Grid

Load forecasting has become an independent part of Energy Management System(EMS)，To electrical power system’s plan development, the electric power production and thedispatch, had the vital significance, the electrical network load forecasting technology alsoobtained more and more applications. the existing load forecasting cannot meet the practicalwork need completely, and need to further consummate in some aspects enhances.This paper describes the recent research and development on loadforecasting.Thecommon methods of short-term load forecasting(STLF)are described andanalyzed on strongpoint and short point.The costs of electricity is introduce in the short-termload forecasting by analyzing the constituent,characteristic of electric load and several aspectswhich will affect the accuracy of load forecasting.The application of artificial neural network(ANN)in load forecasting in power system isstudied and analysis.The forecast results with ANN may be more accurate than other methodsand come close to practical engineering,with a wider range of practical prospects. AlthoughANN used in load forecasting in the scientific research,but there are still many problems haveto be solved in the actual implementation.Such as the selection and processing of inputelements,and the determination of hidden layer nodes are varies.Another outline ofshortcoming of ANN is that the algorithm is easy to converge to the local optimal value.A novel algorithm which combine genetic algorithm with artificial neural networks isproposed to solve the seep up the convergence of ANN and prevent it from converging to alocal minimum value. The application of the grouped modeling method based on mix geneticalgorithm and artificial neural networks shows that,the forecasting model and input value areappropriate,and learning speed and optimization capability is better than the BP network.Theresults also denote that the daily average load forecasting error is of nearly1percentage pointlower and the largest forecast error has been greatly reduced compared with single modelingmethod.In the second period of improveing forecasting system，by increasing knowledge base、model base to improve software’s intelligence and auto-adapted, and increase self evaluationfunction of model to enables the software in each kind of situation,choses the correspondingmodel carries on the forecast,Thus Enhances software’s usability and may thepromotion,achieved the forecasting result rate of accuracy to be high forecast that thereference was good, reduced the hand regulation work load effect.