Research On Power Load Time Series Forecasting Methods Based On Chaotic Characteristic

Electric short-term load forecasting is a quite important work in the Power System. Since Power System is a huge system having a dynamic behaviour, with the development making it complicated day by day, especially with the marketization going into deep, the factors effecting load become various. To research short-term load forecasting methods adapted to the characteristic and development of electric load is an attentional question in the field.Chaotic and phase-space reconstruction theories are expounded at first, and three methods of computing the phase-space reconstruction parameters are proposed. The phase-space reconstruction leads chaotic theory into time series analysis to recover chaotic attracts in high dimension space and to expose the regularity of chaotic system.Next, to improve the forecasting precision , and to solve the key question on confirming the nearest point, after the study of short-term load forecasting, the chaotic forecasting model of the largest Lyapunov exponent and adding-weight one-rank local are introduced based on phase-space reconstruction. The simulated experiment and analysis are also processed by using actual electric load. Then, specifically for the lack of Euclid distance on the similarity measurement, in order to overcome the effect of excursion , noise etc and materialize every element in each vector taking different action on the similarity measurement, and to improve the forecasting model, improvement on Euclid distance and Multiple Correlation Coefficient adding-weight method on Euclid distance are proposed and taken simulated experiment. The experiment results have indicated the effects of these two improvements are good.Finally, based on the three cubed multinomial of electric load curve, a data smoothness expressions is deduced and carried on the original forecasting result to reduce the error of the forecasted data, It has proved that the expressions could improve the precision of load forecasting farther.