Study On Short-term Load Forecasting Based On Multi -theory Fusion

The Short-term Load Forecasting (STLF) plays an important role in powerindustry. Economic and reliable operation of power systems depends significantlyon the accuracy of the load forecasting. In a deregulated and competitive powermarket, it requires that the precision of STLF improved effectively and steadily. The main work and achievement about the research of the STLF model basedon the multi-theory in this paper is generalized as follows. Through the particular analysis of the study trend of STLF at home and abroad,the existent problems of all kinds of principle and method are analyzed in essence,and the necessity and significance of study work in this paper is explained. The load series of electric power systems is a complicated nonlinear time seriesand it possess chaotic characteristic, and several STLF models are proposed basedon the multi-theory. The testing results show that the proposed models can improveeffectively and steadily the precision of STLF. The Optimal Neighbor Points (ONP)Approach is creatively presented, and it can eliminate some false neighbor pointsthrough identifying exponential separating rate of time evolutional trajectory andcan improve the precision of load forecasting. The STLF model based on themulti-theory and the mechanism of the global dynamical behavior is first proposed.It can be ensured not only by the structure of the model, but also by the inherentperformance of dynamical behavior. The STLF model based on the Improved Chaotic Neural Networks (ICNN) isfirstly proposed in this paper, and the ICNN model is built based on modified Aiharachaotic neuron. The ICNN model possesses the sensitivity to the initial load valueand to the walking of the whole chaotic track. And it can characterize complicateddynamics behavior and has global searching optimal ability, and its forecastingperformance is preferable to other dynamic neural networks. The STLF approach isfurther proposed based on the fusion of ONPA and ICNN model, and it is the fusionof several innovative points, and the testing results show that the proposed modeland its algorithm can improve effectively and steadily the precision of STLF.The clustering analysis approach of electric load series based on improved datamining arithmetic is creatively proposed in the paper, and it proposed the notion ofthe variance of difference sequence. In the basis of clustering result of load series inthe summer, the correction model based on the distributed fuzzy neural network andthe correction model based on the distributed DGA-neural network are proposed.The testing results prove the proposed model and approach can improve effectivelyand stably the precision of STLF. This research acquires the effective progressionand practical significance in the prediction engineering.