Research Of The Variable Weighting Combination Forecasting Model For Power Load

Power load forecasting is the basis of planning and operation of electric power system. It is also an important part of electric power market performing. In the course of electric power marketing, the accuracy of power load forecasting is directly related to the interests of all parties. Currently, there are a lot of models for power load forecasting. Only one model cannot totally reflect the changing rules and information of power load. Combination model of load forecasting has been a new direction of developing research.The concept, characteristics of load forecasting and several common forecasting models are introduced in the paper. Linear combination forecasting models are given, and several methods of weighting are discussed. Simulation results show that accuracy of the combination forecasting model is higher than any one sole model. Then three variable weighting combination forecasting models, the combination forecasting model based on artificial neural network (ANN), the combination forecasting model based on support vector machine (SVM), and fuzzy variable weighting(FVW) combination forecasting model are presented and discussed. Simulation results show that the variable weighting combination forecasting models are effective.In order to improve the feasible of forecasting by neural network, a new method is presented in the paper. The forecasting results of the three common neural networks are combined. The method can avoid the risk that the accuracy of forecasting decreases too many when overfitting occurs in single neural network. In order to enhance the load forecasting precision, a variable weighting combination forecasting model based on five layers fuzzy neural networks (FNN) is proposed in the paper. The inputs of the FNN are error and change of the error, and the output is weight. By fusing the strong points of fuzzy logic and neural networks, the fuzzy reference and defuzzification of this model are realized by neural networks. Simulation results indicate that accuracy of the combination forecasting model based on FNN proposed in the paper is higher than any one sole model and the fuzzy variable weighting combination forecasting model. According to the defects of the selection input variables of neural network, a new method that only considers the load processed of chaos theory is presented in the paper. But the method that not considers the factors of affecting the load can't increase the forecasting accuracy exactly. A method that only considers handling factors is introduced in the paper. The results of these two methods are combined for prediction. Simulation results show that the combination of these two methods is feasible and effective.