A combined model of wavelet and neural network for short term load forecasting

Considering the importance of the peak load to the dispatching and management of the system, the error of peak load is proposed in this paper as criteria to evaluate the effect of the forecasting mode. And a new model is proposed which combining the wavelet analysis and neural networks for electric load forecasting. Using wavelet multi-resolution analysis, the load serial is decomposed to different sub-serials, which show the different frequency characteristics of the load. Then an artificial neural network is constructed to predict each sub-serial according to its characteristics. An improved L-M algorithm is used to accelerate the training of neural network and to improve the stability of the convergence. The forecasting result is achieved by reconstructing all predicted results of sub-serials together. A marked improvement has been observed by testing the model in a practical system. Especially, the error of peak load also has been reduced remarkably.