The research of short-term load forecast based on wavelet neural network

Analyzing the deficiency of the traditional BP algorithm, especially forecasting the peak values, the paper presnts a new methods of short-term load forecast of power system based on wavelet multiresolution analysis and neural networks. Through wavelet multiresolution analysis, load series are decomposed into series of different frequency characteristic, choosing appropriate wavelet base and decomposed levels, the decomposed load vectors exhibit stronger regularity. According to the character of each vector adopting different neural network models to predict each vector using L-M algorithm quicken the train and improve stability. By wavelet reconstructed, we can get full load forecasting. For some area power supply load of power corporation in Province Si Chuan, a short-term load forecast is simulated based on three methods namely back propagation algorithm, L-M optimizing BP algorithm and wavelet neural network algorithm. Comparing the simulation results, it shows the wavelet transform neural network algorithm has better forecast precision and adaptive capacity.