Abstract: Based on the historical data of Chinese monthly electric energy demand,this paper researches the forecasting methods of seasonal increasing mid-long term electric indexes. It adopts discrete wavelet transform to decompose the sample series and reconstruct the decomposed results separately. After discarding the stochastic series,the long term increasing andfluctuant vectors are extended by RBF neural network. Adding the extended values together,it gets the forecasting results of electric energy demand. Empirical results show that after the decomposition by discrete wavelet transform,fluctuant trends of sample for RBF neural network are simplified. When it simulates the nonlinear trends,the generalization capability of RBF neural network is improved and the forecasting results are of good performance.

Key words: seasonal increasing trend, discrete wavelet transform, RBF neural network, generalization capability