Wavelet-neural networks-PARIMA method for power system short term load forecasting

A wavelet-neural networks-PARIMA (Periodical Auto-Regressive Integral Moving Average) method is proposed based on the quasi-periodicity, nonstationarity, nonlinearity of the load of power systems and applied to the forecast of the power system short term load. Every kind of hidden periodicity and nonlinearity of the load can be extracted and separated by using the wavelet transform. There is a rule that decomposition data decrease doubly while scales increase doubly. According to this rule and the characteristics of wavelet decompositions can be build MLP neural network model. Each scale wavelet transform is modeled and forecasted by using the PARIMA and a MLP. The scale transform of the most scale of the original signal is modeled and forecasted by using the PARIMA model. Finally, these forecasts in scale domains is synthesized to the system load forecasts in scale domains is synthesized to the system load forecasts by using a RBF neural network. The results of a practical example shows that the proposed method is able to reveal the quasi-periodicity, nonstationarity and nonlinearity of the power system load.