Medium-term Load Forecast Based on Sequence Decomposition and Neural Network

For the accuracy of decomposition forecasting for non-stationary, nonlinear mid-term electricity demand load sequence, a singular spectrum analysis and neural network based decomposition forecasting method is proposed. With it taken into consideration the long-term trend and seasonal cyclical fluctuation of medium-term load sequence, the main periodic components of the sequence are determined with spectral analysis on the basis of sequence trend extraction and the singular spectrum analysis is introduced to filter the main periodic components of the sequence. Neural network models are established to carry out forecast for each sub-sequence. The results of each subsequence are superimposed to give the forecasting result of the electricity demand in the next year. Taking historical data in certain region as an example, the forecast result of the empirical mode decomposition-neural network method is compared with the method proposed in this paper, the results show that a more stationary and accurate result can be obtained with the proposed method for medium-term electricity demand forecasting.