Intelligent Hybrid Wavelet Models for Short-Term Load Forecasting

A wavelet decomposition based load forecast approach is proposed for 24-h and 168-h ahead short-term load forecasting. The proposed approach is applied to and compared with representative load forecasting methods such as: time series in traditional approaches and RBF neural network and neuro-fuzzy forecaster in nontraditional approaches. The other forecasters, such as multiple linear regression (MLR), time series, feed forward neural network (FFNN), radial basis function neural network (RBFNN), clustering, and fuzzy inference neural network (FINN), reported in the literature are also compared with the present approach. The process of the proposed wavelet decomposition approach is that it first decomposes the historical load and weather variables into an approximate part associated with low frequencies and several detail parts associated with high frequencies components through the wavelet transform. The historical data are smoothened by deleting the high frequency components and fed as input to the proposed models for the prediction. A comparison of wavelet and non-wavelet based approaches shows the superiority of proposed wavelet based approach over non-wavelet methods for the same set of data of the same utility.