A hybrid wavelet-Kalman filter method for load forecasting

This paper presents a wavelet transform method for load forecasting. The stochastic nature of the wavelet coefficients for the daily load variation is studied by the decomposition scheme of multiresolution analysis (MRA). The study indicates that the stochastic process of the wavelet coefficients can be modeled as a random walk process. Therefore, the wavelet coefficients are modeled as the state variables of Kalman filters. The best estimation of the wavelet coefficients is obtained by the recursive Kalman filter algorithm. The predicted daily load is the inverse of the predicted wavelet coefficients. Based on the above procedure, two models (weather insensitive and sensitive models) are presented in this paper. Results from an actual system are also presented. (C) 2000 Elsevier Science S.A. All rights reserved.