Hour-ahead wind power prediction for power systems using Hidden Markov Models and Viterbi Algorithm

Abstract

This paper presents a new stochastic method for very short-term (1 hour) wind prediction in electrical power systems. The method utilizes Hidden Markov Models (HMM) and the Viterbi Algorithm (VA). Past wind farm power production data are required to develop the HMM model. The accuracy of the predictions improves drastically if hourly weather forecast data are used as pseudo-measurements. Computer simulations using Northwestern weather recordings from the Bonneville Power Administration (BPA) website show good correlation between our predictions and the actual data.