Spatio‐temporal Markov chain model for very‐short‐term wind power forecasting

Abstract

Wind power forecasting (WPF) is crucial in helping schedule and trade wind power generation at various spatial and temporal scales. With increasing number of wind farms over a region, research focus of WPF methods has been recently moved onto exploring spatial correlation among wind farms to benefit forecasting. In this study, a spatio-temporal Markov chain model is proposed for very-short-term WPF by extending the traditional discrete-time Markov chain and incorporating off-site reference information to improve forecasting accuracy of regional wind farms. Not only are the transitions between the power output states of the target wind farm itself considered in the forecasting model, but also the transitions from the output states of reference wind farms to that of the target wind farm are introduced. The forecasting results derived from multiple spatio-temporal Markov chains regarding different reference wind farms over the same region are optimally weighted using sparse optimisation to generate forecasts of the target wind farm. The proposed method is validated by comparing with both local and spatio-temporal WPF methods, using a real-world dataset.