An improved Markov chain model for hour-ahead wind speed prediction

Markov Chain (MC) models are widely used in wind speed and wind power prediction. Classification of wind data to construct MC states plays a key role in MC models but hasn´t been paid much attention to. This paper presents a Spectral-analysis-based K-means Clustering (SKC) method to classify wind data in a data set containing few variables. Experimental results show that clusters distribute more properly than both the traditional Equal-interval Classification (EC) method and the Spectral Clustering (SC) approach. Based on the SKC method, prediction by a MC Transition-Probability-Matrix (MC-TPM) performs better than the one based on an EC approach in terms of Root Mean Square Error (RMSE) and Mean Absolute Error (MAE). Moreover, the convergence property of transition probabilities has been discovered and proved, which points out the limitation of MC models.