Synthetic High-Resolution Wind Data Generation Based on Markov Model

Abstract

We propose three types of Markov models to generate high-resolution data every five minutes. Experimental data was obtained from a monitor station in southwest China, as well as near rivers of varying land-river connectivity in the Hampshire Avon catchment in England. The comparison of the original high-resolution wind speed and the synthetic high-resolution data from three types of models shows that the statistical characteristics including mean value, autocorrelation, maximum value, minimum value, amplitude probability distribution and variance are satisfactorily reproduced. The amplitude probability density distribution of synthetic data aligns with the Weibull distribution to a good extent. We demonstrate that the Kullback-Leibler divergence of synthetic data from the duplex algorithm model is reduced by 16.7% and 28.6% compared to a second-order and a first-order Markov model, which significantly reduces information loss. The generalization of the duplex algorithm model shows errors are small and within acceptable limits. The result shows that the model generalizes well in some areas.