A New Wind Speed Evaluation Method Based on Pinball Loss and Winkler Score

Abstract

To reduce the adverse effects of the inherent stochastic volatility and uncontrollability of new energy on wind energy forecasting, this paper starting from the two aspects of improving the deterministic forecast and enhancing the predictability of volatility risk, the combination of variational modal decomposition (VMD), neural network and statistical model is applied to point forecasting, and the forecast model selection is based on the statistical characteristics of the components to enhance the degree of preciseness of wind speed forecasting. Then Monte Carlo-Markov Chain (MCMC) stimulation based on different quantiles is proposed to make interval prediction, and a new interval evaluation method is introduced, pinball loss function and Winkler score, to select the best interval prediction results for achieving precise control of wind power within a certain period of time. Finally, through experimental case verification, the performance of the advanced hybrid deterministic forecasting model is more advantageous than that of the traditional model. At the same time, the proposed interval prediction method better quantifies the uncertainty risk of wind power, makes up for the lack of a single evaluation method in the current interval prediction research, and can provide information support for the stable operation.

Keywords

• wind speed forecasting
• variational modal decomposition
• neural network
• monte carlo-markov chain stimulation
• winkler score