Markov switching autoregressive modeling of wind power forecast errors

Abstract

Forecast errors constitute the main hurdle to integrating variable renewable energies into electrical power systems. Errors are inherent to forecasting, although their magnitude varies significantly with respect to both the method adopted and the time horizon. Their dynamic and stochastic modeling is mandatory for power systems to efficiently balance out these errors. A Markov Switching Autoregressive – MS-AR – approach is proposed herein for wind power forecast errors. This particular model is able to identify weather regimes according to the forecast reliability. Such regimes are controlled by a Markov chain whose state – not directly observable – determines the AR model parameters. The statistical features of the data artificially generated by this model are very similar to those of the actual forecast error. This model is used to solve the optimal management of a storage associated with a wind farm. The resolution is performed by means of stochastic dynamic programming while comparing the proposed MS-AR approach with several other models. In this illustrative problem, a 15% reduction in operating costs is derived from a fine model of forecast errors.