Forecasting of Wind Power Generation with the Use of Artificial Neural Networks and Support Vector Regression Models

Abstract The stochastic character of wind power generation suggests limitations on the increased shares of wind energy in electricity systems and challenges market integration of wind power, mainly due to the fact that nowadays, new wind parks are set to cope with more dynamic pricing mechanisms. In this new environment, where advanced bidding strategies need to be adopted from wind power actors, the introduction of novel elements to support wind power generation and address the inherent impact of variability is thought to be a prerequisite. To this end, the current study expands the work of previous studies by examining different methods of prediction with regards to wind power forecasting. More specifically, both Artificial Neural Networks (ANNs) and Support Vector Regression (SVR) models are trained and tested on the basis of different prediction horizons, using as case study real wind speed and wind power generation measurements from a wind park operating in the Greek territory. Models are trained using an in-house forecasting tool, with results obtained reflecting the better fit of the SVR method overall, especially for time horizons longer than 6 hours ahead. At the same time, an effort is made in order to optimize prediction of wind power generation through the combination of both prediction approaches via clustering of prediction areas. This novel approach results in an improvement of the predictions obtained, despite the fact that the SVR method already performs sufficiently even for 24 hours ahead.