Short-term wind speed prediction in wind farms based on banks of support vector machines

Wind speed prediction is a key point in the management of wind farms because it is directly related to the power produced by each of a farm's turbines. Wind speed prediction is usually one of the most important tasks in wind farming, and companies that manage these farms invest large amounts of money to improve their prediction systems. In this paper, we propose an improvement to an existing wind speed prediction system, using banks of regression Support Vector Machines (SVMr) for a final regression step in the system. Several novel SVMr structures are proposed in this paper to manage the diversity in input data arising from the use of different global forecasting models and several parameterizations of a mesoscale model, included in the basic version of the prediction system. We show that the system implementing SVMr banks outperforms the basic system without taking into account diversity in the input data. It also performs better than a similar system using banks of multi-layer perceptrons. All the tests are carried out using real data from several wind turbines on a wind farm in southeast Spain.