Wind power prediction with different artificial intelligence models

In this paper different prediction models based on methods of the artificial intelligence are studied for wind power prediction of single wind farms. The used methods are neural networks, mixture of experts, support vector machines and nearest neighbour search with a superior particle swarm optimization. We build for day-ahead prediction and for short-term prediction with a prediction horizon of one hour. As input variables for these prediction methods weather data of a numerical weather prediction model are used. For the short-term prediction with a prediction horizon of one hour we use additionally measured power values of the wind farms as model input. The performance of the presented methods is compared to predictions from neural networks and to persistence. As results we get improvements of the predictions compared to the neural network based predictions. An additional improvement is possible by using ensembles of the models.