Wind direction forecasting with artificial neural networks and support vector machines

We propose two methods for short term forecasting of wind direction with the aim to provide input for tactic decisions during yacht races. The wind direction measured in the past minutes is used as input and the wind direction for the next two minutes constitutes the output. The two methods are based on artificial neural networks (ANN) and support vector machines (SVM), respectively. For both methods we optimise the length of the moving average that we use to pre-process the input data, the length of the input vector and, for the ANN only, the number of neurons of each layer. The forecast is evaluated by looking at the mean absolute error and at a mean effectiveness index, which assesses the percentage of times that the forecast is accurate enough to predict the correct tactical choice in a sailing yacht race. The ANN forecast based on the ensemble average of ten networks shows a larger mean absolute error and a similar mean effectiveness index than the SVM forecast. However, we showed that the ANN forecast accuracy increases significantly with the size of the ensemble. Therefore increasing the computational power, it can lead to a better forecast.