Feedforward Neural Networks for Very Short Term Wind Speed Forecasting

Since 2007 wind has become the major source of renewable energy in the UK. Moreover, increasing oil costs are driving researchers in the marine transport field to develop innovating wind ships. In order for wind power to be effectively and efficiently exploited, reliable forecasts on wind speed are needed. These will allow saving curtailments costs, improving safety, reducing damages due to extreme weather conditions, etc. Also, short and very short wind forecasts are critical for energy trading. In this study we present a short-term wind forecast based on artificial neural networks, which are mathematical structures able to model complex non-linear systems. In particular, we used a multilayer perceptron that predicts future wind speed values given the past and current recorded values. Data sampled every ten minutes was used to forecast up to one hour ahead, with an uncertainty ranging from 5%, for ten minutes ahead forecast, to 21%, for one hour ahead forecast.