Predictive Analysis of Machine Learning Schemes in Forecasting Offshore Wind Speed

High variability of wind in the farm areas causes a drastic instability in the energy markets. Therefore, precise forecast of wind speed plays a key role in the optimal prediction of offshore wind power. In this study, we apply two deep learning models, i.e. Long Short-Term Memory (LSTM) and Nonlinear Autoregressive EXogenous input (NARX), for predicting wind speed over long-range of dependencies. We use a four-month-long wind speed/direction, air temperature, and atmospheric pressure time series (all recorded at 10 m height) from a meteorological mast (Vigra station) in the close vicinity of the Havsul-I offshore area near Alesund, Norway. While both predictive methods could efficiently predict the wind speed, the LSTM with update generally outperforms the NARX. The NARX suffers from vanishing gradient issue and its performance declines by abrupt variability inherited in the input data during training phase. It is observed that this sensitivity will significantly decrease by integrating, for example, the wind direction at low frequencies in the learning process. Generally, the results showed that the predictive models are robust and accurate in short-term and somewhat long-term forecasting of wind.