Improving supervised wind power forecasting models using extended numerical weather variables and unlabelled data

A variety of supervised forecasting models using numerical weather prediction data have been utilised for short‐term wind power forecasting. These forecasting models only use meteorological variables of the target wind farm as essential features. This study proposes a novel method to improve existing supervised forecasting models such as support vector machines, artificial neural networks, and Gaussian processes (GPs) for higher forecasting accuracy. The proposed method develops a data‐driven feature extraction procedure to utilise unlabelled numerical weather data, and the feature extraction procedure transforms extended numerical weather variables into supplementary input features which then can be used for supervised forecasting models. The only modification to an existing supervised forecasting model is the addition of these supplementary input features, and thus does not alter the training algorithm of the supervised forecasting model. For illustrative purposes, the GP is used as the supervised forecasting model to be improved. Numerical evaluation of the proposed method was performed on a subset of data provided in the 2012 Global Energy Forecasting Competition (GEF 2012). Evaluation results reveal that the proposed method achieves higher forecasting accuracy for all wind farms.