Feature extraction of numerical weather prediction results toward reliable wind power prediction

Wind power prediction is necessary for stable operation of a power grid under the introduction of significant wind power generation. Wind power prediction approaches based on the results of numerical weather prediction (NWP) have been developed successfully in recent years. However, the high dimensionality of NWP results can be a major obstacle when training models. This paper proposes a feature extraction scheme based on convolutional neural networks to compress high-dimensional NWP results by deriving critically important low-dimensional information for wind power prediction. The experimental results show that the proposed feature extractor can contribute to the improvement of wind power prediction accuracy.