A bilateral branch learning paradigm for short term wind power prediction with data of multiple sampling resolutions

Abstract

This paper revisits the data-driven short-term wind power prediction (WPP) modeling process with introducing a new data usage paradigm, using SCADA data of both high and low sampling resolutions as model inputs. To organically incorporate data of two sampling resolutions as inputs and tackle the extreme high dimension of features induced by using high sampling resolution data in WPP modeling, we propose a novel bilateral branch learning based WPP modeling framework, which includes two data feature engineering branches and one prediction module. One branch for processing the high sampling resolution data is developed based on a novel deep learning structure stacking multiple convolutional and sparse pooling layers as well as long short term memory networks (CNN-SP-LSTM). Another branch adopts the classical feature selection to process low sampling resolution data. The bilateral branch engineered features are concatenated as one input for the final wind power prediction. To justify the validity of the proposed WPP data usage and modeling framework, extensive computational experiments are conducted based on real wind farm SCADA data. Via analyzing computational results, we show that it is advantageous to incorporate data of the high sampling resolution into WPP tasks. Via a comprehensive comparative analytics, we identify the machine learning techniques most suitable on the feature selection and prediction module development parts of the proposed modeling framework. Via an ablation study, we verify the contribution and advantage of the proposed CNN-SP-LSTM module. Based on results of a comprehensive computational study, we verify that our proposed framework achieves the state-of-the art performance as it beats a large set of classical data-driven and recent deep learning based WPP methods considered in this study.