Volume 49

Short-Term Combination Prediction of Wind Power Considering Meteorological Complexity and Wind Power Volatility Qianyu Ma,Haiyun Wang,Xiaochen Su,Jiahui Wu

Abstract

Wind power is a significant part of renewable energy and plays a crucial role in modern power networks. Precise wind power prediction is essential for grid scheduling. Numerous studies have been undertaken to forecast wind energy. The traditional single prediction model is limited as it fails to consider the complexity of meteorological data, specifically the correlation between meteorological data and wind energy. Additionally, it overlooks the volatility of wind power. As a result, there is a decrease in prediction accuracy. This study introduces a novel model for prediction short-term wind power combinations, incorporating meteorological feature selection and signal decomposition to overcome current model constraints. Firstly, maximal information coefficient (MIC) is used for meteorological feature selection. Secondly, for the volatility of wind power, complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) is used to decompose the wind power time series data. Subsequently, the meteorological data that underwent feature selection, along with the two frequency signals, were fed into the sample convolution and interaction network (SCINet), bidirectional long short-term memory (BiLSTM), and gated recurrent unit (GRU) models via three separate channels for prediction. The weight coefficients for the meteorological data prediction results and the signal decomposition prediction results were determined using sequential least squares programming (SLSQP). These weight coefficients were then used to get the final prediction results through a weighted combination. The experimental results show that the prediction accuracy of the model proposed in this study is significantly improved compared to the traditional single prediction model.

Keywords wind power prediction, renewable energy, feature selection, signal decomposition, SCINet