Short Term Wind Speed Forecasting Based on Feature Extraction by CNN and MLP

Abstract

At present, most of the short-term wind speed forecasting researches directly use the original data as the input or break them down, and take the decomposed series as the input for forecasting model. There is a lack of feature analysis of the original data and the decomposed series. In this paper, from the perspective of feature analysis of wind speed, Ensemble Empirical Mode Decomposition (EEMD) and Convolutional Neural Networks (CNN) are used to decompose the sequence and extract features, and Multilayer Perceptron (MLP) is used to predict the wind speed. Firstly, EEMD is used to decompose the wind speed into a series of subsequences; Secondly, CNN is used to extract the features of each decomposition layer, and the input variables of each decomposition layer are constructed; Finally, MLP is used to predict each decomposition layer; At the same time, Adam is used to optimize the parameters of CNN and MLP. The results of case study and comparison show that EEMD-CNN-MLP-Adam has high prediction and good generalization, which can provide reference for wind speed prediction in different regions and periods.