Wind-speed Forecasting based on Smoothing Ensemble Empirical Mode Decomposition and LSTM

Abstract

One of the main challenges in power generation in wind farms is to forecast wind accurately. This is due to the non-stationary and non-linearity characteristics of wind data. These characteristics make it difficult for the normal statistical methods and the common statistical and computational intelligence methods to provide adequate predictions for wind speed. Empirical Mode Decomposition (EMD) is designed to de-compose non-stationary and non-linear data into their embedded components. In this study, we used a hybrid method composed of an improved variation of EMD, Smoothing Ensemble Empirical Mode Decomposition (SEEMD), and long-short-term memory neural networks (LSTM) to predict wind data. The results of this show that the proposed method provides better forecasting compared to the existing ones.