Wind Speed Prediction Using a SEEMD-LSTM Model

Although wind energy appears to be a very promising replacement for fossil fuel energy, it faces some challenges during implementation. Producing precise predictions for wind energy is of utmost significance in energy strategic planning. This is because wind data is non-stationary and non-linear in nature; therefore, common statistical and computational methods cannot produce adequate predictions for wind. In this study, we utilized a hybrid method that combines a variation of Empirical Mode Decomposition, Smoothing Ensemble Empirical Mode Decomposition (SEEMD), and long-short-term memory neural networks (LSTM) to predict wind speed. The results of this study provide a promising forecasting method for predicting non-stationary and non-linear phenomena such as wind. The hybrid method proposed in this study successfully predicted the wind-speed data with an R2 value of 0.9353 a mean absolute error (MAE) of 0.0504 as well as the wind-direction data with an R2 value of 0.9711 and a mean absolute error (MAE) of 0.1626.

I. Introduction

Wind energy stands out as one of the most efficient forms of renewable energy and plays a pivotal role in reducing greenhouse gas emissions, representing a crucial step toward a sustainable and clean energy future. Accurate wind forecasting at various time scales is essential for the effective operation and safety of any wind farm. For instance, short-term forecasting (minutes/hours and days ahead) is necessary for regulation/load-following and day-ahead scheduling, while longer time scales are vital for wind farm planning and development, including resource allocation and economic viability [1].