Enhanced Short-Term Wind Speed Prediction by ICEEMDAN and Feature Engineering

Accurate wind speed prediction are essential for ensuring the safety of marine engineering, fishing, and other offshore operations. Due to the nonlinear properties of wind speed, however, it is often difficult for conventional prediction models to produce accurate prediction. This paper proposes a hybrid prediction model that combines the ICEEMDAN decomposition algorithm and feature engineering to solve this problem. This model decomposes the original wind speed series into subsequences with different frequencies and then eliminates the high-frequency subsequences that contain white noise. In addition, current research on wind speed prediction focuses primarily on the wind speed series itself. Using feature engineering, this paper extracts the statistical features within the sliding window and combines them with the time series to generate the characteristic time series input for training the LSTM model. The short-term wind speed predictions of this fusion model are more accurate than those of traditional models.