Hybrid Wind Speed Prediction Framework Using Data Pre-Processing Strategy Based Autoencoder Network

Wind speed forecasting is required for better wind energy grid integration. Predicting wind speed becomes difficult due to the stochastic nature of the wind. So, this study implemented a hybrid forecasting framework using the improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN) based convolutional bidirectional long short-term memory (Bi-LSTM) autoencoder for wind speed forecasting (WSF). ICEEMDAN is used for eliminating residual auxiliary noise of the raw wind speed data. Then, a convolutional network encodes the optimal features from the denoised data and the Bi-LSTM decoder interprets the encoded representation to forecast the wind speed effectively. Existing benchmark approaches fail to provide consistency for the different time horizons, addressing this, the proposed framework is tested using 5-min, 10-min and 30min ahead wind speed data. Two experiments have been performed utilizing data from wind farms in Idalia and Garden city to evaluate the novelty and proposed framework's performance. The experimental results of the proposed framework are compared with seven different state-of-the-art models. The proposed approach is evaluated using various performance metrics and the results of experiments I and II demonstrate that the proposed hybrid method outperformed the state-of-the-art WSF models by 21% and 48%, respectively.