Enhanced Miss Forest and Multivariate Time Series Prediction of Wind Speed Using Deep Learning

As environmental concerns become increasingly serious, renewable energy sources are being prioritized to meet growing demands, among which wind energy stands out as a technologically efficient option. However, the unpredictable nature of offshore wind energy continues to pose challenges for reliable integration into power grids. This study tackles this problem by focusing on improving wind speed prediction by employing a modified Miss Forest algorithm for time series imputation, which achieved the lowest NRMSE: 0.0079, 0.0156 and 0.0730 among the other imputation methods when compared across three datasets. Alongside the Miss Forest, deep learning models such as Bayesian Optimized Stacked Long Short-Term Memory (BO-SLSTM), Deep Long Short-Term Memory (DLSTM), Gated Recurrent Unit (GRU), Convolutional Neural Network (CNN) and a hybrid Convolutional Long Short-Term Memory (CNN-LSTM) model were applied on real-time wind speed data from an offshore site in Gujarat, India. Furthermore, the Friedman Test was conducted to assess the statistical difference in model performance, yielding a p-value of 0.721, indicating no significant difference among the models. Among the models, the DLSTM demonstrated the best performance solely in terms of the error metrics, while CNN proved to be the most computationally efficient.