A novel point-interval prediction model for wind speed based on hybrid deep learning and RIME optimization algorithms

Accurate wind speed forecasting is critical for optimizing wind power systems and utilizing sustainable energy. However, the uncertainty and complexity of wind speed remain major challenges in forecasting. Therefore, a hybrid forecasting system integrating point prediction and interval prediction methods is innovatively proposed in this paper. The system comprehensively considers the correlations and dependencies among meteorological features, significantly enhances the model's feature extraction and fusion capabilities, and quantifies the uncertainty in wind speed forecasting. Specifically, the system first employs the maximal information coefficient (MIC) to select key meteorological variables as inputs, eliminating irrelevant features and improving computational efficiency. Next, it constructs a novel hybrid deep learning model that integrates temporal convolutional networks (TCN), long short-term memory (LSTM), and attention mechanism (ATT). The model further utilizes the RIME optimization algorithm to enhance prediction performance. Furthermore, the system adopts the adaptive boundary kernel density estimation (ABKDE) method to predict the upper and lower bounds of wind speed, thereby enhancing prediction accuracy. Finally, the effectiveness of the proposed system was validated using data from Haikou City, Hainan Province. The experimental results demonstrated that, compared with other benchmark models, the point prediction performance metrics of the proposed system were significantly improved, with reductions of 8.13 %-50.67 % in mean absolute error (MAE), 18.83 %-60.94 % in mean absolute percentage error (MAPE), and 8.41 %-53.30 % in root mean square error (RMSE), along with an increase of 0.64 %-13.63 % in the coefficient of determination (R²). Regarding interval prediction, the proposed system was also proven to outperform all comparative models. The experimental results demonstrate that the proposed hybrid system exhibits robust prediction performance. • Innovative construction of a point-interval dual-mode integrated prediction framework. • Maximum Information Coefficient detects non-linearity and selects key inputs. • The RIME optimization algorithm adaptively tunes ensemble model parameters. • Interval prediction technology based on probabilistic inference.