Wind-Driven Traffic Energy Collaborative Optimization: Multi-modal Probability Prediction and Distributed Dynamic Charging

Under the background of the “double carbon” target and energy transformation, transportation-energy integration has become a core path for coordinated development, with high-precision wind power prediction serving as a key technology to support the system’s efficient operation. This paper proposes a wind power probability prediction method based on the CT-LightNGBoost fusion model (CNN-Transformer-LightGBM-NGBoost) and constructs a hierarchical distributed electric vehicle collaborative optimization framework. The model innovatively integrates the spatio-temporal feature extraction capabilities of CNN-Transformer, the cross-advantages of multi-modal features from LightGBM, and the non-linear relationship modeling and uncertainty quantification mechanisms of NGBoost. It enhances data homogeneity through a spatio-temporal feature pyramid, achieving significant improvements in prediction accuracy and reliability. Experiments conducted on the measured dataset from a wind farm in Gansu Province demonstrate that the model reduces the 12-h prediction RMSE to 43.46, a substantial improvement over benchmark models. Furthermore, the hierarchical ADMM optimization framework and the wind power-aware dynamic charging strategy are combined to provide a high-precision prediction-decision integration solution for the traffic-energy integration system.