WOA-FCM-CNN-WNN-informer: An advanced hybrid deep learning model for ultra-accurate PV power forecasting in electric mobility

Effective prediction of photovoltaic (PV) power generation is essential for enhancing energy management in solar-powered electric vehicles. This study introduces an innovative hybrid forecasting framework that combines Fuzzy C-Means (FCM) clustering, Convolutional Neural Networks (CNN), Wavelet Neural Networks (WNN), the Informer architecture, and the Whale Optimization Algorithm (WOA) to improve prediction accuracy. This approach introduces a condition-aware, end-to-end FCM-CNN-WNN-Informer pipeline tailored for PV dynamics, where: (i) similar-day fuzzy clustering normalizes weather heterogeneity before learning; (ii) wavelet-based multi-scale features are injected into a long-horizon Informer; (iii) a global, cross-module hyperparameter search via Whale Optimization Algorithm (WOA) jointly tunes all stages; (iv) a Generalization Index (GI) is proposed for robust model selection; and (v) Monte-Carlo dropout quantifies predictive uncertainty for practical deployment.The proposed WOA-FCM-CNN-WNN-Informer model is evaluated on a comprehensive dataset of 70,080 hourly PV power recordings gathered over eight years in Tunisia. Results show superior performance compared to standard deep learning models like LSTM and BiLSTM. The framework reduces Mean Absolute Percentage Error (MAPE) by as much as 98.52% and Root Mean Squared Error (RMSE) by 93.84%, while maintaining a high coefficient of determination (R2=0.98) across varying meteorological conditions. These outcomes underscore the model’s robustness and its promise for advancing energy utilization, refining charging strategies, and supporting intelligent route planning in solar-electric transportation systems.