Reliable solar power forecasting is critical for expanding energy access and maintaining grid stability in Sub-Saharan Africa, where electrification rates remain low and climatic variability is substantial. This paper introduces one of the first systematic AI-based forecasting studies for Chad, employing a hybrid architecture that integrates Long Short-Term Memory (LSTM) with attention and Extreme Gradient Boosting (XGBoost). Leveraging hourly PV and meteorological data from three representative cities (Pala, Mao, and Amdjarass) across the Sudanian, Sahelian, and Saharan zones, the framework demonstrates forecasting errors consistently below 3% of average hourly PV output. Model interpretability, provided through SHAP analysis, underscores solar irradiance, temperature, and temporal indicators as dominant features, thereby strengthening transparency and user confidence. The findings extend beyond methodological contributions by revealing region-specific dynamics: rainfall-induced fluctuations in Sudanian areas highlight the need for storage and backup capacity, while the stable Saharan climate favors large-scale PV integration. By translating forecasting accuracy into practical design and policy implications, the study supports mini-grid planning, investment prioritization, and fossil-fuel displacement. These outcomes align with global sustainability objectives and highlight the role of explainable AI in enabling resilient and equitable electrification pathways in data-scarce regions.
