<p>The intermittent nature of solar power generation has a significant impact on power management systems. Accurate forecasting of global horizontal irradiance (GHI) is essential for resource allocation and reliable system planning. Classical AI methods have shown limitations in capturing complex relationships under unstable conditions. To overcome this challenge, hybrid models have emerged as more effective solutions to improve forecasting accuracy. This study proposes a hybrid deep learning model combining Long Short-Term Memory and Convolutional Neural Network (LSTM–CNN) for hourly GHI forecasting. LSTM can effectively extract temporal features from time-series irradiance data, while CNN captures spatial dependencies by analyzing the correlation matrix of meteorological variables such as temperature, precipitation, relative humidity, and wind speed. The model is trained using datasets from three locations in southern Morocco and evaluated over one year using several metrics such as RMSE, MAE and R². Results show that the hybrid model improves forecasting skill by 0.7–6.9% compared to standalone models, including Support Vector Machines, Artificial Neural Networks, LSTM, CNN. The findings suggest that the proposed LSTM–CNN hybrid strategy is a reliable alternative for short-term GHI forecasting, providing higher predictive accuracy across diverse seasonal weather conditions compared to standalone approaches and offering a framework for future research in renewable energy forecasting.</p>