LSTM-Based Deep Learning Model for Solar Irradiance Prediction Across Diverse Climate Zones

Solar energy has become a key pillar of renewable energy strategies due to its sustainability and widespread availability. Accurate forecasting of solar irradiance is critical for photovoltaic energy generation, smart grid management, and agricultural planning-especially amid increasing climate variability. This paper presents a BiLSTM-LSTM sequential model developed for solar irradiance prediction. The model is trained on Typical Meteorological Year (TMY) data from four distinct climate zones: tropical, dry, temperate, and continental. Compared to standard LSTM, BiLSTM, and other hybrid models, the proposed architecture achieves superior performance, recording the lowest MSE (0.0018-0.0106), MAE (0.0257-0.0597), and RMSE (0.0428-0.1028) values in three of the four regions. It also delivers near-perfect R2 scores-0.9782 in dry and 0.9175 in continental zones-while maintaining strong accuracy in tropical (R2 = 0.8382) and temperate (R2 = 0.8961) climates. These results demonstrate the model's effectiveness and adaptability in forecasting solar irradiance across diverse climate conditions.