Hybrid AI Models for Forecasting and Optimizing Solar Energy Generation Under Varying Weather Conditions

Abstract

Accurate solar power forecasting is crucial for maximizing renewable energy integration and grid stability. This study reviews and proposes advanced hybrid AI models that combine convolutional neural networks (CNNs) with long short-term memory networks (LSTMs) to forecast photovoltaic (PV) generation under diverse weather conditions. We highlight how weather factors (irradiance, cloud cover, temperature, humidity) affect PV output and why traditional methods struggle to capture these nonlinear effects. Hybrid CNN-LSTM architectures extract spatial features (e.g. cloud patterns) and learn temporal dependencies in time-series data, yielding higher accuracy than standalone models. For example, Ladjal et al. (2025) report a CNN-LSTM model with R² ≈ 0.9993, far exceeding simpler ANN or SVR approaches. We describe the data preprocessing steps, model structures, and evaluation metrics (MSE, RMSE, MAE, MAPE) used in public solar datasets (e.g. NASA’s POWER data). Experiments on benchmark PV datasets demonstrate that the hybrid model consistently achieves lower errors (e.g. MAPE ≈1-7%) compared to feed-forward ANNs. Moreover, we discuss optimization applications: forecast-informed control (tilt adjustment, energy storage scheduling) can increase energy yields and reduce costs by anticipating weather variability.