AI-Optimized Solar PV Tracking: Low-Cost Neural Network for Variable Climates

Reliable solar radiation forecasting is essential for improving the performance, efficiency, and operational planning of photovoltaic (PV) systems, particularly in regions with fluctuating weather conditions. With the global shift towards sustainable energy solutions, there is an increasing demand for intelligent, accurate, and cost-effective forecasting models that can support better i4ntegration and utilization of solar energy. In this research, we introduce an innovative approach to solar radiation forecasting called the Stacked Long Short-Term Memory with Attention for Solar Radiation Prediction (SLA-SRP) model. By combining stacked LSTM networks with an attention mechanism, the model harnesses the power of deep learning to effectively learn long-range temporal patterns while dynamically identifying and emphasizing the most relevant input features. To evaluate the effectiveness of the SLA-SRP model, we conducted comprehensive experiments comparing its performance against several well-established machine learning models, including Linear Regression, Random Forest, Support Vector Regression (SVR), and XGBoost. The models were assessed using standard performance metrics such as Root Mean Squared Error (RMSE) and the coefficient of determination (R²). Results show that SLA-SRP significantly outperforms all baseline models, achieving an RMSE of 90.34 and an R² value of 0.9456, indicating high prediction accuracy. The incorporation of the attention mechanism enables the model to dynamically focus on the most relevant features in the time series, further enhancing its forecasting capability. These results highlight the potential of the SLA-SRP model as a robust and practical tool for improving solar radiation prediction and supporting more efficient solar energy system operation.