Enhancing Solar Energy Forecasting Accuracy Using LSTM Networks for Global Horizontal Irradiance

This paper introduces an approach for enhancing solar energy predictions through Long Short-Term Memory (LSTM) networks, utilizing Global Horizontal Irradiance (GHI) data. The generation of solar energy is increasingly vital in tackling the global energy crisis and facilitating the shift to renewable energy sources. Thus, accurate forecasting of solar irradiance trends, especially GHI, is essential for optimizing the efficiency and dependability of solar photovoltaic (PV) systems. Current techniques often struggle to account for the intricate temporal relationships in GHI data, resulting in less accurate forecasts. Our study aims to address these shortcomings by leveraging LSTM networks, which are specifically crafted to model long-term dependencies in time-series data. We explored two models: one relying solely on historical GHI data and another that incorporates additional weather factors such as Global Tilt Irradiance (GTI), temperature, Diffuse Horizontal Irradiance (DHI), and Direct Normal Irradiance (DNI), all of which have shown strong correlations with GHI trends. Both models were trained and assessed using a dataset from Solcast, focusing on the Shah Alam area in Malaysia. In addition, simulations of photovoltaic (PV) energy generation were conducted based on GHI predictions to illustrate the practical use of these forecasts in real-world applications. The proposed models were evaluated with performance metrics including Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE). The results indicate that incorporating additional meteorological parameters enhances prediction accuracy, with the second model demonstrating lower training and validation losses. These findings imply that LSTM models, like the one proposed, hold significant promise for improving the accuracy and reliability of solar energy forecasts, serving as a valuable asset for renewable energy systems.