Enhancing Solar Irradiance Forecasting Using LSTM and Meteorological Data

Introduction: Solar energy is an abundant and sustainable resource that plays a crucial role in the global shift towards renewable energy. However, solar irradiance, a key factor in solar power generation, fluctuates throughout the day, affecting the efficiency of solar energy systems. Accurate prediction of solar irradiance is essential for optimizing solar energy generation and ensuring grid stability. Objectives: This study aims to improve the accuracy of solar irradiance forecasting using a Long Short-Term Memory (LSTM) model. The objective is to address the limitations of traditional forecasting models and explore the integration of various meteorological inputs for more precise predictions. Methods: The LSTM model was developed using historical solar irradiance data and meteorological parameters, including temperature, humidity, and wind speed, sourced from NASA’s POWER Solar database. Data preprocessing techniques like MinMaxScaler normalization were applied, and the model was trained using 70% of the data and tested on 30%. The LSTM network incorporated layers with 256, 128, and 64 units, optimized using techniques like EarlyStopping and ReduceLROnPlateau to avoid overfitting. Results: The LSTM model demonstrated strong predictive performance, achieving an RMSE of 47.58, MAE of 22.64, and an R-square value of 96.94%. Compared to traditional Support Vector Regression (SVR), the LSTM model outperformed with a 28.79% improvement in RMSE and a 48.81% improvement in MAE. The model's ability to capture temporal dependencies and nonlinear interactions in solar irradiance data was confirmed through evaluation metrics. Conclusions: The LSTM model successfully enhanced solar irradiance forecasting, providing more accurate predictions for renewable energy applications. Despite its promising results, future research can explore additional weather parameters and hybrid machine learning models to further improve accuracy and generalizability.