Optimizing Renewable Energy Utilization: A Study on Solar Radiation Forecasting with LSTM-Conv1D Models in Libya

Abstract

Prediction of solar system output power is important for efficient power grid's operation and the management of energy flows within the system. Before forecasting the solar system's output, it is imperative to focus on predicting solar irradiance. This study tries to predict daily global solar radiation data in Libya and provide an overview of forecasting methods for solar irradiation using a deep learning approach like the LSTM model with Conv 1D. To enhance prediction performance, a combined model (LSTM with Conv 1D) can be applied to the forecasting approach. So we proposed three deep hybrid models namely Simple Conv1D-LSTM, Complex Conv1D-LSTM (series), Parallel Conv1D-LSTM. The dataset was collected from the National Radiation Database, which is maintained by the United States Environmental Protection Agency (EPA). The present research centers on the utilization of deep-learning models for analyzing solar radiation. In our study, we focus on using deep hybrid LSTM-Conv 1D (series and parallel approaches) from a simple hybrid model to a complex because LSTM networks are specifically designed to handle sequential data. Complex LSTM-Conv 1D in the first experimental is the best approach in all evaluation methods, including Mean Squared Error (MSE), Mean Absolute Error (MAE), Root Mean Squared Error (RMSE) and R2. © 2024 IEEE.