Hybrid CNN-LSTM Architecture for Enhanced Short-Term Solar Irradiance Forecasting Utilizing NASA POWER Data

This paper proposes a hybrid deep learning framework for short-term solar irradiance forecasting, integrating convolutional neural networks (CNNs) and long short-term memory (LSTM) units. The model is trained using publicly available meteorological data from the NASA POWER API, including parameters such as temperature, wind speed, and humidity. CNN layers are used to extract spatial and temporal features from weather sequences, which are subsequently passed to LSTM layers to capture sequential dependencies in global horizontal irradiance (GHI) data. The hybrid architecture was evaluated on a three-year dataset from California (۲۰۲۰–۲۰۲۲), achieving a mean absolute error (MAE) of ۱.۰۸ kWh/m² and root mean square error (RMSE) of ۱.۳۷ kWh/m². These results demonstrate the model’s capacity to provide accurate and stable forecasts suitable for operational deployment in PV system planning and grid integration. The use of freely accessible data and efficient network architecture further enhances its applicability in regions lacking dense sensor infrastructure.

کلیدواژه ها:

solar forecasting ، CNN-LSTM ، NASA POWER ، GHI prediction ، deep learning ، hybrid model ، time series ، photovoltaic planning