Multivariate Deep Learning Bidirectional LSTM Model for Forecasting Solar Radiation considering Different Climate Zones over India

---

Abstract

The growing need for sustainable energy is leading to the widespread development of solar power plants worldwide. The power generation of solar plants depends on solar radiation, which is impacted by weather conditions. Therefore, understanding and predicting weather patterns is crucial for estimating solar radiation availability and ensuring reliable energy production. Precisely predicting solar radiation is essential for the smooth and efficient integration of solar power plants into the grid. This study focuses on designing a multivariate deep learning (DL) Bidirectional Long Short-Term Memory (Bi-LSTM) model to predict solar radiation. The model is developed for various climatic conditions on different climate zone over India. To build the DL model, the wind speed and ambient temperature along with solar radiation have been taken as input. The presented DL model demonstrates superior performance metrics, for hot and dry climate zones with RMSE of 0.0051, R² values of 0.92, MAE of 0.0204, and MSE of 0.0018. The high-quality meteorological data for different climate zones are used to develop the model. A comparative analysis is carried out with other DL models to validate the obtained results. The proposed model is found to be accurate, and it can be useful for smart grid applications to maintain grid stability, reliability, and power quality.

---

Keywords

Renewable Energy Resources. Solar Radiation; Deep Learning; Gated Recurrent Unit; Bi- Directional Long Short-Term Memory.