Accurate solar power forecasting is critical for grid stability, but variability under tropical climatic conditions poses constraints. This multi-national study proposes a hybrid deep learning model that integrates GRU and LSTM, leveraging the models' complementary learning mechanisms. The GRU layer extracts short-term dependencies before utilizing the LSTM layer for long-term contextual representation. The hybrid framework was evaluated using operational data collected under tropical urban conditions at the Universiti Malaya. The model achieved superior predictive capability for Poly-crystalline (Array 1), with an R² value of 0.9961. It also demonstrated high accuracy for Mono-crystalline (Array 2), achieving an R² of 0.9952 and an RMSE of 14.82 W. Incorporating the GRU yielded performance improvements up to 16.41%, providing a scalable solution for improving PV power prediction.
Hybrid deep-learning model improves PV power forecasting reliability
Universiti Malaya published a study detailing a hybrid GRU-LSTM deep learning model that achieved superior PV power forecasting accuracy in tropical urban environments.
/solarbytes/media/media_files/2025/12/06/2025-12-06-tech-gks-2025-12-06-21-32-43.jpg)
Advertisment
/solarbytes/media/agency_attachments/2025/01/13/2025-01-13t112055287z-solarbytes.png)
/solarbytes/media/agency_attachments/2025/01/13/2025-01-13t112030439z-solarbytes.png)