Transductive-Transfer Learning Based Deep Neural Networks for Day-Ahead PV Power Forecasting in Smart Grid Application: A Comparative Analysis

For effective grid management and the integration of renewable energy sources, accurate forecasting of photovoltaic (PV) power output is essential. This paper presents a novel approach to PV power forecasting using deep learning models with Transductive Transfer Learning (TTL) assistance. The proposed models, namely LSTM, Bi-LSTM, and Feed Forward Neural Networks, are trained and tested on a dynamic PV sites dataset that captures the diverse and dynamic nature of PV power generation. The efficiency of day-ahead power forecasting is evaluated on Winter and Summer days to account for seasonal variations. A detailed comparative analysis is conducted, comparing the performance of the proposed models with and without transfer learning assistance, as well as other conventional deep learning models. Additionally, comprehensive statistical analysis is performed to assess the sensitivity and reliability of the proposed models. The results demonstrate that TTL significantly improves the accuracy and reliability of the forecasting models, leading to more effective PV power management. The findings of this research contribute to the field of PV power forecasting and provide valuable insights for practical implementation in renewable energy systems.