Short-Term Solar PV Forecasting in Microgrids Using Cloud Top Temperature and Vision Transformer Based Models

Surasak Surathunmanun^1*Weerakorn Ongsakul^1*Jai Govind Singh¹Kamyar Mehran²

• ¹Asian Institute of Technology, Khlong Luang District, Thailand
• ²Queen Mary University of London, London, United Kingdom

Expanding clean-energy microgrids in remote areas is essential for achieving global decarbonisation and energy transition goals. Accurate short-term solar PV forecasting reduces diesel reliance, improves battery scheduling, and supports reliable renewable integration. However, forecasting remains challenging in developing countries that lack irradiance sensors, cloud cameras, or real-time monitoring infrastructure. This study proposes a novel forecasting framework, CTT–ViT–Transformer, integrating Generative AI to improve short-term solar PV forecasting in sensor-constrained microgrids. The model uses Cloud Top Temperature (CTT) imagery, which reflects cloud height and thermal characteristics, with a Vision Transformer (ViT) for spatial feature extraction and a Transformer for time-series prediction. Results show that the standard Transformer outperformed LSTM and CNN-LSTM (MAE = 23.4518 kW, RMSE = 28.2390 kW, R² = 0.9270). The proposed CTT–ViT–Transformer further improved performance (MAE = 15.9868 kW, RMSE = 24.2782 kW, R² = 0.9744) and surpassed models using RGB imagery (MAE = 17.1945 kW, RMSE = 25.9424 kW, R² = 0.9626). The framework maintained high accuracy across four-step-ahead forecasts (R² > 0.96). Notably, the approach requires no ground-based irradiance sensors, reducing barriers for sensor-limited microgrids while remaining compatible with sensor data where available. Its scalability supports proactive energy management in the carbon-neutral microgrid on Koh Paluay Island, enabling more efficient scheduling of renewable generation and storage, displacing fossil fuels, and lowering operational costs. The CTT–ViT–Transformer provides a reliable, practical forecasting pathway that advances equitable access to advanced energy tools for developing regions. By enabling affordable forecasting and supporting decarbonisation, this framework aligns with SDG 7 (Affordable and Clean Energy) and SDG 13 (Climate Action) to promote a just, sustainable energy transition worldwide.