Time series Foundation Models based on Physics-Informed Synthetic Histories for Cold-Start Photovoltaic Forecasting
Researchers have developed a novel pipeline for photovoltaic (PV) forecasting that addresses the challenge of cold-start scenarios where historical site data is unavailable. This method generates synthetic production histories using plant metadata and meteorological data, enabling time-series foundation models (TSFMs) to forecast energy output. The approach significantly outperforms traditional baselines, achieving up to a 2x improvement in accuracy across various climate conditions and PV sites. AI
IMPACT Enables more accurate renewable energy forecasting in challenging cold-start scenarios, potentially improving grid stability and energy management.