Organization: Open Climate Fix
Work Repository: TabAdjust
Open Climate Fix’s (OCF) Quartz Solar forecasting model produces short-term solar photovoltaic (PV) generation forecasts. These forecasts are post-processed using a simple rule-based adjuster, which computes the mean of recent forecast errors and applies it to correct the incoming forecast. While effective in stable weather conditions, this rule-based approach is rigid and can perform poorly during:
- Rapid weather transitions (e.g., sudden cloud cover changes)
- Atypical error patterns (e.g., persistent underestimation or overestimation)
The aim of this project was to investigate whether TabPFN, a transformer-based foundation model for tabular data, or its time-series variant (TabPFN-TS), can outperform the existing adjuster in terms of **