Risk-Aware Rolling-Horizon Scheduling of a PV–Battery–EV Charging Station via Adaptive Conformal Prediction

Abstract

To mitigate the impact of inaccurate forecasts of electric vehicle charging load and photovoltaic power generation on the operation of PV–storage–charging systems, a risk-aware adaptive conformal rolling-horizon scheduling method is proposed. First, a net-load point forecast is constructed from the forecasts of electric vehicle charging load and photovoltaic generation, and a one-sided conformal upper bound is derived using historical forecast residuals. Subsequently, the conformal calibration parameter is updated online based on the miscoverage outcomes of completed forecasts, enabling the net-load upper bound to adapt to recent changes in forecast errors. Finally, the dynamic upper bound is incorporated into model predictive control to achieve rolling-horizon optimization of battery charging and discharging as well as grid power purchase and sale. The results show that, compared with the fixed conformal method, the proposed method improves the event-period coverage rate and reduces the number of event-period miscoverage instances. Compared with the point-forecast method, it reduces low-SOC exposure and enhances the system’s capability to cope with net-load forecast errors at a relatively low cost.