Forecast‐Informed Rolling Conformal Robust Scheduling of Source–Load–Storage Systems With Electric Vehicle Flexibility

Time‐varying photovoltaic (PV) and load forecast errors complicate the coordinated operation of source–load–storage systems, particularly when electric vehicle (EV) flexibility must be used without compromising charging service. A forecast‐informed rolling robust scheduling framework is developed by combining multistep probabilistic forecasting, trajectory conformal calibration, and two‐stage adaptive robust optimization (ARO). A Ridge‐assisted residual quantile deep neural network (R‐QRDNN) provides PV and load forecasts, which are converted into a rolling joint PV–load conformal box using only fully observed historical trajectories. The scheduling model coordinates gas‐turbine commitment, energy storage, EV charging and vehicle‐to‐grid (V2G) response, demand response, and network‐constrained recourse. The resulting finite‐vertex robust problems are solved through scenario‐accumulating column‐and‐constraint generation (C&CG), followed by first‐period implementation and nonlinear AC power‐flow assessment. R‐QRDNN achieved a load RMSE of 0.0521 MW, while the fixed joint calibration attained an empirical calibration‐sample trajectory coverage of 0.9015. The rolling conformal strategy showed its clearest benefit under large forecast errors: on the maximum forecast‐error day, operating cost and emergency grid import were reduced by 46.36% and 72.38%, respectively, relative to deterministic scheduling, with no emergency load shedding. Under rapid PV ramps, the economic difference was modest, with no emergency actions required. Across the five evaluated days, rolling conformal scheduling avoided emergency grid import and achieved the lowest mean operating cost among the compared methods. These results indicate that rolling calibration is most valuable when recent forecast‐error characteristics depart from those represented by a fixed historical calibration, whereas its economic benefit under more regular operating conditions is comparatively limited.