Regime-Conditioned Reserve Calibration and Hydrogen Flexibility for Virtual Power Plant Scheduling

Abstract

High-renewable virtual power plants (VPPs) exhibit asymmetric operational risks stemming from net-load forecast errors: positive errors require upward reserve capacity, while negative errors can cause grid export saturation and renewable curtailment. This paper proposes RC-CVaR-H2, a linear rolling-horizon framework combining regime-conditioned day-block conformal reserve floors, empirical conditional value-at-risk (CVaR), carried resource states and flexible electrolysis. A nested matched comparison separates reserve calibration from surplus absorption under common inputs. The framework is evaluated on a two-year Inner Mongolia synthetic benchmark with 2016 out-of-sample hourly tests. Relative to empirical CVaR, regime-conformal CVaR raises dispatch-reserve coverage from 88.99% to 92.71% and reduces upward shortfall by 60.66%. Relative to pooled conformal CVaR, regime conditioning reduces shortfall by 38.27% with 3.1% more mean reserve. The paired daily reduction is 1.838 MWh, with a 95% interval from 0.940 to 2.777 MWh. At the fixed 35 MW export cap, hydrogen flexibility lowers curtailment from 233.40 to 71.43 MWh, or 69.40%, with unchanged headroom-based reliability metrics. Both effects retain their direction across five generated realizations. The full method costs 9.40% less than deterministic day-ahead scheduling, mainly through rolling updates, and each 4 h update takes 5.52 ms. These results quantify complementary calibration and absorption effects within the disclosed synthetic forecasts, fixed parameters and aggregate resource model.