Abstract
When a virtual power plant (VPP) participates in energy and reserve markets, day-ahead bids, reserve commitments, and real-time delivery share the same internal resource constraints, while aggregate models cannot readily trace internal responsibility allocation or identify local shortfalls. This study develops a fixed-responsibility partitioning framework for day-ahead bidding and real-time delivery. Fixed membership mappings are screened under a unified budget; resource-type-specific reserve certificates and sustained-delivery constraints are embedded in a finite-support Wasserstein distributionally robust optimization (WDRO) model; and hierarchical model predictive control executes the market commitments. The 365-day evaluation retains three non-dominated candidates, K∈{6,7,8}, whose mean operational-fitness objective values are reduced by 25.5%–43.1% relative to the corresponding independent spectral-clustering baselines. In a common 30-day ablation study, detailed partition responsibilities increase mean daily adjusted operating profit by CNY 6442–8162 and reduce settlement deviation by 1.19–1.23 MW relative to two equivalent-responsibility rules. Here, adjusted operating profit is profit net of balancing-correction cost, settlement deviation is the schedule–settlement difference, and partition tracking error is the deviation between partition targets and aggregated resource response. The directions of these effects remain consistent across two independent 10-day windows, and hierarchical real-time coordination further reduces settlement deviation and partition tracking error. Public-aggregate-data replays show that the profit and deviation effects of WDRO relative to sample average approximation (SAA) vary with market settings and node-downscaled inputs. As an engineering extension, bounded contracted-capability factors, defined as multipliers on market-committable capability, increase mean daily adjusted operating profit by 6.00% in the 2024 holdout set, and attribution analysis indicates that most of this gain is associated with the average contracted-capability factor and released fast-response headroom. Under the tested settings, fixed-responsibility partitioning enables the continuous transfer of resource-type-specific day-ahead responsibilities into real-time resource control and makes the trade-offs among profit, delivery quality, and execution cost explicit.