Hybrid-MAS modeling method for capacity pricing and ancillary service optimization in the Southern China electricity market

Capacity-price formation and ancillary-service dispatch are strongly coupled in regional electricity markets with high renewable penetration, yet they are commonly optimized by separate models and at different time scales. This separation can produce inconsistent price and dispatch signals when net-load forecasts change rapidly or flexible resources compete across energy, reserve, and regulation products. This paper develops a Hybrid-MAS modeling method that formulates capacity pricing and ancillary-service allocation as a constrained partially observable Markov game. Generation, storage, and load-aggregation agents generate local dispatch and bidding proposals under centralized training and decentralized execution. A genetic-search layer updates block-wise capacity-price coefficients, a particle-swarm layer refines regulation and reserve allocation, and a quadratic feasibility-projection layer maps joint actions to the physically feasible operating region. The framework is evaluated on a modified IEEE 118-bus system calibrated with quarter-hour operating samples from an anonymized southern regional profile and is compared with rule-based, mixed-integer, and learning-based schedulers. Comparative, ablation, disturbance, missing-data, and scalability tests show that the coordinated architecture improves capacity matching and ancillary-service fulfillment, stabilizes capacity-price corrections, reduces operating cost and reserve shortage, and maintains executable market actions under multiple uncertainty conditions. These results support Hybrid-MAS as a reproducible market-simulation and dispatch-decision framework for coordinating short-horizon capacity-price correction with ancillary-service allocation.