A low-carbon economic bidding strategy for power generators considering multi-market coupling: a multi-agent deep reinforcement learning approach

The coupling of electricity, carbon, and tradable green certificate markets is expected to reshape generation companies’ (GENCOs’) bidding strategies. To support power-sector decarbonization and improve strategic bidding decisions, this paper proposes a multi-layer electricity market clearing framework integrating carbon trading and tradable green certificates (TGC). A non-convex multi-segment bidding model is developed for GENCOs to capture the joint effects of generation costs and environmental compliance costs on bidding offers. In addition, a Beta-distribution-based multi-agent proximal policy optimization method (MAPPO-Beta) is adopted to solve the complex bidding problem under incomplete information and improve GENCOs’ market payoffs. Case studies on the IEEE 39-bus system show that MAPPO-Beta outperforms benchmark algorithms in reward performance and training stability, with an average reward improvement of 10.2%. Its effectiveness is further verified on the IEEE 118-bus system. Comparative analysis indicates that carbon trading is more effective in constraining high-emission thermal power generators, whereas TGC improves the bidding competitiveness of renewable energy generators. The results reveal clear strategy evolution after introducing carbon trading and TGC: low-emission thermal generators tend to submit more competitive bids with higher declared capacities and lower prices, high-emission units tend to raise bidding prices and maintain capacity withholding, and renewable generators tend to submit lower bids as environmental value is monetized. Compared with the traditional bidding method, the proposed model increases aggregate GENCO profit by 39.3% and reduces carbon emissions by 2.4%. These findings provide insights for GENCO bidding and coordinated low-carbon market design.