Decision-making and cost models of generation company agents for supporting future electricity market mechanism design based on agent-based simulation

The large-scale surge in renewable energy installations has transformed the capacity mix of power systems and the roles of generation companies (GENCOs). For example, some thermal generators are now operated at low output levels to ensure the generation capacity and ramping capability of the power system. As a result, the nonlinear characteristics of GENCOs’ marginal generation costs have gradually become prominent, rendering some previously linear assumption-based models obsolete. It is essential to reexamine the decision-making and cost models of GENCOs to support the equilibrium solution and mechanism design of the electricity market during this transition. This paper analyzes the impact of different cost model assumptions on GENCOs, thereby examining the relationship between GENCOs’ bidding models and cost models. We propose standardized expressions for GENCOs’ linear bidding models and piecewise step bidding models in multi-agent simulations of the electricity market. The applicability of different bidding models is analyzed. To address the issue of overly compressed decision space for GENCOs in previous studies, we propose a Multi-worker decision model based on (deep) reinforcement learning. This allows the decision space of GENCOs’ piecewise step bidding to fully cover the bidding space in actual market rules. Finally, various electricity market experiments based on multi-agent simulations are conducted. On the one hand, our proposed GENCOs decision model more effectively reproduces GENCOs’ behavior in actual electricity markets. On the other hand, using real mechanism design as an example, previous GENCOs models may lead to incorrect conclusions in simulations. The decision model proposed in this paper, employing piecewise step bidding and polynomial cost functions, makes the simulation results more consistent with actual rules, thereby effectively supporting future-proof market design.