Abstract:

It is of great significance to study the bidding strategy of thermal power enterprises under the scenario of high proportion of new energy penetration for guaranteeing the normal operation of thermal power enterprises and promoting the construction of new power system. Based on the multi-agent modeling framework, this paper establishes a power spot market simulation model and an unit self-learning decision model. In the environment module, a spot market clearing model with multiple participation of wind-solar-fire-storage is established considering the uncertainty of both sources and loads. The bidding decision-making process of thermal power units is described as a partially observed Markov decision-making process in the agent module and solved by improved Deep Deterministic Policy Gradient algorithm. Finally, a HRP-38 node system is simulated to clarify the market trading strategy for thermal power enterprises under a high proportion of new energy. The results show that, when the auxiliary services provided by thermal power units are not considered, with the increase of new energy penetration, some thermal power units with unique locations and cost advantages are still competitive; the increase of prediction error will make the bidding strategy of large-capacity thermal power units tend to be conservative, while the bidding strategy of small-capacity units is opposite; the thermal power units have implicit collusion tendency in all kinds of scenarios, that is, they will increase the quotation at the same time while hiding information from each other.

Key words: electricity market, multi-agent modeling, reinforcement learning, quotation strategy, auxiliary decision