Bidding strategy evolution analysis based on multi-task inverse reinforcement learning

Benefiting from the opening of electricity market data in recent years, the analysis of bidding behaviors based on actual data has gradually attracted increasing attention. The data-driven analysis methods could overcome the ideal assumption problem compared with traditional optimization-based methods. In this paper, we focus on the dynamic changes in bidding strategies and propose a multi-task inverse reinforcement learning-based analysis framework. It can identify several bidding objectives adopted by the participant in different time periods and label the adopted objective in each day, according to historical bidding records and market status. Moreover, analyzing the relationship between bidding objective functions and their influencing factors will enrich our understanding of the bidding mechanism and make the market operation simulation closer to reality. An empirical analysis conducted on Australian electricity market data demonstrates the feasibility and effectiveness of our framework. The results also show how a typical hydropower station changes its bidding strategies according to water storage.

Automated summary

This paper proposes a bidding behavior analysis framework based on multi-task inverse reinforcement learning, which is demonstrated to be feasible and effective through empirical analysis on electricity market data.