Coevolutionary fuzzy multiagent bidding strategies in competitive electricity markets

Following the development of online markets, trading practices as dynamic pricing, online auctions and exchanges have become relevant to a variety of markets. In this paper we suggest a machine learning approach to find a suitable bidding strategy for an auction participant using information commonly available in online auction settings. We take the electricity auction as the main application example, due to its importance as an experimental instance of the suggested approach. In previous works we evolved successful fuzzy bidding strategies. Here we introduce a coevolutionary algorithm to study how the evolving strategies react to each other in a more dynamic environment. By enabling a fuzzy system to learn trough an evolutionary algorithm one expects to find effective and transparent bidding strategies. By adopting a coevolutionary approach a more realistic representation of the agents participating in an auction based electricity market allows the evolutionary bidding strategies interact. The results show that the coevolutionary approach can improve agents profits at the cost of increasing system hourly price paid by demand.