Convergence of learning behaviors in games of Generation Companies in electricity market

Abstract

In electricity markets with repeated games, generation companies' (GenCos) bidding strategies and learning behaviors are complex dynamic problems. The theories of learning in games are fundamental to research on the evolution process of games and its equilibrium, one branch of those is named as the belief learning model which demonstrates the following learning behaviors: each participant sets his strategic choice correspond to the best response to his perceptions of the circumstances that surround him. To simulate and study GenCos' learning behaviors and the dynamic evolution process of electricity markets, the idea of learning in games is imported and several new types of GenCos' bidding strategies based on the belief learning model are presented, among them we distinguish a) Cournot best response dynamic; b) partial adjustment to the best responses; and c) fictitious play. Three propositions are established for describing and solving the convergence problems of GenCos' learning behaviors. A simple example with two GenCos in unilateral electricity market has been used to demonstrate those problems, and the simulation results are consistent with our propositions.