Agent-Based Analysis of Capacity Withholding and Tacit Collusion in Electricity Markets.

This paper employs agent-based simulation to study energy market performance and, in particular, capacity withholding and the emergence of tacit collusion among the market participants. The energy market is formulated as a repeated game, where each stage game corresponds to an hourly energy auction. Each hourly energy auction is cleared using locational marginal pricing. Generators are modeled as adaptive agents capable of learning through the interaction with their environment, following a reinforcement learning algorithm. The SA-Q-learning algorithm, a modified version of the popular Q-Learning, is used. Test results on a two-node power system with two and eight competing generator-agents, demonstrate the development of tacit collusion among generators even under competitive conditions. [ABSTRACT FROM AUTHOR]