Strategic bidding in Colombian Electricity market using a multi-agent learning approach

Abstract

In this paper, a multi-agent model of an electricity market is proposed using the Agent-based Computational Economics (ACE) methodology. The proposed methodology for modeling the bidding price behavior of Generation Companies (GENCOs) is based on a reinforcement learning algorithm (QLearning) that uses some soft computing techniques to face the discovery of a complex function among bidding prices, states and profits. The proposed model also comprise the power system operation of a large-scale system by simulating Optimal DC Power Flows (DCOPF) in order to obtain real dispatches of agents and a mapping from action space (bidding strategies) to quantities dispatched. In this model, agents are provided with learning capabilities so that they learn to bid depending on market prices and their risk perception so that profits are maximized. The proposed methodology is applied on colombian power market and some results about bidding strategies dynamics are shown. In addition, a new index defined as rate of market exploitation is introduced in order to characterize the agents bidding behavior.