Multi-agent based modeling and learning approach for intelligent day-ahead bidding strategy in wholesale electricity market

This work considers a restructured electricity market where the agents are considered as entities, and each agent has a learning capability. Also, the Independent System Operator (ISO) adopted the nodal pricing technique for managing transmission line congestion. In this multi-agent platform, the objective of each Generating Company (GenCo) agent is maximizing its net earnings. GenCo exhibits economic capacity withholding to enhance its daily net income ports updated cost parameters to the ISO. The basic model free policy-based reinforcement algorithms fail to calculate the probability that the rewards obtained are negative or zero. To overcome this drawback, the co-learning agent-based modeling approach implements an interactive learning algorithm applied to GenCo's for reward consideration in all respects and thereby applies the strategic approach in improvising marginal pricing, commitments, and net incomes. The co-learning effect of GenCo is accomplished with the optimum value of the learning parameters by deploying a novel interactive Variant Roth - Erev (VRE) Reinforcement learning technique for efficient electricity market operation. The introduction of various parameters in learning helps to improve the mutual interaction among the agents and provides better experimentation implementable by GenCo's meeting the social welfare and smart bidding. The result presented in this paper suggests that the learning capability of the agents is having an impact on the marginal pricing, which supports the operator in managing the congestion in the transmission line. The analysis showcases the effectiveness of the co-learning approach adopted by the agents in the electricity market which is tested on a standard IEEE system and the performance of the technique is validated on a realistic Indian grid network.