DER Control: Harnessing DDPG and MILP for Enhanced Performance in Active Energy Management

Abstract

Energy communities (EC) have emerged as a significant approach for enhancing the efficiency of renewable energy use in the low-power private sector. The increasing penetration of decentralized renewable energy resources (DERs), such as battery storage, photovoltaic (PV) systems, and heat pumps, has generated a need for advanced community-wide control mechanisms. Traditional model-based controllers, while effective, often require labor-intensive mathematical modeling and linear approximations, limiting their flexibility and accuracy. This paper introduces a deep reinforcement learning (DRL) approach, specifically the Deep Deterministic Policy Gradient (DDPG) algorithm, for the autonomous and optimal management of DERs within an EC. The algorithm’s ability to enhance DER control is demonstrated by comparing its performance against the results of a model-derived optimization task solved by mixed integer linear programming (MILP). The results yielded by both algorithms are found to be comparable in quality in shaving demand peaks and power grid-friendly power consumption curves. The study underscores the potential of EC power consumption optimization and intermediate energy storage, thereby contributing to grid stability and carbon emission reduction.