A Multi-Agent Reinforcement Learning Framework for Voltage-Constrained Incentive Demand Response in PV-Rich Low-Voltage Distribution Systems

Distributed generation can help to balance energy production and consumption in stable power systems. An imbalance in energy supply and demand or intermittent renewable-based distributed generation management might increase utility company and consumer costs and incapacitate the electrical grid. To solve these issues, this paper proposes an incentive based demand response system that integrates photovoltaic (PV) systems and voltage support capability into the control algorithm. The introduced system utilizes a Multi-Agent Reinforcement Learning (MARL) approach to decrease power consumption and coordinates demand response measures during periods of peak PV generation, while ensuring that each node complies with voltage standards. The proposed approach involves the aggregator agent and a household agent with a home energy management system that employs disjunctively constrained knapsack problem optimization to schedule home appliances, taking into account the user’s dissatisfaction. A simulation is performed on a feeder containing 25 residential houses to assess the performance of Voltage-Controlled Incentive-Based Demand Response (VC-IBDR). The simulation results indicate a clear improvement in both individual consumer demand and overall system performance. On average, the feeder demand is reduced by 4.20%, while the peak power requirement decreases by 7.01%. In addition, photovoltaic sources contribute approximately 30% of the total energy supply over the daily operating period. The voltage profile at all nodes is maintained within acceptable limits, ranging from 0.96 to 1.02 p.u. at every time step. These findings confirm that the proposed approach effectively enhances the operational performance of low-voltage distribution systems with high PV penetration.