Coordination of Smart PV Inverters in DistributionSystems with Deep Reinforcement Learning

This research proposes a multi-tasking DeepReinforcement Learning (DRL) algorithm capable of minimizing powerlosses and mitigating voltage fluctuations in high photovoltaicpenetration (PV) distribution grids without curtailing their activeoutput power. The problem is formulated as a Markov Decision Process(MDP) with an optimized multi-purpose reward function. Two modelfreedata-driven DRL algorithms, Soft Actor-Critic (SAC) and TrustRegion Policy Optimization (TRPO), are trained. The proposedalgorithms are implemented on the IEEE ۳۷-bus test case distributionsystem, featuring a significant penetration of photovoltaic (PV) sources.Their efficacy is assessed using authentic real-world data. Thesealgorithms demonstrate a robust ability to proficiently handleuncertainties associated with PV generation while ensuring compliancewith standard operational constraints.

کلیدواژه ها:

Photovoltaic ، Deep Reinforcement Learning ، Voltage Regulation ، Smart Inverter ، Volt-Var.