Robust Voltage Control in Distribution Networks via CVaR-Based Bayesian Optimization

The rapid proliferation of distributed solar photovoltaic systems has intensified voltage fluctuations and uncertainty in distribution networks. Traditional Volt/VAR control strategies often struggle with robustness against extreme scenarios and impose high communication overheads. To address these challenges, this paper proposes a Bayesian Evolutionary Optimization with Conditional Value at Risk (BEO-CVaR) framework for optimizing Volt/VAR control rules. This novel approach integrates Conditional Value at Risk (CVaR) into the objective function to explicitly mitigate tail risks arising from grid uncertainties. Furthermore, it employs Bayesian Evolutionary Optimization (BEO) utilizing Gaussian process surrogate modeling to efficiently solve the computationally expensive, black-box optimization problem. Validation on a standard IEEE test feeder demonstrates that BEO-CVaR achieves superior voltage regulation, strict adherence to safety standards, and significantly reduced communication requirements compared to conventional decentralized strategies. Additionally, the framework’s scalability and robustness are verified through extensive experiments across varying dimensions of decision spaces, confirming its effectiveness in complex multi-inverter coordination scenarios.