ML-Assisted Hyperparameter Tuning for Distributed Voltage Control Algorithm in Distribution Systems

Traditionally, power system utilities have relied on centralized or local Volt-Var approaches for voltage regulation in distribution grids. However, with recent increase in integration of Distributed Energy Resources (DERs) with advanced power electronics control and computation, distributed voltage control algorithms offer several advantages, including scalability, robustness against communication failures, data privacy, and fast computation. A significant drawback of distributed approaches, however, is their heavy reliance on the proper selection of algorithm parameters for better performance. If these parameters are not chosen correctly, it can result in hundreds or even thousands of iterations, potentially leading to non-convergence. This paper proposes a Machine Learning based hyperparameter tuning approaches for a distributed voltage control algorithm to achieve the fastest convergence. Three methods-Random Search, Grid Search, and Genetic Algorithm-are employed for this purpose. The hyperparameter-tuned distributed voltage control algorithm is validated in the unbalanced multiphase IEEE 123-bus feeder, using OpenDSS as the power network modeler. Results demonstrate superior results using the hyperparameter tuning methods.