A Multi-agent Based Distributed Voltage Control Scheme Using Machine Learning and Smart Inverters

A new distributed voltage control strategy for PV power systems that does not need support from centralized SVCs is proposed. The methodology uses smart inverters, agent-based coordination, and machine learning-based forecasting to offer a scalable and economical solution for decoupling voltage variations in the context of high penetration of PV. Each inverter acts as an autonomous agent that regulates its reactive power output using local voltage measurements and short-term irradiance predictions derived from a Long Short-Term Memory (LSTM) model. The agents cooperate with their neighbors, utilizing a consensus algorithm for coordinated voltage control throughout the network. This decentralized strategy enables fast, adaptive, and cost-effective voltage stabilization without relying on hardware-intensive centralized devices. The effectiveness and reliability of the proposed control strategy are verified through a simulation study using a five-bus radial distributed generation (DG) system with high PV penetration. Simulation results on a five-bus radial distribution feeder show better voltage stability, fault recovery, and reactive power utilization as compared with conventional and existing distributed control strategies. The findings confirm the feasibility of software-defined, inverter-based voltage regulation as a practical alternative for future smart grids. In addition, the proposed framework offers extensibility to hybrid renewable energy systems, such as wind and storage, supporting the transition toward resilient, low-carbon, and data-driven energy infrastructures.