Autonomous Voltage and Reactive Power Control of Grid-Forming Inverters Using Physics-Guided ANN-Based Virtual Impedance

Voltage control is problematic in an islanded microgrid, as small and mismatched feeder impedances lead to inaccurate reactive power sharing among grid-forming inverters and potential instability under conventional droop control. Existing adaptive virtual impedance solutions often depend on communication links, creating a system vulnerability. This study introduces an autonomous control strategy to enhance reactive power sharing without requiring communication. The proposed method utilizes an artificial neural network (ANN) consisting of an offline and online phase to determine the optimal virtual impedance locally at each grid-forming inverter. During an offline phase, a physics-aware recursive least-squares (RLS) algorithm is used to generate a training data set. In online operation, the trained ANN is a lightweight model that uses only local measurements to calculate the required voltage compensation. This ANN-based virtual impedance is a practical and adaptable solution for autonomous voltage and reactive power control. By eliminating communication dependency, this strategy enhances microgrid stability, reliability, and scalability, offering a significant improvement over communication-based methods in terms of cybersecurity. MATLAB/SIMULINK simulations validate the approach, showing that the controller achieves precise reactive power sharing under varying loads and eliminates steady-state errors. Significantly, it maintains robust performance during communication failures and seamlessly adapts to the grid changes.