Eliminating Cascaded Control Loops: A Neural Network Approach for Grid-Forming Inverters

Grid-forming inverters (GFMIs) are fundamental in ensuring voltage and frequency stability in islanded microgrids. Traditional control architectures rely on cascaded voltage and current control loops, which require precise tuning and may exhibit limitations in dynamic response and adaptability. In this work, we propose a neural network-based control strategy that eliminates the conventional voltage and current control loops, directly generating the modulation index in the dq frame of reference. This approach simplifies the control structure while enhancing adaptability to system variations. The neural network is trained to learn the nonlinear mapping between system states and optimal modulation signals, resulting in improved transient response and robustness against disturbances. Simulation re-sults demonstrate that the proposed methodology achieves the expected system behavior and maintains a low mean squared error (M S E = 3%), ensuring stable and reliable operation. The findings indicate that neural network-based control offers a promising alternative to conventional methods, paving the way for intelligent inverter control in resilient microarids.