Modified Model-Free Predictive Control for Reliable Operation of Multiple Parallel Grid-Forming Inverters

The increasing integration of parallel grid-forming (GFM) inverters within islanded AC microgrids introduces significant challenges in maintaining robust voltage regulation and accurate power sharing. Central to controlling the output voltage is the need for a reference voltage, which can be derived using the virtual synchronous generator (VSG) concept. The VSG not only ensures active power sharing but furthermore provides inertia emulation, reducing the rate of change of frequency (ROCOF) in response to sudden load variations, thereby enhancing system stability and frequency support. To achieve precise voltage regulation, a model predictive control (MPC) strategy is commonly employed, particularly the finite control set-MPC (FCS-MPC). However, FCS-MPC relies on an accurate system model, which can limit performance under parameter mismatches or dynamic changes. To overcome this limitation, this paper proposes a modified model-free predictive control (MFPC) based on the dual second-order generalized integrator-quadrature signal generator (DSOGI-QSG) strategy. The MFPC eliminates the need for an explicit system model by using an ultra-local model (ULM) for prediction, with the unknown function in the ULM playing a critical role in the accuracy of output voltage prediction. Algebraic identification methods are typically used to estimate this unknown function. Nevertheless, as the sampling time increases, the estimation becomes increasingly distorted, leading to prediction inaccuracies. The DSOGI-QSG is introduced to extract a filtered version of the unknown function, improving the quality of the estimation and, consequently, the prediction accuracy. Simulation results using MATLAB Simulink demonstrate the effectiveness of the proposed method, showing enhanced robustness, improved dynamic response, and reduced harmonic distortion compared to conventional MFPC.