Design and Analysis of Modular Neural Network-Based PI-Controller Ensemble With Karush-Kuhn-Tucker Conditions for Grid-Connected Photovoltaic Systems Under Ground Fault Conditions

The increasing penetration of grid-connected solar PV system necessitates roboust and adaptive control strategies to mitigate the adverse effects of fluctuating grid voltages, variable solar irradiance, and ground faults. Traditional PI controllers through their widely used simplicity lack adaptability under non-linear and time varying conditions, leading to poor dynamic performance, increased total harmonic distortion, and compromised grid stability. To address these challenges, this paper proposes a Modular Neural Network (MNN)-based PI controller, which dynamically tunes proportional and integral gains in real time, ensuring optimal system performance. This work bridges the gap between fixed-gain PI controllers and heuristic optimization techniques by introducing a data-driven, real-time adaptive control methodology. Experimental validation through hardware implementation demonstrates a 45% reduction in steady-state error, a 30% reduction in THD, and improved reactive power compensation and grid synchronization. The findings establish the MNN-PI controller as a superior alternative for enhancing the reliability, efficiency, and grid compliance of SPV systems, particularly under dynamic operating conditions and fault scenarios.