This paper presents a data-driven adaptive control framework that integrates Long Short-Term Memory (LSTM) neural networks with dVOC to enable real-time optimization of control parameters based on sequential system measurements. The LSTM network is trained on time-domain simulation data to predict optimal dVOC gains and PI controller parameters from dynamic system states, including voltages, currents, and power flows. The proposed framework is validated in a MATLAB/Simulink environment using a two-inverter microgrid testbed under three operational scenarios: islanded operation, parallel coordination, and fault recovery. Simulation results demonstrate superior performance compared to fixed-parameter approaches, with the adaptive system maintaining stable voltage regulation during load variations, achieving effective power sharing between parallel inverters, and exhibiting rapid fault recovery with voltage restoration times of 0.45 seconds during disconnection events. Integrating machine learning with physics-based control eliminates manual parameter tuning requirements while enhancing system resilience and demonstrating autonomous parameter adaptation capabilities essential for reliable microgrid operation.
