Grid-connected inverters typically employ a virtual impedance loop to facilitate power decoupling control. However, designing the control parameters to achieve optimal performance is a challenging task. This paper introduces a novel power decoupling index as a quantitative measure for evaluating the effectiveness of power decoupling control. The training dataset is generated under 800 sets of parameters by using Real Time Digital Simulators (RTDS). A deep neural network (DNN) surrogate model is developed to establish the relationship between control parameters and the power coupling index. To optimize the control parameters effectively, a Genetic Algorithm (GA) optimization process is employed based on the trained surrogate model. Simulation results are provided to verify the effectiveness of the control parameter optimization method.