The increasing use of distributed generation has revealed limitations in conventional power electronic converters, which are unable to provide sufficient inertia and damping support to the grid. As a result, virtual synchronous generator (VSG) has gained widespread adoption for regulating the output voltage and frequency. However, VSG may encounter challenges such as generating large inrush currents and power fluctuations during on-grid switching, significantly reducing the efficacy of virtual synchronous control strategies. Therefore, this study optimizes the dynamic performance of VSG based on grid-connected switching control strategy using radial basis function neural network (RBFNN) integrated nonlinear active disturbance rejection control (NLADRC) approach. In comparison with the conventional pre-synchronization control strategy, the proposed strategy effectively suppresses system variable oscillations through the NLADRC approach. This facilitates the rapid restoration of system output frequency, voltage, and power to the steady state, thereby enhancing transient stability. Moreover, the RBFNN-NLADRC approach leverages the robust fitting capability of the network for obtaining dynamic parameter information, which allows for gain parameter tuning, further enhancing the effectiveness of the proposed strategy. Finally, verifications conducted in MATLAB/Simulink and a Starsim hardware-in-the-loop environment illustrate the superiority and feasibility of the proposed strategy.