Abstract: The conventional grid-connected control method for fractional-order virtual synchronous generators (FOVSG) suffers from a critical trade-off in dynamic response performance between grid-connected active power and output frequency (hereinafter referred to as “power-frequency”) under typical disturbances, such as step changes in active power reference and grid frequency. To overcome this inherent limitation, this paper explicitly proposes an improved grid-connected control strategy for FOVSG based on active power differential feedback (APF), designated as APF-FOVSG. A power dynamic compensation link, which extracts the derivative of the active power and passes it through a first-order filter, is incorporated into the existing FOVSG control structure, thereby forming the APF-FOVSG. This additional feedback path effectively introduces a structural damping term into the rotor motion equation, enhancing the system’s ability to dampen power oscillations without sacrificing frequency response quality. The small-signal mathematical model of the APF-FOVSG grid-connected system is rigorously derived, explicitly accounting for the fractional-order dynamics of the virtual inertia. Based on this model, the closed-loop transfer functions relating active power and frequency responses to both internal (active power command) and external (grid frequency) disturbances are obtained. Furthermore, a comprehensive parameter stability analysis method is developed, which includes frequency-domain criteria (Bode plots, phase margin) and consistent time-domain validation. To facilitate practical implementation, the Oustaloup recursive filter is employed to approximate the fractional-order operator s^μ within the frequency range of interest (0.01 to 1000 rad/s) using a fifth-order rational transfer function, and the approximation accuracy is verified through Bode diagram comparison, showing excellent matching in both magnitude and phase. The parameter selection guidelines for the APF gain k_a and filter time constant τ are discussed, and their impact on system stability and dynamic performance is analyzed. Finally, a 100 kV·A FOVSG grid-connected system is implemented both in MATLAB/Simulink simulation and on a physical experimental platform. Extensive comparative tests are conducted under two representative disturbance scenarios: a step increase in active power reference from 20 kW to 60 kW, and a step decrease in grid frequency from 50 Hz to 49.95 Hz. The simulation and experimental results consistently demonstrate that, unlike the conventional FOVSG, the existing APF-based integer-order VSG (APF-VSG), and the standard VSG, the proposed APF-FOVSG method achieves simultaneous suppression of active power oscillations and frequency overshoot. In the active power step test, the APF-FOVSG exhibits no dynamic oscillation in either active power or frequency, with a frequency overshoot amplitude of only 0.061 Hz (compared to 0.123 Hz for VSG, 0.151 Hz for FOVSG, and 0.064 Hz for APF-VSG). In the grid frequency step test, the APF-FOVSG again shows no oscillation, with an active power overshoot of merely 4.36% (versus 55.81% for VSG, 21.41% for FOVSG, and 16.32% for APF-VSG). These results conclusively validate the superiority of the proposed APF-FOVSG method in enhancing power-frequency dynamic response performance, effectively resolving the trade-off inherent in existing FOVSG approaches. The proposed method offers higher control flexibility, improved damping characteristics, and robust stability under various disturbances, making it a promising solution for grid-connected inverters requiring both inertia support and high-quality transient response.