The vibration and radiation noise caused by the mechanical power equipment running on the ship have great harm, and seriously reduce the stealth performance and combat ability of the ship. The feedforward control algorithm which depends on the precise model will fail due to the adverse factors such as the long running of the power plant without stopping or the external impact. The traditional method of on-line system identification using auxiliary white noise not only reduces the control performance, but also increases the convergence time of the identification process. The method proposed in this paper uses the control signal to model the controlled system required by the FxLMS algorithm online in the noise frequency band, with faster convergence speed and identification accuracy. When the controlled system changes abruptly, that is, when the phase frequency characteristics of the controlled system change beyond ±90°, the algorithm can also track the changes of the system in real time and maintain the stability of the control. The active vibration control of the single-layer power unit vibration isolation platform was studied. The experimental results showed that the online identification of FxLMS control algorithm achieved 20.44 dB noise reduction at the motor operating frequency (50 Hz) when there was no secondary path model. The on-line identification algorithm can also maintain the control stability and quickly identify the changes in the phase frequency characteristics of the system after the mutation of secondary path.