When ships are sailing on the sea,roll motion will greatly reduce the safety of ships and cargo,as well as the health of the crew.Therefore,the ship roll stabilization device has become one of the indispensable equipment on the ship.As an active roll reduction device,fin stabilizer is widely used for roll reduction due to its good anti-rolling performance.In this paper,an online learning optimal controller based on action dependent heuristic dynamic programming (ADHDP) is proposed for the ship fin stabilizer system.The input and output data,instead of a system model,are used in the design to obtain the system state.Two back propagation neural networks,including a critic network and an action network,are used to approximate the performance function and obtain the control law,respectively.The two neural networks can use real-time measurement data,and reduce internal model error and the uncertainty disturbance,thus improve the robustness of the system.Finally,the effectiveness of the proposed ADHDP controller is validated by simulation results.