Water-jet propulsion has become the preferred propulsion method for high-speed unmanned surface vessels. To enhance rotational speed and power density, a highly integrated, long-lifespan, lubrication-free single-winding bearingless pump-thruster motor is employed as an effective solution. To address the limitations of conventional suspension and rotation control methods, which struggle to adapt to the complex and variable operating conditions of such platforms, a reinforcement learning-based dynamic tuning strategy for control parameters is proposed. First, a motor dynamic model is established, and the additional disturbance forces generated by the platforms’ motion are analyzed. A reward function integrating speed error, displacement error, and torque ripple is then designed, together with an operating-condition recognition mechanism. This enables the control parameters to remain fixed during steady-state operation, while allowing autonomous real-time optimization under varying operating conditions. Simulation results based on PID and LADRC controllers demonstrate that the proposed strategy effectively improves system adaptability and robustness under dynamic disturbances.