To achieve the automatic cleaning of insulators in the catenary along railway lines and ensure the safety of the power supply system of electrified railways, an insulator water rinsing robot tracking rinsing system based on multi-variable model predictive control (MPC) was proposed. Based on the MPC control principle, a prediction model was established for the servo power part of the water flushing robot, and a multi-variable MPC control strategy for the rotating water cannon was designed to achieve comprehensive control of the rotation position and speed of the servo motor of the rotating water cannon. The terminal error weight matrix was optimized to maintain the stability of the flushing action throughout the tracking process, and the finite set model of the current loop was integrated for predictive control to improve the dynamic response speed of the entire system. The simulation experiments show that compared with the traditional PI cascade control, this control strategy shortens the forward response distance of the water rinsing robot when aiming at the target insulator by more than 50%, and reduces the average switching time in multi-target scenarios by 0.215 s. The field experiments verified the effectiveness of the system at an operating distance from 3 m to 15 m. The maximum aiming error is reduced from 47 mm to 24 mm, meeting the flushing requirement that the aiming error in actual operation is no more than 30 mm.