Abstract: Aiming at the problems of poor formation synchronization and unsatisfactory pollination effect caused by the lack of an accurate cooperative operation control system in the mechanical rope-impact pollination for hybrid rice seed production, a cooperative pollination control method for the machine group based on the Physics-Informed Neural Network(PINN) is proposed. This algorithm equips each vehicle in the pollination unit with an independent Distributed Model Predictive Controller(DMPC) to address the complex constraints and nonlinear issues of the system in farmland operation scenarios, ensuring the independence and flexibility of each vehicle. Kinematic information is integrated into the PINN network in the form of partial differential equations. Leveraging PINN's advantages of combining data-driven and physical law constraints, by learning the actual motion states and physical movement laws of vehicles in the real world, the PINN network can predict the state quantities of the next time step by inputting the current time, current state quantities, and current control quantities. Combined with the self-circulation strategy, it quickly meets the DMPC's computational requirements for future state quantities, effectively solving the pain points of low solution efficiency and insufficient real-time performance of traditional model predictive control in nonlinear systems. Finally, the predicted state quantities are input into the DMPC for rolling optimization. By constructing a cost function including tracking accuracy error, control stability error, and terminal stability error, the optimal control quantities for the next moment are obtained under the premise of satisfying the vehicle's physical performance constraints, realizing the precise tracking of the target position and formation coordination of the pollination unit, and completing the pollination operation. To comprehensively verify the tracking accuracy, formation coordination ability, and actual operation adaptability of the algorithm, simulation experiments, ground tests, and field experiments were sequentially carried out in this study. The simulation results show that compared with the Adaptive-PID and Pure-Pursuit algorithms, the PINN-DMPC algorithm exhibits better performance in tracking error and asymptotic convergence. The tracking error is within \text± 0.032 m with a standard deviation of 0.0098 m, enabling precise and stable path tracking, and the yaw angle error is stabilized within 0.022 rad, providing a reliable directional reference for position tracking. In terms of solution efficiency, the average single-step solution time of PINN-DMPC is 0.011 s, which is 54.2% less than that of the original DMPC, significantly improving the real-time response capability of the controller and reducing the performance and cost requirements for control equipment. In the ground test, the effective operation range of the actual simulation can cover 84% of the entire road section in terms of ground formation coordination performance, and rapid adjustment can be achieved in the case of position deviation. The longitudinal distance error of the formation is controlled within 0.1 m, and the included angle between the unit and the rice parental rows does not exceed 9°, ensuring the stability of formation coordination. This study provides a set of cooperative control algorithms suitable for pollination in hybrid rice seed production. The algorithm has precise path tracking performance and cooperative operation performance, laying a foundation for the modernization and automation of hybrid rice seed production operations.