本文提出了一种船舶航迹保持的在线神经网络控制器。该控制器能用舵角控制航迹偏差，它去除了对船舶运动模型的辨识过程，它也不需“教师”或离线训练。为克服船舶过冲引起的航迹偏差等缺点，还设置了适当的训练目标。仿真结果表明该控制器具有很好的控制精度、动态特性和鲁棒性。 This paper presents an online neural network controller for ship track keeping. The controller can control the track deviation with the rudder angle, which removes the identification of the ship’s motion model and does not require “teacher” or off-line training. To overcome the shortcomings of track deviation caused by ship overshoot, but also set the appropriate training objectives. Simulation results show that the controller has good control accuracy, dynamic characteristics and robustness.