基于泊位坐标的无人船自主靠泊仿真研究

针对神经网络自主靠泊控制器只能完成特定港口的靠泊,而无法拓展到其他没有训练数据的泊位这一问题,提出一种以泊位顶点为原点,岸线为y轴的坐标系(泊位坐标系)。采用泊位坐标系中相对位置来训练神经网络控制器,在V.Dragon-5000航海模拟器中选取集装箱船“银河号”,在大连港进行靠泊训练后,成功将靠泊控制器拓展到没有训练数据的深圳蛇口港与新加波樟宜港。仿真验证了采用新建立的坐标系可以拓展神经网络自主靠泊控制器适用范围,减小训练数据所需的成本。

Research on Autonomous Berthing for Unmanned Ship Based on Berth Coordinates

Aiming at the problem that the neural network autonomous docking controllers can only complete the docking of a specific ports, but cannot extend to other ports without training data, a coordinate system (berth coordinates with the berth vertex as theorigin and the shoreline as the Y Axis) is proposed. The relative position is used to train the controller. In V.Dragon-5000 navigation simulator, the container ship “Yinhe” is selected. After docking training at Dalian port, the docking controller is successfully extended to Shenzhen Shekou port and Singapore Changi port without training data. The simulation verifies that the application of the newly established coordinate system can greatly expand the application range of the neural network autonomous docking controller and reduce the cost required for training data.