Research On Trajectory Prediction And Path Planning Methods For Intelligent Collision Avoidance Of Vessels

As one of the most important modes of transportation for international trade,sea freight continues to grow in volume in the context of the era of trade globalization.However,vessel collisions have occurred frequently in recent years,resulting in tragic casualties and serious environmental pollution.In order to solve these problems,more and more researches have been conducted to make the maritime navigation of vessels more automated and intelligent by applying artificial intelligence and sensor technology,and intelligent collision avoidance of vessels has become a hot spot of international attention.The purpose of this thesis is to study two core areas in vessel intelligent collision avoidance technology: vessel trajectory prediction and vessel path planning.Considering the influence of the time interval difference between trajectory data points in different regions on the trajectory prediction accuracy and the degradation of the existing vessel trajectory prediction methods in different regions,an improved sequence-to-sequence vessel trajectory prediction method based on AIS(Automatic Identification System)is proposed;considering the influence of the simulated simulation environment and Considering the influence of the difference between the simulated environment and the real environment on the path planning and the problem that the existing traditional path planning algorithm involves a large amount of storage and calculation in the path planning process,which leads to the long path planning time and high accident rate,we propose the double-layer vessel path planning method based on rasterization.The main research work is summarized as follows:(1)An improved sequence-to-sequence vessel trajectory prediction method based on AIS is proposed.The method mainly consists of two parts: trajectory data preprocessing and DCNN,an improved sequence-to-sequence vessel trajectory prediction model based on double convolution.In the trajectory data preprocessing,firstly,the complete vessel trajectory data with prediction value is extracted by restricting the time interval between vessel trajectory data points and navigation speed characteristics.Then,fixed time interval restriction and mean interpolation complementation strategy are used to eliminate the influence of inconsistent time interval size of data in different regions on trajectory prediction accuracy.Finally,the trajectory data normalization method is used to eliminate the dimensional differences in the trajectory data attributes to reduce the computational complexity of the method and improve the prediction accuracy.In DCNN,two modules of global time convolution and local time convolution are designed in the encoder stage according to the characteristics of continuity and abrupt change of vessel trajectories to extract the trajectory data features of different time scales and generate the predicted vessel trajectory sequences in the decoder.The experimental results show that the method can effectively improve the trajectory prediction accuracy on different regional trajectory data sets.(2)A rasterization-based two-layer vessel path planning method is proposed.The method mainly consists of three parts: rasterized sea area scene modeling based on Arc GIS,global path planning algorithm based on deep reinforcement learning and local path planning algorithm based on improved A*.Firstly,the rasterized map of real sea area is generated based on the geographic information data of real sea area using Arc GIS.Second,the global path planning algorithm based on deep reinforcement learning is used on the global rasterized map to generate the global planning path with the help of neural network alternative value function method to generate directional decision.Then,the CRI(Collision Risk Index)is calculated by quantifying the collision risk factors and combined with the trajectory prediction method to identify the vessel collision risk in the encounter scenario,so as to switch the path planning algorithm from global to local.Finally,on the local rasterized map,the A* algorithm heuristic function is improved according to the fuzzy calculation of vessel domain boundary,collision risk and search bias to generate local planning paths.The experimental results show that the use of neural network alternative value functions to generate directional decisions can effectively reduce global path planning time and reduce the incidence of planned path accidents;the combination of vessel domain boundary fuzzy calculation,collision risk and search bias improvement heuristic function can realize the local path planning of A* algorithm in dynamic scenes.