Research On Autonomous Collision Avoidance Method Of USV Based On Machine Learning Method

Unmanned surface vehicle (USV), as a high-autonomous surface ship, has great prospects for development in civilian areas, military and research of ocean exploration and development because of its strong endurance and intelligent. The precondition of USV's safe task is its ability of avoiding obstacles autonomous. This capability is based on the followings: firstly, USV's collision avoidance model and collision risk model are built based on the actual situation; secondly, USV searches data of dynamic and static obstacles,and estimate dynamic obstacles' motion state; finally,intelligent algorithms are used assist the USV develop collision avoidance. In this paper, the following content are mainly studied.Firstly, the paper builds a three-degree model of USV. Based on ship modeling theory,two common coordinate systems are established, hull coordinate system and northeast coordinate system. Then, on the basis of the kinematic model, USV's surface mathematical model with three degrees of freedom is established combined with ship motion modeling theory for USV's features, then effectiveness and stability of USV are verified by direct and rotary test.Secondly, the paper builds a collision's avoidance decision model of USV. This paper deeply studies theories in the field of USV's collision avoidance, analysis collision avoidance procedures and encounter situation between USV and obstacles, then, propose appropriate measures to avoid obstacles according to different encounter situations.Parameters involved in the process of collision avoidance are calculated, and collision avoidance model is established. Then the paper calculates the degree of danger of collision between the obstacles and USV by building a collision risk model, and design examples to verify the feasibility of the model.Thirdly, the paper constructs outlines of static obstacles and estimates motions of dynamic obstacles. The paper clusters the static obstacles in USV's sailing environment by using Support Vector Clustering Method, construct a macro outline of static obstacles,exclude outliers of the obstacles, and reduce the randomness of obstacles' data. Then build a prediction model by using Elman network,predict obstacles'changes of movement based on the data of their existing position of motion by using predictive control, by the long time domain's optimization, obtain obstacles' information of motion state closest to the actual.Those studies are strong guarantees for USV's safe avoiding obstacles for the next step.Finally, the paper makes USV avoiding collisions of static obstacles based on improved artificial potential field method. The paper researches the basic principles of Artificial Potential Field, and describes the defect of the basic potential field, improve function of repulsion field, add two factors, the distance between USV and the target point,the estimated time of collision between USV and the obstacles, complete avoidance of static and dynamic obstacles by using the force USV receives in the improved artificial potential field, solve the problem of USV's local shock caused by falling into a global minima, and the problem that path between near obstacles can not be found and USV can not reach the preset goals and other issues.