Collision Avoidance Algorithm For Ships In Narrow Channels Based On Q-learning Algorithm

In the field of modern shipping,narrow waterway navigation has always been a very important issue.Due to factors such as space limitations and changes in water currents when ships navigate narrow waterways,there are higher safety risks in narrow channel navigation.Therefore,path planning of narrow channel ships has always been one of the hot research directions in the field of ship automation control.The traditional narrow channel ship path planning method is usually a rule-based method,such as the guide line method,anchorage method,etc.These methods can achieve good results in some cases,but there are also some problems,such as the need for different rules in different environments,and the design of these rules needs to consider many factors,which makes these methods have certain limitations in practical applications.In view of these problems,this thesis proposes a ship narrow channel collision avoidance algorithm combined with the Q-learning algorithm of artificial potential field method,so as to realize the collision avoidance operation of ships in narrow waterway under the constraints of the International Maritime Collision Avoidance Regulations(the "Rules").Firstly,this thesis comprehensively considers the heading and speed limitations of ships in narrow waterways,combines the ship collision risk model and water environment model,and uses the artificial potential field method to establish an overall environmental model,and divides it according to the risk value and the situation of the narrow waterway.Then,combined with the requirements of the Rules,the constraints that should be met in different collision avoidance situations are analyzed.Finally,the gravitational and repulsive models constructed by the artificial potential field method are used as the reward data of the reinforcement learning Q-learning algorithm for training,so as to obtain the final collision avoidance algorithm.In terms of the application of the Rules,the artificial potential field method is used to construct a reasonable repulsive field and a special reward function is designed to apply the Rules.In this thesis,the following work is done: the gravitational repulsive field is designed by using the artificial potential field method combined with the ship collision model to meet the input of the Q-learning algorithm;In order to solve the situation that the artificial potential field method is prone to jitter in narrow waterways,special training is carried out by using deep learning to avoid the problem of multiple turns in narrow waterways.The Q-learning algorithm is designed to avoid the local minimum of the artificial potential field method.In order to verify the feasibility of the narrow channel collision avoidance system designed in this thesis,the scenario simulation of the Yangtze River waterway is carried out at the end of the thesis,and the simulation scenario is designed to simulate avoidance,and the simulation results show that the narrow channel collision avoidance algorithm proposed in this thesis can ensure that the actions taken by the experimental ship can not only comprehensively consider the current navigation environment,but also comply with the relevant requirements of the Rules for collision avoidance in narrow waterways.