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

With the need of military,resources,scientific research and other fields,people are increasingly aware of the exploration of the ocean.Unmanned Surface Vehicle(USV)is an Unmanned Surface Vehicle that can make autonomous planning and decision-making navigation on the sea Surface,and accomplish boring and dangerous tasks for people.Both in military,civilian and scientific research fields have broad prospects for development.Ship collision accident is the main component of maritime traffic accident,and autonomous collision avoidance of unmanned boat is the foundation of maritime safe navigation.Based on unmanned craft of maritime navigation safety as the primary factors to consider,using machine learning algorithms for clustering static obstacles to track prediction of dynamic obstacles,the clustering of static obstacles turns the obstacles that need a large number of calculations to avoid collision into obstacles with a small number.The trajectory prediction of dynamic obstacles can obtain the trajectory of dynamic obstacles in the future,lighten the unmanned craft collision complexity,enhance the accuracy of collision avoidance,centered on safety collision avoidance,study the position of obstacles and trajectory,mainly completed the following content:First,a framework is established for the knowledge of obstacle avoidance of unmanned craft.The collision avoidance process of the unmanned boat against obstacles and the encounter situation between the unmanned boat and the dynamic and static obstacles were studied in depth.The basic flow of obstacle avoidance and the collision avoidance stage were determined,and the encounters were subdivided into different angles according to the different encounter situations.The kinematics model of the unmanned boat is established and various parameters of the ship are calculated.The risk model of the unmanned boat collision is established so that when the unmanned boat is in danger of collision,it can judge when to respond according to specific parameters.Secondly,the machine learning algorithm is used to cluster the static obstacles of the unmanned boat.Clustering of the selected machine learning algorithm for the peak density algorithm,using density peak algorithm to static obstacle clustering can make the scattered distribution of reefs and small islands,the obstacles of small area is divided into a class,as an obstacle,the obstacle of dense area simplification,reduce the difficulty of collision avoidance,the unmanned sailing boat more secure.Then,the dynamic obstacle trajectory of the unmanned vehicle is predicted.The machine learning algorithm chosen is RBF neural network.The principle of RBF neural network is studied in depth,and the learning process of RBF neural network is optimized by using the improved Wolf pack algorithm,the parameters of each part of the neural network are determined step by step,the nonlinear trajectory is predicted,and the improved Wolf pack algorithm is compared with that before the improvement,which shows the effect of the improvement.Predicting the trajectory of dynamic obstacles can make the timing and movement of collision avoidance more accurate.Finally,the dynamic and static obstacles of the unmanned boat are avoided.The time and action of avoiding collision of unmanned craft are studied.Based on the speed of obstacles,the geometric model of avoiding collision is established,and the steering Angle is calculated.The whole process of the algorithm can be completed independently by the unmanned boat,realizing the intelligent navigation of the unmanned boat on the sea.