Research On The Stability Test Of Ship Autonomous Collision Avoidance Algorithms

With the rapid development of intelligent ships,autonomous collision avoidance algorithms of ships have been widely concerned.Before the autonomous collision avoidance algorithm is put into use,safety,reliability and stability must be tested.Based on machine learning theory,this thesis analyzes the performance of ship autonomous collision avoidance algorithm in boundary test scenarios,and conducts a simulation study on the stability test method of ship autonomous collision avoidance algorithm in the case of two ships meeting in open waters.The main work of this thesis is as follows:Firstly,the types and value ranges of the component factors of the scenario of two ships’ encounter in open waters are studied.After sample scenarios are selected,the compliance evaluation of the autonomous collision avoidance algorithm of ships is carried out and the sample database is established.On this basis,gauss process regression is used to expand the sample database and build the test database.An example is given to test the process of database establishment to verify the feasibility of the method.Then,the mean shift unsupervised clustering algorithm is used to cluster the score data contained in the test database,and the performance score clustering of the autonomous collision avoidance algorithm in different scenarios is obtained.A density-based unsupervised clustering algorithm with noise is used to cluster each scene again,and the sub-data set under each scene cluster is obtained.The Euclidean distance between the sub-data sets under different scene clustering was calculated to identify the boundary scenes,and the required boundary scene pairs were obtained.An example experiment of boundary scene identification is carried out to verify the feasibility of the method.Finally,the stability of collision avoidance algorithm is analyzed according to its performance in boundary scenes.Based on the ship maneuvering simulator of Dalian Maritime University,the stability test data interface of collision avoidance algorithm was standardized,and the stability simulation test platform of ship autonomous collision avoidance algorithm was built.In this thesis,245 pairs of boundary scenarios are generated through the establishment process of test database and identification of boundary scenarios,and the stability analysis is carried out by randomly selecting four pairs of boundary scenarios corresponding to the four typical encounter scenarios: two-ship pursuit,left-crossing encounter,right-crossing encounter and encounter.It is found that although the compliance evaluation score of the tested collision avoidance algorithm is not zero in all encounter scenarios,that is,all the tested collision avoidance algorithms successfully avoid in all encounter scenarios.However,in the selected four pairs of boundary scenes,the collision avoidance performance of the tested algorithm still has a large mutation in one of the boundary scene pairs,indicating that the stability of the tested collision avoidance algorithm is poor.Therefore,even if the tested collision avoidance algorithm can meet the requirements of collision avoidance,its stability needs to be tested to prevent the collision avoidance performance of the tested collision avoidance algorithm from mutation in some scenarios,resulting in a relatively dangerous situation in the collision avoidance process.This proves the necessity,feasibility and rationality of the stability test method proposed in this thesis,thus laying a certain foundation for the construction of the method system of intelligent ship navigation function test and verification,which is of great significance to promote the theoretical research and commercial application of ship autonomous collision avoidance algorithm.