Analysis And Evaluation Of Human Risk In Autonomous Ships Based On STPA And BP Neural Networks

With the rapid development of independent technology at home and abroad,after the advent of autonomous vehicles and autonomous aircrafts,autonomous ships have also begun to appear in the public ’ s vision and gradually put into use.People can use autonomous ships to carry out navigation tasks in some complex traffic routes and changeable climate environment,so as to improve the efficiency of navigation.For example,through autonomous transport ships to transport goods to reduce crew costs and time costs.For the use of autonomous ships,the most important point is the safety problem.The degree of safety hazard and the solution determine the prospect of autonomous ships.According to the structure and system of autonomous ships,scholars conduct simulation and hypothesis analysis to study the risk factors of autonomous ships and carry out early prevention,so that we can enjoy the convenience of autonomous ships,and also contribute to improving the safety of ships sailing at sea,reducing the incidence of maritime accidents and reducing casualties.In this paper,the human risk factors of MASS I,II and III defined by IMO are studied and analyzed.In order to find the key prevention and control objects and formulate correct preventive measures,the human risk quantitative analysis model of autonomous ships in different navigation stages is proposed,and the weight analysis and evaluation of different human factors are carried out.The main work is as follows :Firstly,this paper defines the system-level risk of autonomous ships through the STPA method.According to the working process of autonomous ships and related equipment,the safety control structure diagram in the navigation stage is constructed.The unsafe control behaviors of crew and SCCs personnel are identified through the four control scenarios proposed by STPA,and the human risk factors of autonomous ships are obtained.Subsequently,637 ship accident investigation reports published by the China Maritime Administration from 2015 to 2020 were collated and analyzed.596 accident investigation reports most likely to occur on autonomous ships were selected from the safety control structure diagram for data mining.The key information of the report was extracted and the information was grouped according to the time and stage of the accident.At the same time,the BP neural network based on RPROP algorithm is constructed by using MATLAB neural network function.The data source is obtained by matrixizing and normalizing the data after grouping,and the data source is introduced into the network for training.At the same time,the neural network is continuously optimized and debugged.Finally,the effectiveness of the network is tested to make the effectiveness reach the expected effect.After the success of network training,the weights of artificial risk factors of autonomous ships are calculated by using the weights of neurons in the neural network.Finally,the human risk assessment system of autonomous ships is built by the way of separation of the front and back ends.Vue.js is used as the main system framework,Java Script language is used as the driver,Echarts.js is used as the auxiliary,Node.js and Ali cloud object storage server are combined to complete the development of the assessment application system.Through the design and development of intuitive UI interface and friendly interactive function,the accident data record and the visualization of the weight of human risk factors of autonomous ships are realized.In order to expand the accident database,enrich the data source,improve the accuracy and accuracy of artificial risk weight analysis of autonomous ships by neural network,prepare for the dynamic analysis of artificial risk of autonomous ships,provide a theoretical basis for the safety analysis of autonomous ships before they are officially put into use in the future,and provide suggestions and references for the training of SCCs personnel and autonomous ship crew.