Risk Research Of Maritime Traffic Accidents Based On TAN-Bayes

Maritime traffic safety is an important guarantee to ensure the sustainable development of maritime transportation.With the development of economic globalization,international trade is becoming more and more frequent,and the volume of maritime freight is also increasing.At the same time,the risk of maritime accidents is also increasing.The consequences of maritime traffic accidents are often very serious,which may lead to casualties,ship damage,economic losses and serious marine environmental pollution.So,the problem of maritime traffic safety can not be ignored.The purpose of this thesis is to construct a new risk analysis method based on Bayesian network to analyze the relationship between maritime traffic accident risk and risk factors.First of all,the accident data are collected and processed from the global maritime accident report 2018-2021 recorded in GISIS.In fact,the accident report does not record the normal traffic condition,and there is a problem of data imbalance in the research process,which may lead to the deviation of the research results.Therefore,this thesis uses the generative adversarial neural network(GAN)to generate sample data close to the real data,and construct a new database together with the collected data.Secondly,according to the previous literature research and combined with the accident report,identify the risk factors of maritime traffic accidents,regard them as the nodes of the risk model,and divide the state of each risk factor.The tree augmented naive Bayesian algorithm is used to learn the network structure,and the expectation maximization algorithm is used to calculate the conditional probability table of each node to construct the TAN-Bayes network model.Through data training and machine learning,the established marine traffic accident database is analyzed to reveal the impact of related factors on accident risk and its mechanism.Finally,the model is verified and the results are analyzed.Through the correctness test,prediction performance and actual cases to evaluate the effect of the model,the sensitivity analysis is carried out by using mutual information,joint probability and real risk influence.Among them,the stronger the correlation between the corresponding variables and the accident risk,the greater the mutual information value;the joint probability is used to measure the impact of each state of important factors on different accident risk levels;in addition,the impact of important factors on each accident risk can be effectively obtained by ranking the risk factors of marine traffic accidents according to the TRI value.And through the most likely explanation(MPE),it explains the possible situations in different situations and identifies the related potential risks.It is found that accident type,ship type,deadweight,wind speed,gross tonnage and location are the most important factors in accident risk.Fire / explosion,capsizing / sinking and drowning are all most likely to cause very serious accidents.In addition,fishing vessels and other small vessels are more likely to have very serious accidents than other types of vessels.By using TAN Bayes network for analysis and research,we can determine which factors are most likely to cause maritime traffic accidents and have serious risk,which will help to analyze and predict maritime accidents and take targeted preventive measures,help to improve maritime traffic safety,reduce the risk of accidents,protect the safety of crew and cargo,and benefit shipowners and other maritime stakeholders.