Research On Risk Interaction And Prediction Of Maritime Accidents For Seafarers’ Unsafe Acts

With the accelerated construction of "strong maritime country","the Belt and Road " initiative,the establishment of "maritime community with a shared future " and a series of new marine development strategies,the marine economy has been developed rapidly and the demand for maritime transportation has increased significantly.However,the increasing maritime transportation activities inevitably bring about maritime accidents,which cause serious economic losses and casualties,and even lead to marine pollution and other environmental problems.80% of maritime accidents are attributed to seafarers’ unsafe acts.Seafarers are the only part of the maritime transportation system with subjective initiative.The coupling and superposition of factors from organization,technology,environment,and individual traits on individual seafarer makes the generation of seafarers’ unsafe acts uncertain,sudden and random.In order to minimize the occurrence of maritime accidents caused by seafarers’ unsafe acts,it is necessary to conduct in-depth research on the complex interaction characteristics of seafarers’ unsafe acts and their influencing factors involved in maritime accidents from a risk management perspective.Based on the systematic cognition of seafarers’ unsafe acts and their influencing factors,a systematic analysis and classification model is developed to provide the basis for the interaction analysis of seafarers’ unsafe acts and their influencing factors,and then the complex risk evolution mechanism of seafarers’ unsafe acts is explored in depth and identify the key nodes in risk management,thereby achieving the interpretable prediction of maritime accident type and severity under the interaction of seafarers’ unsafe acts and their influencing factors,which is of great significance to improve the safety level of maritime transportation.In this regard,the present study carried out the following researches:Firstly,based on the systematic identification of seafarers’ unsafe acts and their influencing factors,a systematic analysis and classification model is developed.For the five typical types of maritime accidents,476 maritime accident investigation reports are collected and screened,31 typical seafarers’ unsafe acts and 38 influencing factors are identified based on grounded theory,and the analysis and classification model of seafarers’ unsafe acts and their influencing factors is developed by improving the HFACS model;Further,the dataset of seafarers’ unsafe behaviors and their influencing factors is established,and statistical analysis of seafarers’ unsafe acts and their influencing factors is conducted.The results show that the risk interaction process of seafarers’ unsafe acts is different,and it is necessary to investigate the interaction of seafarers’ unsafe acts and their influencing factors in order to take targeted safety management measures;Then,the éclat algorithm is used to mine the frequent itemsets between seafarers’ unsafe acts and their influencing factors,and with the help of the analysis and classification model of seafarers’ unsafe acts and their influencing factors,the risk association pattern of seafarers’ unsafe acts is explored,so as to find that the risk association pattern has certain regularity and lay the foundation for the following research on the interaction of seafarers’ unsafe acts and their influencing factors.Secondly,the interaction network of seafarers’ unsafe acts and their influencing factors is developed,and the risk evolution characteristics and laws of seafarers’ unsafe acts are explored from a multidimensional perspective.Based on the results of frequent itemsets between seafarers’ unsafe acts and their influencing factors,valid association rules are generated and introduced into the complex network to develop a directed weighted interaction network of seafarers’ unsafe acts and their influencing factors with 53 nodes and 260 edges,which avoids the defect that the weight of edges is over-reliance on experts’ subjective experience during the traditional complex network construction process;Further,through the global network topology feature analysis,it is determined that the interaction network of seafarers’ unsafe acts and their influencing factors has small-world network characteristics,and the weighted Page Rank algorithm is adopted to obtain the importance of nodes in the network and clarify the risk propagation capabilities of seafarers’ unsafe acts and their influencing factors in the interaction network;Then,based on the community detection method of random walking,the local community structure in the network is extracted to explore the typical interaction patterns of seafarers’ unsafe acts and their influencing factors,and clarify the risk propagation paths of seafarers’ unsafe acts,which provide vital guidance for interrupting the occurrence of seafarers’ unsafe acts.Finally,selective ensemble learning technique is introduced to enhance the prediction accuracy of maritime accident type and severity under the interaction of seafarers’ unsafe acts and their influencing factors.Applying the node importance results in the interaction network of seafarers’ unsafe acts and their influencing factors for feature selection,multiple heterogeneous individual learners are built and hyperparameter optimization is performed using random search and grid search algorithms,respectively;Then,a selective ensemble learning method that balances model accuracy and diversity is proposed,and the maritime accident type and severity prediction models are constructed based on the Stacking ensemble learning algorithm.Through model performance comparison and robustness analysis,the prediction accuracy,stability and generalization ability of the proposed model are proved;Further,the SHAP method is introduced to clarify the effective mapping association and the association degree between the model prediction results and seafarers’ unsafe acts and their influencing factors,and to achieve the interpretability of the prediction model;Finally,the simulation data of the interaction between seafarers’ unsafe acts and their influencing factors are used to demonstrate the practical application of the maritime accident type and severity prediction model under the interaction of seafarers’ unsafe acts and influencing factors.