关键词: 水路运输, 海上通道, 风险预警, 动态贝叶斯网络, 突发事件, 敏感性分析

Abstract: Sea lanes is an important carrier of cargo transport, and its safety is related to the development of China's maritime trade. With the continuous development of ocean shipping transportation, its risk level is constantly increasing. Risk early warning is an effective way to reduce the probability of emergency. In order to ensure the safety of sea lanes, it is of great significance to study the risk early warning of sea passage. The existing research on the risk of sea lanes mainly includes risk analysis, evaluation, management and emergency response, which is little on risk early warning. Therefore, this paper constructs a risk early warning model of sea lanes based on DBN to provide a decision reference for ensuring the safety of sea transport and improving the risk early warning and emergency response capability of sea lanes.
BN is one of the methods of expression and reasoning of uncertain knowledge in the field of artificial intelligence. DBN introduces time variable on the basis of BN, which can carry out dynamic analysis and early warning of emergencies. EM algorithm and Viterbi algorithm are used to learn the parameters of DBN model and predict the risk of sea lanes respectively. EM algorithm is an iterative algorithm, mainly used to calculate the mode or maximum likelihood estimation of the posterior distribution, and is widely used in the so-called incomplete data statistical inference problem such as missing data, truncated data, cluster data, data with unwelcome parameters. EM algorithm is simple and stable, and can find the optimal convergence value very reliably. Viterbi algorithm is a dynamic programming algorithm, which is used to find the hidden state sequence that is most likely to produce the observed event sequence, and is widely used in dynamic prediction and programming.
A risk early warning model based on DBN is proposed to predict the risk of sea lanes. The factors are identified according to the statistics of historical emergencies. The initial network structure of DBN is worked out based on the risk characteristics of sea lanes. The historical data of emergencies happening in Indian Ocean from 2008 to 2017 is used for warning the risk of sea lanes. The case study data in this paper come from the risk events in the Indian Ocean, and the observation data come from wind speed from Remote Sensing Systems, sea weather forecast, ship traffic statistics, military exercises, and reports published by the Labor Market Association.
A comparative analysis of DBN model using BN model and Markov model is conducted to verify the effectiveness of the proposed model. F-measure, accuracy, precision and recall are used as indicators to evaluate the accuracy of risk warning results. The sensitivity analysis is conducted to determine the sensitivity between sea passage risks and influencing factors.
The results show that the risk of sea lanes is within a small range but has a downward trend overall. Compared with the BN and Markov models, the results of the DBN model have higher accuracy, which is 9.3% and 9.2% higher than that of the BN and Markov models, respectively. The risk early model established in this paper can effectively alert the risk of sea lanes, identify the key risk factors, and provide decision-making references for improving the risk early warning and emergency management capabilities of sea lanes.
This study applies the DBN-based model to predict the risk of sea lanes. The expansion of DBN in risk early warning and the applicability of further development algorithms to other types of early warning research can be studied in the future. Future works can also be extended in applying the DBN-based model to other safety management problems.

Key words: waterway transport, sea lanes, risk early warning, dynamic Bayesian network, emergencies, sensitivity analysis