关键词: 交通工程, 航道, 交通事故, 贝叶斯网络, 风险, 主动评估

Abstract: In order to assess sailing risk of real-time ship traffic in waterway and give early warning of accident, the Bayesian network was built by using Gaussian mixture model and maximum expectation algorithm. Through our study and training about Bayesian network structure and parameters and based on the data from ship detectors and the record of traffic accident occurred in Caoxiexia waterway of Yangtze River, BN distributor was established. BN classification models in relation to 8 group data of ship traffic flow were set up respectively. Results show that the model using data taken from the two ship detectors closest to traffic scene within 20 to 40 minutes just before occurrence of accident generates optimum prediction results with accuracy up to 78.13%.The result of comparison among BN, BP neural network and K close neighbor proves that BN prediction model generates a more accurate prediction result and is thus a good method of risk evaluation of sailing ships in real-time. 

Key words: traffic engineering, waterway, traffic accident, Bayesian network, risk, proactive evaluation