A Detection Method for Maritime Traffic Accidents Based on AIS Communication Volume
-
摘要: 基于数据驱动的交通事故检测对水上交通事故的快速救援与降低事故损失具有重要作用。为实现无自主报告情形下的交通事故自动检测，研究了基于船舶自动识别系统(automatic identification system，AIS) 通信量的水上交通事故检测方法。通常情况下，突发的水上交通事故会扰乱船舶正常航行秩序，事故引起的船舶运动状态改变会导致AIS通信量短时间内产生突变，为挖掘AIS通信量随水上交通事故演变的内在规律，并降低噪声影响，以凸显检测指标的突变特征，引入AIS通信量和船舶总数比值构建水上交通事故检测指标；为确保检测的时效性，采用滑动窗口模型用于划分检测指标数据片段与设定更新时间的间隔，并构建基于卡尔曼滤波的水上交通事故检测模型进行检测指标的短期预测；为保证检测结果的准确性，采用云模型进行检测模型阈值范围的快速划分。使用长江武汉段水域内AIS数据对基于AIS通信量的水上交通事故检测方法进行模型验证和仿真研究，实验结果表明：与采用标准正态偏差和多尺度直线拟合算法的检测模型相比，提出的基于卡尔曼滤波算法的检测模型可在耗时最短的情形下获得最高命中率与最低误检率，分别为97.25%与0.42%；在进一步的仿真实验中，针对3种不同的船舶事故场景，提出的基于AIS通信量的水上交通事故检测方法均能在5 min内检测出水上交通事故的发生。
-
关键词:
- 交通安全 /
- 水上交通事故 /
- 水上交通事故检测方法 /
- 卡尔曼滤波 /
- AIS通信量
Abstract: The data-driven approach for traffic accident detection plays a crucial role in the rapid rescue and reduction of losses in maritime accidents. To achieve automatic detection of maritime accidents without autonomous reporting, a method based on Automatic Identification System (AIS) communication volume is proposed. Normally, sudden maritime accidents disrupt the normal navigation patterns of vessels, leading to sharp changes in AIS communication volume within a short time due to changes in vessel movement states during the accidents. To extract the inherent laws of AIS communication volume during the evolution of maritime accidents and reduce noise interference to highlight the abrupt features of detection indicators, the AIS communication volume-to-total vessel count ratio is introduced as an indicator for maritime accident detection. To ensure timely detection, a sliding window model is used to segment detection indicator data and set update time intervals. Furthermore, a maritime accident detection model based on Kalman filtering is developed for short-term prediction of detection indicators. To ensure the accuracy of detection results, a cloud model is employed for rapid division of detection model threshold ranges. Validations and simulations are conducted using AIS data from the Yangtze River Wuhan section to verify the AIS communication volume-based maritime accident detection method. Results show that the proposed detection model based on Kalman filtering achieves the highest hit rate and lowest false alarm rate in the shortest time, namely 97.25% and 0.42%, respectively, compared to models employing standard normal deviation and multi-scale linear fitting algorithms. In further simulated experiments involving three different accident scenarios, the proposed AIS communication volume-based maritime accident detection method successfully detects accidents within 5 minutes. -
图 1 船舶群组内船舶数占比变化对AIS通信量影响
Figure 1. The influence of changes in the proportion of ships in a ship group on AIS communication volume
图 2 AIS通信量及船舶总数变化趋势图
Figure 2. Trend chart of AIS communication volume and total number of ships
图 3 AIS通信量-船舶总数散点分布图
Figure 3. AIS communication volume - total number of ships scatter distribution map
图 4 AIS通信量和水上交通事故检测指标的变化曲线
Figure 4. Variation curves of AIS communication volume and maritime traffic accident detection indicators
图 5 AIS通信量Ne(t) 分布统计
Figure 5. Distribution statistics of AIS communication volume Ne(t)
图 6 水上交通事故检测指标NeM(t) 分布统计
Figure 6. Distribution statistics of NeM(t) detection index for maritime traffic accidents
图 7 基于云模型的检测阈值确定流程
Figure 7. Detection threshold determination process based on cloud model
图 8 基于AIS通信量的水上交通事故检测方法流程
Figure 8. Maritime traffic accident detection method process based on AIS communication volume
图 10 3种算法检测结果示意图
Figure 10. Schematic diagram of the detection results of the three algorithms
图 11 水上交通事故检测仿真流程图
Figure 11. Flow chart of maritime traffic accident detection simulation
图 13 船载A类和B类AIS设备的数量关系
Figure 13. The relationship between the number of class A and class B AIS devices on ships
图 17 场景1输出的水上交通事故检测指标及预测残差
Figure 17. The detection indicators and prediction residuals of maritime traffic accidents output in Scenario 1
图 18 场景2输出的水上交通事故检测指标及预测残差
Figure 18. The detection indicators and prediction residuals of maritime traffic accidents output in Scenario 2
图 19 场景3输出的水上交通事故检测指标及预测残差
Figure 19. The detection indicators and prediction residuals of maritime traffic accidents output in Scenario 3
表 1 AIS动态信息更新间隔表
Table 1. AIS dynamic information update interval table
AIS终端类型 船舶状态 更新间隔 A类AIS 抛锚、停泊且速度≤ 3 n mile/h 3 min 抛锚、停泊且速度 > 3 n mile/h 10 s 航行速度0~14 n mile/h 10 s 航行速度0~14 n mile/h并改变航向 3.3 s 航行速度14~23 n mile/h 6 s 航行速度14~23 n mile/h并改 2 s 变航向 2 s 航行速度 > 23 n mile/h 2 s 航行速度 > 23 n mile/h并改变航向 3 min B类AIS(SOTDMA) 航行速度≤ 2 n mile/h 3 min 辅助导航 30 s 航行速度2~14 n mile/h 15 s 航行速度14~23 n mile/h 5 s 航行速度 > 23 n mile/h 3 min B类AIS(CSTDMA) 航行速度≤ 2 n mile/h 30 s 航行速度 > 2 n mile/h 
下载: 导出CSV
表 2 突变数据流片段设定
Table 2. Mutation data flow fragment setting
突变数据流片段 开始时间 结束时间 片段1 09∶20 10∶10 片段2 12∶40 13∶30 片段3 16∶00 16∶50
下载: 导出CSV
表 3 3种算法的阈值范围
Table 3. Threshold ranges for the three algorithms
参数名称 SND检测算法 多尺度直线拟合算法 Kalman滤波算法 期望 -1.7×10-2 -2.6×10-5 -4.2×10-4 熵 1.274 4 0.016 5 0.146 3 超熵 0.060 1 4.7×10-4 0.005 3 阈值上限 3.806 2 0.047 8 0.438 6 阈值下限 -3.840 3 -0.047 9 -0.439 4
下载: 导出CSV
表 4 3种检测算法模拟测试的评价指标统计表
Table 4. Statistical table of evaluation indicators for simulation tests of three detection algorithms
算法 命中率/% 误检率/% 时耗/s SND算法 92.00 1.25 0.027 多尺度直线拟合算法 66.67 1.67 2.199 卡尔曼滤波算法 97.25 0.42 0.005
下载: 导出CSV
表 5 事故时间及位置设置
Table 5. Accident time and location setting
场景设置 事故发生时刻 事故消失时刻 事故位置 场景1 08：00 08：30 114.3384°E，30.6233°N 场景2 11：27 11：55 114.3671°E，30.6436°N 场景3 16：22 16：55 114.4177°E，30.6632°N
下载: 导出CSV
表 6 船舶分配及航速设置
Table 6. Ship allocation and speed setting
船舶类型 船舶总数/艘 具有救援能力船舶/艘 初始航速/(n mile/h) A类AIS设备船舶 9 2 7 B类AIS设备船舶 21 3 4
下载: 导出CSV
表 7 仿真场景1中预设时间和检测结果时间对比
Table 7. Comparison of preset time and detection result time in simulation scenario 1
时间设定 预设时间 报警时间 差值/min 产生时间 08：00 08：03 +3 恢复时间 08：30 08：32 +2
下载: 导出CSV
表 8 仿真场景2中预设时间和检测结果时间对比
Table 8. Comparison of preset time and detection result time in simulation scenario 2
时间设定 预设时间 报警时间 差值/min 产生时间 11：27 11：31 +4 恢复时间 11：55 11：58 +3
下载: 导出CSV
表 9 仿真场景3中预设时间和检测结果时间对比
Table 9. Comparison of preset time and detection result time in simulation scenario 3
时间设定 预设时刻 报警时刻 差值/min 产生时刻 16：22 16：25 +3 恢复时刻 16：55 16：56 +1
下载: 导出CSV
-
[1] 李文杰, 于凇凌, 杜洪波, 等. 内河航运需求与腹地经济产业结构的相关性分析[J]. 水运工程, 2022(4): 88-93.LI W J, YU S L, DU H B, et al. Correlation analysis of inland waterway shipping demand and industrial structure of hinterland economy[J]. Port and Waterway Engineering, 2022(4): 88-93. (in Chinese) [2] 刘儿七. 国内外内河航运发展现状和趋势[J]. 港口科技, 2019, (5): 45-48.LIU E Q. Current situation and trends of inland navigation development at home and abroad[J]. Science & Technology of Ports, 2019, (5): 45-48. (in Chinese) [3] CHEN C, WU Q, GAO S. Mining of inland water traffic accident data using a biclustering algorithm: A case study of the Yangtze River[J]. Journal of Risk and Reliability, 2019, 233 (1): 48-57. [4] 郑中义, 吴兆麟, 杨丹. 港口船舶事故致因的灰色关联分析模型[J]. 大连海事大学学报, 1997, (2): 62-65.ZENG Z Y, WU Z L, YANG D. Analysis model of accident's main causes on port vessels incidence by grey system theory[J]. Journal of Dalian Maritime University 1997, (2): 62-65. (in Chinese) [5] 王海燕, 刘清. 水上船舶交通事故人为因素致因机理[J]. 中国航海, 2016, 39(3): 41-44.WANG H Y, LIU Q. Accident-causing mechanism of human errors in marine navigation[J]. Navigation of China, 2016, 39 (3): 41-44. (in Chinese) [6] FAN L, WANG M, YIN J. The impacts of risk level based on PSC inspection deficiencies on ship accident consequences[J]. Research in Transportation Business and Management, 2020, 33: 100464. [7] RAHMAN S. Introduction of Bayesian network in risk analysis of maritime accidents in Bangladesh[C]. The 1st International Conference on Mechanical Engineering and Applied Science, Dhaka, Bangladesh: AIP Publishing LLC, 2017. [8] BEATRIZ N, RAFET E. Marine accident learning with fuzzy cognitive maps: a method to model and weight human-related contributing factors into maritime accidents[J]. Ship and Offshore Structure, 2020, 17(3): 555-563. [9] 李奕良. 基于贝叶斯网络的干散货船舶自沉事故致因分析[D]. 大连: 大连海事大学, 2020.LI Y L. Cause analysis of ship foundering accident of dry bulk carrier based on Bayesian network[D]. Dalian: Dalian Maritime University, 2020. (in Chinese) [10] DEBNATH A K, CHIN H C. Navigational traffic conflict technique: a proactive approach to quantitative measurement of collision risks in port waters[J]. The Journal of Navigation, 2010, 63(1): 137-152. doi: 10.1017/S0373463309990233 [11] OROVI B, DJUROVI P. Research of marine accidents through the prism of human factors[J]. Promet - Traffic - Traffico, 2013, 25(4): 369-377. doi: 10.7307/ptt.v25i4.1210 [12] REKHA A G, PONNAMBALAM L, ABDULLA M S. Predicting maritime groundings using support vector data description model[C]. International Symposium on Intelligence Computation and Applications. Singapore: Springer, 2015. [13] 徐东星, 尹勇, 张秀凤, 等. 基于改进三参数灰色模型的海上交通事故预测[J]. 中国航海, 2020, 43(1): 12-17.XU D X, YIN Y, ZHANG X F, et al. Improved three-parameter grey model for prediction of marine traffic accidents[J]. Navigation of China, 2020, 43(1): 12-17. (in Chinese) [14] 范中洲, 赵羿, 周宁, 等. 基于灰色BP神经网络组合模型的水上交通事故数预测[J]. 安全与环境学报, 2020, 20(3): 857-861.FAN Z Z, ZHAO Y, ZHOU N, et al. Integrated model for forecasting waterway traffic accident based on Gray-BP neural network[J]. Journal of Safety and Environment, 2020, 20 (3): 857-861. (in Chinese) [15] 张逸飞, 付玉慧. 基于ARIMA-BP神经网络的船舶交通事故预测[J]. 上海海事大学学报, 2020, 41(3): 47-52.ZHANG Y F, FU Y H. Prediction of ship traffic accidents based on ARIMA-BP neural network[J] Journal of Shanghai Maritime University, 2020, 41(3): 47-52. (in Chinese) [16] 黄琛, 陈德山, 吴兵, 等. 船舶航行交通事件实时检测技术研究现状与展[J]. 交通信息与安全, 2022, 40(6): 1-11. doi: 10.3963/j.jssn.1674-4861.2022.06.001HUANG C, CHEN D S, WU B, et al. A real-time detection of nautical traffic events: A review and prospect[J]. Journal of Transport Information and Safety, 2022, 40(6): 1-11. (in Chinese) doi: 10.3963/j.jssn.1674-4861.2022.06.001 [17] 刘人杰, 刘畅, 黄习刚. 船舶自动识别系统的信息传输[J]. 中国航海, 2002(3): 43-46.LIU R J, LIU C, HUANG X G. The Information transmission of marine AIS[J]. Navigation of China, 2002(3): 43-46. (in Chinese) [18] 刘彤, 吴建华, 雷金平. AIS通信系统性能分析[J]. 交通科技, 2004(4): 134-136.LIU T, WU J H, LEI J P. Analysis of the AIS communication system performance[J]. Transportation Science & Technology, 2004(4): 134-136. [19] TRUONG C, OUDRE L, VAYATIS N. Selective review of offline change point detection methods[J]. Signal Processing, 2019, 167: 107299. [20] 庞景月. 滑动窗口模型下的数据流自适应异常检测方法研究[D]. 哈尔滨: 哈尔滨工业大学, 2013.PANG J Y. Adaptive anomoly detection for data stream of sequence-based slinding windows model[D]. Harbin: Harbin Institute of Technology. 2013. (in Chinese) [21] 程小洋. 交通事件检测算法的阈值自适应调整与优化[D]. 南京: 东南大学, 2021.CHEN X Y. A thesis submitted to southeast university for the academic degree of master of engineering[D]. Nanjing: Southeast University, 2021. (in Chinese) [22] 王国胤, 李德毅, 姚一豫. 云模型与粒计算[M]. 北京: 科学出版社, 2012.WANG G Y, LI D Y, YAO Y Y. Cloud model and granular computing[M]. Beijing: Science Press, 2012. (in Chinese). [23] CHANG L X, GUO Y W, QING H Z. A new multi-step backward cloud transformation algorithm based on normal cloud model[J]. Fundamenta Informaticae, 2014, 133(1): 55-85. -
点击查看大图
计量
- 文章访问数: 446
- HTML全文浏览量: 238
- PDF下载量: 11
- 被引次数: 0