论文总字数：29171字

摘 要

随着中国船舶数量快速上涨，航运业也因此进入了发展的快车道，这就使得水上交通安全问题日益明显，尤其是在船舶流量大、航行条件复杂的港口水域，航行安全问题更加明显。而由于基于大数据对船舶航行安全风险进行分析判断的研究目前还比较匮乏，很多避碰工作仍需要依靠值班驾驶员与VTS监管人员的个人经验和直觉判断，传统助航系统(雷达,ARPA等)对交通流密集水域中的航行态势，特别是对船舶会遇过程中的避让行为实时辨识能力依旧十分欠缺。而船舶自动识别系统(AIS)通过广播方式定期发送船位、速度和航向等船舶动态信息，具有数据信息种类齐全、数据集丰富等特点。因此如果能运用数据挖掘、机器学习等信息技术对船舶航行数据深入分析和处理，提取交汇水域中船舶的行为变化规律，并对其量化建模,分析典型的会遇行为，则可为船舶航行风险判断提供技术支持。

本文的主要内容涵盖：

1）会遇场景与行为特征提取

选取交汇水域的船舶轨迹信息作为研究对象，针对不同船舶会遇场景和过程进行特征分析，同时利用时空约束关系构建会遇行为参量提取方法；获取包含会遇船舶位置、速度及航向变化、CPA 在内的参量信息，在此基础上，构建参量的量化表征方法，并在轨迹上进行映射，形成船舶会遇行为的多元特征序列。

2）会遇风险辨识模型与预警方法

在提取特征序列的基础上，融合多种会遇风险指标，对会遇船舶之间的碰撞风险进行全阶段量化，基于深度循环神经网络 LSTM 形成会遇行为特征与风险等级之间的映射模型，对未来会遇走势及风险进行预判，并利用行为特征数据对模型进行参数优化估计。

3）方法验证与优化

对特定的港口水域进行方法验证。评估相关分析方法有效性。

关键词：船载自动识别系统（AIS）；船舶会遇态势；会遇特征；异常检测

Abstract

With the rapid increase in the number of ships in China, the shipping industry has entered the fast lane of development, which makes the problem of water traffic safety increasingly obvious, especially in the port waters with large ship flow and complex navigation conditions. However, due to the lack of research on the analysis and judgment of ship navigation safety risks based on big data, a lot of collision avoidance work still needs to rely on the personal experience and intuitive judgment of the pilot on duty and VTS supervisors. The traditional navigation aids (radar, ARPA, etc.) still lack the ability to identify the navigation situation in the areas with dense traffic flow, especially the real-time identification of collision avoidance behavior in the process of ship encounter. On the other hand, the automatic ship identification system (AIS) sends ship dynamic information such as position, speed and course regularly by broadcast, which has the characteristics of complete kinds of data information and rich data sets. Therefore, if the information technology such as data mining and machine learning can be used to deeply analyze and process the ship navigation data, extract the behavior change law of the ship in the confluence water area, model it quantitatively, and analyze the typical encounter behavior, it can provide technical support for ship navigation risk judgment.

The main contents of this paper include:

1. Encounter scene and behavior feature extraction

Select the specific water area as the research object, carry on the spatio-temporal analysis of different ship encounter scenes and processes, use the spatio-temporal constraint relationship to construct the encounter trajectory extraction method, obtain the behavior characteristic parameters such as ship position, speed and course change, CPA and so on, construct the quantitative representation method of the parameters, and map them on the trajectory to form a multiple feature sequence of ship encounter behavior

1. Identification model and early warning method of encounter risk

On the basis of extracting the feature sequence, combining a variety of encounter risk indicators, the collision risk between encounter ships is quantified in all stages. Based on the deep cyclic neural network LSTM, a mapping model between encounter behavior characteristics and risk grade is formed to predict the future encounter trend and risk, and the behavioral characteristic data are used to optimize the parameters of the model..

1. method verification and optimization

Carry on the method verification to the specific port water area. Evaluate the effectiveness of related analysis methods.

Key Words：Automatic Identification System, (AIS); ship encounter situation; meeting characteristics；Anomaly Detection；

请支付后下载全文，论文总字数：29171字