基于轨迹意图概率建模的船舶行为模式识别研究

Ship Trajectory and Navigational Environment Information contain rich knowledge of ship movements. Identifying and understanding ship behavior pattern from massive trajectory data becomes important contents of intelligent maritime development. Traditional methods of ship behavior pattern recognition distinguish different behaviors based on trajectory structure similarity. The limitations of these methods are the lack of representation and correlation analysis of ship behavior semantics, that make it difficult to meet the demand of maritime situation awareness in large complex waters. This study intends to develop new method of ship behavior pattern recognition on semantic level. Firstly, key-point based semantic association of ship trajectory and navigational environment information is proposed to produce semantic feature representation of ship trajectory. Moreover, probabilistic generative model is used for mining semantic features of ship trajectory and identifying behavior intention hidden in different trajectories. Ship trajectories then can be classified based on similarity measurement of ship behavior intention. After that, textual association rule analysis method is applied to mine behavior sequential pattern of ships with particular intention and mixed intention. This study is an interdisciplinary research of geographic information science and maritime traffic engineering. It plans to provide new idea and method for data driven pattern recognition of ship behavior and supports for intelligent maritime applications.

船舶轨迹及其通航环境信息蕴含丰富的船舶活动知识，从大规模轨迹数据中识别和理解船舶行为模式是智慧海事建设的重要内容。传统船舶行为模式识别强调轨迹结构相似性区分船舶行为，缺少船舶行为语义的表达及潜在相关性分析，无法满足大型复杂水域的海事态势感知需求。本项目拟发展语义层次的船舶行为模式识别方法，利用基于关键点的船舶轨迹-通航环境语义关联，构建船舶轨迹的行为语义特征表达，进一步采用概率生成模型对船舶轨迹行为语义特征进行挖掘，辨识不同轨迹隐含的船舶意图并进行一致性测度，最后结合文本关联规则分析挖掘特定意图和混合意图下的船舶行为序列模式。本项目是地理信息科学和海上交通运输工程的交叉研究，可为数据驱动的船舶行为模式识别提供新的思路和方法，支撑海事安全保障、航线定制优化等智慧海事应用。

船舶轨迹及其通航环境信息蕴含丰富的船舶活动知识，从大规模轨迹数据中识别和理解船舶行为模式是智慧海事建设的重要内容。在港口水域，项目从行为认知的角度，构建了船舶原子行为语义模型，结合港口水上交通规则提出了港口水域通用的船舶行为概念模型，并基于本体建模理论建立了船舶行为本体模型，结合《国际海上避碰规则》和语义网规则语言实现了船舶行为的语义推理；在内河水域，构建了多层次轨迹时空特征框架，提出了轨迹特征驱动的船舶停留行为提取方法与分类方法，建立了基于空间拓扑分析的船舶轨迹-通航环境语义关联模型，结合空间网格模型、八方向锥形模型、船舶操纵模型等构建了船舶轨迹的语义表征；采用隐含狄利克雷分布主题模型对船舶语义轨迹进行主题挖掘，识别具有相同活动意图的船舶轨迹集合；构建了基于最小公共子序列的轨迹相似性模型，实现了船舶伴随行为模式的语义挖掘。通过本项目研究，揭示了数值化船舶轨迹隐含的船舶行为语义及特征，提高了水上交通船舶活动的认知能力，研究结果可为船舶异常行为识别、船舶智能航行研究提供借鉴。