船舶航跡模式分析與異常偵測

Abstract

目前行政院海岸巡防署執行海上監控任務，是藉由監控人員的經驗來分析與判斷船舶之移動行為的異常與否。面對不斷產生與增加的大量船舶移動軌跡資料，使得監控人員執行海上交通監偵的任務更加繁重。另一方面，對於相對少數的異常移動行為，愈難以透過人力來偵測與判斷。然現行船舶航行態樣是否異常端靠監控人員全程監控分析研判；若人員監控目標有疏漏狀況或經驗不足等情形，易導致異常目標漏失，致使查緝勤務單位決策人員錯失防範或查緝佈署良機。 因此，本研究主要以海上船舶航行監控航跡資料(巨量空間移動GIS資料)為研究對象，運用空間資訊資料探勘(Geo-spatial Data Mining)與空間分群技術，進行空間資料分群模型建立，分析產出船隻航行航跡，進而以航跡資料特徵點提取與進行群聚、利用航跡特徵空間模式與航跡簡約化為基礎，進行正常與異常航跡的判斷。期能運用本研究發展的方法，建立異常軌跡模式偵測(Pattern Anomaly Detection)相關方法與技術。 本研究採用海洋大學基隆港附近AIS觀測資料進行研究，以Python程式語言進行AIS資料前處理，將2017年10-11月份AIS資料匯入PostgreSQL資料庫中，再從AIS船位資料表格，產出船隻航行航跡。並以QGIS為工具，配合PostGIS功能，實作提取航跡資料特徵點、進行航跡資料特徵點群聚分析、建立航跡特徵空間模式與航跡簡約化等功能，建立基隆港附近基於AIS船舶航行資料的船隻航跡模型雛型架構。 本研究以2017年12月份AIS資料進行模型驗證，經初步驗證結果證實，大部分的船隻航跡態樣均符合本研究建立船隻航跡模型，故可做為未來運用此模型，據以建立異常軌跡模式偵測方法。

Currently, the Coast Guard Administration of the Executive Yuan carries out maritime surveillance tasks by analyzing and judging the abnormality of the ship's movement behavior through the experience of the monitoring personnel. In the face of the continuous generation and increase of many ship movement trajectory data, the task of monitoring personnel performing marine traffic monitoring is even more onerous. On the other hand, for relatively few abnormal movements, the more difficult it is to detect and judge through human resources. However, whether the current navigation status of the ship is abnormal depends on the monitoring and analysis of the monitoring personnel throughout the entire process. If the personnel monitoring target has omissions or inexperienced situations, it may easily lead to the loss of an abnormal target, causing the decision-makers of the service unit to miss the precautions or check the deployment opportunities. Therefore, this study mainly focuses on marine ship navigation and monitoring data (large-scale mobile GIS data) and uses spatial data mining and spatial grouping techniques to establish a spatial data grouping model and analyze production. The ship's trajectory was tracked, and the normal and abnormal tracks were judged based on the feature extraction and clustering of track data and the use of track feature space model and track simplification. During the period, we can use the algorithms developed in this research to establish methods and techniques related to Pattern Anomaly Detection. This study uses the AIS observation data around the Keelung Harbor of the National Taiwan Ocean University to conduct AIS data preprocessing in the Python programming language, importing AIS data from October to November 2017 into the PostgreSQL database, and outputting ships from the AIS position data. Using QGIS as a tool, with the help of Post-GIS function, it implements the function of extracting the feature points of the trajectory, analyzing the characteristic points of the track data, establishing the track feature space model and trajectory generalization, and establishing a vessel tracing model prototype architecture based on the AIS ship navigation data near the Keelung Harbor. In this study, AIS data in December 2017 were used to verify the model. After preliminary verification, it was confirmed that most of the ship's track conditions are in line with the study to establish a ship track model. Therefore, this model can be used in the future to establish an abnormal trajectory model. Detection method.