A Method for Detecting Abnormal Behavior of Ships Based on Multi-Dimensional Density Distance and an Abnormal Isolation Mechanism
Lixiang Zhang,Yian Zhu,Jie Ren,Wei Lu,Ye Yao
DOI: https://doi.org/10.3934/mbe.2023620
2023-01-01
Mathematical Biosciences and Engineering
Abstract:Abnormal ship behavior detection is essential for maritime navigation safety. Most existing abnormal ship behavior detection methods only build A ship trajectory position outlier detection model; however, the construction of a ship speed outlier detection model is also significant for maritime navigation safety. In addition, in most existing methods for detecting a ship's abnormal behavior based on abnormal thresholds, one unsuitable threshold leads to the risk of the ship not being minimized as much as possible. In this paper, we proposed an abnormal ship behavior detection method based on distance measurement and an isolation mechanism. First, to address the problem of traditional trajectory compression methods and density clustering methods only using ship position information, the minimum description length principle based on acceleration (AMDL) algorithm and Multi-Dimensional Density Clustering (MDDBSCAN) algorithm is used in this study. These algorithms not only considered the position information of the ship, but also the speed information. Second, regarding the issue of the difficulty in determining the anomaly threshold, one method for determining the anomaly threshold based on the relationship between the velocity weights and noise points of the MDDBSCAN algorithm has been introduced. Finally, due to the randomness issue of the selected segmentation value in iForest, a strategy of selectively constructing isolated trees was proposed, thus further improving the efficiency of abnormal ship behavior detection. The experimental results on the historical automatic identification system data set of Xiamen port prove the practicality and effectiveness of our proposed method. Our experiment results show that the proposed method achieves an improvement of about 10% over the trajectory outlier detection based on the local outlier fraction method, about 14% over the isolation-based online anomalous trajectory method in terms of the accuracy of ship position information anomaly detection, and about 3% over the feature fusion method in terms of the accuracy of ship speed anomaly detection. This method improves algorithm efficiency by about 5% compared to the traditional isolation forest anomaly detection algorithm.