舰船航行数据具有高维度、多模态与强动态等特征，使得异常数据与正常航行数据的区分难度增加，因此，提出聚类分析的舰船航行异常数据识别方法。依据舰船航行异常数据分布特性，引入全局-局部融合机制，采用k近邻聚类分析方法和密度峰值聚类方法分别计算航行数据全局离散异常值和局部离散异常值，结合2个异常值的计算结果，获取该数据异常得分，以此判定异常航行数据；在此基础上，采用最小熵K-均值算法对检测出的异常数据进行自适应分类，完成异常数据的分类识别。测试结果显示：该方法通过全局与局部异常得分综合计算，准确识别不同航行异常数据，并划分不同的异常数据类别；不同类别异常数据划分的标准化互信息均高于0.938，保证多类别异常数据的划分效果。
Ship navigation data has the characteristics of high dimension, multi-mode and strong dynamics, which makes it more difficult to distinguish abnormal data from normal navigation data. Therefore, a clustering analysis method for identifying abnormal data of ship navigation is proposed. According to the distribution characteristics of abnormal data of ship navigation, the global-local fusion mechanism is introduced, and the global discrete abnormal value and local discrete abnormal value of navigation data are calculated by K nearest neighbor clustering analysis method and density peak clustering method respectively, and the abnormal score of the data is obtained by combining the calculation results of the two abnormal values, so as to judge abnormal navigation data; On this basis, the minimum entropy K- means algorithm is used to adaptively classify the detected abnormal data to complete the classification and identification of abnormal data. The test results show that this method can accurately identify different navigation abnormal data and classify different abnormal data categories through comprehensive calculation of global and local abnormal scores. The standardized mutual information of different types of abnormal data division is higher than 0.938, which ensures the division effect of multi-category abnormal data.
2025,47(21): 173-177 收稿日期：2025-5-9
DOI：10.3404/j.issn.1672-7649.2025.21.028
分类号：L166；TP391
作者简介：宋伟(1982-),女,硕士,副教授,主要研究方向为基础数学及数学教育
参考文献：
[1] 石悦, 罗贺, 蒋儒浩, 等. 融合数据质量增强和时空信息编码网络的船舶海上轨迹预测方法[J]. 模式识别与人工智能, 2025, 38(1): 51-67
SHI Y, LUO H, JIANG R H, et al. Ship maritime trajectory prediction method integrating data quality enhancement and spatio-temporal information encoding network[J]. Pattern Recognition and Artificial Intelligence, 2025, 38(1): 51-67
[2] 陈蜀喆, 王子威, 龚彪. 基于滑动窗口算法的船舶避碰转向点数据挖掘模型[J]. 中国航海, 2025, 48(1): 124-131.
CHEN S Z, WANG Z W, GONG B, et al. Data mining model of ship collision avoidance turning points based on sliding window algorithm[J]. Navigation of China, 2025, 48(1): 124-131.
[3] 祁文娟, 刘志恒, 周绥平, 等. 基于改进TRACLUS算法的船舶轨迹聚类研究[J]. 系统工程与电子技术, 2025, 47(4): 1214-1221.
QI W J, LIU Z H, ZHOU S P, et al. Research on ship trajectory clustering based on improved TRACLUS algorithm[J]. Systems Engineering and Electronics, 2025, 47(4): 1214-1221.
[4] 李费旭, 周利, 丁仕风, 等. 基于改进LSTM的船体监测数据异常处理方法[J]. 船舶工程, 2024, 46(7): 90-102,121
LI F X, ZHOU L, DING S F. Exception handling method for hull monitoring data based on improved LSTM[J]. Ship Engineering, 2024, 46(7): 90-102,121
[5] 黄滔, 陈冬梅, 杨勇兵. 船舶柴油机运行参数异常检测及分析[J]. 船海工程, 2024, 53(4): 66-70
HUANG T, CHEN D M, YANG Y B. Detection and analysis of abnormal operating parameters of marine diesel engines[J]. Ship & Ocean Engineering, 2024, 53(4): 66-70
[6] 李可欣, 郭健, 王宇君, 等. 一种基于降维密度聚类的船舶异常轨迹识别方法[J]. 包装工程, 2023, 44(11): 284-292
LI K X, GUO J, WANG Y J, et al. Trajectory anomaly identification method of vessels based on dimensional-density reduction clustering[J]. Packaging Engineering, 2023, 44(11): 284-292
