数据驱动的船舶异常行为识别方法
Data-Driven Identification of Abnormal Behavior of Ships
-
摘要: 为对不同环境影响因素下的船舶异常行为进行有效识别,提出一种综合考虑船舶位置和船舶航行状态的多角度船舶异常行为识别方法。将船舶航行状态分为停留、直行和转向等3类,对网格化水域内的船舶航行状态进行统计,获得船舶正常航行状态的区域分布情况;利用核密度估计算法对船舶位置特征进行提取,获得正常航行位置的区域分布;利用正常船舶航行状态和船舶位置分布情况对船舶异常行为进行识别。选取曹妃甸水域的船舶轨迹数据,用以验证异常行为识别模型的检验效果。试验结果表明:船舶位置异常识别取决于阈值的设定,宽松的阈值识别的异常位置包含船舶较少的航线轨迹,严格的阈值识别的异常位置反映船舶的危险行为;在船舶航行状态异常识别中,该方法可以对航向大幅度波动和航速剧烈变化的船舶异常航行行为状态进行有效的识别。Abstract: A comprehensive method of identifying abnormal behavior of ships concerning both ship's navigation status and position is introduced. The navigation status is sorted into 3 categories, go straight, turning and staying, and the right distribution of ship's navigation status over the grid cells of the water area. The position feature of ships is extracted with kernel density estimation method and the normal position distribution of ships over the grid is found. These two distributions form the reference to find abnormal behavior of ships. The method is verified in the water area of Caofeidian. The method is proved effective in finding ship's abrupt change of course or/and peed, given the threshold has been appropriately set.