【作者】 李爽；

【导师】 黄耀倞；

【作者基本信息】 大连海事大学 ， 交通运输工程（专业学位）， 2020， 硕士

【摘要】 AIS大数据中蕴含着大量的隐藏在数据当中的无法直接依靠单一设备获取到的重要信息。现阶段航海领域对AIS数据挖掘工作尚处在一个浅层次挖掘的阶段,对于AIS数据的挖掘不够充分,忽视了船舶交通流之间的关联因素,对复杂数据信息的利用率较低。目前对于船舶AIS数据的挖掘重点主要集中在船舶位置信息在不同港口空间位置的宏观表现,对于轨迹时间等因素还处于辅助决策信息的环节,对船舶AIS数据没有做到充分挖掘及合理的应用,对这一类的研究还有待探索与创新。本文综合考虑了 AIS数据表征船舶运动状态的特性,采用基于改进谱聚类算法的数据挖掘方法进行船舶轨迹聚类以及航迹的预测,具体内容主要包括以下几点:(1)在船舶轨迹预处理阶段对AIS数据进行压缩处理,采用改进Sliding Window算法的数据压缩方式,结合AIS数据特点,加入船舶航向作为压缩变量参数之一,改变以往单纯使用经纬度变量造成的不符合航海实际、压缩不合理等情况,提高数据挖掘的精度和速度,降低AIS轨迹数据的内存。(2)结合谱聚类算法特性得出较为合理的算法复杂度及时间复杂度,改善传统聚类方法过度依赖参数导致出现陷入局部最优的情况,为船舶挖掘有效信息提供支持。(3)改进DTW距离使轨迹间的对比结果不因轨迹长度的不一致受到影响;在k-means聚类算法环节,借鉴DBSCAN聚类算法中密度聚类的优势,采用点集密度的思想,选取密度峰值点,改善单纯k-means算法容易陷入局部最优的弊端。(4)应用改进的谱聚类算法,以天津港AIS数据作为样本数据进行实验验证,得出轨迹聚类结果并能准确提取和划分某水域主要航迹段,为航路辨识等提供理论支持。(5)应用GRU神经网络,对船舶航迹进行预测。该算法在时间复杂度及计算复杂度上均有良好表现。通过神经网络的辅助功能,可反向验证船舶聚类轨迹等一系列数据挖掘结果。更多还原

【Abstract】 AIS big data contains a lot of important information hidden in the data that cannot be directly obtained by a single device.At present,the mining of AIS data in the fiel d of navigation is still in a shallow mining stage.The mining of AIS data is not suffic ient,and the related factors between ship traffic flows are ignored,and the utilization r ate of multiple data information is low.At present,the focus of the mining of ship AI S data is mainly on the macro performance of ship position information in different po rt spaces.The trajectory time and other factors are still in the stage of assisting decisio n-making information.The ship AIS data has not been fully excavated and properly ap plied.This kind of research still needs to be explored and innovated.This paper comprehensively considers the characteristics of AIS data to represent t he ship’s motion state,and adopts a data mining method based on an improved spectral clustering algorithm for ship trajectory clustering and trajectory prediction.The specifi c contents mainly include the following points:(1)Compressing the AIS data during the ship trajectory pre-processing stage,adop t the data compression method of improved Sliding Window algorithm,combine the ch aracteristics of AIS data,adding the ship course as one of the compression variable par ameters,and changing the previous situation that the simple use of latitude and longitu de variables is not in line with the actual nautical conditions and the compression is un reasonable,etc.,which improves the accuracy and speed of data mining and reduces th e memory of AIS trajectory data.(2)Combining the characteristics of spectral clustering algorithm to obtain a more reasonable algorithm complexity and time complexity,improving the traditional clusteri ng method’s excessive dependence on parameters,resulting in a situation of local optim ization,and providing support for the effective information mining of ships.(3)Improving the DTW distance so that the comparison results between the traject ories were not affected by the inconsistency of the trajectory length;in the k-means clu stering algorithm,the advantages of density clustering in the DBSCAN clustering algori thm were used,and the idea of point set density was used to select Density peak point,which improves the disadvantage that the simple k-means algorithm is easy to fall into the local optimal.(4)The improved spectral clustering algorithm is applied,and the AIS data of Tia njin Port is used as the sample data for experimental verification.The trajectory cluster ing results are obtained and the main trajectory segments of the water area can be accu rately extracted and divided,providing theoretical support for route identification.(5)Applying the GRU neural network to predict the ship’s trajectory.The algorith m has good performance in both time complexity and calculation complexity.Through the auxiliary function of neural network,a scries of data mining results such as ship cl uster trajectory can be verified in reverse.更多还原