贺琪,曹万万,黄冬梅,郝增周,杜艳玲,耿立佳.面向海洋时序数据异常模式发现的多视图协同可视分析[J].海洋通报,2022,(6):
面向海洋时序数据异常模式发现的多视图协同可视分析
Multi-view collaborative visual analysis for anomaly detection of marine environment time series data
投稿时间:2021-11-19  修订日期:2022-01-26
DOI:10.11840/j.issn.1001-6392.2022.06.001
中文关键词:  海洋多要素数据,多维度标度算法,密度聚类,平行坐标,异常模式
英文关键词:marine multi-factor data  multi-dimensional scaling algorithm  density clustering  parallel coordinates  anomaly pattern
基金项目:海洋环境安全保障项目(202115)
作者单位E-mail
贺琪 上海海洋大学信息学院,上海市 201306 qihe@shou.edu.cn 
曹万万 上海海洋大学信息学院,上海市 201306  
黄冬梅 上海电力大学 上海市 200090  
郝增周 自然资源部第二海洋研究所 杭州 310012  
杜艳玲 上海海洋大学信息学院,上海市 201307 yldu@shou.edu.cn 
耿立佳 国家海洋局 东海标准计量中心 上海 201306  
摘要点击次数: 34
全文下载次数: 0
中文摘要:
      如何发现多要素海洋环境时序数据中蕴含的由自然现象导致的异常模式,进而实现对海洋事件发生的有效预测,是一个亟待解决的问题。本文提出了一种面向海洋环境时序数据异常模式挖掘的多视图协同可视分析方法,首先计算出多要素数据间的相似性矩阵,通过多维标度法(Multi-Dimensional Scaling, MDS)投影降维,将投影结果通过密度聚类(Density-Based Spatial Clustering of Applications with Noise, DBSCAN)生成时序MDS聚类视图,表达多要素数据的时序特征,用于发现多要素叠加后的异常模式;其次,再基于相似性矩阵计算每个要素熵值,生成与时序MDS聚类视图对应的多要素信息熵视图,表达每个要素在时序上的不确定性,用于确定不同要素对异常模式的贡献度;最后,针对异常模式,提供由对应原始数据投影生成的焦点平行坐标视图,进一步的分析要素之间的相关性强弱和数据内部具体的变化趋势。将本文方法应用于东山台站(23.9N、117.5E)、遮浪台站(22.6N、115.5E)附近海洋数据,发现数据由台风造成的异常模式和要素之间的相关性,证明本文提出的多视图可视分析方法的有效性,方法具备发现多要素时序数据蕴含的异常模式的能力。
英文摘要:
      How to find the abnormal patterns caused by natural phenomena contained in multi-element marine environment time series data, and then realize the effective prediction of marine events, is an urgent problem to be solved. In this paper, a multi view collaborative visual analysis method for mining abnormal patterns of marine environment time series data is proposed. Firstly, the similarity matrix between multi-element data is calculated, the dimension is reduced by multi-dimensional scaling (MDS) projection, and the projection results are processed by density based spatial clustering of applications with noise (DBSCAN) Generate a temporal MDS clustering view to express the temporal characteristics of multi-element data, which is used to find the abnormal pattern after multi-element superposition; Secondly, the entropy value of each element is calculated based on the similarity matrix, and the multi-element information entropy view corresponding to the time-series MDS clustering view is generated to express the uncertainty of each element in time-series, which is used to determine the contribution of different elements to the abnormal pattern; Finally, for the abnormal pattern, the focus parallel coordinate view generated by the projection of the corresponding original data is provided to further analyze the correlation between the elements and the specific change trend within the data. This method is applied to the marine data near Dongshan Station (23.9N, 117.5E) and ZHELANG station (22.6N, 115.5E). The correlation between the abnormal patterns and elements caused by typhoon is found, which proves the effectiveness of the multi view visual analysis method proposed in this paper. The method has the ability to find the abnormal patterns contained in multi-element time series data.
查看全文    下载PDF阅读器
关闭