高级检索

    宋文燏, 周海波, 吴宗培, 李海员, 袁玉波. 图记忆诱导的大气排污时序数据异常检测算法[J]. 华东理工大学学报(自然科学版), 2023, 49(3): 341-350. DOI: 10.14135/j.cnki.1006-3080.20221118001
    引用本文: 宋文燏, 周海波, 吴宗培, 李海员, 袁玉波. 图记忆诱导的大气排污时序数据异常检测算法[J]. 华东理工大学学报(自然科学版), 2023, 49(3): 341-350. DOI: 10.14135/j.cnki.1006-3080.20221118001
    SONG Wenyu, ZHOU Haibo, WU Zongpei, LI Haiyuan, YUAN Yubo. Image Memory Induced Anomaly Detection Algorithm for Atmospheric Pollutant Emission Time-Series Data[J]. Journal of East China University of Science and Technology, 2023, 49(3): 341-350. DOI: 10.14135/j.cnki.1006-3080.20221118001
    Citation: SONG Wenyu, ZHOU Haibo, WU Zongpei, LI Haiyuan, YUAN Yubo. Image Memory Induced Anomaly Detection Algorithm for Atmospheric Pollutant Emission Time-Series Data[J]. Journal of East China University of Science and Technology, 2023, 49(3): 341-350. DOI: 10.14135/j.cnki.1006-3080.20221118001

    图记忆诱导的大气排污时序数据异常检测算法

    Image Memory Induced Anomaly Detection Algorithm for Atmospheric Pollutant Emission Time-Series Data

    • 摘要: 大幅降低环境污染是国家“碳中和”和“碳达峰”战略的核心目标,降低大气排污是其关键。如何有效评估有关企业的污染排放数据质量是一个技术难题。本文以时间序列异常数据检测技术为基础,提出了图记忆诱导的大气排污时序数据异常检测算法(IMI-TSA);给出了异常时间序列的数学定义,将图记忆方法用于对时间序列的编码,建立了基于图结构特征的序列数据记忆模式,并利用样本间的特征与类别的关联性通过记忆来获得无标签样本的类别,同时利用有标签样本与无标签样本构建图记忆网络实现了时间序列异常检测任务;在生态环保领域采集了8个代表企业的大气排污数据,完成了相应异常检测。实验结果表明IMI-TSA算法准确率均达到了80%以上,该算法可用于构建大气排污数据监管平台。

       

      Abstract: Effectively evaluating the quality of pollution emission data from relevant enterprises is a significant technical problem. Based on time-series anomaly detection, IMI-TSA (Image Memory Induced Time-Series Anomaly), an image memory induced anomaly detection algorithm for atmospheric pollutant emission time-series data is proposed. Firstly, the mathematical definition of abnormal time-series is presented. Then, the image memory method is used to encode the time series and establish a memory mode based on the sequence data's image structure characteristics. Therefore, the category of unlabeled samples is obtained by the memory mode, using the correlation between features and categories of the samples. Finally, the image memory network is constructed with labeled and unlabeled samples to realize the time-series anomaly detection task. For ecological environmental protection, pollution discharge data from 8 representative enterprises are collected and anomaly detection for the data is conducted. The experiments show that the accuracy exceeds 80%, which means this algorithm can be used to build an atmospheric pollution data supervision platform.

       

    /

    返回文章
    返回