供水管网流量监测数据异常值检测方法对比分析  

Comparison of Methods for Flow Monitoring Data Outlier Detection in Water Distribution Network

作　　者：胡诗苑 高金良 钟丹 武睿 刘路明 HU Shi‑yuan;GAO Jin‑liang;ZHONG Dan;WU Rui;LIU Lu‑ming(School of Environment,Harbin Institute of Technology,Harbin 150090,China;Guangdong Yuehai Water Investment Co.Ltd.,Shenzhen 518021,China;National Engineering Research Center of Urban Water Resources Co.Ltd.,Harbin Institute of Technology,Harbin 150090,China)

机构地区：[1]哈尔滨工业大学环境学院,黑龙江哈尔滨150090 [2]广东粤海水务投资有限公司,广东深圳518021 [3]哈尔滨工业大学水资源国家工程研究中心有限公司,黑龙江哈尔滨150090

出　　处：《中国给水排水》2024年第3期53-59,共7页China Water & Wastewater

基　　金：国家重点研发计划项目(2022YFC3203800);国家自然科学基金资助项目(51978203);黑龙江省重点研发计划项目(2022ZX01A06);揭榜制科研项目(CE602022000203)。

摘　　要：随着信息化技术的发展,水务企业迎来了智慧化转型升级。数据采集与预处理作为水务企业实现智慧管理的重要前序步骤,为后续数据挖掘、运营管理、调度决策提供了基础。由于环境的影响、管网中的随机扰动、管网事故等原因,监测数据的质量问题广泛存在,因此寻求有效的供水管网流量监测数据的异常值检测方法至关重要。基于此,首先根据供水管网流量监测数据的基本特征和时间维度的相关性,将常见异常归纳为3个类型;其次,以东南沿海某市的真实小区流量监测数据为例,分别探究基于统计、密度和预测的Boxplot、LOF与Prophet异常值检测模型在不同类型异常数据检测中的性能。结果表明,Boxplot与LOF模型能够较准确地识别出异常数据,但Boxplot对异常的判断标准较宽泛,容易将部分非异常数据识别为异常点,Prophet对于不稳定性较高的流量数据识别效果有限。With the development of information technology,water enterprises are undergoing intelligent transformation and upgrading.Data collection and preprocessing is an important pre‑step for water enterprises to realize intelligent management,and provides a foundation for subsequent data mining,operation management and scheduling decision.Due to the reasons such as environmental factors,random disturbance in the pipe network and pipe network accident,monitoring data quality issues exist widely,making it is very important to find an effective method for flow monitoring data outlier detection in water distribution network.The common anomalies were firstly classified into three categories according to the basic characteristics and temporal correlation of flow monitoring data in water distribution network.Then,the performance of Boxplot,LOF and Prophet outlier detection models based on statistics,density and prediction in the detection of different types of real flow monitoring data outliers was explored in a southeast coastal city of China.Boxplot and LOF models identified outliers more accurately.However,Boxplot had broad criteria for outlier identification,and it was easy to identify some non‑abnormal data as outliers.Prophet had limited effectiveness in identifying unstable flow data.

关 键 词：流量监测数据 异常值检测 Boxplot LOF PROPHET 

分 类 号：TU991[建筑科学—市政工程]