管网水质多指标动态关联异常检测方法

Contamination event detection method based on dynamic correlation analysis of multiple water quality parameters

为了提高城市供水管网水质监测系统的污染检测能力,利用污染物所引起多个指标变化之间的关联特性,提出基于多常规水质指标动态关联分析的水质异常检测方法.应用动态时间规整算法(DTW)衡量多个常规水质指标时间序列间的动态距离,刻画各指标波动的相似程度和动态关联特性.利用D-S证据理论融合各指标单独的异常概率,将融合后得到的供水管网水质异常概率与所设定的多指标融合异常概率阈值进行比较,作出当前时刻水质是否存在水质异常的综合判断.依托课题组模拟供水管网实验系统,设计不同浓度的硫酸铜和铁氰化钾污染物的注入实验,利用在线监测的pH值、浊度、余氯、溶解氧等8种常规水质指标进行动态关联分析和水质异常检测,方法的可行性和异常检测性能通过受试者工作特征曲线(ROC)进行验证.

A multivariate correlation analysis method was proposed by exploring the internal correlationwithin conventional water quality parameters before and after the occurrence of contamination event in orderto improve the performance of the existing water quality anomaly detection methods. The dynamic distancebetween each two monitored parameters was calculated to define the fluctuation correlation of thetwo time series by using the dynamic time warping （DTW ） method. The correlation coefficient was fusedwith univariate basic abnormal probability based on D-S evidence theory in order to obtain the fused probability.The synthesis alarm decision was made by comparing the fused probability with the threshold. Theproposed method was tested with experimental monitoring data collected from the laboratory pipeline system.Different concentrations of copper sulfate and potassium ferricyanide were separately injected into thepipeline system. Eight conventional monitoring parameters were measured by sensors installed along thepipeline. The collected monitoring data was applied to correlation analysis and probability fusion based onthe proposed method. The ROC analysis was introduced to verify the performance and validity of the techniques.