基于多因子融合的水质异常检测算法  

Multi-parameters fusion algorithm for detecting anomalous water quality

作　　者：何慧梅[1] 侯迪波[1] 赵海峰[1] 黄平捷[1] 张光新[1] 

机构地区：[1]浙江大学工业控制技术国家重点实验室,浙江杭州310027

出　　处：《浙江大学学报（工学版）》2013年第4期735-740,共6页Journal of Zhejiang University：Engineering Science

基　　金：国家自然科学基金资助项目(41101508);国家重大科技专项资助项目(2008ZX07420-004)

摘　　要：为了及时有效地发现偶发或蓄意的水质异常,保障饮用水供水安全,在总结现有异常检测算法的基础上,提出基于自回归(AR)模型和模糊C-均值聚类(FCM)的多因子融合水质异常检测算法.通过AR模型实现水质背景信号的高精度跟踪,采用FCM算法融合多种水质指标的AR预测残差,与设定阈值作比较并判断异常.实验结果表明,与单因子算法、直接融合水质指标监测值的算法、利用多维欧氏距离融合AR预测残差的算法相比,提出的多因子融合水质异常检测算法具有更高的异常检出率和较低的误报率.Rapid detection of anomalous water quality within a water quality early warning system is desirable for the protection of drinking water against both accidental and malevolent contamination events. Using existing data from in-situ water quality sensors, a multi-parameters fusion algorithm was presented based on AR model and fuzzy C-means clustering algorithm （FCM）. AR model uses previous observations of the water quality to predict future water quality values. Then prediction deviations can be caculated and classificated into two clusters within FCM. To discriminate between normal and anomalous water quality, the distance between a new prediction deviation and the anomalous cluster center was calculated and compared to a constant threshold. Results show that this fusion algorithm produces the lowest false alarm rate （FAR） and highest positive detection （PD） for all cases of simulated event strength, which has better performance than single-parameter algorithm, multi-dimension algorithms fused water quality data directly and multi-dimension Euclidean distance algorithm fused predictive residuals.

关 键 词：多因子融合 水质异常检测算法 自回归模型 模糊C-均值聚类 受试者工作特征曲线(ROC) 

分 类 号：X832[环境科学与工程—环境工程]