污染物警告系统（CWS）的主要目标是快速检测可能的污染物 在配水系统内发生事故，以便采取措施将后果降至最低。这个 美国环境保护局（USEPA）的水安全（WS）倡议（前身为WaterSentinel）旨在设计、部署和评估 模型CWS。WS-CWS模型结合了多种监控策略，包括 在线水质监测、采样和分析、增强的安全监测、消费者 投诉监控和公共卫生监控。 在线水质监测是本文评价的重点。 这 该组件由布置在整个配水系统战略性位置的监测站组成 包含持续监控各种WQ参数的传感器。安装传感器是不切实际的 直接监测每种可能的污染物，因为存在太多的潜在污染 污染物和传感器技术的缺乏，甚至可以覆盖这些污染物的一小部分。因此 在WS-CWS设计中，通过监测标准WQ间接检测污染 试图识别水质异常或与既定水质标准的偏差的参数 质量基准状态，这可能表示污染。 一个或多个WQ参数中的异常可以提供污染预警，但前提是 可以从嘈杂的背景数据中提取。配水系统水质数据自然是可变的 而且在很大程度上是不可预测的，收集数据的传感器硬件不完善也增加了不确定性。 因此，WS-CWS的在线监控组件依赖于事件检测算法 区分水质的正常变化和由污染引发的水质变化 异常情况。在本文中，事件检测系统（EDS）指的是一个 包括事件检测算法。在实践中，EDS工具将通过接收数据来近乎实时地工作 通过水务公司的监控和数据采集（SCADA）进行远程遥测 系统，执行近实时分析，并返回结果（即，是否发出警报，或 在操作员屏幕上显示事件发生的概率）。而这种异常检测工具 在其他应用中，几乎没有将其应用于异常检测的经验 饮用水系统。 WS-CWS试点项目继续提供一个独特的机会来收集严格测试所需的数据 EDS工具的评估。本文描述了作为WS的一部分开发的评估方法 倡议包括实验矩阵和使用的性能度量，评估了两个EDS工具 并部署在WS-CWS试点实用程序中，部分评估结果在 2007年春天。该评估的主要目标包括选择一个或多个EDS工具进行评估 在pilot utility上部署，建立一个模拟正常情况的EDS工具评估流程 公用设施运行条件，提高对EDS工具功能的理解，并量化 EDS工具的性能。包括11个参考文献、表格、图表。

The primary goal of a Contaminant Warning System (CWS) is to quickly detect a possible contamination incident within a water distribution system so that action can be taken to minimize consequences. The US Environmental Protection Agency's (USEPA's) Water Security (WS) Initiative (formerly WaterSentinel) seeks to design, deploy, and evaluate a model CWS. The WS-CWS model combines multiple monitoring and surveillance strategies including online water quality monitoring, sampling and analysis, enhanced security monitoring, consumer complaint surveillance, and public health surveillance. Online water quality (WQ) monitoring is the focus of the evaluation described in this paper. This component consists of monitoring stations placed strategically throughout the water distribution system that contain sensors that continuously monitor various WQ parameters. It is impractical to install sensors that directly monitor for each possible contaminant because there are simply too many potential contaminants and a lack of sensor technologies to cover even a fraction of these contaminants. Therefore, under the WS-CWS design, contamination is detected indirectly by monitoring standard WQ parameters in an attempt to identify water quality anomalies, or deviations from an established water quality base state, that might be indicative of contamination. Anomalies in one or more WQ parameters can provide early warning of contamination, but only if they can be picked out of noisy background data. Distribution system water quality data is naturally variable and largely unpredictable, and the imperfect sensor hardware that collects the data adds to the uncertainty. Therefore, the online monitoring component of WS-CWS relies upon event detection algorithms to distinguish between normal variations in water quality and changes in water quality triggered by abnormal conditions. In this paper, event detection system (EDS) refers to a software package that includes event detection algorithm(s). In practice, EDS tools will work in near real-time by receiving data via remote telemetry through the water utility's supervisory control and data acquisition (SCADA) system, perform an analysis in near real-time, and return a result (i.e., either sound an alarm or not, or indicate the probability of an event on an operator's screen). While such abnormality detection tools have been used in other applications, there is little experience applying them to detection of anomalies in drinking water systems. The WS-CWS pilot continues to provide a unique opportunity to collect data necessary for a rigorous evaluation of EDS tools. This paper describes the evaluation approach developed as part of the WS Initiative including the experimental matrix and performance measures used, the two EDS tools evaluated and deployed at the WS-CWS pilot utility, and a portion of the results of the evaluation conducted in the Spring of 2007. The primary objectives of this evaluation included selecting one or more EDS tool for deployment at the pilot utility, establishing a process for evaluating EDS tools which mimics normal utility operating conditions, improving understanding of EDS tool capabilities, and quantifying the performance of EDS tools. Includes 11 references, tables, figures.