National Academy of Agricultural Science, Rural Development Administration, 249 Seodun-dong, Gwonson-gu, Suwon 441-707, Republic of Korea.
Water Res. 2013 Sep 1;47(13):4630-8. doi: 10.1016/j.watres.2013.04.018. Epub 2013 Apr 22.
Assuming a scenario of a hypothetical pathogenic outbreak, we aimed this study at developing a decision-support model for identifying the location of the pathogenic intrusion as a means of facilitating rapid detection and efficient containment. The developed model was applied to a real sewer system (the Campbell wash basin in Tucson, AZ) in order to validate its feasibility. The basin under investigation was divided into 14 sub-basins. The geometric information associated with the sewer network was digitized using GIS (Geological Information System) and imported into an urban sewer network simulation model to generate microbial breakthrough curves at the outlet. A pre-defined amount of Escherichia coli (E. coli), which is an indicator of fecal coliform bacteria, was hypothetically introduced into 56 manholes (four in each sub-basin, chosen at random), and a total of 56 breakthrough curves of E. coli were generated using the simulation model at the outlet. Transport patterns were classified depending upon the location of the injection site (manhole), various known characteristics (peak concentration and time, pipe length, travel time, etc.) extracted from each E. coli breakthrough curve and the layout of sewer network. Using this information, we back-predicted the injection location once an E. coli intrusion was detected at a monitoring site using Artificial Neural Networks (ANNs). The results showed that ANNs identified the location of the injection sites with 57% accuracy; ANNs correctly recognized eight out of fourteen expressions with relying on data from a single detection sensor. Increasing the available sensors within the basin significantly improved the accuracy of the simulation results (from 57% to 100%).
假设出现假设性的病原体爆发情况，我们旨在开发一种决策支持模型，以确定病原体入侵的位置，从而便于快速检测和有效遏制。该模型应用于真实的污水系统（亚利桑那州图森市的坎贝尔污水盆），以验证其可行性。所研究的盆地被分为 14 个子盆地。使用 GIS（地理信息系统）对与污水管网相关的几何信息进行数字化，并将其导入城市污水管网模拟模型，以生成出口处的微生物突破曲线。假设有定量的大肠杆菌（E. coli），它是粪便大肠菌群的指标，被假设引入 56 个检查井（每个子盆地四个，随机选择），并使用模拟模型在出口处生成总共 56 个大肠杆菌突破曲线。根据注射点（检查井）的位置、从每个大肠杆菌突破曲线中提取的各种已知特征（峰值浓度和时间、管道长度、旅行时间等）以及污水管网的布局对传输模式进行分类。利用这些信息，一旦在监测点检测到大肠杆菌入侵，我们就使用人工神经网络（ANNs）回溯预测注射位置。结果表明，ANNs 以 57%的准确率识别了注射点的位置；ANNs 在仅依靠单个检测传感器数据的情况下，正确识别了 14 个表达式中的 8 个。增加盆地内可用的传感器可显著提高模拟结果的准确性（从 57%提高到 100%）。
