Machine Learning Models for Drinking Water Quality Monitoring based on Sensor Data

即使在美国和加拿大这样的发达国家，饮用水中毒也不罕见。像碳酸钙，氯化钠，铅，K12大肠杆菌，真菌等污染物可以进入我们的水系统的每一个阶段，自来水厂，管道或末端水龙头。因此，仅从源头监测水质是不够的。从历史上看，在最终用户端安装监控设备是不可能的或非常昂贵的。只有像医院、食品加工公司等大型实体才能负担得起水质监测设备。现在传感器技术的发展使得在每个单独的水龙头上安装传感器成为可能和经济的。然而，单个水龙头上的传感器受到很大的未知因素和噪声的影响。毕竟，厨房不是一尘不染的科学实验室。这使得对监测不足的饮用水是否受到污染进行分类变得更加困难。我们需要一种新的强大的机器学习模型和算法，能够在这样一个嘈杂的环境中准确地触发警报。大数据技术和机器学习算法的进步使这一原本不可能实现的目标极有可能实现。

Drinking water poisoning are not rare even in well developed countries like USA and Canada. Pollutants likeCaCO3, Sodium Chloride, Lead, K12 E.Coli, Fungal could enter our water system in every stage, water plant,pipe, or the end faucet. Hence it is not enough to monitor water quality only at the source. Historically it isimpossible or prohibitively expensive to install monitoring devices at the end user side. Only big entities likehospitals, food processing companies, etc. can afford water quality monitoring devices. Now the developmentof sensor technology makes it possible and economical to install sensors at each individual faucet. However,sensors at individual faucets are subject to great unknown factors and noises. After all, a kitchen is not aspotless scientific lab. This causes greater difficulties in classifying whether or not the drinking water undermonitoring is polluted or not. We need a new robust machine learning model and algorithm, which canaccurately trigger alarms such a noisy environment. The advancement of big data technologies and machinelearning algorithms make this otherwise impossible goal highly likely.