Development of smart data analytics tools to support wastewater treatment plant operation
Development of smart data analytics tools to support wastewater treatment plant operation
复制标题
DOI：
10.1016/j.chemolab.2018.03.006
复制
发表时间：
2018-06-15
影响因子：
3.9
通讯作者：
Saint, Christopher P.
中科院分区：
文献类型：
--
作者：
Chow, Christopher W. K.;Liu, Jixue;Saint, Christopher P.
进行了应用化学计量学方法、k-means 聚类算法来开发使用在线测量的实时工业过程预警系统的案例研究。 2013年至2015年间，在污水处理厂入口处安装了在线分光光度计，进行了为期18个月的监测研究。在此期间，内部开发了一个基于网络的原型门户，具有针对复杂在线数据集的数据集成、可视化、预测和异常检测功能，以评估分光光度计与其他数据库（例如降雨量和温度）获取的光谱数据。使用关联分析和特征选择等多种化学计量学选项从获取的数据中提取有用的操作信息。本文详细描述了异常检测功能，包括模式学习和比较算法以及强大的用户界面。通过使用这些功能，可以从处理厂入口处的光谱数据成功检测到过程异常。然后将检测到的事件/扰动与处理厂日志进行比较，发现它们吻合良好，这证明异常检测技术是有效的，并且有可能为协助工厂操作员提供决策信息。此外，所提出的异常检测技术也是一种灵活的算法，可与任何类似的时间序列数据一起检测其他过程相关问题，以提供实时警告来支持处理厂的运营。
A case study of applying chemometrics approach, k-means, a clustering algorithm to develop a real-time industrial process early warning system using online measurements was conducted. An online spectrophotometer was installed for an eighteen-month monitoring study between 2013 and 2015 at the inlet of a wastewater treatment plant. During this time a web-based prototype portal with data integration, visualization, prediction and anomaly detection functions for complex online data sets was developed in-house to assess the spectral data acquired by the spectrophotometer together with other databases (such as rainfall and temperature). Several chemometrics options, such as association analysis and feature selection, were used to extract useful operational information from the acquired data. In this paper, the anomaly detection function which includes pattern learning and comparison algorithms and a powerful user interface was described in detail. By using the functions, process upsets were successfully detected from the spectral data at the inlet of the treatment plant. The detected events/ upsets were then compared with the treatment plant logs and they were found aligned well, which proved that the anomaly detection technique was effective and has the potential to inform decision to assist plant operators. In addition, the proposed anomaly detection technique is also a flexible algorithm which works with any similar time series data to detect other process related issues to provide real-time warning to support treatment plant operations.