基于高频在线水质数据异常的突发污染预警

发布时间：2018-10-08 19:41

【摘要】：在高频水质自动监测背景下,建立了基于软测量和水质时间序列异常检测的水体突发污染预警预报技术.假定突发污染事故会引起典型自动监测水质参数变化,采用回归分析建立水质参数和在线高频监测水质参数间的线性关系进行软测量,采用人工神经网络预测短程水质变化,建立基于预测残差的异常判断最小阈值,最终通过有序监督聚类进行水质突变检测从而对突发污染事故进行预警.采用美国弗吉尼亚州的Potomac River流域在线监测数据进行算法验证和案例分析.分析受试者工作曲线(ROC)表明:该方法对2倍异常和3倍异常水平的检测准确率分别为62.7%和92.5%,且随着异常水平的增加准确率增加,通常突发污染事故中特定污染物浓度水平一般明显高于3倍,该方法具有较高的准确率.较其他突发污染水质预警技术,该技术有效缩短了平均检测时间,为流域污染预警预报和快速应急响应提供了新途径.
[Abstract]:Under the background of high frequency water quality automatic monitoring, the technology of water burst pollution early warning and prediction based on soft sensing and water quality time series anomaly detection is established. Assuming that the sudden pollution accident will cause the change of the typical automatic monitoring water quality parameters, the linear relationship between the water quality parameters and the on-line high frequency monitoring water quality parameters is established by regression analysis, and the short range water quality change is predicted by artificial neural network. The minimum threshold of abnormal judgment based on predicted residual error is established, and the sudden change detection of water quality is finally carried out by orderly supervised clustering to warn the sudden pollution accident. The online monitoring data of Potomac River watershed in Virginia are used for algorithm verification and case analysis. The analysis of the operating curve (ROC) showed that the detection accuracy of the method for 2 and 3 times abnormal level was 62.7% and 92.5%, respectively, and the accuracy rate increased with the increase of abnormal level. Usually, the concentration of specific pollutants in sudden pollution accidents is obviously higher than 3 times, and this method has a high accuracy. Compared with other technologies of water quality warning for sudden pollution, this technology can effectively shorten the average detection time and provide a new way for early warning and forecasting of river basin pollution and rapid emergency response.
【作者单位】：哈尔滨工业大学环境学院;南方科技大学环境科学与工程学院;哈尔滨工业大学城市水资源与水环境国家重点实验室;
【基金】：中国博士后科学基金资助项目(2014M551249) 国家自然科学基金资助项目(51779066) 中央高校基本科研业务费专项基金资助项目(HIT.NSRIF.2017060)
【分类号】：X832

【相似文献】