基于传统在线监测指标的污染事件探测方法研究

发布时间：2018-04-01 12:24

本文选题：水环境监测　切入点：污染事件探测　出处：《清华大学》2015年硕士论文

【摘要】：水源地的水质质量对保证城市和农村居民的饮用水安全有至关重要的作用。然而近年来,水源地突发污染事件发生的频次和危害程度呈迅速上升趋势。针对我国面临的建立与加强水环境监测预警建设的实际问题,同时考虑到实际建设中的可操作性,本研究旨在基于目前较为普及的传统在线监测仪器,建立一套切实可行的突发污染事件探测预警方法。目前多数学者的研究集中于不同预警算法的建立与优化,但其测试数据均为计算机模拟得到的数据,对真实的污染事件无法起到良好的预警效果。本研究通过搭建中试系统,模拟了15种不同污染物引起的污染事件,以此实验数据作为探测方法建立的基础。本研究首次提出了基于皮尔逊相关系数的污染事件探测方法,并利用非支配排序遗传算法求解最优参数组合。对10种污染物共计32组实验数据进行求解,获得最优解集的帕累托前锋曲线,为决策者提供了参数设置选择。当参数取值兼顾准确率和错报率时,探测准确率为95.5%,错报率为4.4%;当参数取值倾向于低错报率时,探测准确率为93.0%,错报率为3.1%,均具有良好的探测效果。基于优化后的探测方法,分别利用已知污染事件测试集、未知污染事件测试集和日常基线测试集进行可靠性评价。结果表明,已知污染事件测试集的平均准确率为93.3%,未知污染事件测试集的平均准确率为93.6%,日常基线的平均错报率为0.35%,探测效果良好,具有较强的可靠性。同时,引入拉丁超立方抽样方法,分别利用已知污染事件测试集与未知污染事件测试集对提出的探测方法进行鲁棒性分析。结果表明,46组数据的平均鲁棒性为0.8520,鲁棒性良好。最后,本研究还原了以往学者提出的三种污染事件探测方法,并利用本研究的实验数据对这三种方法进行了测试与评价。结果表明,时间序列增量判别法和自回归模型判别法探测效果较差,多变量空间距离判别法探测效果较好;本研究提出的探测方法,效果远优于这三种方法。
[Abstract]:The quality of water quality in water sources is crucial to ensuring the safety of drinking water for both urban and rural residents. However, in recent years, The frequency and harm degree of sudden pollution events in water sources are rising rapidly. In view of the practical problems of establishing and strengthening water environment monitoring and warning construction in China, and considering the maneuverability of practical construction, The purpose of this study is to establish a set of feasible methods for detecting and warning pollution emergencies based on traditional on-line monitoring instruments. At present, most scholars' research focuses on the establishment and optimization of different early warning algorithms. But the test data are all the data obtained by computer simulation, which can not play a good early warning effect on the real pollution events. In this study, the pollution events caused by 15 different pollutants were simulated by building a pilot system. Based on the experimental data, the method of detecting pollution events based on Pearson correlation coefficient is proposed for the first time. The optimal parameter combination is solved by non-dominated sorting genetic algorithm. The Pareto front curve of the optimal solution set is obtained by solving 32 groups of experimental data of 10 kinds of pollutants. When the parameter value takes into account the accuracy rate and misstatement rate, the detection accuracy rate is 95.5 and the misstatement rate is 4.40.When the parameter values tend to be low misstatement rate, The detection accuracy is 93.0 and the false report rate is 3.1, which has good detection effect. Based on the optimized detection method, the test set of known pollution events is used separately. The reliability evaluation of unknown pollution event test set and routine baseline test set is carried out. The results show that, The average accuracy of the test set of known pollution events is 93. 3, the average accuracy of the test set of unknown pollution events is 93. 6, the average misreporting rate of daily baseline is 0. 35, the detection effect is good, and the reliability is strong. At the same time, the Latin hypercube sampling method is introduced. By using known pollution event test set and unknown pollution event test set respectively, the robustness of the proposed detection method is analyzed. The results show that the average robustness of 46 groups of data is 0.8520, and the robustness is good. In this study, three pollution detection methods proposed by previous scholars were reduced, and the experimental data were used to test and evaluate the three methods. The results show that, The detection effect of time series incremental discriminant method and autoregressive model discriminant method is poor, and the detection effect of multivariable spatial distance discriminant method is better than that of these three methods.
【学位授予单位】：清华大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：X832

【相似文献】