| 基于光谱和色谱特征数据融合的化工园区地表水污染源识别技术 |
| 摘要点击 1420 全文点击 199 投稿时间：2024-01-09 修订日期：2024-04-12 |
| 查看HTML全文
查看全文 查看/发表评论 下载PDF阅读器 |
| 中文关键词 化工园区 三维荧光光谱（EEMS） 气相色谱-质谱（GC-MS） BP神经网络（BPNN） 支持向量机（SVM） 污染源识别 |
| 英文关键词 chemical park 3D excitation emission matrix spectrometry（EEMS） gas chromatography-mass spectrometry（GC-MS） back propagation neural network（BPNN） support vector machine（SVM） pollution source identification |
| 作者 | 单位 | E-mail | | 赵远 | 常州大学环境科学与工程学院, 常州 213164 | zhaoyuan@cczu.edu.cn | | 殷新育 | 常州大学环境科学与工程学院, 常州 213164
浙江清华长三角研究院生态环境研究所, 浙江省水质科学与技术重点实验室, 嘉兴 314006 | | | 刘小凤 | 中环清科(嘉兴)环境技术研究院有限公司, 嘉兴 314006 | | | 金梦 | 浙江清华长三角研究院生态环境研究所, 浙江省水质科学与技术重点实验室, 嘉兴 314006 | | | 兰亚琼 | 浙江清华长三角研究院生态环境研究所, 浙江省水质科学与技术重点实验室, 嘉兴 314006 | | | 刘锐 | 浙江清华长三角研究院生态环境研究所, 浙江省水质科学与技术重点实验室, 嘉兴 314006 | 1393612924@qq.com |
|
| 中文摘要 |
| 化工园区内工业企业多,且各企业排水组分复杂,很多企业之间排水组成具有相似性.因此,当园区内地表水发生污染时,快速识别污染源的难度很大.为此,以嘉兴市某国家级化工园区为研究对象,收集该园区内7家重点企业10个批次的排水样本,并对其进行三维荧光光谱（EEMS）和气相色谱-质谱（GC-MS）分析.利用平行因子法从同一企业不同批次排水的EEMS图谱中提取共有组分,并构建EEMS特征数据矩阵.同时,从企业排水的GC-MS数据中筛选出具有高检出率或能够有效区分不同企业排水的特征物质,并构建GC-MS特征数据矩阵.比较研究基于不同数据矩阵以及不同模型的污染源识别效果.结果表明,无论是基于EEMS原始数据矩阵、EEMS特征数据矩阵还是GC-MS特征数据矩阵,BP神经网络模型对污染源的识别准确率都不高,分别仅为71.43%、76.19%和71.43%,略高于支持向量机模型（76.19%、76.19%和57.14%）.当使用EEMS与GC-MS特征数据融合矩阵时,污染源识别性能显著提升.支持向量机模型对7家企业排水的识别准确率、宏精确率、宏召回率以及宏调和平均值分别为95.24%、96.43%、95.24%和95.10%,而BP神经网络模型的识别性能更佳, 4项指标均接近100%.研究结果为化工园区等地表水污染源识别提供了一种有效方法. |
| 英文摘要 |
| Identification of the pollution source of surface water in a chemical park was difficult because many industrial enterprises with complex wastewater components and similar characteristics are located there. Therefore, a national-level chemical park in Jiaxing City was studied, and wastewater samples from ten batches of seven key enterprises in the park were collected and analyzed using 3D excitation emission matrix spectrometry （EEMS） and gas chromatography-mass spectrometry （GC-MS）. Parallel factor analysis was used to extract common components of EEMS spectra from different batches of drainage in the same enterprise to construct an EEMS characteristic data matrix. Furthermore, specific substances with high detection rates or that could effectively distinguish other enterprise drainage were screened out from the GC-MS data to construct a GC-MS characteristic data matrix. Pollution source identification was attempted with different models based on different data matrices. The results showed that regardless of whether being based on the EEMS original data matrix, the EEMS characteristic data matrix, or the GC-MS characteristic data matrix, the identification accuracy of the BP neural network model was not high, only 71.43%, 76.19%, and 71.43%, respectively, which was slightly higher than that of the support vector machine model （76.19%, 76.19%, and 57.14%）. However, when the EEMS and GC-MS characteristic data fusion matrix were used, the pollution source identification performance was significantly improved. The identification accuracy, macro precision, macro recall, and macro harmonic mean of the support vector machine model for the wastewater of the seven enterprises were 95.24%, 96.43%, 95.24%, and 95.10%, respectively, while the performance of the BP neural network model was better, with all four indicators close to 100%. The study provides an effective method for identifying surface water pollution sources in chemical parks. |
|
|
|
