卷积神经网络技术在入海河流污染溯源中的应用研究

Application of Convolutional Neural Network in Pollution Source Identification for Sea-Discharging River

中文关键词: 三维荧光  卷积神经网络  水污染溯源  地表水

英文关键词: Three-dimensional fluorescence  Convolutional neural network  Water pollution traceability  Surface water

中文摘要:

      入海河流污染严重、溯源难度大，是目前流域管控面临的关键环保难题。为实现流域污染水体的准确、高效溯源识别，以Z市X河及流域上游7家重点企业作为研究对象，于2023年12月至2024年10月分别对各企业污水处理站进、出水和X河上、中、下游水样进行6次采样，并配制企业废水与河流水样混合的模拟企业事故水样，水样经三维荧光测试后，利用基于卷积神经网络构建的溯源算法模型对荧光光谱数据进行特征学习和溯源分类。溯源准确性验证结果表明，构建的溯源算法模型能够准确识别各企业实际污水和低浓度模拟企业事故水样，溯源验证准确率达84.38%，该方法可有效提升流域环保部门的监管执法效率，成为保护流域水质和生态的有效工具。

英文摘要:

      Severe pollution in rivers flowing into the sea and the difficulty in tracing pollution sources present critical environmental challenges for current watershed management. To achieve accurate and efficient source identification of polluted water body in watershed, the paper took X River in Z City and seven key enterprises upstream of the watershed as research subjects. From December 2023 to October 2024, six rounds of sampling were conducted on inlet and outlet from each enterprise’s sewage treatment station, as well as water samples from the upper, middle, and lower reaches of X River. Simulated accident water samples were prepared by mixing each enterprise wastewater with river water samples. Through three-dimensional fluorescence test, a source identification algorithm model based on convolutional neural network was applied to extract features from the fluorescence spectral data and perform source classification. Verification results of accuracy showed that the algorithm model can accurately identify both actual wastewater from enterprises and low-concentration simulated enterprise accident water samples, achieving a source identification accuracy of 84.38%. This method can significantly enhance regulatory enforcement efficiency of watershed environmental protection departments, serving as an effective tool for protecting water quality and ecosystems.