非点源污染河流水质的人工神经网络模拟  

ANN approach for modeling and prediction of water quality in non-point sources polluted river

作　　者：陈丁江[1] 吕军[1] 沈晔娜[1] 金树权[1] 

机构地区：[1]浙江大学环境与资源学院资源科学系,浙江杭州310029

出　　处：《水利学报》2007年第12期1519-1525,共7页Journal of Hydraulic Engineering

基　　金：国家"973"计划项目(2002CB410807);国家自然科学基金资助项目(40571070);浙江省科技厅计划项目(2004C33067)

摘　　要：本文应用非点源污染河流水质的BP人工神经网络模型,模拟长乐江水体的总氮、总磷和溶解氧浓度的变化。通过模拟的水质参数相关性分析,协同非点源污染河流的机理性水质模型分析,确定适当的BP网络模型结构。采用实测的水质、水文逐月数据资料,对不同结构的BP网络模型进行了训练与验证。结果表明,相关性与机理性模型协同分析的方法,能较好地解决BP网络输入层参数的选择问题,所选择的参数较全面地表达了流域非点源污染发生的主要驱动因素和河流中污染物自净过程的主要影响因素。BP网络模型可以较精确地模拟非点源污染河流的水质变异,各水质参数模拟结果的平均相对误差在±10%以内。单隐含层结构的BP网络模型模拟结果比多隐含层结构模型结果更准确;单参数输出结构的网络模型模拟结果,优于多参数输出结构模型的模拟结果。The 1-D model using BP algorithm for simulating the total nitrogen （TN）, total phosphorus （TP） and dissolved oxygen （DO） in non-point sources polluted river was presented. The input parameters of BP network were selected through coupling analyses on correlation between hydro-parameters and water quality parameters and mechanism modeling of river water quality. Sequentially, the conceptual models for BP network for different simulated parameters were established, and the BP models with different structures were trained and validated using monthly data set. The results show that the coupling analyses method can efficiently ascertain appropriate input parameters for BP network. The selected parameters comprehensively express the main factors for non-point pollution occurrence in a given watershed and pollutant purification process. It is concluded that the BP network can precisely simulate the water quality variation due to non-point sources pollution with relative errors of water quality parameters less than ± 10%.

关 键 词：非点源污染 水质模拟:人工神经网络 BP算法:河流 

分 类 号：X522[环境科学与工程—环境工程]