利用回归和人工神经网络预测城市雨水径流总磷模型的有效性

利用人工神经网络(ANN),探讨在不无监测系统的集水区城市降水质量预测的适用性.预测使用常规的气候和地理数据集,通过构建背景传播的神经网络和回归联合模型,克服利用逐步回归的方法对数据进行分析时违背独立数据假设的问题.研究通过交叉验证用于确定停止降水时间为输入变量参数,利用地区平均浓度(EMC)作为独立的变量,构建的模型比用负荷量构建的模型更精确.数据域和输入变量的选择对回归模型的准确性也有较大影响.但计算效率、动量和隐节点数目的选择等因素,对人工神经网络模型准确性的影响较小.同时,回归和人工神经网络模型的降水质量预测结果十分相似,但在不无监测系统的集水区域城市降水质量的预测方面,回归模型更有实效性.

Validity of Modelling Total Phosphorus in Urban Stormwater Runoff Using Regression and Artificial Neural Networks

This paper investigates the applicability of using artificial neural networks (ANNs) to predict urban stormwater quality at unmonitored catchments.Back-propagation neural networks and regression models were constructed using a set of general climatic and geographic data.Violation of the assumption of data independence lead to the inclusion of insignificant variables when the data was analysed using stepwise regression.To overcome this problem, cross validation was used to determine when to cease input variable entry.Models constructed using event mean concentration (EMC) as the dependent variable were more accurate than those using load.The data domain and selection of input variables had a significant effect upon the accuracy of the regression models.Whereas the choice of learning rates, momentum and number of hidden nodes had an insignificant effect upon the accuracy of the ANN models.Regression and ANN models yielded similar predictions.However, the efficiency of the regression models made them a more pragmatic approach for predicting urban stormwater quality at unmonitored sites.