文献类型：期刊文献

中文题名：两种模型对潜流湿地出水中总氮含量的预测能力对比

英文题名：Comparison of Predictive Ability on Contents of Total Nitrogen in the Effluent of Subsurface Flow Constructed Wetlands by Two Kinds of Models

作者：魏佳明[1,2,3] 崔丽娟[1,2,3] 李伟[1,2,3] 雷茵茹[1,2,3] 平云梅[1,2,3] 朱利[4]

第一作者：魏佳明

机构：[1]中国林业科学研究院湿地研究所;[2]湿地生态功能与恢复北京市重点实验室;[3]北京汉石桥湿地生态系统国家定位观测研究站;[4]北京汉石桥湿地自然保护区

年份：2017

卷号：15

期号：2

起止页码：281-286

中文期刊名：湿地科学

外文期刊名：Wetland Science

收录：CSTPCD;;Scopus;北大核心:【北大核心2014】；CSCD:【CSCD2017_2018】；

基金：中央级公益性科研院所基本科研业务专项项目(CAFYBB2014QA029)资助

语种：中文

中文关键词：潜流湿地;广义回归神经网络模型;遗传算法;BP神经网络模型;总氮

外文关键词：subsurface flow constructed wetlands; general regression model; genetic algorithm; back propagation neural network model; total nitrogen

分类号：X824

摘要：以北京顺义汉石桥湿地自然保护区中水处理厂的潜流湿地为例,选取2014~2015年的水质监测数据,以电导率、溶解性固体总量、氧化还原电位、p H、水温和总输入氮含量为输入层,比较遗传算法优化的BP神经网络模型和广义回归神经网络模型对多处理单元潜流湿地出水中的总氮含量预测能力。研究结果表明,遗传优化的BP神经网络模型的拟合优度R2可达到0.835,平均相对误差百分比为12.89%,说明其对出水中的总氮含量有一定的预测能力,但精度较差;广义回归神经网络模型的平均相对误差百分比为4.46%,精度较高。利用广义回归神经网络模型对潜流湿地出水中的总氮含量进行预测较适宜。
Taking the subsurface flow constructed wetlands of reclaimed water treatment factory in Han Shiqiao Nature Reserve as the research area, the study analyzed the water quality monitoring data in 2014 and2015. Conductivity, total dissolved solids, oxidation-reduction potential, p H, water temperature and contents of total input nitrogen were selected as input layers, to compare the predective ability of contents of total nitrogen in the effluent of subsurface flow constructed wetlands by using general regression neural network model and back propagation neural network model optimized by genetic algorithms. The results showed that the goodness of fit of back propagation neural network model optimized by genetic algorithms was 0.835 and the average error percentage was 12.89%. But the average error percentage of general regression neural network model was only 4.46%. The results indicated that the general regression neural network model could be better to predict the total nitrogen in subsurface wetlands than back propagation neural network model optimized by genetic algorithms.