水资源与水工程学报

文章摘要

查木哈, 卢志宏，翟继武，张福顺.双隐含层BP神经网络模型在老哈河水质预测中的应用Journal of Water Resources and Water Engineering[J].,2018,29(2):56-61

双隐含层BP神经网络模型在老哈河水质预测中的应用

Using double-suppressed BP neutral network model to predict water quality in Laoha River

DOI：10.11705/j.issn.1672-643X.2018.02.10

中文关键词: 双隐含BP神经网络河流水质水质指标缺失数据拉格朗日插值法水质预测老哈河

英文关键词: double-suppressed BP neutral network river water quality water quality index missing data Lagrange interpolation water quality prediction Laoha River

基金项目:中国农业科学院创新工程牧草病虫害灾变机理与防控团队（CAAS-ASTIP-IGP）；内蒙古自然科学基金项目（2017MS0380）

Author Name	Affiliation
CHA Muha¹, LU Zhihong²， ZHAI Jiwu¹, ZHANG Fushun³	(1.赤峰市环境监测中心站，内蒙古赤峰 024000 2.铜仁学院，贵州铜仁 554300；3.中国农业科学院草原研究所，内蒙古呼和浩特 010010)

Hits: 1815

Download times: 893

中文摘要:

为了快速准确预测老哈河水质，采用老哈河2011-2015年水质监测数据，运用拉格朗日插值法补充缺失值，分别对化学需氧量、生化需氧量、高锰酸盐指数和总磷浓度建立Levenberg - Marquardt 优化的双隐含层BP神经网络模型，利用2011-2014的数据建立训练网络，以2015年的数据进行验证与测试。结果表明：五日生化需氧量预测模型，第一隐含层节点数为 4 , 第二隐含层节点数为 12 时，决定系数 0.751 6（P=0.000 3），平均相对误差25.73%；化学需氧量预测模型，第一隐含层节点数为 12, 第二隐含层节点数为 10时，决定系数 0.887 5（P<0.000 1），平均相对误差27.69%；高锰酸盐预测模型，第一隐含层节点数为6, 第二隐含层节点数为3时，决定系数 0.854 7（P<0.000 1），平均相对误差28.90%；总磷预测模型，第一隐含层节点数为12,第二隐含层节点数为12时，决定系数0.889 2（P<0.000 1），平均相对误差17.94%。应用拉格朗日插值法对缺失数据进行补充后建立的双隐含层BP神经网络模型相对误差均小于28.90%，模型的预测效果较好，其中总磷浓度预测效果最好。通过拉格朗日插值，可以建立老哈河赤峰段甸子点位污染指标的双隐含层人工神经网络模型进行水质预测。

英文摘要:

In order to quickly and accurately predict the water quality of the Laoha River, the monitoring data of the Laoha River for the period from 2011 to 2015 was used to supplement the missing values using the Lagrange interpolation method. The values of chemical oxygen demand, biochemical oxygen demand, permanganate index, and total phosphorus concentration were respectively used to establish a Levenberg-Marquardt optimized double hidden layer BP neural network model.By using the 2011-2014 data, a training network was established, and the validation and test were conducted with the data of 2015. The results show that: for the five-day BOD forecasting model, when the first hidden layer node number is 4 and the second hidden layer node number is 12, the determination coefficient is 0.751 6 (P=0.000 3), and the average relative error is 25.73%. For the chemical oxygen demand forecasting model, when the number of nodes in the first hidden layer is 12 and the number of nodes in the second hidden layer is 10, the coefficient of determination is 0.887 5 (P<0.000 1), and the average relative error is 27.69%. For the permanganate prediction model, when the number of nodes in the first hidden layer is 6, and the number of nodes in the second hidden layer is 3, the coefficient of determination is 0.854 7 (P<0.000 1), and the average relative error is 28.90%. For the TP prediction model, when the number of nodes in the first hidden layer is 12, and the number of nodes in the second hidden layer is 12, the coefficient of determination is 0.889 2 (P<0.000 1), and the average relative error is 17.94%. The relative error of the BP neural network model of double hidden layer established by complementing the missing data using Lagrange interpolation method is less than 28.90%. The prediction effect of the model is good, and the total phosphorus concentration has the best prediction effect. Through Lagrange interpolation, a dual-hidden artificial neural network model can be established to predict the water quality in the Laihe River Chifeng segment.

View Full Text View/Add Comment Download reader