BP神经网络在长江水质COD预测中的应用  

Application of BP Neural Network in Predicting COD of Yangtze River

作　　者：郭庆春[1,2,3] 郝源[1] 李雪[2,3] 杜北方 张向阳[1] 

机构地区：[1]陕西广播电视大学,陕西西安710119 [2]中国科学院地球环境研究所黄土与第四纪国家重点实验室,陕西西安710075 [3]中国科学院大学,北京100049

出　　处：《计算机技术与发展》2014年第4期235-238,242,共5页Computer Technology and Development

基　　金：国家重点基础研究发展规划项目(2010CB833406);国家自然科学基金资助项目(40975020;41075067);陕西省教育科学研究计划项目(12JK0123;12JK0414)

摘　　要：水质变化具有非线性、突变性,且含有噪声,传统线性预测模型不能全面反映其变化规律,预测精度低,误差大。针对水质变化规律复杂,影响因素间非线性程度高的问题,为了提高水质预测精度,将改进算法的BP神经网络引入化学需氧量(COD)预测预报领域,以pH、溶解氧(DO)、氨氮(NH3-N)为输入向量,以COD为输出向量,建立了COD的预测模型并对效果进行检验。结果表明:检验样本中COD的预测值与实测值的线性相关系数为0.991。BP神经网络模型预测精度高,收敛速度快,具有良好的泛化能力,能较好地反映COD和影响因子的变化规律。Water quality change is of nonlinear and dynamicity ,it is a kind of complex time series data, therefore, the traditional linear pre- diction model cannot reflect the variation rule,and the prediction accuracy is low. For the problems of complex water quality change rule and high degree of nonlinear between factors,in order to improve the water quality prediction accuracy,introduce the BP neural network of improved algorithm into a model of COD, with pH, DO, NH3 -N as input and COD as output, the prediction model of COD is estab- lished and tested, The research results show the linear correlation coefficient of COD between forecasting and the monitoring in the test samples is 0.991. BP neural network has high forecast precision,fast convergence rate and the good generalization ability,which can bet- ter reflect the change rule between COD and impact factors.

关 键 词：神经网络 水质 化学需氧量 溶解氧 氨氮 

分 类 号：X522[环境科学与工程—环境工程]