氧化沟系统出水COD预报的神经网络模型

摘要：以漯河市污水净化中心的Carrousel氧化沟（以下简称氧化沟）系统为考察对象，针对该系统进水水质复杂，控制滞后的难点，引入人工神经网络的理论和方法，对其模拟分析，建立了基于BP网络的氧化沟系统出水COD预报模型。模型性能检验和灵敏度检验表明，建成的模型准确度高，适应性强，可直接用于该系统出水COD预报，这为氧化沟工艺在线控制提供了一条简便的途径。

关键词：人工神经网络 氧化沟系统 出水COD

The ANN Model Predicting Effluent COD of Carrousel Oxidation Ditch System.

Abstract: The carrousel oxidation ditch system in Luohe Center of Wastewater Treatment is difficult to control on-line because the influent characteristics are complex and vary significantly. To resolve the problem, advanced artificial neural network (ANN) was employed to simulate the correlation between water parameters of oxidation ditch system and a BPNN model predicting effluent COD was built up. Sentivity and performance tests showed that the model can adapt to different situations and has good ability to generalize. It can be directly used to predict effluent COD concentration, which is very helpful to oxidation ditch system control on-line.

Keywords: ANN; oxidation ditch system; effluent COD

模型训练及检验

建模数值试验参数调整范围设定：E:0.2～0.4；G:0.001～0.5；隐节点数H:4～14；隐含层激活函数:tansig或logsig；输出层激活函数:logsig;训练最大迭代次数:1000。从72,000次搜索训练中筛选出最佳的一组解:网络结构7-6-1(三层神经网络每层节点数), E=0.3, G=0.15,H=6,隐含层函数:tansig。模型训练经过18次迭代达到稳定，训练总平方误差0.13,图2为误差下降曲线；模型模拟及检验(预报)结果见图3。

图2 误差下降曲线

Fig 2. The Error Curve of Training

模型性能指标值见表1。

表1 模型性能指标

Tab 1 The Values of Model Performance Testing

指标

C

R N A 模拟

0.9517

5.0152

0.1309

0.1214

预报

0.7399

9.9225

0.2645

0.2436

综合

0.8709

7.5127

0.1978

0.1736

从学习样本集检验合格的样本中任取一组样本,对应输入矢量X1…X7分别为:{14.3,273,292,33,5479,3394,44},考察网络输出随单项输入变化而改变的趋势，灵敏度曲线见图4

图3 出水COD原始值与模拟/预报结果

Fig.3. Observed and Simulated/Predicted Results of Effluent COD

图4 模型灵敏度曲线

Fig 4. The Sentivity Curves of Model