模糊神经模型对废水处理过程COD的预测及控制  

Research on Forecasting COD in Wastewater Treatment Process and Controlling Addition Dosage Using Fuzzy Neural-Net Model

作　　者：马邕文[1] 黄明智[1] 万金泉[1] 王艳[1] 

机构地区：[1]华南理工大学环境科学与工程学院,广东广州510640

出　　处：《中国造纸学报》2008年第4期113-118,共6页Transactions of China Pulp and Paper

基　　金：广东省科技厅重大专项攻关项目(项目编号2003A30406);广州市科技计划项目(项目号2004Z3-D0271)

摘　　要：基于提高工业废水处理自动化程度、保证出水水质的考虑，通过正交实验法获得了用于FNN模型训练和测试的样本数据，并建立了相应的FNN预测和控制模型；结合模糊C均值聚类和混合算法完成网络的结构辨识和参数辨识，仿真结果表明，预测模型具有很好的学习能力和泛化能力，而测试数据的相对误差范围为1．2％～8％；建立好的预测控制模型与MCGS组态软件结合应用于实验室的造纸废水处理控制，改变原水COD和进水流量的大小，控制系统会自动计算出该时刻的加药量，其出水CODcr维持在400mg／L左右，同人工恒定加药量相比平均相对误差小很多，只有1．98％，结果表明MCGS和控制算法结合可以有效控制废水处理过程。In order to improve automatic level in industrial wastewater treatment and have a high treated effluent quality, a fuzzy neural-based model with Takagi-Sugeno inference-based prediction and a conventional network for system control are presented, and the data samples are collected through orthogonal test to train and test the FNN model obtained. Fuzzy C-means clustering to identify model＇s architecture and a hybrid learning rule to identify parameters are also introduced, and the simulative results indicate that the model has good ability both in learning and generalization, with relative errors of test data are 1.2% - 8%. The control of papermaking wastewater treatment process in the laboratory under the combination of FNN model established and MCGS shows that the expectation CODcr value of effluent from integrative reactor was set in 400 mg/L,the dosage was worked out by the intelligent control system when CODcr value or flux of influent was changed to make sure CODer value of effluent from the high effective reactor was 400 mg/L. Compared with invariable addition dosage, the average relative error of this model was only 1.98 %, the results indicated that the combination of MCGS and control algorithms is virtual to control the wastewater treatment process.

关 键 词：模糊神经网络 工业废水处理 预测控制 

分 类 号：X793[环境科学与工程—环境工程]