序批式活性污泥法中污水化学需氧量的软测量研究
|
摘要
随着经济的发展和社会的进步,人类的需求不断增加,水资源的污染问题越来越严重。序批式活性污泥法(SBR)是一种污水的生物处理方法,广泛运用于世界各国污水处理厂。SBR法的特点之一是它严重依赖现代自动化技术。化学需氧量(COD)是表示水中有机物总量的一个综合性指标,是关于天然水体环境质量分级的重要指标之一,也是导致水体发黑发臭的主要因素。COD是SBR反应过程的重要控制参数。传统的COD测量是采用化学方法,虽然准确,但时效性很差,成本高,测试过程还会造成二次污染。随着污水处理工艺技术发展的日趋成熟,污水处理研究的重点已经转移到提高污水处理工艺过程的自动控制水平,改善出水水质,加强运行监控等方面。因此,研究SBR工艺中COD的快速准确测量具有重要理论意义和现实价值。
本文首先通过介绍软测量技术确定了实时准确测量COD参数的研究方向,并介绍了软测量模型的建立方法。结合软测量的建模方法,详细分析了SBR工艺中与COD参数有关的水质参数,找到了软测量建模的辅助变量,并建立了基于水质参数变化率的SBR污水COD软测量模型和基于水质参数时间窗的SBR污水COD软测量模型。之后,详细介绍了如何通过BP神经网络实现软测量的功能,并针对神经网络的泛化能力和局部最小问题提出了优化方案。
其次,针对COD软测量建模中有限种类辅助变量造成的矛盾数据问题和神经网络学习的局部最小问题,提出了一种支持向量机和神经网络联合软测量SBR污水处理中COD的方法。该方法通过引入支持向量机对COD值进行预估计,再根据COD的变化规律使用BP神经网络和Elman神经网络分别估计污水COD值,再进行数据融合得到COD的估计值。实验表明,该方法的软测量结果优于单一神经网络的软测量结果。
最后,在提出单一BP神经网络建模和采用支持向量机与神经网络联合软测量的模型基础上,介绍了多神经网络集成的概念。多神经网络集成中,详细介绍了AdaBoost算法,提出了一种AdaBoost的多神经网络COD软测量模型。该模型在预测准确性和稳定性上均有了进一步提高。
With economic and social development and the increasing human needs, water pollution is becoming a more and more serious problem. Sequencing batch reactor (SBR) is a biological wastewater treatment method, which is widely used in the wastewater treatment plants worldwide. The features of SBR make it rely heavily on the modern automation technology. Chemical oxygen demand (COD) is a comprehensive indicator of the total organic matter amount in water and an important indicator of the natural water quality classification, and also one of the main factors causing the deterioration of waterbodies.It largely contributes to the smelly and black water. COD is an important control parameter of SBR process. Traditionally, COD is measured with chemical methods, which is accurate with the shortage of poor timeliness and high cost, and the testing process is usually carried on with the production of other pollutants. As the sewage treatment technology has matured, the focal points of sewage treatment research has changed to how to improve the level of automation in the sewage treatment process, improve the water quality and enhance the operation monitoring and so on. Therefore, the fast and accurate measuring of COD in SBR process is both theoretically and practically important.
This paper set the research direction of real-time accurate COD parameter measurement by introducing soft measurement technology, and then introduces the soft measurement modeling approaches. With the modeling approaches, the paper gave a detailed analysis on the COD parameter-related water-quality parameter in the SBR process, found out the secondary variables in soft measurement modeling, and established the COD soft measurement modeling in SBR process based on the change rate of water quality parameter and the time window of water quality parameter. And then, the paper gave details of how to carry out the soft measurement function by BP neural network, and brought forward the optimization program of neural network on generalization problem and local minimum problem.
Considering the contradictory data caused by limited types of auxiliary variables in COD soft measurement modeling and local minimum problem, the paper put forward a method combining a support vector machine and neural network for COD soft measurement in SBR process.This method can have a COD pre-estimation value by using the support vector machine, then under the COD changing rule, this method can estimate the COD value by using the BP neural network and Elman neural networks respectively. Then the estimated COD value can be got after data fusion. Experiments show that the soft measurement results of this method is better than ones of single neural network measurements.
Finally, after proposing unilateral BP neural network modeling and a method jointing a support vector machine and neural networks for COD soft measurement modeling, the paper introduced the concept of integrated multiple neural network and AdaBoost, and put forward an AdaBoost multi-neural network COD soft measurement model.The experiment results show that the model haves better results on both prediction accuracy and stability
本文首先通过介绍软测量技术确定了实时准确测量COD参数的研究方向,并介绍了软测量模型的建立方法。结合软测量的建模方法,详细分析了SBR工艺中与COD参数有关的水质参数,找到了软测量建模的辅助变量,并建立了基于水质参数变化率的SBR污水COD软测量模型和基于水质参数时间窗的SBR污水COD软测量模型。之后,详细介绍了如何通过BP神经网络实现软测量的功能,并针对神经网络的泛化能力和局部最小问题提出了优化方案。
其次,针对COD软测量建模中有限种类辅助变量造成的矛盾数据问题和神经网络学习的局部最小问题,提出了一种支持向量机和神经网络联合软测量SBR污水处理中COD的方法。该方法通过引入支持向量机对COD值进行预估计,再根据COD的变化规律使用BP神经网络和Elman神经网络分别估计污水COD值,再进行数据融合得到COD的估计值。实验表明,该方法的软测量结果优于单一神经网络的软测量结果。
最后,在提出单一BP神经网络建模和采用支持向量机与神经网络联合软测量的模型基础上,介绍了多神经网络集成的概念。多神经网络集成中,详细介绍了AdaBoost算法,提出了一种AdaBoost的多神经网络COD软测量模型。该模型在预测准确性和稳定性上均有了进一步提高。
With economic and social development and the increasing human needs, water pollution is becoming a more and more serious problem. Sequencing batch reactor (SBR) is a biological wastewater treatment method, which is widely used in the wastewater treatment plants worldwide. The features of SBR make it rely heavily on the modern automation technology. Chemical oxygen demand (COD) is a comprehensive indicator of the total organic matter amount in water and an important indicator of the natural water quality classification, and also one of the main factors causing the deterioration of waterbodies.It largely contributes to the smelly and black water. COD is an important control parameter of SBR process. Traditionally, COD is measured with chemical methods, which is accurate with the shortage of poor timeliness and high cost, and the testing process is usually carried on with the production of other pollutants. As the sewage treatment technology has matured, the focal points of sewage treatment research has changed to how to improve the level of automation in the sewage treatment process, improve the water quality and enhance the operation monitoring and so on. Therefore, the fast and accurate measuring of COD in SBR process is both theoretically and practically important.
This paper set the research direction of real-time accurate COD parameter measurement by introducing soft measurement technology, and then introduces the soft measurement modeling approaches. With the modeling approaches, the paper gave a detailed analysis on the COD parameter-related water-quality parameter in the SBR process, found out the secondary variables in soft measurement modeling, and established the COD soft measurement modeling in SBR process based on the change rate of water quality parameter and the time window of water quality parameter. And then, the paper gave details of how to carry out the soft measurement function by BP neural network, and brought forward the optimization program of neural network on generalization problem and local minimum problem.
Considering the contradictory data caused by limited types of auxiliary variables in COD soft measurement modeling and local minimum problem, the paper put forward a method combining a support vector machine and neural network for COD soft measurement in SBR process.This method can have a COD pre-estimation value by using the support vector machine, then under the COD changing rule, this method can estimate the COD value by using the BP neural network and Elman neural networks respectively. Then the estimated COD value can be got after data fusion. Experiments show that the soft measurement results of this method is better than ones of single neural network measurements.
Finally, after proposing unilateral BP neural network modeling and a method jointing a support vector machine and neural networks for COD soft measurement modeling, the paper introduced the concept of integrated multiple neural network and AdaBoost, and put forward an AdaBoost multi-neural network COD soft measurement model.The experiment results show that the model haves better results on both prediction accuracy and stability
引文
[1]顾夏声,黄铭荣,王占生等,水处理工程[M],清华大学出版社,1985
[2]王浩,我国水资源合理配置的现状和未来[J],水利水电技术,2006.2,Vol37,No.2,P7-14
[3]WORLD WATER COUNTIL, World Water Vision 2025[M].Earthscan Publications Ltd 2000
[4]钱正英,张光斗,中国可持续发展水资源战略研究(综合报告及各专题报告)[M],北京：中国水利水电出版社,2001
[5]中国工程院“21世纪中国可持续发展水资源战略研究”项目组,中国可持续发展水资源战略研究综合报告[J],中国工程科学学报,2000.2,Vol.8,P1-17
[6]张利平,夏军,胡志芳,中国水资源状况与水资源安全问题分析[J],长江流域环境与资源,2009.2,Vol.18,No.2,P116-120
[7]2001年中国水环境公报[R],中华人民共和国水力部,2002
[8]刘昌明,王红瑞,浅析水资源与人口、经济和社会环境的关系,自然资源学报[J],2003,Vol.18,No.5,P635-643
[9]王德旺,21世纪污水利用对策研究[J],山西水利,2002,Vol.2,P41-42
[10]刘兴平,郝晓美,城市污水处理工艺及其发展[J],水资源保护,2003,Vol.18,No.6,P35-39
[11]李亚新,活性污泥法理论与技术[M],中国建筑工业出版社,2007
[12]Nicholas P.Cheremisinoff,Biotechnology for Waste and Wastewater Treatment[M],Noyse Publications,1996
[13]高廷耀,顾国维,水污染控制工程[M],高等教育出版社,2005
[14]李海,孙瑞征,陈振选,城市污水处理技术及工程实例[M],化学工业出版社,2002
[15]申秀英,影响活性污泥脱氮效率的因素[J],环境科学与技术,1993,Vol.2,P15-19
[16]朱怀兰,SBR生物除磷工艺的研究[J],上海环境科学,1993,Vol.2,P8-13
[17]Bjorn R,Helge E,Sequencing Batch Reactor for Nutrient Removal at Small Wastewater Treatment Plants[J],Water Science And Technology,1993, Vol.28,P233-242
[18]Heinrich D,Wastewater Treatment in A Company with Advanced Demands for Water Quality[J],Water Science And Technology,1995,Vol.32, No.7, P143-150
[19]Azedine Charef, Antoine Ghauch, Patrick Baussand and Michel Martin-Bouyer, Water Quality Monitoring Using A Smart Sensing System[J],Measurement, 2000,Vol.28,No.3,P219-224
[20]Dong-Jin Choi and Heekyung Park, A Hybrid Artificial Neural Network As A Software Sensor For Optimal Control of A Wastewater Treatment Process[J], Water Research,2001,Vol.35,No.16,P3959-3967
[21]Dae Sung Lee, Che Ok Jeon, Jong Moon Park, Kun Soo Chang, Hybrid Neural Network Modeling of A Full-scale Industrial Wastewater Treatment Process[J], Biotechnology and Bioengineering,2002, Vol.78,No.6,P670-682
[22]张文艺,钟掩英,蔡建安,活性污泥法系统人工神经网络模型[J],给水排水,2002,Vol.28,No.6,P12-15
[23]刘载文,崔莉凤,祁国强等,SBR出水BOD值的RBF软测量法[J],中国给水排水,2004,Vol.20,No.5,P17-20
[24]管秋,王万良等,基于神经网络的污水处理指标软测量研究[J],环境污染与防治,2006,Vol.28,No.2,P156-158
[25]Ming T. Tham, Gary A. Montague, A. Julian Morris, Paul A. Lant,Soft Sensors for Process Estimation and Inferential Control.Process Control[J], 1991,Vol.1,No.1,P3-14
[26]俞金寿,刘爱伦,张克进等,软测量技术及其在石油化工中的应用[M],化学工业出版社,2000
[27]T.J.McAvoy,Contemplative Stance for Chemical Process[J],Automatic, 1992,28(2):441-442
[28]汪永生,软测量理论方法、软件包及其工业应用研究[D],上海交通大学,2000,P12-19
[29]卿晓霞,余建平,软测量技术及其在污水处理系统中的应用[J],工业水处理,2005,Vol.25,No.3,P13-16
[30]管秋,基于人工神经网络的污水水质指标软测量方法的研究[D],浙江工业大学,2004
[31]穆秀春,基于统计回归的污水处理出水水质的软测量研究[D],浙江工业大学,2005
[32]曾薇,王淑莹,高景峰等,SBR法处理啤酒废水COD与DO的相关关系[J], 给水排水,2000,Vol.26,No.5,P28-30
[33]彭永臻,张立东,王淑莹等,SB、R法曝气量的模糊控制[J],哈尔滨建筑大学学报,2002,Vol.35,No.1,P53-57
[34]彭永臻,邵剑英,周利等,利用ORP作为SBR法反应时间的计算机控制参数[J],中国给水排水,1997,Vol.13,No.6,P6-9
[35]高景峰,彭永臻,王淑莹,SBR法反硝化模糊控制参数pH和ORP的变化规律[J],环境科学,2002,Vol.23,No.1,P
[36]杨岸明,王淑莹,杨庆等,以pH和ORP作为脉冲SBR工艺的实时控制参数[J],环境污染治理技术与设备,2006,Vol.7,No.12,P32-35
[37]丁峰,彭永臻,徐学清等,PH值对SBR法处理工业废水的影响[J],给水排水,2001,Vol.27,No.2,P55-57
[38]高景峰,彭永臻,王淑莹,SBR法去除有机物硝化和反硝化过程中PH变化规律[J],环境工程,2001,Vol.19,No.5,P21-24
[39]李淼,神经网络结构自适应设计及其在COD软测量中的应用研究[D],北京工业大学,2009
[40]袁亚湘,孙文瑜,最优化理论与方法[M],科学出版社,1997
[41]Marquardt D W,An Algorithm For Least-squares Estimation of Nonlinear Parameters[J],Applied Mathematics,1963,Vol.11,No.2,P431-441
[42]刘载文,崔莉凤,祁国强等,SBR出水BOD值的RBF软测量法[J],中国给水排水,2004,Vol.20,No.5,P17-20
[43]Boser B E,Guyon I M,Vapnik V N,A Training Algorithm for Optimal Margin Classifiers[J],The 5th Annual ACM Workshop on COLT,Pittsburgh: ACM Press,1992,P144-152
[44]Cortes C,Vapnik V N,Support Vector Networks[J],Machine Learning, 1995,Vol.20,No.3,P273-297
[45]Drucker H,Burges C J C,Kaufman L,Support Vector Regression Machines[M],Advances in Neural Information Processing Systems Cambridge:MIT Press,1997,P155-161
[46]Vapnik V N,Golowich S.Smola A. Support Vector Method for Function Approximation,Regression Estimation and Signal Processing[M] Advances in Neural Information Processing Systems,Cambridge:MIT Press, 1997,P281-287
[47]许建华,张学工,李衍达,支持向量机的新发展[J],控制与决策,2004,Vol.19,No.5,P481-485
[48]Christopher J. C.Burges, A Tutorial on Support Vector Machines for Pattern Recognition[J],Data Mining and Knowledge Discovery,1998,Vol.2,No.2, P121-167
[49]Pham D T, Liu X, A Comparison of Three Types of Neural Networks For System Identification[J],MACS International Symposium on Signal Processing, Robotics and Neural Networks,1994, P568-571
[50]Wolpert D H,Stacked Generalization[J],Neural Networks,1992,Vol.5, No.2,P241-259
[51]Cho S B,Kim J H,Combining Multiple Neural Networks by Fuzzy Integral for Recognition[J],IEEE Trans on Syst,Man and Cybern,1995,Vol.25, No.2,P380-384
[52]Xiong Zhihua,Wang Xiong,Xu Yongmao,Nonlinear Software Sensor Modeling Using Multiple Neural Network[J],Control and Decision,2000, Vol.15,No.2,P173-176
[53]路刚,陈永,范永欣,胡城,Boosting算法研究[J],电脑知识与技术,2008,Vol.4,No.9,P2698-2699
[54]Yoav Freund and Robert E.Schapire,A Decision-theoretic Generalization of On-line Learning and An Application to Boosting[J],Computer and System Sciences,1997,no.55,P119-139
[55]Drucker H, Improving Regressors Using Boosting Techniques[J],The Fourteenth International Conference on Machine Learning, Morgan Kaufmann, USA,1997
[2]王浩,我国水资源合理配置的现状和未来[J],水利水电技术,2006.2,Vol37,No.2,P7-14
[3]WORLD WATER COUNTIL, World Water Vision 2025[M].Earthscan Publications Ltd 2000
[4]钱正英,张光斗,中国可持续发展水资源战略研究(综合报告及各专题报告)[M],北京：中国水利水电出版社,2001
[5]中国工程院“21世纪中国可持续发展水资源战略研究”项目组,中国可持续发展水资源战略研究综合报告[J],中国工程科学学报,2000.2,Vol.8,P1-17
[6]张利平,夏军,胡志芳,中国水资源状况与水资源安全问题分析[J],长江流域环境与资源,2009.2,Vol.18,No.2,P116-120
[7]2001年中国水环境公报[R],中华人民共和国水力部,2002
[8]刘昌明,王红瑞,浅析水资源与人口、经济和社会环境的关系,自然资源学报[J],2003,Vol.18,No.5,P635-643
[9]王德旺,21世纪污水利用对策研究[J],山西水利,2002,Vol.2,P41-42
[10]刘兴平,郝晓美,城市污水处理工艺及其发展[J],水资源保护,2003,Vol.18,No.6,P35-39
[11]李亚新,活性污泥法理论与技术[M],中国建筑工业出版社,2007
[12]Nicholas P.Cheremisinoff,Biotechnology for Waste and Wastewater Treatment[M],Noyse Publications,1996
[13]高廷耀,顾国维,水污染控制工程[M],高等教育出版社,2005
[14]李海,孙瑞征,陈振选,城市污水处理技术及工程实例[M],化学工业出版社,2002
[15]申秀英,影响活性污泥脱氮效率的因素[J],环境科学与技术,1993,Vol.2,P15-19
[16]朱怀兰,SBR生物除磷工艺的研究[J],上海环境科学,1993,Vol.2,P8-13
[17]Bjorn R,Helge E,Sequencing Batch Reactor for Nutrient Removal at Small Wastewater Treatment Plants[J],Water Science And Technology,1993, Vol.28,P233-242
[18]Heinrich D,Wastewater Treatment in A Company with Advanced Demands for Water Quality[J],Water Science And Technology,1995,Vol.32, No.7, P143-150
[19]Azedine Charef, Antoine Ghauch, Patrick Baussand and Michel Martin-Bouyer, Water Quality Monitoring Using A Smart Sensing System[J],Measurement, 2000,Vol.28,No.3,P219-224
[20]Dong-Jin Choi and Heekyung Park, A Hybrid Artificial Neural Network As A Software Sensor For Optimal Control of A Wastewater Treatment Process[J], Water Research,2001,Vol.35,No.16,P3959-3967
[21]Dae Sung Lee, Che Ok Jeon, Jong Moon Park, Kun Soo Chang, Hybrid Neural Network Modeling of A Full-scale Industrial Wastewater Treatment Process[J], Biotechnology and Bioengineering,2002, Vol.78,No.6,P670-682
[22]张文艺,钟掩英,蔡建安,活性污泥法系统人工神经网络模型[J],给水排水,2002,Vol.28,No.6,P12-15
[23]刘载文,崔莉凤,祁国强等,SBR出水BOD值的RBF软测量法[J],中国给水排水,2004,Vol.20,No.5,P17-20
[24]管秋,王万良等,基于神经网络的污水处理指标软测量研究[J],环境污染与防治,2006,Vol.28,No.2,P156-158
[25]Ming T. Tham, Gary A. Montague, A. Julian Morris, Paul A. Lant,Soft Sensors for Process Estimation and Inferential Control.Process Control[J], 1991,Vol.1,No.1,P3-14
[26]俞金寿,刘爱伦,张克进等,软测量技术及其在石油化工中的应用[M],化学工业出版社,2000
[27]T.J.McAvoy,Contemplative Stance for Chemical Process[J],Automatic, 1992,28(2):441-442
[28]汪永生,软测量理论方法、软件包及其工业应用研究[D],上海交通大学,2000,P12-19
[29]卿晓霞,余建平,软测量技术及其在污水处理系统中的应用[J],工业水处理,2005,Vol.25,No.3,P13-16
[30]管秋,基于人工神经网络的污水水质指标软测量方法的研究[D],浙江工业大学,2004
[31]穆秀春,基于统计回归的污水处理出水水质的软测量研究[D],浙江工业大学,2005
[32]曾薇,王淑莹,高景峰等,SBR法处理啤酒废水COD与DO的相关关系[J], 给水排水,2000,Vol.26,No.5,P28-30
[33]彭永臻,张立东,王淑莹等,SB、R法曝气量的模糊控制[J],哈尔滨建筑大学学报,2002,Vol.35,No.1,P53-57
[34]彭永臻,邵剑英,周利等,利用ORP作为SBR法反应时间的计算机控制参数[J],中国给水排水,1997,Vol.13,No.6,P6-9
[35]高景峰,彭永臻,王淑莹,SBR法反硝化模糊控制参数pH和ORP的变化规律[J],环境科学,2002,Vol.23,No.1,P
[36]杨岸明,王淑莹,杨庆等,以pH和ORP作为脉冲SBR工艺的实时控制参数[J],环境污染治理技术与设备,2006,Vol.7,No.12,P32-35
[37]丁峰,彭永臻,徐学清等,PH值对SBR法处理工业废水的影响[J],给水排水,2001,Vol.27,No.2,P55-57
[38]高景峰,彭永臻,王淑莹,SBR法去除有机物硝化和反硝化过程中PH变化规律[J],环境工程,2001,Vol.19,No.5,P21-24
[39]李淼,神经网络结构自适应设计及其在COD软测量中的应用研究[D],北京工业大学,2009
[40]袁亚湘,孙文瑜,最优化理论与方法[M],科学出版社,1997
[41]Marquardt D W,An Algorithm For Least-squares Estimation of Nonlinear Parameters[J],Applied Mathematics,1963,Vol.11,No.2,P431-441
[42]刘载文,崔莉凤,祁国强等,SBR出水BOD值的RBF软测量法[J],中国给水排水,2004,Vol.20,No.5,P17-20
[43]Boser B E,Guyon I M,Vapnik V N,A Training Algorithm for Optimal Margin Classifiers[J],The 5th Annual ACM Workshop on COLT,Pittsburgh: ACM Press,1992,P144-152
[44]Cortes C,Vapnik V N,Support Vector Networks[J],Machine Learning, 1995,Vol.20,No.3,P273-297
[45]Drucker H,Burges C J C,Kaufman L,Support Vector Regression Machines[M],Advances in Neural Information Processing Systems Cambridge:MIT Press,1997,P155-161
[46]Vapnik V N,Golowich S.Smola A. Support Vector Method for Function Approximation,Regression Estimation and Signal Processing[M] Advances in Neural Information Processing Systems,Cambridge:MIT Press, 1997,P281-287
[47]许建华,张学工,李衍达,支持向量机的新发展[J],控制与决策,2004,Vol.19,No.5,P481-485
[48]Christopher J. C.Burges, A Tutorial on Support Vector Machines for Pattern Recognition[J],Data Mining and Knowledge Discovery,1998,Vol.2,No.2, P121-167
[49]Pham D T, Liu X, A Comparison of Three Types of Neural Networks For System Identification[J],MACS International Symposium on Signal Processing, Robotics and Neural Networks,1994, P568-571
[50]Wolpert D H,Stacked Generalization[J],Neural Networks,1992,Vol.5, No.2,P241-259
[51]Cho S B,Kim J H,Combining Multiple Neural Networks by Fuzzy Integral for Recognition[J],IEEE Trans on Syst,Man and Cybern,1995,Vol.25, No.2,P380-384
[52]Xiong Zhihua,Wang Xiong,Xu Yongmao,Nonlinear Software Sensor Modeling Using Multiple Neural Network[J],Control and Decision,2000, Vol.15,No.2,P173-176
[53]路刚,陈永,范永欣,胡城,Boosting算法研究[J],电脑知识与技术,2008,Vol.4,No.9,P2698-2699
[54]Yoav Freund and Robert E.Schapire,A Decision-theoretic Generalization of On-line Learning and An Application to Boosting[J],Computer and System Sciences,1997,no.55,P119-139
[55]Drucker H, Improving Regressors Using Boosting Techniques[J],The Fourteenth International Conference on Machine Learning, Morgan Kaufmann, USA,1997