多核SVR在污水处理出水指标建模中的应用研究

发布时间：2018-06-25 12:52

本文选题：污水处理 + 支持向量回归机　；参考：《湖南工业大学》2015年硕士论文

【摘要】：污水处理厂为保护水资源和防治水污染做出了重大贡献,但由于污水处理系统是一个高度非线性、强耦合、多变量和大滞后的复杂系统,其机理研究还不够成熟,关键指标参数不能实现实时测量,且污水处理的效果依赖于污水出水水质的好坏。因此建立一个高效、合理的污水出水指标模型,将预测结果作为指导污水厂运行的依据,动态调整污水处理过程中各工序运行状态,具有一定的现实意义和应用价值。本文以活性污泥法污水的出水水质为研究对象,对于污水处理过程水质参数数据分布较复杂,采用单一核函数支持向量回归机模型建模精度不理想的问题,在前人研究成果的基础上,把多核和智能算法相结合,建立了出水水质参数的多核支持向量回归机(MK-SVR)模型,研究内容有以下几方面:首先了解污水处理工艺流程及方法,对影响污水处理过程的水质参数及相关排放标准进行了分析,利用主成分分析(PCA)法对污水处理过程的影响因素进行降维处理,提取新主元作为支持向量回归机的输入,建立了出水COD、BOD、SS和TN的MK-SVR模型。其次,由于模型自身参数问题的影响,在分析智能算法中粒子群算法(PSO)具有编程方便结构简单易于实现、搜索速度快、收敛能力强等特点,提出了利用PSO算法对MK-SVR模型进行参数寻优,并针对基本PSO算法的不足对其进行了改进。最后为所提出的模型更具说服力,对几种不同模型——SVR单核模型、SVR多核模型、基于PCA分析的SVR多核模型与单核模型、基于PCA分析的PSO+MK-SVR模型及基于PCA分析的改进PSO+MK-SVR模型的预测效果进行了对比,从平均相对误差、均方误差及相关系数等几个性能指标进行了分析,结果表明,基于PCA分析的改进PSO+MK-SVR模型预估效果最好,泛化性能最强,为污水厂的实时高效运行提供强劲的理论支撑。
[Abstract]:Sewage treatment plants have made great contributions to the protection of water resources and the prevention of water pollution. However, because the sewage treatment system is a highly nonlinear, strongly coupled, multivariable and lag complex system, the study of its mechanism is not mature enough. The key parameters can not be measured in real time, and the effect of sewage treatment depends on the quality of effluent. Therefore, it is of practical significance and practical value to establish an efficient and reasonable effluent index model, to take the prediction results as the basis for guiding the operation of the wastewater treatment plant, and to dynamically adjust the operation state of each process in the process of sewage treatment. In this paper, the effluent quality of activated sludge wastewater is taken as the research object. For the complex distribution of water quality parameter data in the process of sewage treatment, the modeling accuracy of single kernel function support vector regression model is not ideal. On the basis of previous research results, a multi-core support vector regression model (MK-SVR) for effluent quality parameters is established by combining multi-core and intelligent algorithms. The research contents are as follows: firstly, the process and methods of wastewater treatment are understood. The influence of water quality parameters and relevant discharge standards on wastewater treatment process is analyzed. Principal component analysis (PCA) method is used to reduce the dimension of wastewater treatment process, and a new principal component is extracted as the input of support vector regression machine. The MK-SVR model of effluent CODDSS and TN was established. Secondly, due to the influence of the model's own parameter problem, Particle Swarm Optimization (PSO) has the advantages of simple and easy programming, fast searching speed and strong convergence ability in the analysis of intelligent algorithm. PSO algorithm is used to optimize the parameters of MK-SVR model, and the basic PSO algorithm is improved. Finally, the proposed model is more persuasive. For several different models, SVR multi-core model, SVR multi-core model based on PCA analysis and single core model, The prediction results of PSO MK-SVR model based on PCA analysis and improved PSO MK-SVR model based on PCA analysis are compared. Several performance indexes, such as average relative error, mean square error and correlation coefficient, are analyzed. The improved PSO MK-SVR model based on PCA has the best prediction effect and the strongest generalization performance, which provides a strong theoretical support for the real-time and efficient operation of the wastewater treatment plant.
【学位授予单位】：湖南工业大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：X703;TP181

【参考文献】