基于模拟退火算法的支持向量机在MBR膜污染中的应用研究

发布时间：2019-07-02 13:27

【摘要】：膜生物反应器(MBR)是污水处理工艺中一种重要的方式,具有处理效率高、出水水质好、容易实现自动控制等优点,应用范围不断扩大,规模也在逐年增加,越来越受到世界上各个国家的重视。但随之而来的是膜污染问题正在逐步成为阻碍MBR快速发展的一个主要因素,因为膜污染会直接导致膜通量的减小,因此如何有效的降低MBR膜污染开始成为一个热点研究问题。本文在研究了之前MBR领域中的各种模型的基础上,针对传统的神经网络模型在MBR膜污染研究中存在容易陷入局部极小值、参数难以确定的不足,阅读并参考了大量的文献,提出使用模拟退火算法和支持向量机建立模型,即基于SA-SVM支持向量机的MBR膜污染预测模型。首先使用模拟退火算法对支持向量机的三个重要参数惩罚因子、不敏感系数和核参数进行全域优化搜索,然后将得到的最优参数作为支持向量机的初始参数并建立MBR膜污染预测模型,最后使用主成分分析法对影响膜污染的因素进行分析,选取主要影响因素作为模型的输入,膜通量的大小作为输出,进行预测。实验研究表明,基于SA-SVM支持向量机的MBR膜污染预测模型在对膜通量进行预测时,有较好的拟合效果和较高的预测精确度,在稳定性和泛化能力方面也比之前的神经网络模型也有了一定的提高。在使用模拟退火算法对MBR膜污染预测模型进行初始参数寻优的过程中,我们也发现了一些问题,SA算法存在收敛速度慢、初始参数设置比较敏感等缺点。于是我们引入一种同时结合了 SA和GA算法的混合优化算法对模型的参数进行优化,该算法既保留了 GA算法全局搜索能力强的特点,又拥有SA算法局部寻优的优势。实验研究表明,通过与SA-SVM模型相比较,ASAGA-SVM模型具有较高的拟合效果,平均相对误差为0.0263,由此我们可以得出,在面对小样本的膜通量数据时,基于ASAGA-SVM的MBR膜污染预测模型比SA-SVM模型具有更好地预测精度。
[Abstract]:Membrane Bioreactor (MBR) is an important way in wastewater treatment process, which has the advantages of high treatment efficiency, good effluent quality and easy automatic control. The scope of application is expanding and the scale is increasing year by year. More and more countries in the world pay attention to it. However, the problem of membrane fouling is gradually becoming a major factor hindering the rapid development of MBR, because membrane fouling will directly lead to the reduction of membrane flux, so how to effectively reduce MBR membrane fouling began to become a hot research issue. Based on the study of various models in the field of MBR, aiming at the shortcomings of the traditional neural network model in the study of MBR membrane fouling, which is easy to fall into local minimum and difficult to determine the parameters, this paper reads and refers to a large number of literatures, and proposes to use simulated annealing algorithm and support vector machine to establish the model, that is, the prediction model of MBR membrane fouling based on SA-SVM support vector machine. Firstly, the simulated annealing algorithm is used to search the three important parameter penalty factors, insensitive coefficients and kernel parameters of support vector machine, and then the optimal parameters are taken as the initial parameters of support vector machine and the prediction model of MBR membrane fouling is established. Finally, the factors affecting membrane fouling are analyzed by principal component analysis, and the main influencing factors are selected as the input of the model and the size of membrane flux as the output. Make a prediction. The experimental results show that the MBR membrane fouling prediction model based on SA-SVM support vector machine has good fitting effect and high prediction accuracy, and also improves the stability and generalization ability of the previous neural network model. In the process of optimizing the initial parameters of MBR film fouling prediction model by simulated annealing algorithm, we also find some problems. SA algorithm has some shortcomings, such as slow convergence speed, sensitive initial parameter setting and so on. Therefore, we introduce a hybrid optimization algorithm which combines SA and GA algorithm to optimize the parameters of the model. The algorithm not only preserves the strong global search ability of GA algorithm, but also has the advantage of local optimization of SA algorithm. The experimental results show that compared with the SA-SVM model, the ASAGA-SVM model has a higher fitting effect, and the average relative error is 0.0263. It can be concluded that the MBR membrane fouling prediction model based on ASAGA-SVM has better prediction accuracy than the SA-SVM model when facing the membrane flux data of small samples.
【学位授予单位】：天津工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：X703;TP18

【参考文献】