PTA氧化过程中4-CBA含量的软测量建模研究

发布时间：2018-10-18 18:03

【摘要】：PTA氧化过程是石油化工生产的重要化学反应过程,其反应产物是生产聚酯产品的重要化工原料,4-CBA是氧化过程中的主要副产物。PTA氧化过程反应条件苛刻,反应机理和反应过程复杂,采用软测量技术来对反应过程进行实时预测。软测量技术采用一些可测变量来对不可测变量预测,本文将对PTA氧化过程进行研究,以4-CBA含量为研究对象,通过AdaBoost算法建立软测量模型。论文将通过以下几个方面展开研究。AdaBoost算法是一种组合算法,将一组训练差异较大的弱学习器通过组合成强学习器,本文选择BP神经网络和支持向量机作为弱学习器。为解决AdaBoost算法中训练弱化的问题,采用双阈值的方法更新样本权重,减小误差较大的样本对弱学习器的影响,采用轮盘赌的方法对样本进行重采样,并通过非线性函数拟合来证明改进算法的可行性。针对PTA氧化过程中4-CBA含量软测量模型,分别以BP神经网络和支持向量机做为弱学习器,以改进的AdaBoost算法做为强学习器建立软测量模型,通过MATLAB训练仿真对4-CBA含量进行预测。并且分别与单一弱学习器建模和未改进AdaBoost算法建模进行分析比较,结论证明改进的AdaBoost算法建立的软测量模型在这几种模型中预测精度更高。
[Abstract]:The PTA oxidation process is an important chemical reaction process in petrochemical production. The reaction product is an important chemical raw material for the production of polyester products. 4-CBA is the main by-product in the oxidation process. The reaction conditions of the PTA oxidation process are harsh. The reaction mechanism and reaction process are complex, and the soft sensing technique is used to predict the reaction process in real time. Soft sensing technology uses some measurable variables to predict unmeasurable variables. In this paper, the oxidation process of PTA is studied. Taking the content of 4-CBA as the research object, the soft sensing model is established by AdaBoost algorithm. The AdaBoost algorithm is a combination algorithm, which combines a group of weak learning devices with different training into strong learning devices. In this paper, BP neural network and support vector machine are selected as weak learning devices. In order to solve the problem of training weakening in AdaBoost algorithm, the method of double threshold is used to update the weight of samples, to reduce the influence of the samples with large errors on the weak learner, and the method of roulette is used to resample the samples. The feasibility of the improved algorithm is proved by nonlinear function fitting. Aiming at the soft sensing model of 4-CBA content in the process of PTA oxidation, BP neural network and support vector machine are used as weak learning devices, and the improved AdaBoost algorithm is used as strong learning device to establish soft sensor model. The 4-CBA content is predicted by MATLAB training simulation. And compared with the single weak learner model and the unimproved AdaBoost algorithm, it is proved that the improved soft sensor model based on the improved AdaBoost algorithm is more accurate in these models.
【学位授予单位】：南京邮电大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP18;O633.14

【参考文献】