基于UKF的未建模过程状态估计及其在腈纶聚合中的应用

发布时间：2019-01-05 05:16

【摘要】：对于大多数复杂非线性过程,由于很多需要控制的中间变量无法直接测量,这直接影响到过程监控系统的实现。非线性滤波技术为复杂系统状态估计提供了有力的基础,因此得到了广泛的关注与研究。目前,不敏卡尔曼滤波Unscented Kalman filter, UKF)算法已经成功地解决了很多实际的非线性系统状态估计问题。但由于很多不确定因素的存在,使得精确而稳定的状态估计仍具有较大的挑战。论文以腈纶聚合过程作为研究背景,对UKF算法进行深入地研究。具体包括以下内容：首先,简单介绍了二步法腈纶水相聚合工艺过程。对腈纶聚合过程中典型非线性环节(连续搅拌反应釜,pH中和过程)的动力学模型进行分析,总结出复杂非线性系统的一般特征,为之后滤波算法的改进提供了方向。接着,介绍了解决非线性滤波的典型算法。对UKF算法的具体步骤进行分析研究,总结出UKF算法在滤波过程中面临的两大挑战：系统模型未知和噪声统计特性未知。针对UKF要求系统模型精确已知,而实际的非线性系统往往因其复杂性及不确定性等原因导致系统模型难以建立的问题,将UKF和神经网络结合(NN-UKF)解决了一类过程模型未知且输出为状态线性组合的非线性过程状态估计问题。通过实例的仿真,验证了NN-UKF算法具有较好的估计效果。将该算法应用于腈纶聚合的连续搅拌反应釜过程,解决了连续搅拌反应釜在非线性模型未知时浓度与温度的估计问题。针对非线性系统噪声统计特性难以获得,而UKF算法对噪声信息不准的鲁棒性较差导致滤波精度急剧下降,甚至滤波发散的问题,提出了一种基于Cauchy鲁棒函数的UKF改进算法(CR-UKF).以UKF的测量先验值与其实际值的残差作为基准,采用联合权函数对滤波过程的噪声估计值进行实时修正,降低噪声不准的估计值的权重,提高了UKF算法的精度。两个实例的仿真结果表明,CR-UKF算法对提高噪声估计不准时UKF的状态估计精度非常有效。将CR-UKF算法应用于腈纶聚合过程的pH中和过程,一方面提高了中和反应pH值的监控精度,另一方面提高了离子浓度的估计精度。最后,对本文所提出的改进算法进行总结,并对之后的工作做了进一步的展望。
[Abstract]:For most complex nonlinear processes, many intermediate variables that need to be controlled can not be measured directly, which directly affects the realization of process monitoring system. Nonlinear filtering technology provides a powerful basis for state estimation of complex systems, so it has been widely studied. At present, the nonsensitive Kalman filter (Unscented Kalman filter, UKF) algorithm has successfully solved many practical nonlinear system state estimation problems. However, due to the existence of many uncertain factors, accurate and stable state estimation still poses a great challenge. In this paper, the polymerization process of acrylic fiber is used as the research background, and the UKF algorithm is deeply studied. The main contents are as follows: firstly, the water phase polymerization process of two-step acrylic fiber is briefly introduced. The dynamic models of typical nonlinear links (continuous stirred tank reactor, pH neutralization process) in the polymerization process of acrylic fiber were analyzed, and the general characteristics of the complex nonlinear system were summarized, which provided a direction for the improvement of filtering algorithm. Then, the typical algorithms to solve nonlinear filtering are introduced. The concrete steps of UKF algorithm are analyzed and studied, and two major challenges in filtering process of UKF algorithm are summarized: unknown system model and unknown statistical characteristics of noise. In view of UKF, the system model is required to be accurately known, but the actual nonlinear system is often difficult to establish because of its complexity and uncertainty. Combining UKF with neural network (NN-UKF), the problem of state estimation for a class of nonlinear processes with unknown process model and linear state combination is solved. The simulation results show that the NN-UKF algorithm has a good estimation effect. The algorithm is applied to the process of acrylonitrile polymerization and the estimation of concentration and temperature of continuous stirred tank reactor is solved when the nonlinear model is unknown. It is difficult to obtain the statistical characteristics of noise in nonlinear systems, but the poor robustness of UKF algorithm to noise information leads to a sharp decline in filtering accuracy and even filtering divergence. An improved UKF algorithm (CR-UKF) based on Cauchy robust function is proposed. Based on the residual error between the measured prior value and the actual value of UKF, the joint weight function is used to modify the noise estimation value in the filtering process in real time, which reduces the weight of the noise estimation value and improves the accuracy of the UKF algorithm. The simulation results of two examples show that the CR-UKF algorithm is very effective to improve the accuracy of the state estimation of the noise estimation unpunctual UKF. The CR-UKF algorithm is applied to the pH neutralization process of acrylic polymerization. On the one hand, the monitoring accuracy of the neutralization reaction pH value is improved; on the other hand, the accuracy of ion concentration estimation is improved. Finally, the improved algorithm proposed in this paper is summarized, and the future work is prospected.
【学位授予单位】：华东理工大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TQ342.31;TN713

【参考文献】