基于最优边界划分的非线性系统多模型辨识方法
发布时间：2018-07-27 14:28
【摘要】：随着工业过程的日益复杂,控制系统往往具有多变量、非线性、工况范围广等特点。为提升复杂动态系统的控制性能,基于多模型的非线性系统辨识与控制方法愈发受到关注。本文面向非线性动态系统,分别提出了在单维调度变量及多维调度变量情况下基于最优边界划分的辨识方法,通过调节子模型边界参数优化多模型输出误差,突出了样本点与子模型之间的对应关系,并从模型精度和控制性能两方面说明该辨识模型的优点。本文的主要贡献如下:1)对于多模型的调度变量为一维的情况,提出一种使输出误差最小的最优边界划分辨识方法。该方法使用基于局部模型参数向量的聚类方法初始化划分数据集,充分考虑了多模型中子模型边界对模型精度带来的影响,在精准划分数据的基础上辨识子模型参数。通过与一般的聚类方法和基于工作点线性化的模型相对比,在同样的调度变量下,该方法有效地提升了模型精度。2)对于调度变量为多维的情况,基于Softmax分类方法,提出一种使输出误差最小的最优边界划分辨识方法。该方法解决了多维调度变量下子模型边界不易于初始化表达和子模型区域不完全划分的两个难题,使基于最优边界划分的辨识方法在调度变量为多维的情况下得到推广。3)基于辨识得到的PWA模型,设计了基于MLD框架的多模型预测控制器。通过对比不同辨识模型在预测控制器下的控制效果,进一步验证本文所提出辨识方法在控制性能上的优势。
[Abstract]:With the increasing complexity of industrial processes, control systems are often multivariable, nonlinear, and have a wide range of operating conditions. In order to improve the control performance of complex dynamic systems, the identification and control methods of nonlinear systems based on multiple models have attracted more and more attention. In this paper, for nonlinear dynamic systems, an identification method based on optimal boundary partition is proposed in the case of single dimensional scheduling variables and multidimensional scheduling variables, and the output errors of multiple models are optimized by adjusting the boundary parameters of submodels. The corresponding relationship between the sample points and the sub-model is highlighted, and the advantages of the identification model are illustrated in terms of model precision and control performance. The main contribution of this paper is as follows: (1) in the case that the scheduling variable of multiple models is one-dimensional, an optimal boundary partition identification method is proposed to minimize the output error. In this method, the partitioned data set is initialized by clustering method based on local model parameter vector. The influence of the boundary of multi-model neutron model on model accuracy is fully considered, and the submodel parameters are identified on the basis of accurate partitioning data. Compared with the general clustering method and the model based on working-point linearization, under the same scheduling variables, this method effectively improves the precision of the model .2) for the multi-dimensional scheduling variables, based on the Softmax classification method, An optimal boundary partition identification method with minimum output error is proposed. This method solves the two difficult problems of multi-dimensional scheduling variable model boundary is not easy to initialize the representation and sub-model area is not completely partitioned. The identification method based on optimal boundary partition is extended to 3. 3) based on the identified PWA model, a multi model predictive controller based on MLD framework is designed. By comparing the control effect of different identification models under predictive controller, the superiority of the proposed identification method in control performance is further verified.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP273
本文编号：2148119
[Abstract]:With the increasing complexity of industrial processes, control systems are often multivariable, nonlinear, and have a wide range of operating conditions. In order to improve the control performance of complex dynamic systems, the identification and control methods of nonlinear systems based on multiple models have attracted more and more attention. In this paper, for nonlinear dynamic systems, an identification method based on optimal boundary partition is proposed in the case of single dimensional scheduling variables and multidimensional scheduling variables, and the output errors of multiple models are optimized by adjusting the boundary parameters of submodels. The corresponding relationship between the sample points and the sub-model is highlighted, and the advantages of the identification model are illustrated in terms of model precision and control performance. The main contribution of this paper is as follows: (1) in the case that the scheduling variable of multiple models is one-dimensional, an optimal boundary partition identification method is proposed to minimize the output error. In this method, the partitioned data set is initialized by clustering method based on local model parameter vector. The influence of the boundary of multi-model neutron model on model accuracy is fully considered, and the submodel parameters are identified on the basis of accurate partitioning data. Compared with the general clustering method and the model based on working-point linearization, under the same scheduling variables, this method effectively improves the precision of the model .2) for the multi-dimensional scheduling variables, based on the Softmax classification method, An optimal boundary partition identification method with minimum output error is proposed. This method solves the two difficult problems of multi-dimensional scheduling variable model boundary is not easy to initialize the representation and sub-model area is not completely partitioned. The identification method based on optimal boundary partition is extended to 3. 3) based on the identified PWA model, a multi model predictive controller based on MLD framework is designed. By comparing the control effect of different identification models under predictive controller, the superiority of the proposed identification method in control performance is further verified.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP273
【参考文献】
相关期刊论文 前3条
1 史运涛;杨震安;李志军;孙德辉;刘大千;;基于数据驱动的混杂系统建模与优化控制研究[J];系统仿真学报;2013年11期
2 陈贵敏;贾建援;韩琪;;粒子群优化算法的惯性权值递减策略研究[J];西安交通大学学报;2006年01期
3 李少远;薛振框;王昕;李柠;;非线性系统多模型控制方法的研究[J];系统仿真技术;2005年01期
相关博士学位论文 前1条
1 杜静静;基于非线性度量和MLD-MPC的多模型方法研究[D];浙江大学;2010年
相关硕士学位论文 前2条
1 王林涛;混杂系统优化控制数值求解及其在非线性系统控制中的应用[D];浙江大学;2011年
2 周芦文;基于聚类分析方法的非线性系统多模型预测控制算法研究[D];浙江大学;2006年
，本文编号：2148119
本文链接：https://www.wllwen.com/kejilunwen/zidonghuakongzhilunwen/2148119.html
