Abstract: This paper has induced the algorithm of Multi-dimension Cyclic Subspace Regression (MCSR), an approach as linear regression which can deal with multi-dimension dependent variables. By combining the Radial Basis Functions (RBF) with MCSR, it provides a better approach of modeling non-linear systems in chemical engineering process in which its dependent variables are multi-dimensional. It adopts the frame of the RBFN, which can express a complicated non-linear relationship, and combines with the MCSR to avoid a variety of troubles of the design and training of the network. Thus, it can search the optimal model in an extensive solution space and the model has a briefly analytic form. Its good performance is demonstrated by an example of modeling the equipment of isomerization of xylene to paraxylene as compared with the RBF-PLSR.

摘要： 推导了多因变量循环子空间回归 (MCSR)算法 ,并将MCSR集成于径向基网络 (RBFN)的输出端 ,由此提出了RBF—MCSR方法 ,它能表达复杂的非线性关系 ,而且在更为宽广的解空间内选取具有简明的解析形式的最优模型。将该法应用于二甲苯异构化装置 ,效果良好 ,与现有的RBF -PLSR比较 ,显示出其非线性建模的优势