一种基于加权性能指标的多模型辨识算法及其在热工过程中的应用.pdfVIP

Vol.31, No.3 ACTA AUTOMATICA SINICA May, 2005 A Multi-model Identification Algorithm Based on Weighted Cost Function and Application in Thermal Process1) XUE Zhen-Kuang LI Shao-Yuan (Institute of Automation, Shanghai Jiaotong University, Shanghai 200030) (E-mail: syli@sjtu.edu.cn) Abstract Most existing algorithms for identifying multi-model system are based on minimizing the square of bias between global outputs of the actual system and the identified model, but the resultant model lacks of robustness. In order to solve this problem, this paper considers some other algorithms in which local models are identified independently and presents a multi-model identifica- tion algorithm based on weighted cost function, which uses the idea of local weighted regression and local approximation while keeps the model structure of global identification algorithm. The result of application to a 300MW unit boiler superheater illustrates that the multi-model generated by the proposed algorithm has better trade-off between global fitting and local interpretation. Key words Multi-model, identification, nonlinear, local model network, thermal process 1 Introduction Local model network (LMN) is a typical multi-model used for modeling nonlinear systems, in which global outputs are described as the sum of local ones multiplied by their corresponding basis functions[1]. Usually modeling LMN can be divided into two tasks: first to find optimal number and position of basis functions using clustering algorithm, then to identify the parameters of local models. The clustering a