A key to the analysis and design of a dynamic system is to establish a suitable mathematical model of the system. This paper investigates the parameter optimization problem of a class of radial basis function-based multivariate hybrid models. Taking into account the high dimensions of the models and different forms of the parameters, the original identification model is separated into several regressive sub-identification models according to the characteristics of model outputs. Some auxiliary models are constructed to solve the unmeasurable noise terms in the information matrices. For the purpose of eliminating the redundant computation and to deal with the associate terms caused by the model decomposition, inspired by the coupling concept, a partially-coupled nonlinear parameter optimization algorithm is proposed for the multivariate hybrid models. Through the computational efficiency analysis and numerical simulation verification, it is shown that the proposed algorithm has low computational complexity and high parameter estimation accuracy.

动态系统分析和设计的关键是建立合适的系统数学模型。本文研究了一类基于径向基函数的多元混合模型的参数优化问题。考虑到模型维数高，参数形式不同，根据模型输出的特点，将原始识别模型拆分为若干个回归子识别模型。构建了一些辅助模型来解决信息矩阵中不可测量的噪声项。为了消除冗余计算和处理模型分解引起的关联项，受耦合概念启发，针对多元混合模型提出了一种部分耦合的非线性参数优化算法。通过计算效率分析和数值仿真验证，表明该算法计算复杂度低，参数估计精度高。