用于模型结构和参数联合辨识的共轭梯度追踪辨识方法

Conjugate gradient pursuit identification algorithm for combined model structure and parameter identification

中文关键词  参数辨识  结构辨识  共轭梯度  共轭梯度追踪

英文关键词  parameter identification  structure identification  conjugate gradient  conjugate gradient pursuit

中文摘要

      如何利用有限量的数据同时辨识系统的结构和参数是辨识实践的一大挑战.本文将共轭梯度优化算法和 贪婪算法相结合,提出了一种共轭梯度追踪稀疏辨识方法.首先,将系统模型转化为稀疏参数辨识模型;其次,利用 贪婪搜索和共轭梯度优化,挑选并估计非零参数的位置和数值;然后,根据稀疏参数结构获得系统阶次和时滞的估 计. 仿真实验表明,该方法能够在有限的采样数据条件下,有效地辨识系统的阶次、时滞和参数,且具有迭代速度 快、计算量小的特点.与现有的贪婪辨识方法相比,其计算量远小于基于正交匹配追踪的辨识方法,迭代次数低于 基于梯度追踪的辨识方法,且模型的精度高于基于梯度追踪的辨识方法.

英文摘要

      It is a significant challenge to simultaneously identify the model structure and parameters with limited sam pling data. By combining the conjugate gradient optimization algorithm and a greedy algorithm, a conjugate gradient pursuit based sparse identification method is proposed. Firstly, the system model is transformed into a sparse parameter identification model. Then, the greedy search and conjugate gradient optimization are used to select and estimate the posi tions and values of non-zero parameters, system orders and delays are then obtained based on the sparse parameter structure. Simulation examples show that this method can utilize limited sampling data to simultaneously identify the structure and parameters of the system, with the advantages of fast iteration speed and low computational complexity. Compared with existing greedy identification methods, it has lower computational complexity than the orthogonal matching pursuit based method and fewer iterations than the gradient pursuit based method, while achieving a higher model accuracy than the gradient pursuit based method.