A factor graph approach to parameter identification for affine LPV systems
A factor graph approach to parameter identification for affine LPV systems
复制标题
仿射 LPV 系统参数识别的因子图方法
DOI：
--
复制
发表时间：
2017
期刊：
影响因子：
--
通讯作者：
P. Rostalski
中科院分区：
文献类型：
--
作者：
C. Hoffmann;Andreas Isler;P. Rostalski
因子图是可因式分解函数的通用图形表示。作为一种概率图模型，它们允许可视化条件独立性，可以利用它通过沿着图的节点传递消息来有效地解决推理问题。在本文中，提出了一种用于仿射线性参数变化系统识别的基于期望最大化的估计技术的新颖因子图公式。此外，适用于跟踪预定义线性参数变化系统描述所解释和未解释的时变变化的算法的递归重新表述是直接从其基于因子图的表述中得出的。
Factor graphs are a versatile graphical representation of factorizable functions. As a probabilistic graphical model they allow to visualize conditional independence, which can be exploited for efficiently solving inference problems by means of message passing along the nodes of the graph. In this paper, a novel factor graph formulation of the expectation maximization-based estimation technique for affine linear parameter-varying system identification is presented. Furthermore, a recursive reformulation of the algorithm suitable for tracking time-varying changes both accounted and unaccounted for by a pre-defined linear parameter-varying system description is immediate from its factor graph-based formulation.