Identification of Switched Gated Recurrent Neural Networks Using the EM Algorithm

In the domain of nonlinear hybrid dynamic system modeling, the effectiveness of switched autoregressive exogenous (SARX) systems may face certain restrictions. To address this issue, this paper presents an enhanced switched system framework. In this framework, all SARX subsystems are replaced with g...

Full description

Saved in:

Bibliographic Details
Published in:	2023 5th International Conference on Data-driven Optimization of Complex Systems (DOCS) pp. 1 - 6
Main Authors:	Bai, Wentao, Gu, Suhang, Yan, Chao, Zhang, Haoyu, Jiang, Chunli, Guo, Fan
Format:	Conference Proceeding
Language:	English
Published:	IEEE 22.09.2023
Subjects:	EM algorithm gated recurrent neural networks Heuristic algorithms hybrid dynamic systems Logic gates Probabilistic logic Recurrent neural networks SARX Switched systems Switches Switching systems
Online Access:	Get full text
Tags:	Add Tag No Tags, Be the first to tag this record!

Description
Summary:	In the domain of nonlinear hybrid dynamic system modeling, the effectiveness of switched autoregressive exogenous (SARX) systems may face certain restrictions. To address this issue, this paper presents an enhanced switched system framework. In this framework, all SARX subsystems are replaced with gated recurrent neural networks, aiming to overcome these limitations. Importantly, the proposed switched system does not rely on any prior assumptions about the knowledge of operating modes. Finally, a new identification method is proposed based on the expectation-maximization (EM) algorithm, and its effectiveness is validated through a simulation example.
DOI:	10.1109/DOCS60977.2023.10294958