An improved Hammerstein system identification method using Stein Variational Inference and sampling technology. (April 2023)

• Record Type: Journal Article
• Authors: Zhang, Limin; Jin, Di; Zhao, Jia
• Is Part Of: Journal of process control. Volume 124(2023)
• Page Start: 25
• Page End: 35
• Publication Date: 2023-04
• Subjects: Hammerstein system -- Parameters identification -- Stein variational inference -- Reversible jump markov chain monte carlo; Process control -- Periodicals; Fabrication -- Contrôle -- Périodiques; Process control; Periodicals; Electronic journals; 660.281
• DOI: 10.1016/j.jprocont.2023.02.005
• Languages: English
• ISSNs: 0959-1524

Abstract

Abstract: This paper considers the identification of the Hammerstein system with immeasurable process noise. The complexity of the Hammerstein system makes it difficult to obtain accurate mathematical expressions of the parameters, or even impossible to obtain accurate mathematical expressions at all. In this contribution, we cast the Hammerstein system parameter identification problem as a posterior parameter estimation problem and take a sampling and Stein variational inference viewpoint to solve it. Improved Stein variational gradient descent(ISVGD)algorithm is proposed in posterior parameter calculation. Compared with other methods, not only the prior distribution of parameters but also the proposal distribution of parameters is considered. In other words, The posterior information is enriched and the parameter identification accuracy is improved. At the same time, ISVGD and reversible jump markov chain monte carlo (RJMCMC) algorithms are used (called ISVGD-RJMCMC algorithm) in the structure detection problem. In the algorithm, the correct basis function number k can be found in the parameter estimation. It tends to be accurate but is slow to converge. Three simulation examples were given to demonstrate the proposed algorithm's effectiveness. Furthermore, the performances of these approaches were analyzed, including parameter estimation accuracy and error, system order estimation and parameter convergence analysis. Highlights: Stein variational inference is used in Hammerstein system identification. Improved Stein variational gradient descent(ISVGD)algorithm is proposed. ISVGD-RJMCMC algorithm is used in the structure detection problem.