A Non-Parametric LPV Approach to the Indentification of Linear Periodic Systems⁎This work is financed by National Funds through the Portuguese funding agency, FCT - Fundação para a Ciência e a Tecnologia within project: UID/EEA/50014/2019. Issue 8 (2021)

• Record Type: Journal Article
• Authors: dos Santos, Paulo Lopes; Perdicoúlis, T-P Azevedo
• Is Part Of: IFAC-PapersOnLine. Volume 54:Issue 8(2021)
• Page Start: 13
• Page End: 19
• Publication Date: 2021
• Subjects: System Identification -- LTP Systems -- LPV Systems -- Arx -- Gaussian Process Regression -- Kernel; Automatic control -- Periodicals; 629.805
• DOI: 10.1016/j.ifacol.2021.08.574
• Languages: English
• ISSNs: 2405-8963

Abstract

Abstract: A non-parametric identification algorithm is proposed to identify Linear Time Periodic (LTP) systems. The period is unknown and can be any real positive number. The system is modelled as an ARX Linear Parameter Varying (LPV) system with a virtual scheduling signal consisting of two orthogonal sinusoids (a sine and a cosine) with a period equal to the system period. Hence, the system parameters are polynomial functions of the scheduling vector. As these polynomials may have infinite degree, a non-parametric model is adopted to describe the LPV system. This model is identified by a Gaussian Process Regression (GPR) algorithm where the system period is a hyperparameter. The performance of the proposed identification algorithm is illustrated through the identification of a simulated LTP continuous system described by a state-space model. The ARX-LTP discrete-time model estimated in the noiseless case was taken as the true model.