Affiliations: School of Instrumentation Science and Opto-Electronics Engineering, Beihang University, Beijing, China
Correspondence:
[*]
Corresponding author: Tong Wei, School of Instrumentation Science and Opto-Electronics Engineering, Room B606 New Main Building, Beihang University, 37 Xueyuan Road, Beijing 100191, China. E-mail:weitong@buaa.edu.cn
Abstract: System identification is an important basis of adaptive
control and fault identification of active magnetic bearing (AMB) system.
Among all the identification algorithms, least square method (LS) is a
comparatively simple one with small calculation amount but a poor
performance as well under colored noise, and particle swarm optimization
algorithm (PSO) has a good performance under colored noise but a huge
calculation amount. Taking advantage of the complementation of LS and PSO,
the identification based on LS-PSO algorithm was proposed, which combines
the briefness of LS and the high precision of PSO. In the novel algorithm,
PSO updates the poles of identified transfer function model, and LS
estimates the numerator optimally. The output error variance is minimized by
repeating the procedure. In simulation, the calculation amount declined by
91.4% on average comparing with PSO, and the experiments validated that
the error variance of LS-PSO declined by 92.92% relative to LS. The
experimental results show that LS-PSO retains high precision of PSO under
colored noise, while its calculation amount is greatly reduced by
introducing LS, which is significant to practical application.
Keywords: Magnetic bearing, model identification, least square method, particle swarm optimization
