Title: Observer/Kalman filter identification with support vector machines
Authors: Sinchai Chinvorarat; Chien-Hsun Kuo
Addresses: King Mongkut's University of Technology North Bangkok, Bangkok, 10800, Thailand ' Kao Yuan University, Kaohsiung City, Taiwan
Abstract: This paper presents a novel identification filter by utilising the SVM technique with the OKID to identify the nonlinear dynamic system. The proposed filter increases the identifiability and accuracy of the state space realisation model. By integrating the SVM with the OKID in a single identification block, the rich training input and output data from the dynamic system are fed into the identification block and start computing the hyperplane classifier with a radial basis kernel function as a nonlinear mapping function. The algorithm can determine nonlinear signals (noise) from the hyperplane parameters. The modified input-output data determined by filtering out nonlinear signals from the dynamic system output are used for the OKID filter. Since the proposed SVM/OKID filter identifies the discrete model from the nearly un-noise signals, it demonstrates high accuracy in identifying the nonlinear dynamic system over the regular OKID.
Keywords: SVM/OKID filter; Kernel function; Markov parameters.
DOI: 10.1504/IJMIC.2021.123434
International Journal of Modelling, Identification and Control, 2021 Vol.39 No.2, pp.107 - 119
Received: 30 Mar 2021
Accepted: 18 Jun 2021
Published online: 20 Jun 2022 *