Noise Reduction for Data-driven Control by Correlation Analysis and High-order ARX Identification

Abstract

FRIT and VRFT are widely applied as representatives of Data-driven control, but the actual closed-loop response cannot be predicted. Recently, Data-driven prediction methods including V-Tiger that predict the actual closed-loop response have been proposed, but assume that there is no noise in one-shot experimental data. System identification attempts to separate the one-shot experimental data into dynamic response and noise. For the purpose of noise reduction in the data, there is no problem even if the model order is large. Therefore, we focuse on the Correlation analysis and High-order ARX identification, which are less related to the model order among the identification methods. The Correlation analysis estimates impulse response of stable plant. The waveform of the impulse response is mainly in the interval from the start of the response to its decay. This paper proposes (1) Estimate the impulse response by the Correlation analysis to concentrate the waveform in the interval of the initial response, (2) Cut out the interval and perform High-order ARX identification, and (3) Apply the output response of the identified model to Data-driven control especially V-Tiger as the experimental data with noise removed.