Novel adaptive methods for output-only recursive identification of time-varying systems subject to gross errors

Abstract

Gross errors are generally used to model intermittent sensor failures and occasional data packet losses or corruption, which arise in many engineering communities. In this work, we propose to deal with the problem of output-only recursive identification of time-varying systems subject to gross errors by using an adaptive weighting and forgetting combined strategy. Under the assumption that gross errors are unknown and can be of arbitrarily large magnitude, time-dependent autoregressive model-based adaptive recursive identification methods are proposed by minimizing the sum of norm errors and achieving a sparse prediction error sequence. The adaptive weighting strategy is used to deemphasize data observations contaminated by gross errors, and the forgetting mechanism is used to deemphasize data from the remote past, allowing the proposed methods to track time-varying dynamics of the system subject to gross errors. The proposed methods are numerically and experimentally tested, and the comparative results demonstrate the superior time-varying tracking capability of the proposed methods in extremely challenging gross error circumstances.

Multivariate recursive Bayesian linear regression and its applications to output-only identification of time-varying mechanical systems

“…The proposed multivariate recursive BLR method is first applied to identify a time series model with gradual parameter evolution as follows (Ma et al, 2020b) x½t…”

“…The proposed multivariate recursive BLR method is first applied to identify a time series model with gradual parameter evolution as follows (Ma et al, 2020b)where the true AR parameters are given byand e[t] is an innovation sequence with zero mean and identity covariance matrix, that is e[t]∼N(0,I).…”

“…In recent years, many parametric time-domain methods for TV system identification have been proposed by using parameterized representations, such as time-dependent autoregressive moving average (TARMA) models (Ma et al, 2018b(Ma et al, , 2020bPoulimenos and Fassois, 2006;Spiridonakos and Fassois, 2014;Yang et al, 2015;Zhou et al, 2018). Output-only recursive methods within this class also have a rich history in nonstationary signal modeling and analysis (Ljung, 1999;Ljung and Soderstrom, 1983;Niedzwiecki, 2000), and the main idea of these methods is summarized as follows: data observations are measured and considered one at a time, and the TARMA model parameters are recursively updated in real time after each such presentation (Ma et al, 2020b). Over the past decades, many efforts have been undertaken to develop output-only recursive methods.…”

“…Ma et al (2018b) proposed an improved pseudo-linear regression form of the vector TARMA model and significantly reduced the computational complexity of TARMA model parameter estimation when processing multivariate response signals. Ma et al (2020b) further proposed adaptive methods for output-only recursive identification of TV systems subject to gross errors by using an adaptive weighting and forgetting combined strategy.…”

Mass and stiffness identification of particle beam system based on a dynamic effective mass method

“…A mathematical model is still needed to explain the mechanism of the particle beam system in the article, such as a bouncing ball model (Jiang et al, 2007;Wang et al, 2016). There are lots of methods, such as time-and frequency-domain methods (Julián et al, 2017;Urgessa, 2010), time-frequency analysis (Feldman, 2007;Ma et al, 2020), and modal methods (Simon et al, 2018), to identify the nonlinear behaviors of dynamic systems. For the complicated frequency-dependent particle system, however, it is difficult.…”