Abstract
This paper is concerned with the identification problem of Markov jump autoregressive exogenous systems with unknown slow-varying time-delay. With the implementation of variational Bayesian inference approach, the distribution characteristics of local model parameters are obtained. Meanwhile, the transition probability matrix and the unknown time-delay are determined. Furthermore, a numerical example and a simulated continuous fermentation reactor example are performed to verify the effectiveness of the proposed algorithm.
Similar content being viewed by others
Notes
In the following context, both the probability mass functions and probability density functions are expressed as p, without influencing the derivations and readers’ understanding.
References
O. Cappé, Online EM algorithm for hidden Markov models. J. Comput. Graph. Stat. 20(3), 728–749 (2011)
S.P. Chatzis, D.I. Kosmopoulos, A variational Bayesian methodology for hidden Markov models utilizing Student’s-t mixtures. Pattern Recognit. 44(2), 295–306 (2011)
F. Chen, H. Garnier, M. Gilson, Robust identification of continuous-time models with arbitrary time-delay from irregularly sampled data. J. Process Control 25, 19–27 (2015)
J. Chen, B. Huang, F. Ding, Y. Gu, Variational Bayesian approach for ARX systems with missing observations and varying time-delays. Automatica 94, 194–204 (2018)
J. Chen, Y. Liu, Variational Bayesian-based iterative algorithm for ARX models with random missing outputs. Circuits Syst. Signal Process. 37, 1594–1608 (2018)
X. Chen, S. Zhao, F. Liu, Identification of time-delay Markov jump autoregressive exogenous systems with expectation-maximization algorithm. Int. J. Adapt. Control Signal Process. 31(12), 1920–1933 (2017)
E. Cinquemani, R. Porreca, G. Ferrari-Trecate, J. Lygeros, A general framework for the identification of jump Markov linear systems. In: IEEE Conference on Decision & Control, pp. 5737–5742 (2007)
O.L.V. Costa, R.P. Marques, M.D. Fragoso, Discrete-Time Markov Jump Linear System (Springer, Heidelberg, 2004)
A.P. Dempster, N.M. Laird, D.B. Rubin, Maximum likelihood from incomplete data via the EM algorithm. J. R. Stat. Soc. Ser. B (Methodol.) 39(1), 1–38 (1977)
N.H. El-Farra, P.D. Christofides, Coordinating feedback and switching for control of hybrid nonlinear processes. AIChE J. 49(8), 2079–2098 (2003)
L. Fan, H. Kodamana, B. Huang, Robust identification of switching Markov ARX models using EM algorithm. In: Proceedings of the 20th IFAC World Congress, vol. 50(1), pp. 9772–9777 (2017)
J.K. Gugaliya, R.D. Gudi, S. Lakshminarayanan, Multi-model decomposition of nonlinear dynamics using a fuzzy-CART approach. J. Process Control 15(4), 417–434 (2005)
M.A. Henson, D.E. Seborg, Nonlinear control strategies for continuous fermenters. Chem. Eng. Sci. 47(4), 821–835 (1992)
X. Jin, B. Huang, D.S. Shook, Multiple model LPV approach to nonlinear process identification with EM algorithm. J. Process Control 21(1), 182–193 (2011)
X. Jin, B. Huang, Identification of switched Markov autoregressive eXogenous systems with hidden switching state. Automatica 48(2), 436–441 (2012)
Q. Lin, R. Loxton, C. Xu, K.L. Teo, Parameter estimation for nonlinear time-delay systems with noisy output measurements. Automatica 60, 48–56 (2015)
S. Liu, J. Jia, Y.D. Zhang, Y. Yang, Image reconstruction in electrical impedance tomography based on structure-aware sparse Bayesian learning. IEEE Trans. Med. Imaging 37(9), 2090–2102 (2018)
S. Liu, Y.D. Zhang, T. Shan, R. Tao, Structure-aware Bayesian compressive sensing for frequency-hopping spectrum estimation with missing observations. IEEE Trans. Signal Process. 66(8), 2153–2166 (2018)
L. Ljung, Perspectives on system identification. Annu. Rev. Control 34(1), 1–12 (2010)
A. Logothetis, V. Krishnamurthy, Expectation maximization algorithms for MAP estimation of jump Markov linear systems. IEEE Trans. Signal Process. 47(8), 2139–2156 (1999)
A. Masmoudi, F. Bellili, S. Affes, A. Stéphenne, A non-data-aided maximum likelihood time delay estimator using importance sampling. IEEE Trans. Signal Process. 59(10), 4505–4515 (2011)
A. Masmoudi, F. Bellili, S. Affes, A. Stéphenne, A maximum likelihood time delay estimator in a multipath environment using importance sampling. IEEE Trans. Signal Process. 61(1), 182–193 (2013)
J. Na, X. Ren, Y. Xia, Adaptive parameter identification of linear SISO systems with unknown time-delay. Syst. Control Lett. 66, 43–50 (2014)
U. Orguner, M. Demirekler, Maximum likelihood estimation of transition probabilities of jump Markov linear systems. IEEE Trans. Signal Process. 56(10), 5093–5108 (2008)
E. Özkan, F. Lindsten, C. Fritsche, F. Gustafsson, Recursive maximum likelihood identification of jump Markov nonlinear systems. IEEE Trans. Signal Process. 63(3), 754–765 (2015)
S. Paoletti, A.L. Juloski, G. Ferrari-Trecate, R. Vidal, Identification of hybrid systems: a tutorial. Eur. J. Control 13(2–3), 242–260 (2007)
J.P. Richard, Time-delay systems: an overview of some recent advances and open problems. Automatica 39(10), 1667–1694 (2003)
N. Sammaknejad, B. Huang, Operating condition diagnosis based on HMM with adaptive transition probabilities in presence of missing observations. AIChE J. 61(2), 477–493 (2015)
L. Xie, H. Yang, B. Huang, FIR model identification of multirate processes with random delays using EM algorithm. AIChE J. 59(11), 4124–4132 (2013)
A. Xue, H. Wang, R. Lu, Event-based H\(\infty \) control for discrete Markov jump systems. Neurocomputing 190, 165–171 (2016)
X. Yang, B. Huang, H. Gao, A direct maximum likelihood optimization approach to identification of LPV time-delay system. J. Frankl. Inst. 353(8), 1862–1881 (2016)
X. Yang, H. Gao, Multiple model approach to linear parameter varying time-delay system identification with EM algorithm. J. Frankl. Inst. 351(12), 5565–5581 (2014)
X. Yin, L. Zhang, Z. Ning, D. Tian, A. Alsaedi, B. Ahmad, State estimation via Markov switching-channel network and application to suspension systems. IET Control Theory Appl. 11(3), 411–419 (2017)
X. Yin, Z. Li, L. Zhang, M. Han, Distributed state estimation of sensor-network systems subject to Markovian channel switching with application to a chemical process. IEEE Trans. Syst. Man Cybern. Syst. 48(6), 864–874 (2018)
Y. Zhao, A. Fatehi, B. Huang, A data-driven hybrid ARX and Markov-chain modeling approach to process identification with time varying time delays. IEEE Trans. Ind. Electron. 64(5), 4226–4236 (2017)
Y. Zhao, A. Fatehi, B. Huang, Robust estimation of ARX models with time varying time delays using variational Bayesian approach. IEEE Trans. Cybern. 48(2), 532–542 (2018)
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
This work was supported by the National Natural Science Foundation of China (Grant Numbers 61773183 and 61833007), and supported by National First-Class Discipline Program of Light Industry Technology and Engineering (LITE2018-25).
Appendices
A Appendix
The optimization problem expressed in Eq. (9) is solved in this section. The lower bound of the evidence can be derived as
Then, a Lagrange multiplier \(\lambda _1\) is introduced as
Employing the variational derivative, the above Lagrange function is derived as follows:
Solving Eq. (39), the expression of \(q\left( {I,\tau } \right) \) can be obtained as
When Eq. (40) is substituted into the constraint \(\int \limits {q\left( {I,\tau } \right) \mathrm{d}I\mathrm{d}\tau } = 1\), the term relating to the Lagrange multiplier \(\lambda _1\) is obtained as
Finally, the expression of the joint variational posterior of the latent variable I and the time-delay \(\tau \) is obtained as
B Appendix
The derivation of Eq. (11) is detailed here. Considering the law of total probability, the joint variational posterior of mode identity \(I_k\) and \(I_{k-1}\) is expressed as
Then, Eq. (10) is substituted into the above equation, in which only the instants 1 to k are taken into consideration.
The numerator of Eq. (44) can be decomposed into the following expression.
where \(K_2 = p\left( {{u_k}|{y_{1:k - 1}},{u_{1:k - 1}},{I_{1:k - 2}},{I_{k - 1}} = j,\tau _{1:{k-1}},\varTheta } \right) \) is a constant that unrelated to the integral. The denominator of Eq. (44) has a similar expression as
Then, \(\int \limits _{{\tau _{1:k - 1}}} {\int \limits _{{I_{1:k - 2}}} {\exp {{\left\langle {\log p\left( {{y_{1:k - 1}},{u_{1:k - 1}},{I_{1:k - 2}},{I_{k - 1}} = j,{\tau _{1:k - 1}}|\varTheta } \right) } \right\rangle }_{q\left( \varTheta \right) }}} } \mathrm{d}{I_{1:k - 2}}\mathrm{d}{\tau _{1:k - 1}}\), which appears both in the numerator and denominator of Eq. (44), is simplified in the following context.
When only the instants 1 to \(k - 1\) are taken into consideration for Eq. (10), the variational posterior \(q\left( {{I_{1:k - 2}},{I_{k - 1}} = j} \right) \) can be expressed as
An integral of the variable \({I_{1:k - 2}}\) is performed on the above equation, then,
Equation (48) can be rearranged as follows:
where \(K_3=\int \limits _{{\tau _{1:k - 1}}} {\int \limits _{{I_{1:k - 1}}} {\exp \big [ {{\left\langle {\log p\left( {{y_{1:k - 1}},{u_{1:k - 1}},{I_{1:k - 1}},{\tau _{1:k - 1}}|\varTheta } \right) } \right\rangle }_{q\left( \varTheta \right) }}\big ] \mathrm{d}}}{{{I_{1:k - 1}}\mathrm{d}{\tau _{1:k - 1}}} } \) is a normalization term that unrelated to the integrals of \({\tau _k}\), \({I_k}\), and \({I_{k-1}}\). Therefore, a common factor can be removed from the numerator and denominator of Eq. (44). Finally, the joint variational posterior of \({I_k}\) and \({I_{k-1}}\) can be simplified as
When the integrals in Eq. (50) are transformed into summations, Eq. (11) is obtained.
C Appendix
The derivation of Eq. (14) is presented in this section. According to the property of conditional probability, the variational posterior distribution of the time-delay \(\tau _k\) conditional on the mode identity \(I_k\) can be derived as
Taking advantage of the expression of the joint posterior distribution in Eq. (10), the conditional variational posterior distribution of the time-delay can be further derived as
Then, the above equation can be expressed into a summation form as Eq. (14).
D Appendix
Details of the derivation of Eq. (16) are provided in this section. The Lagrange multiplier \(\lambda _2\) is adopted with respect to the lower bound of the evidence of Eq. (8) as
The variational derivative is then implemented, and the above Lagrange function can be derived as follows:
Employing a similar procedure as Appendix A, the solution of the variational posterior of parameters is obtained as
E Appendix
In this section, the derivation of Eq. (30) is detailed. As no valid prior exists, direct optimization of the lower bound with respect to the transition probability \(\alpha _{ij}\), instead of its distribution, is implemented. A partial derivative of Eq. (29) with respect to \(\alpha _{ij}\) is solved and assigned to be zero as
Then, the transition probability \(\alpha _{ij}\) can be derived as
When the above equation is substituted into the constraint of the transition probability \(\sum \nolimits _{i = 1}^M {{\alpha _{ij}}} = 1\), the Lagrange multiplier \(\lambda _{\alpha }\) can be obtained as
Then, the solution of the transition probability is derived as
Rights and permissions
About this article
Cite this article
Chen, X., Liu, F. A Variational Bayesian Approach for Identification of Time-Delay Markov Jump Autoregressive Exogenous Systems. Circuits Syst Signal Process 39, 1265–1289 (2020). https://doi.org/10.1007/s00034-019-01206-x
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00034-019-01206-x