Control Systems and Computers, N5, 2016, Article 9
DOI: https://doi.org/10.15407/usim.2016.05.076
Upr. sist. maš., 2016, Issue 5 (265), pp. 76-83.
UDC 004.032.26
Bodyanskiy Yevgeniy V., Doctor (Eng.), Prof., Professor of the Department of Artificial Intelligence, Scientific Director of the Problem Research Laboratory of Automated Control Systems, Kharkiv National University of Radio Electronics, E-mail: yevgeniy.bodyanskiy@nure.ua, +380577021890
Vynokurova Olena А., Doctor (Eng.), Prof., Chief Scientist of the Problem Research Laboratory of Automated Control Systems, Professor of the Department of Information Technology Security, Kharkiv National University of Radio Electronics, E-mail: olena.vynokurova@nure.ua,
Kobylin Illya O., PhD student, Kharkiv National University of Radio Electronics, E-mail: ilya.kobylin@nure.ua,
Mulesa Pavlo P., PhD (Eng.), Associate Professor, Department of Cybernetics and Applied Mathematics, State University “Uzhgorod National University”, E-mail: ppmulesa@gmail.com
Robust Adaptive Identification of Non-Stationary Time Series Using Ensemble of Tuning Hybrid Adaptive Models
Introduction. A synthesis of monitoring and control systems for stochastic plants under uncertain conditions requires an identification circuit in these systems, which provides the continuous tuning of mathematical models based on observational data for input and output signals in real time.
Purpose. For solving this problem an effective tools are methods of the informational theory of identification, forming a common approach, which is connected with recurrent algorithms of optimization under uncertain conditions.
These algorithms tune mathematical models based on input and output signals of system. Different types of such systems (both static and dynamic) can be represent as a pseudo-linear regression equation where is scalar output of object in discrete time , is vector of unknown parameters (regression coefficients), which should be estimated, is a vector of input variables, is random noise with unknown function of the density distribution.
Most of procedures and algorithms for the identification usually are based on the hypothesis of Gaussian distribution of noise . It led to using of the least squares method in different modifications. If the data are fed sequentially for processing or the sample size is not fixed, preference should be given to the recurrent least squares method, which, however, for large values of can be numerically unstable, tedious and ultimately leads to “blow-up” of the covariance matrix of the parameters, because a computational procedure of matrix inverse may be unstable.
Conclusions. Therefore, computationally simple and high-speed adaptive algorithms for robust identification of nonstationary non-linear time series are proposed. The distinctive feature of such algorithms is an ability to implement them using the learning models, which are consisted of the elementary arithmetic operations. The possibility of on-line information processing provides a solution of a wide class of problems, which are appeared in Data Stream Mining tasks. Also an ensemble of tuning hybrid adaptive models is proposed. This system allows to choose the best model in the context of accepted quality criterion at each discreet instant of time.
Download full text! (In Russian).
Keywords: adaptive identification, robust function, non-linear non-stationary time series, ensemble of hybrid models, outliers with unknown distribution law.
- Aggarwal, C.C., Reddy, C.K., 2014. Data Clustering. Algorithms and Applications. – Boca Raton: CRC Press, 620 p.
https://doi.org/10.1201/b17320 - Aggarwal, C.C., 2015. Data Mining. Cham: Spinger, 734 p.
- Bifet, A., 2010. Adaptive Stream Mining: Pattern Learning and Mining from Evolving Data Streams. IOS Press., 224 p.
- Lughofer, E., 2011. Evolving Fuzzy Systems. Berlin-Heidelberd: Springer-Verlag, 454 p.
- Tsypkin, YA.Z., 1984. Osnovy informatsionnoy teorii identifikatsii. M.: Nauka, 320 p. (In Russian).
- Ljung, L., 1999. System Identification: Theory for the User. Prentice Hall, 672 p.
- Goodwin, Y.C., Ramadge, P.J., Caines, P.E., 1981. A globally convergent adaptive predictor. Automatica, 17, pp. 135–140.
https://doi.org/10.1016/0005-1098(81)90089-3 - Rey, J. W.W., 1978. Robust Statistic Methods. Berlin–Heidelberd–New York: Springer, 128 p.
https://doi.org/10.1007/BFb0064677 - Polyak, B.T., Tzypkin, Y.Z., 1980. Robust identification. Automatica, 16, pp. 53–63.
https://doi.org/10.1016/0005-1098(80)90086-2 - Khyuber, P., 1984. Robastnost v statistike. M.: Mir, 304 p. (In Russian).
- Li, S.Z., 1995. Markov Random Field Modeling in Computer Vision. Berlin: Springer, 285 p.
https://doi.org/10.1007/978-4-431-66933-3 - Zhang, Zh., 1997. Parameter estimation techniques: a tutorial with application to conic fitting. Image and Vision Comp. J., 15 (1), pp. 59–76.
https://doi.org/10.1016/S0262-8856(96)01112-2 - Ch-Ch. Lee, Yu.-Ch. Chiang, Ch-Yu. Shin et al., 2009. Noisy time series production using M-estimate based on robust radial basic function neural networks with growing and pruning techniques. Expert Systems with Applications, 36. pp. 4717–4724.
https://doi.org/10.1016/j.eswa.2008.06.017 - F.R. Hampel, E.M. Ronchetti, P.J. Rousseeuw et al., 2005. Robust Statistics. The Approach Based on Influence Functions. Wiley, 536 p.
https://doi.org/10.1002/9781118186435 - Bodyanskiy, Ye., Kolodyazhniy, V., Stephan, A., 2001. An adaptive learning algorithm for neural-fuzzy network, Ed. by B. Reusch Computational clutelligence. Theory and Applications. Berlin–Heidelberd–New York: Springer, pp. 68–75.
- Dave, R.N., Krishnapuram, R., 1997. Robust clustering methods: a unified view. IEEE Transaction on Fuzzy Systems, 5, pp. 270–293.
https://doi.org/10.1109/91.580801 - Otto, P., Bodyanskiy, Ye., Kolodyazhniy, V., 2003. A new learning algorithm for a forecasting neuro-fuzzy network. Integrated Computer – Aided Engineering, 10 (4), pp. 399–409.
https://doi.org/10.3233/ICA-2003-10409
Received 20.09.2016
