Two-stage procedure for the synthesis of control of nonlinear non-stationary objects using a multilayer perceptron

The review of neurocontrol methods and analysis of their advantages and disadvantages is presented. The problem of searching of quasioptimal tuning parameters of neurocontrol for nonlinear non-stationary objects in the presence of random disturbances is formulated. A procedure for the synthesis of control for nonlinear non-stationary objects using a multi-layer perceptron, which consists of two stages, is presented. In the first stage the problem of finding a robust neurocontrol vector tuning parameters for adaptation algorithm based on the proposed set of variants of the model is solved. Founded tuning parameters for adaptation algorithm are used in the second stage - model-free neurocontrol, which searching for quasi-optimal tuning parameters for the algorithm of model-free neurocontrol. Stability of tuning parameters search procedure for the algorithm of model-free neurocontrol achieved by using the regularization method. Effectiveness and stability of the proposed procedure for the synthesis of control for nonlinear non-stationary objects are shown using the model example. In the numerical experiment, an object was chosen that was described by a nonlinear differential equation with coefficients that depend on time. At the first stage, 20 variants of the object model were randomly generated, the architecture of the neural network, the tuning coefficients of the adaptation algorithm were found. The neural network includes 2 neurons in the inner layer and uses a sigmoidal activation function. At the second stage, numerical studies of the adaptive control process were carried out. As a result of the adaptation algorithm, the degree of attenuation of transient processes exceeds 50% and the control process is stable with a significant deviation of the object's parameters from the nominal values. The presented method is effective for the control of multiply connected non-stationary nonlinear objects in robotics, transport systems, and chemical industries.

1. Galushkin A.I. Sostoyanie i perspektivy razvitiya robototekhniki, vklyuchaya sistemy upravleniya.. http://2045.ru/expert/276.html (accessed 28.09.2020).

2. Savenkov P.A. Use methods and algorithms for data analysis and machine learning in UEBA/DSS to assist management decisions. Modeling, Optimization and Information Technology. 2020;8(1). Available from: https://moit.vivt.ru/wpcontent/uploads/2018/01/AndreevaTsiruleva_1_1_18.pdf (accessed 28.09.2020).

3. Norsahperi N.M.H., Danapalasingam K.A. Neurocontrol Design for an Aerodynamics System: Simple Backpropagation Approach. 10th International Conference on Robotics, Vision, Signal Processing and Power Applications. Lecture Notes in Electrical Engineering. 2019;547 Available from: https://link.springer.com/chapter/10.1007/978-981-13-6447-1_1 DOI: https://doi.org/10.1007/978-981-13-6447-1_1 (accessed: 28.09.2020).

4. Liu T., Liang S., Xiong Q., Wang K. Adaptive Critic based Optimal Neurocontrol of a Distributed Microwave Heating System using Diagonal Recurrent Network. IEEE Access. 2018;6:68839-68849. Available from: https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=8502748 DOI: 10.1109/ACCESS.2018.2877206 (accessed: 28.09.2020).

5. Mohammadzaheri M., Chen L., Grainger S. A critical review of the most popular types of neurocontrol. Asian Journal of Control. 2012;14(1):1-11. Available from: https://onlinelibrary.wiley.com/doi/full/10.1002/asjc.449 DOI: https://doi.org/10.1002/asjc.449 (accessed: 28.09.2020).

6. Frolov S.V., Frolova T.A., Sindeev S.V. Upravlenie nelinejnymi nestacionarnymi ob"ektami na osnove nejrokontrollera. Promyshlennye ASU i kontrollery. 2012;5:51-56. (In Russ).

7. Szepesvari C. Algorithms for Reinforcement Learning. Morgan & Claypool Publishers. 2009. (accessed: 28.09.2020).

8. Omatu C. Nejroupravlenie i ego prilozheniya. M.: IPRZHR.2000.

9. Omidvar O., Elliott D. Neural Systems for Control. Academic Press .1997.

10. Zmeu K.V., Markov N.A., SHipit'ko I.A., Notkin B.S. Bezmodel'noe prognoziruyushchee inversnoe nejroupravlenie s regeneriruemym etalonnym perekhodnym processom. Intellektual'nye sistemy. 2009;3(21):109-117. Available from: http://ics.khstu.ru/media/2010/N21_14.pdf (In Russ) (accessed 28.09.2020).

11. Ostrem K.Y. Nastrojka i adaptaciya. Pribory i sistemy upravleniya.1997; 9:53-65. (accessed 28.09.2020).

12. Perel'man I.I. Analiz sovremennyh metodov adaptivnogo upravleniya s pozicij prilozheniya k avtomatizacii tekhnologicheskih processov. AiT.1991; 7:3-32. (In Russ). Available from: http://www.mathnet.ru/links/325d5b12c45bd8fa80294f202fbdb262/at4218.pdf (accessed 28.09.2020).

13. Kvan N.V. Modeli i algoritmy robastnogo upravleniya nelinejnymi ob"ektami v sistemah s bystrodejstvuyushchim etalonom. Avtoreferat dissertacii kandidata tekhnicheskih nauk: 05.13.01. Habarovsk, Amurskij gosudarstvennyj universitet. 2011. (In Russ).

14. Al Smadi Prof-Takialddin & Al-Agha, Osman & Alsmadi, Khalid. Overview of Model Free Adaptive (MFA) Control Technology. IAES International Journal of Artificial Intelligence. 2018;7:165-169. Available from: http://ijai.iaescore.com/index.php/IJAI/article/view/14511/pdf DOI: 10.11591/ijai.v7.i4.pp165-169. (accessed: 28.09.2020).

15. Hajkin S. Nejronnye seti: polnyj kurs, 2-e izdanie – M.: Izdatel'skij dom «Vil'yams». 2006.

16. Frolov S.V., Tret'yakov A.A. Sintez matematicheskih modelej dlya promyshlennyh ASU na osnove nejronnyh setej. Promyshlennye ASU i kontrollery.2000;2.28-31. (In Russ).

17. Frolov S.V., Frolova T.A., Somov P.T. Primenenie metoda regulyarizacii dlya obespecheniya ustojchivosti sistem regulirovaniya s nejrokontrollerom. Promyshlennye ASU i kontrollery.2011;5:54-58. (In Russ).

18. Bishop C.M. Pattern recognition and machine learning. Springer. Science+Business Media. LLC, 2006.