novtexmechМехатроника, автоматизация, управлениеMekhatronika, Avtomatizatsiya, Upravlenie1684-64272619-1253Commercial Publisher «New Technologies»10.17587/mau.24.563-572novtexmech-1450Research ArticleСИСТЕМНЫЙ АНАЛИЗ, УПРАВЛЕНИЕ И ОБРАБОТКА ИНФОРМАЦИИSYSTEM ANALYSIS, CONTROL AND INFORMATION PROCESSINGСтруктурно-параметрический синтез нейросетевых регуляторов для объектов управления с ограничителямиStructural and Parametric Synthesis of Neural Network Controllers for Control Objects with LimitersФеофиловС. В.FeofilovS. V.

д-р техн. наук, проф.

Feofilov Sergey V., Dr. of Eng. Sc., Professor

Tula, 300012

svfeofilov@mail.ruКозырьА. В.KozyrA. V.

канд. техн. наук, доц.

Tula, 300012

kozyr_a_v@mail.ruХапкинД. Л.KhapkinD. L.

мл. науч. сотр.

Tula, 300012

dima-hapkin@ya.ruТульский государственный университетРоссияTula State UniversityRussian Federation2023311020232411563572Copyright © Commercial Publisher «New Technologies», 20232023Commercial Publisher «New Technologies»Commercial Publisher «New Technologies»https://mech.novtex.ru/jour/about/submissions#copyrightNoticehttps://mech.novtex.ru/jour/article/view/1450

Излагается методика синтеза цифровых систем управления для нелинейных объектов c ограничителями в условиях неполной информации. Рассматриваются замкнутые следящие системы с отрицательной обратной связью. Для построения регулятора, который включается последовательно с объектом управления, предлагается использовать искусственные нейронные сети.

Такой подход эффективен, когда известные классические методы не позволяют напрямую синтезировать управление. Это происходит, например, в случае, если математическая модель является существенно нелинейной и не полностью определена. Разработанные методы позволяют расширить класс технических систем, для которых возможен прямой (без использования различного рода упрощений) синтез близких к оптимальным законов управления. Кроме того, нейросетевые регуляторы обладают свойствами робастности, адаптивности, являются исходно цифровыми, т. е. имеют те качества, которые очень востребованы на практике. В статье основное внимание уделяется таким проблемам, как выбор структуры нейросети для нейроимитатора и нейрорегулятора, построение обучающей выборки, обеспечение сходимости процесса корректировки весов. Для обучения нейросетей в качестве базового используется метод обратного распространения ошибки.

Следует отметить, что сегодня нейросетевые технологии достаточно широко распространены в различных сферах деятельности. Впечатляют успехи, продемонстрированные в области обработки звука, изображения, автоматического перевода, в системах навигации, при обработке больших данных. Однако их применение в системах автоматического управления не столь широко. Авторы статьи считают, что потенциал искусственных нейронных сетей может быть использован в данном направлении. При этом следует понимать, что применение нейросетей эффективно лишь при определенных условиях и свойствах объекта управления.

The article presents a methodology for the synthesis of digital control systems for nonlinear objects with limiters under conditions of incomplete information. Closed-loop tracking systems with negative feedback are considered. Artificial neural networks are proposed to build a controller, which is included in series with the control object. This approach is effective when known classical methods do not allow to synthesize control. This is the case, for example, if the mathematical model is essentially nonlinear and is not fully defined. The developed methods allow us to expand the class of technical systems, for which the direct (without using various kinds of simplifications) synthesis of control laws that are close to optimal is possible. In addition, neural network controllers possess the properties of robustness, adaptivity, and are initially digital, i.e. those qualities, which are very much in demand in practice. In article main attention is given to such problems, as a choice of neural network structure for neural simulator and neural controller, construction of training sample, ensuring convergence of the process of weights correction. For training neural networks the method of back propagation of error is used as a basic one. The effectiveness of the proposed technique is demonstrated by the example of the synthesis of a neuroregulator for a specific technical object and its comparison with classical control systems. It should be noted that today neural network technologies are widespread enough in various spheres of activity. The successes demonstrated in sound processing, image processing, automatic translation, in navigation systems, in big data processing are impressive. However, their application in automatic control systems is not so widespread. The authors of this article believe that the potential of artificial neural networks can be used in this direction. It should be understood that the use of neural networks is effective only under certain conditions and properties of the control object.

нейросетевой регуляторследящая системаобучающая выборканейроимитаторneural network controllertracking systemlearning samplingneurosimulatorИсследование выполнено за счет гранта Российского научного фонда № 23-29-00609, https://rscf.ru/project/23-29-00609/.The research was carried out at the expense of the grant of the Russian Science Foundation No. 23-29-00609.ReferencesSünderhauf N., Brock O., Scheirer W., Hadsell R., Fox D., Leitner J., Upcroft B., Abbeel P., Burgard W., Milford M., Corke P. The Limits and Potentials of Deep Learning for Robotics // arXiv:1804.06557. arXiv. 2018.Sünderhauf N., Brock O., Scheirer W., Hadsell R., Fox D., Leitner J., Upcroft B., Abbeel P., Burgard W., Milford M., Corke P. The Limits and Potentials of Deep Learning for Robotics, arXiv:1804.06557, arXiv, 2018.Beard R. W., McLain T. W. Small Unmanned Aircraft: Theory and Practice. Small Unmanned Aircraft. Princeton University Press, 2012. P. 317.Beard R. W., McLain T. W. Small Unmanned Aircraft: Theory and Practice. Small Unmanned Aircraft, Princeton University Press, 2012, pp. 317.Kaess M., Johannsson H., Roberts R., Ila V., Leonard J. J., Dellaert F. iSAM2: Incremental smoothing and mapping using the Bayes tree // The International Journal of Robotics Research. 2012. Vol. 31, N. 2. P. 216—235.Kaess M., Johannsson H., Roberts R., Ila V., Leonard J. J., Dellaert F. iSAM2: Incremental smoothing and mapping using the Bayes tree, The International Journal of Robotics Research, 2012, vol. 31, no. 2, pp. 216—235.Huang S., Papernot N., Goodfellow I., Duan Y., Abbeel P. Adversarial Attacks on Neural Network Policies. arXiv:1702.02284. arXiv. 2017.Huang S., Papernot N., Goodfellow I., Duan Y., Abbeel P. Adversarial Attacks on Neural Network Policies, arXiv:1702.02284, arXiv, 2017.Bendale A., Boult T. Towards Open World Recognition // Proceeding of 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 2015. P. 1893—1902.Bendale A., Boult T. Towards Open World Recognition, Proceeding of 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2015, pp. 1893—1902.Fabiani F., Goulart P. J. Neural network controllers for uncertain linear systems // arXiv:2204.13209. arXiv. 2022.Fabiani F., Goulart P. J. Neural network controllers for uncertain linear systems, arXiv:2204.13209, arXiv, 2022.Jabali M. B. A., Kazemi M. H. Uncertain polytopic LPV modelling of robot manipulators and trajectory tracking // International Journal of Control, Automation and Systems. 2017. Vol. 15, N. 2. P. 883—891.Jabali M. B. A., Kazemi M. H. Uncertain polytopic LPV modelling of robot manipulators and trajectory tracking, International Journal of Control, Automation and Systems, 2017. vol. 15, no. 2, pp. 883—891.Li S., Ozturk E., De Wagter C., Croon G. C. H.E. de, Izzo D. Aggressive Online Control of a Quadrotor via Deep Network Representations of Optimality Principles // Proceeding of 2020 IEEE International Conference on Robotics and Automation (ICRA), Paris, France: IEEE, 2020. P. 6282—6287.Li S., Ozturk E., De Wagter C., Croon G. C. H. E. de, Izzo D. Aggressive Online Control of a Quadrotor via Deep Network Representations of Optimality Principles, Proceeding of 2020 IEEE International Conference on Robotics and Automation (ICRA), Paris, France, IEEE. 2020, pp. 6282—6287.Чернодуб А. Н., Дзюба Д. А. Обзор методов нейроуправления // Проблемы программирования. 2011. № 2. С. 79—94.Chernodub A. N., Dzyuba D. A. Review of methods, Problemy programmirovaniya, 2011, no 2, pp. 79—94 (In Russian).Ekman M. Learning Deep Learning: Theory and Practice of Neural Networks, Computer Vision, Natural Language Processing, and Transformers Using TensorFlow. Learning Deep Learning. Addison-Wesley, 2022. 688 P.Ekman M. Learning Deep Learning: Theory and Practice of Neural Networks, Computer Vision, Natural Language Processing, and Transformers Using TensorFlow. Learning Deep Learning, Addison-Wesley, 2022, 688 p.Izzo D., Öztürk E. Real-Time Guidance for Low-Thrust Transfers Using Deep Neural Networks // Journal of Guidance, Control, and Dynamics. 2021. Vol. 44, N. 2. P. 315—327.Izzo D., Öztürk E. Real-Time Guidance for Low-Thrust Transfers Using Deep Neural Networks, Journal of Guidance, Control, and Dynamics, 2021, vol. 44, no. 2, pp. 315—327.Reifman J., Feldman E. E. Nonlinear programming with feedforward neural networks // IJCNN’99. International Joint Conference on Neural Networks. Proceedings (Cat. No. 99CH36339). 1999. Vol. 1. P. 594—598.Reifman J., Feldman E. E. Nonlinear programming with feedforward neural networks, IJCNN’99. International Joint Conference on Neural Networks. Proceedings (Cat. No.99CH36339), 1999, vol. 1, pp. 594—598.Effati S., Baymani M. A new nonlinear neural network for solving quadratic programming problems // Applied Mathematics and Computation. 2005. Vol. 165, N. 3. P. 719—729.Effati S., Baymani M. A new nonlinear neural network for solving quadratic programming problems, Applied Mathematics and Computation, 2005, vol. 165, no. 3, pp. 719—729.Niestroy M. Optimal feedback controller approximation using neural networks and nonlinear programming techniques // 1998 IEEE International Joint Conference on Neural Networks Proceedings. IEEE World Congress on Computational Intelligence (Cat. No.98CH36227). 1998. Vol. 3. P. 2110—2115.Niestroy M. Optimal feedback controller approximation using neural networks and nonlinear programming techniques, 1998 IEEE International Joint Conference on Neural Networks Proceedings. IEEE World Congress on Computational Intelligence (Cat. No.98CH36227), 1998, vol. 3, pp. 2110—2115.Nakamura-Zimmerer T., Gong Q., Kang W. A CausalityFree Neural Network Method for High-Dimensional HamiltonJacobi-Bellman Equations // Proceeding of 2020 American Control Conference (ACC). 2020. P. 787—793.Nakamura-Zimmerer T., Gong Q., Kang W. A Causality-Free Neural Network Method for High-Dimensional Hamilton-Jacobi-Bellman Equations, Proceeding of 2020 American Control Conference (ACC), 2020, pp. 787—793.Nakamura-Zimmerer T., Gong Q., Kang W. Adaptive Deep Learning for High-Dimensional Hamilton-Jacobi-Bellman Equations // SIAM Journal on Scientific Computing. 2021. Vol. 43, N. 2. P. A1221—A1247.Nakamura-Zimmerer T., Gong Q., Kang W. Adaptive Deep Learning for High-Dimensional Hamilton-Jacobi-Bellman Equations, SIAM Journal on Scientific Computing, 2021, vol. 43, no. 2, pp. A1221—A1247.Nakamura-Zimmerer T., Gong Q., Kang W. QRnet: Optimal Regulator Design With LQR-Augmented Neural Networks // IEEE Control Systems Letters. 2021. Vol. 5, N. 4. P. 1303—1308.Nakamura-Zimmerer T., Gong Q., Kang W. QRnet: Optimal Regulator Design With LQR-Augmented Neural Networks, IEEE Control Systems Letters, 2021, vol. 5, no. 4, pp. 1303—1308.Yang X., Liu D., Huang Y. Neural-network-based online optimal control for uncertain non-linear continuous-time systems with control constraints // IET Control Theory & Applications. 2013. Vol. 7, N. 17. P. 2037—2047.Yang X., Liu D., Huang Y. Neural-netw ork-based online optimal control for uncertain non-linear continuous-time systems with control constraints, IET Control Theory & Applications, 2013, vol. 7, no. 17, pp. 2037—2047.Parrilo P. A. Structured Semidefinite Programs and Semialgebraic Geometry Methods in Robustness and Optimization. California Institute of Technology, 2000. P. 234.Parrilo P. A. Structured Semidefinite Programs and Semialgebraic Geometry Methods in Robustness and Opti mization, California Institute of Technology, 2000, pp. 234.Lofberg J. Preand Post-Processing Sum-of-Squares Programs in Practice // IEEE Transactions on Automatic Control. 2009. Vol. 54, N. 5. P. 1007—1011.Lofberg J. Pre- and Post-Processing Sum-of-Squares Programs in Practice, IEEE Transactions on Automatic Control, 2009, vol. 54, no 5, pp. 1007—1011.Feofilov S. V., Khapkin D. L. Synthesis of neural network controllers for objects with non-linearity of the constraint type // Journal of Physics: Conference Series. 2021. Vol. 1958, N. 1. P. 012014.Feofilov S. V., Khapkin D. L. Synthesis of neural network controllers for objects with non-linearity of the constraint type, Journal of Physics: Conference Series, 2021, vol. 1958, no. 1, pp. 012014.Nair V., Hinton G. E. Rectified Linear Units Improve Restricted Boltzmann Machines // Proceedings of the 27th International Conference on International Conference on Machine Learning ICML’10, Madison, WI, USA, Omnipress, 2010, P. 807—814.Nair V., Hinton G. E. Rectified Linear Units Improve Restricted Boltzmann Machines, Proceedings of the 27th International Conference on International Conference on Machine Learning ICML’10, Madison, WI, USA, Omnipress, 2010, pp. 807—814.Галушкин А. И. Синтез многослойных систем распознавания образов. М.: Энергия, 1974. 367 с.Galushkin A. I. Synthesis of multilayer image recognition systems, Moscow, Energiya, 1974, 367 p. (In Russian).Werbos P. J. Backpropagation through time: what it does and how to do it // Proceedings of the IEEE. 1990. Vol. 78, N. 10. P. 1550—1560.Werbos P. J. Backpropagation through time: what it does and how to do it, Proceedings of the IEEE, 1990, vol. 78, no. 10, pp. 1550—1560.Goodfellow I., Bengio Y., Courville A. Deep learning: Adaptive computation and machine learning. Cambridge, Massachusetts: The MIT Press, 2016. P. 775.Goodfellow I., Bengio Y., Courville A. Deep learning: Adaptive computation and machine learning, Cambridge, Massachusetts, The MIT Press, 2016, 775 p.Kingma D. P., Ba J. Adam: A Method for Stochastic Optimization // Adam. arXiv:1412.6980. arXiv. 2017.Kingma D. P., Ba J. A. A Method for Stochastic Optimization, arXiv:1412.6980, arXiv, 2017.Хоанг Ч. К. Оптимизация объемных силовых следящих гидроприводов по быстродействию и по точности режима слежения / Диссертация... кандидата технических наук: 05.13.01. Тула, 2006. 131 с.Hoang C. K. Optimization of volumetric power tracking hydraulic drives in terms of speed and accuracy of the tracking mode, Ph.D dissertation (05.13.01), Tula, 2006, 131 p. (In Russian).

The authors declare that there are no conflicts of interest present.