Birinci mertebe artı zaman gecikmeli sistem dinamiğinin gradyan iniş yöntemine dayali adaptif IIR filtreler ile modellenmesi ve farkli zaman gecikmeleri durumlari için performans analizleri

Öz

Bu çalışmada makine öğrenmesi uygulamalarında sıklıkla kullanılan ve popüler bir nümerik optimizasyon yöntemi olan Gradyan İniş (GD) yöntemine dayalı adaptif sonsuz impuls cevabı (IIR) filtresi ile Birinci Mertebe Zaman Gecikmeli (FOPTD) sistem dinamiğinin zaman bölgesinde giriş-çıkış verisi yardımı ile modellemesi incelenmiştir. FOPTD dinamik sistem modelleri kontrol sitemlerinin modellemesinde kullanılan en temel sistem modelidir. Çalışmada, IIR filtre katsayıları, aynı giriş işareti için IIR filtre cevabının FOPTD dinamik sistem modelinin cevabına yakınsaması için GD yöntemi ile online optimize edilmiştir. Aynı giriş için IIR filtre çıkışının, FOPTD dinamik sistemin çıkışına uzaklığı anlık karesel hata fonksiyonu ile ifade edilmiş ve bu fonksiyonun özyinelemeli gradyan iniş çözümleri FOPTD sistem cevabı ile tasarlanan adaptif IIR filtre cevabı arasındaki çıkış uyumsuzluğunu minimize etmek için kullanılmıştır. Böylece, zaman bölgesinde IIR filtrenin bir FOPTD dinamik sistemin giriş-çıkış dinamiğine yakınsaması GD yöntemi ile elde edilen özyinelemeli filtre katsayı çözümleri ile sağlanmıştır. Adaptif IIR filtre çözümlerinin FOPTD sistemlerin online modellemesinde uygulaması MATLAB-Simulink ortamında gerçekleştirilmiş. Geliştirilen simülasyon ortamında FOPTD dinamik sistemin giriş ve çıkışlarından alınan işaretler, GD optimizasyon bloğunda IIR filtre katsayılarının online olarak optimize edilmesinde kullanılmış. Bu simülasyon ortamında IIR filtre cevabının FOPTD plant modelinin zaman gecikmeli sistem dinamiğine yakınsama performansı farklı zaman gecikme değerleri için incelenmiştir.

Anahtar Kelimeler

• Sistem modelleme
• Doğrusal olmayan optimizasyon
• Gradyan iniş yöntemi
• Adaptif IIR filtre
• Dinamik sistem

Modeling of first order plus time delay system dynamics with adaptive IIR filters based on gradient descent methods and performance analyses for different time delay cases

Öz

In this study, the modeling of First Order Plus Time Delay (FOPTD) dynamics by using adaptive infinite impulse response (IIR) filter based on Gradient Descent (GD) method, which is frequently used in machine learning applications, has been investigated by the help of the input-output data in the time domain. The First Order Time Delay (FOPTD) dynamic system models are the most basic system model that is used in the modeling of control systems. In the study, the IIR filter coefficients are optimized online by using the GD method for convergence of the IIR filter response to the FOPTD dynamic system model response for the same input signal. The distance of the IIR filter output to the output of the FOPTD dynamic system for the same input is expressed by the instant square error function and, recursive GD solutions of this function are used to minimize output mismatches between FOPTD system model and the proposed adaptive IIR filter. Thus, the convergence of the IIR filter to the input-output dynamics of a FOPTD dynamic system is provided in the time domain by performing recursive filter coefficient solutions that are obtained by the GD method. An application of the adaptive IIR filter solutions in the online modeling of FOPTD systems was carried out in MATLAB-Simulink environment. In the developed simulation environment, the collected signals from the inputs and outputs of the FOPTD dynamic system were used to online optimize the IIR filter coefficients in the GD optimization block. In this simulation environment, the convergence performance of the IIR filter response for the time delay system dynamics of the FOPTD plant model is investigated for different time delay values.

Anahtar Kelimeler

• System modeling
• Nonlinear optimization
• Gradient descent method
• Adaptive IIR Filter
• Dynamic system

Kaynakça

1. [1] Fletcher R. Practical Methods of Optimization. 2nd Ed. New York, USA, Wiley, 2000.
2. [2] Tan RK, Bora Ş. “Parameter optimization and used techniques in modeling and simulation”. Journal of Engineering Sciences and Design, 5(3), 685-697, 2017.
3. [3] Lopez D, Manogaran G. “Health data analytics using scalable logistic regression with stochastic gradient descent”. International Journal of Advanced Intelligence Paradigms, 10(1-2), 118-132, 2018.
4. [4] Duchi J,Hazan E,Singer Y. “Adaptive subgradient methods for online learning and stochastic optimization". Journal of Machine Learning Research, 12(7), 2121-2159, 2011.
5. [5] Alagoz BB, Tepljakov A, Ates A, Petlenkov E, Yeroglu C. “Time-domain identification of one noninteger order plus time delay models from step response measurements”. International Journal of Modeling, Simulation, and Scientific Computing, 10(1), 1-22, 2019.
6. [6] Galán O, Romagnoli JA, Palazoǧlu A, Arkun Y. “Gap metric concept and implications for multilinear model-based controller design”. Industrial & Engineering Chemistry Research, 42(10), 2189-2197, 2003.
7. [7] Ruder, S. “An Overview of Gradient Descent Optimization Algorithms”. arXiv, 2017. https://arxiv.org/pdf/1609.04747.pdf.
8. [8] Amari SI. “Backpropagation and stochastic gradient descent method”. Neurocomputing, 5(4-5), 185-196, 1993.

9. [9] Psaltis D, Sideris A, Yamamura A. “A multilayered neural network controller”. IEEE Control Systems Magazine, 8(2), 17-21, 1988.
10. [10] Amasyali MF. “Mini‐batching for artificial neural network training “. TBV Journal of Computer Science and Engineering, 8(1), 25-34, 2015.
11. [11] Xue Y, Wang Y, Liang J. “A self-adaptive gradient descent search algorithm for fully-connected neural networks”. Neurocomputing, 478, 70-80, 2022.
12. [12] Alagoz BB, Alisoy HH, Koseoglu M, Alagoz S. “Modeling and analysis of dielectric materials by using gradient- descent optimization method”. International Journal of Modeling, Simulation, and Scientific Computing, 8(1), 3-45, 2017.
13. [13] Long B, Zhu Z, Yang W, Chong KT, Rodríguez J, Guerrero JM. “Gradient descent optimization based parameter identification for FCS-MPC control of LCL-type grid connected converter”. IEEE Transactions on Industrial Electronics, 69(3), 2631-2643, 2021.
14. [14] Boukis C, Mandic D, Papoulis E, Constantinides A. “A gradient adaptive step size algorithm for IIR filters”. 2003 IEEE International Conference on Acoustics, Speech, and Signal Processing, Proceedings. (ICASSP 03), Hong Kong, China, 06-10 April 2003.
15. [15] Widrow B, Stearns SD. Adaptive signal processing. 1nd ed. Englewood Cliffs, 1985.
16. [16] Guo W, Zhi Y. “Nonlinear spline adaptive filtering against non-gaussian noise”. Circuits, Systems, and Signal Processing, 41(1), 579-596, 2022
17. [17] Shimizu K, Ito S, Suzuki S. “Tracking Control of General Nonlinear Systems by Direct Gradient Descent Method”. IFAC Proceedings Volumes, 31(17), 183-188, 1998.
18. [18] Vinagre BM, Petráš I, Podlubny I, Chen YQ. “Using fractional order adjustment rules and fractional order reference models in model-reference adaptive control “. Nonlinear Dynamics, 29(1), 269-279, 2002.
19. [19] Alagoz BB, Kavuran G, Ates A, Yeroglu C. “Reference-shaping adaptive control by using gradient descent optimizers”. Plos One, 12(11), 1-20, 2017.
20. [20] Yagmur N, Alagoz BB. “Adaptive gradient descent control of stable, first order, time-delay dynamic systems according to time-varying FIR filter model assumption”. 2019 International Artificial Intelligence and Data Processing Symposium (IDAP), IEEE, Malatya, Turkey, 21-22 September 2019.
21. [21] Yagmur N. Gradyan Inis Yontemi ve Kontrol Sistemlerinde Uygulamalari. Master’s Thesis, Inonu University Computer Engineering Department, Malatya, Turkey, 2020.
22. [22] Balamurali A, Feng G, Lai C, Tjong J, Kar NC. “Maximum efficiency control of PMSM drives considering system losses using gradient descent algorithm based on DC power measurement”. IEEE Transactions on Energy Conversion, 33(4), 2240-2249, 2018.
23. [23] Alagoz BB, Tepljakov A, Kavuran G, Alisoy H. “Adaptive control of nonlinear TRMS model by using gradient descent optimizers”. In 2018 International Conference on Artificial Intelligence and Data Processing (IDAP), Malatya, Turkey, 28-30 September 2018.
24. [24] Yagmur N, Alagoz BB. “Comparision of Solutions of Numerical Gradient Descent Method and Continous Time Gradient Descent Dynamics and Lyapunov Stability”. 2019 27th Signal Processing and Communications Applications Conference (SIU), Sivas, Turkey, 24-26 April 2019.
25. [25] Chen X, Tang B, Fan J, Guo X. “Online gradient descent algorithms for functional data learning”. Journal of Complexity, 70, 1-14, 2022.
26. [26] Li S, Li L, Shi D, Zou W, Duan P, Shi L. “Multi-Kernel maximum correntropy kalman filter for orientation estimation”. IEEE Robotics and Automation Letters, 7(3), 6693-6700, 2022.
27. [27] Hochreiter S, Younger AS, Conwell PR. “Learning to Learn Using Gradient Descent”. International Conference on Artificial Neural Networks, Vienna, Austria, 21-25 August 2001.
28. [28] Pankaj S, Kumar JS, Nema RK. “Comparative analysis of MIT rule and Lyapunov rule in model reference adaptive control scheme”. Innovative Systems Design and Engineering, 2(4), 154-162, 2011.
29. [29] Ian Goodfellow and Yoshua Bengio and Aaron Courville. “Deep Learning, MIT Press”. deeplearningbook.org/contents/numerical.htm (02.07.2023).
30. [30] Panda G, Pradhan PM, Majhi B. “IIR system identification using cat swarm optimization”. Expert Systems with Applications, 38(10), 12671-12683, 2011.
31. [31] Matlab Simulink, “The MathWorks”. https://www.mathworks.com/products/ simulink.html (02.07.2023).