LI Jiawei1, 2, SHEN Yongjun1, 2, YANG Shaopu1, 2
1. State Key Lab of Mechanical Behavior and System Safety of Traffic Engineering Structures, Shijiazhuang Tiedao University,
Shijiazhuang 050043, China;
2.Department of Mechanical Engineering, Shijiazhuang Tiedao University, Shijiazhuang 050043, China
Abstract：The gradient descent method is a classical convex optimization algorithm.Here, combined with the theory of fractional-order advanced calculus, the fractional-order gradient descent method was investigated.It was shown that compared with the conventional gradient descent method, the fractional-order gradient descent method with the order larger than 1 has faster convergence speed and lower convergence accuracy, while it with the order less than 1 has higher convergence accuracy and slower convergence speed.In order to combine advantages of different fractional-order gradient descent methods and solve the contradiction between convergence speed and convergence accuracy, 3 improved variable fractional-order gradient descent methods were proposed based on the published study results to obtain better optimization algorithm performances.Typical examples were taken to verify relevant conclusions.
［1］XUE Dingyu, ZHAO Chunna, CHEN Yangquan .Fractional order PID control of a DC-motor with elastic shaft: a case study ［C］∥Proceedings of the 2006 American Control Conference.Minneapolis, 2006.
［2］张小龙, 魏井福, 东亚斌, 等.滚珠自动控制转子系统强迫振动研究(1/2次分数谐波振动)［J］.振动与冲击, 2017, 36(21): 23-27.
ZHANG Xiaolong, WEI Jingfu, DONG Yabin, et al.Forced vibration of the rotor controlled by an automatic ball balancer(1/2 order subharmonic vibration)［J］.Journal of Vibration and Shock, 2017, 36(21): 23-27.
［3］HARTLEY T T, LORENZO C F .Fractional-order system identification based on continuous order-distributions ［J］.Signal Processing,2003,83(11):2287-2300.
［4］林晓然.分数阶非线性系统动力学特性及其图像处理应用研究 ［D］.重庆：重庆大学,2018.
［5］STEPHEN B, LIEVEN V .Convex optimization ［M］.Cambridge: Cambridge University Press, 2004.
［6］WONG Chingchang,CHEN Chiachang .A hybrid clustering and gradient descent approach for fuzzy modeling ［J］.IEEE Transactions on Systems,Man,and Cybernetics,Part B: Cybernetics,1999,29(6):686-693.
［7］LIN Q, LOXTON R, XU C, et al.Parameter estimation for nonlinear time-delay system with noisy output measurements ［J］.Automatica, 2015, 60: 48-56.
［8］KRETSCHMER F，LEWIS B.An improved algorithm for adaptive filters processing ［J］.IEEE Transactions on Aerospace and Electronic Systems,1979, 1(14): 172-177.
［9］BALLA-ARABE S, GAO X B, WANG B.A fast and robust level set method for image segmentation using fuzzy clustering and lattice Boltzmann method［J］.IEEE Transactions on Cybernetics, 2013, 43(3): 910-920.
［10］VAUDREY M A, BAUMANN W T, SAUNDERS W R.Stability and operating constrains of adaptive LMS-based feedback control ［J］.Automatica,2003, 39(4): 595-605.
［11］LIN J Y, LIAO C W.New IIR filter-based adaptive algorithm in active noise control applications:commutation error-introduced LMS algorithm and associated convergence assessment by a deterministic approach ［J］.Automatica, 2008, 44(11): 2916-2922.
［12］PU Y F, ZHOU J L, ZHANG Y, et al.Fractional extreme value adaptive training method:fractional steepest descent approach ［J］.IEEE Transactions on Neural Networks and Learning Systems, 2015, 26(4): 653-662.
［13］CHEN Y Q, GAO Q, WEI Y H, et al.Study on fractional order gradient methods ［J］.Applied Mathematics and Computation, 2017, 314: 310-321.
［14］程松松.分数阶LMS自适应滤波算法研究［D］.合肥：中国科学技术大学，2018.
［15］李大字, 卢婷, 靳其兵.基于分数阶梯度下降法的分数阶非线性系统预测控制［J］.系统仿真技术, 2017, 13(2): 127-131.
LI Dazi, LU Ting, JIN Qiping.Predictive control with fractional gradient descent for fractional-order nonlinear system ［J］.System Simulation Technology, 2017, 13(2): 127-131.
［16］CHENG S S, WEI Y H, CHEN Y Q, et al.An innovative fractional order LMS based on variable initial value and gradient order［J］.Signal Processing, 2017, 133: 260-269.
［17］CUI R, WEI Y H, CHEN Y Q, et al.An innovative parameter estimation for fractional-order systems in the presence of outliers ［J］.Nonlinear Dynamics, 2017, 89(1): 453-463.
［18］CHENG S S, WEI Y H, SHENG D, et al.Identification for Hammerstein nonlinear ARMAX systems based on multi-innovation fractional order stochastic gradient ［J］.Signal Processing, 2018, 142: 1-10.
［19］
CORLISS G, LOWERY D.Choosing a stepsize for Taylor series methods for solving ODE'S［J］.Journal of Computational and Applied Mathematics ,1977,3(4): 251-256.
［20］WEI Y H,GAO Q,CHEN Y Q,et al.Infinite series representation of fractional calculus［C］∥2019 Chinese Automation Congress.Hangzhou: CAC, 2019.
