具有初态偏移的分数阶PID型迭代学习控制
PID-Type Fractional Order Iterative Learning Control with Initial State Shift
摘要: 针对分数阶线性连续时不变系统的初值问题,提出了一种基于初值学习的PI1αDα型迭代学习控制算法。在 λ 范数的意义下,对控制算法的收敛性条件进行了严格证明。理论分析表明,在此算法的作用下,随着迭代次数的增加,能够实现系统输出对期望输出的精确跟踪,保证了跟踪误差的收敛性。相比传统的PID型算法,该算法解决了算法中要求每一次迭代初值都相同的限制,消除了随机初值对系统的影响。数值仿真验证了所提算法的有效性和正确性。
Abstract: For the initial value problem of fractional-order linear continuous-time invariant systems, a PI1αDα iterative learning control algorithm based on initial value learning is proposed, in the sense of λ norm, a rigorous proof of convergence conditions for the control algorithm is established. Theoretical analysis demonstrates that as the number of iterations increases, this algorithm achieves precise tracking of the desired system output and guarantees convergence of tracking errors. Compared with traditional PID-type algorithms, the proposed method overcomes the constraint of requiring identical initial values in each iteration inherent to conventional control algorithms, effectively eliminating the impact of random initial values on system performance. Numerical simulations validate the effectiveness and correctness of the proposed algorithm.
文章引用:高帆. 具有初态偏移的分数阶PID型迭代学习控制[J]. 理论数学, 2025, 15(7): 61-69. https://doi.org/10.12677/pm.2025.157207

1. 引言

迭代学习控制(Iterative Learning Control,简称ILC)是一种模仿人类通过学习经验获取知识的智能技术,适用于具有重复运动特性的被控系统[1] [2]。其基本机理是利用系统当前次运行的输出与期望轨线的误差,修正当前次控制输入,从而得到系统下次运行的控制输入,目的是使控制系统的跟踪性能不断得以改善。ILC特别适合高精度、强重复性的工程场景,尤其是在传统控制方法难以应对非线性、周期性扰动的系统中表现突出。ILC研究成果突飞猛进,并广泛应用于光盘驱动系统、自动驾驶汽车、数控机床加工、半导体制造、航空航天、3D打印等领域[3]-[12]

另一方面,在现实世界中,实际系统大多数是分数阶的,与整数阶相比,分数阶模型能够更好地揭示带有分数阶特性的对象的本质及其行为,获得更优良的控制性能[13]。近年来,分数阶迭代学习控制理论的研究受到国内外学者的青睐[14]-[18]。文献[19]提出了Dα型分数阶迭代学习控制算法,在频域内讨论了算法的收敛性,并将迭代学习控制的应用范围推广到分数阶系统,即分数阶迭代学习控制。文献[20]在时域中讨论了分数阶线性系统迭代过程的收敛性,分析了卷积形式的时变系统的分数阶迭代学习控制,提出了广义分数阶时变系统。之后,文献[21]给出了一类受类齿隙迟滞影响的参数化分数阶系统的自适应迭代学习控制算法,并讨论了其收敛性和稳定性。文献[22]研究了具有非置换常系数矩阵的分数阶脉冲时滞系统的迭代学习控制算法。文章[22]探讨了在一个领导者和固定拓扑约束下,具有重复运动的多智能体机器人系统的一致性控制。

然而,具有初态偏移的分数阶系统的迭代学习控制研究甚少,在现有文献中,大部分都假设所研究的系统在每一次迭代过程中初值与期望的初值相同。例如,文献[23]针对线性分数阶奇异系统提出了P型算法,针对时滞系统给出了PDα型算法,但这些算法都没有考虑初态偏移对系统造成的影响。另外,文献[24]在研究基于频率分析的时滞分数阶线性系统迭代学习控制问题时,也将每次迭代初值假设为期望初值设计算法。而在实际的应用中,很难保持每一次迭代初值与期望初值相同。目前,只有少量文献对分数阶系统的初值问题进行了初步研究。而且主要是研究P型或PDα型算法,文献[25]针对一类具有任意初始状态的分数阶线性连续系统,提出了一种具有初始状态学习的开环和开闭环PDα型分数阶迭代学习控制算法。在Lebesgue-p范数的意义下,利用卷积积分的广义Young不等式在迭代域中给出了PDα型算法收敛的充分条件。文献[26]针对分数阶线性时不变系统的随机初值问题,提出了基于初值学习的PDα型分数阶迭代学习控制算法,利用 λ 范数对控制算法的收敛条件进行了严格证明。

但是PDα型控制器可能会有稳态误差,而PIDα型控制器在响应快、稳定性好的同时能够消除稳态误差,功能更加全面,适合高精度控制,相比于PDα型算法,PIDα型迭代学习控制算法在实际应用中更适合重复操作场景。引入初值学习后,在完成高精度任务和含不确定性的重复任务时有更显著的优势。因此,本文针对具有初态偏移的线性分数阶重复系统,提出了一种基于初值学习的PI1−αDα型分数阶迭代学习控制算法,理论证明了系统跟踪误差单调收敛到零。数值仿真验证了该算法的有效性。

2. 预备知识

定义1 对于连续向量函数 f:[ 0,T ] R n ( f( t )= [ f 1 ( t ), f 2 ( t ),, f n ( t ) ] T ) λ 范数为 f( t ) λ = sup 0tT e λt f( t ) λ>0

引理1 [27] 初值问题 { D τ C t α x( t )=Ax( t )+Bu( t ) x( t 0 )= x 0 的解为

x( t )= Φ α,1 ( t ) x 0 + t 0 t Φ α,α ( tτ )Bu( τ )dτ

其中, A C n×n B C n×p 0<α<1

定义2 函数 f( t ) [ 0,T ] 上的 α 阶分数阶微分的定义为

0 C D T α f( t )= 1 Γ( nα ) 0 T ( tτ ) nα1 f n ( τ )dτ

其中, α R + n1α<n n 为正整数。 Γ( ) 为Gamma函数, Γ( α )= 0 exp( t ) t α1 dt

定义3 对于任意可积函数 f( t ) ,当 α>0 时,左侧分数阶积分和右侧分别定义为

I 0 t α f( t )= 1 Γ( α ) 0 t f( τ ) ( tτ ) 1α dτ , τ[ 0,t )

I t T α f( t )= 1 Γ( α ) t T f( τ ) ( tτ ) 1α dτ , τ[ t,T )

定义4 双参数的Mittag-Leffler函数定义为 E α,β ( Z )= k=0 Z k Γ( αk+β ) α>0 β>0 Z C n×n ,其在分数阶积分中的作用非常重要。

特别地,当 β=1 时,单参数Mittag-Leffler函数定义为

E α,1 ( Z )= E α ( Z )= k=0 Z k Γ( αk+1 ) , α>0 , Z C n×n

引理2 如果函数 f( t ) g( t ) 在区间 [ 0,T ] 上连续,且 D t C T α f( t ) D 0 C t α g( t ) 存在,则

0 T ( D t C T α f( t ) )g( t )dt = 0 T f( t )( D t C T α g( t ) )dt

引理3 [28] Φ α,β ( t )= t β1 E α,β ( A t α ) t[ 0,+ ] ,其中 α>0 β>0 A C n×n ,则函数 Φ α,β ( t ) 具有以下性质:

(i) D τ C t 1α Φ α,1 ( tτ )= Φ α,α ( tτ ) 0<α<1

(ii) d dτ Φ α,1 ( tτ )= Φ α,α ( tτ )A α>0 A C n×n

引理4 Φ α,β ( t )= t β1 E α,β ( A t α ) t[ 0,+ ] ,其中 α>0 β>0 A C n×n ,则 Φ α,1 ( t ) c 0 e A 1 α t Φ α,α ( t ) c 1 e A 1 α t e α ( t ) e A 1 α T =ξ ,所以 Φ α,1 ( t ) c 0 ξ Φ α,α ( t ) c 1 ξ ,其中, c 0 = 1 α c 1 = 1 α A 1α α 0<α<1

3. 问题描述及分析

考虑一类分数阶线性时不变系统

{ D 0 C t α x k ( t )=A x k ( t )+B u k ( t ) y k ( t )=C x k ( t ) (1)

其中 t[ 0,T ] α( 0,1 ) x k ( 0 ) R n u k ( t ) R p y k ( t ) R q 分别为系统第 k 次重复操作的状态向量、控制输入向量和输出向量, A R n×n ,B R n×p C R q×n 都为常数矩阵。

下面给出分数阶系统的一些基本假设。

假设1 分数阶线性时不变系统的期望输出 y d ( t ) [ 0,T ] α 阶微分存在,对于给定的 y d ( t ) ,有唯一期望控制输入 u d ( t ) 和理想状态 x d ( t ) 满足

{ D 0 C t α x d ( t )=A x d ( t )+B u d ( t ) y d ( t )=C x d ( t ) (2)

假设2 CB 为行满秩矩阵。

在现有文献中,大部分都假设所研究的分数阶系统在每一次迭代过程中初态相同,即初态可重置。然而在实际工程应用中,初态偏移是一个常见的问题。

针对系统的控制输入和系统初值,本文设计如下具有初值学习的一阶PI1−αDα型分数阶迭代学习控制算法:

{ u k+1 ( t )= u k ( t )+ L p e k ( t )+ L i I 0 t 1α e k ( t )+ L d D 0 t α e k ( t ) x k+1 ( 0 )= x k ( 0 )+B L d e k ( 0 ) (3)

其中 e k ( t )= y d ( t ) y k ( t ) 为第 k 次迭代学习时对应的跟踪误差, L p L i L d 为学习增益矩阵, x k ( 0 ) 表示初值。

4. 收敛性分析

定理1 当初值学习的PI1−αDα型迭代学习控制算法(3)作用于分数阶线性时不变系统(1)时,若满足条件: ICB L d <1 ,则当 k 时,系统输出 y k ( t ) 一致收敛于期望输出 y d ( t ) ,即 lim k y k ( t )= y d ( t )

证明 根据引理1,由系统(1),有

x k ( t )= Φ α,1 ( t ) x k ( 0 )+ 0 t Φ α,α ( tτ )B u k ( τ )dτ . (4)

由式(3)可得

e k+1 ( t )= y d ( t ) y k+1 ( t ) = e k ( t )C Φ α,1 ( t )( x k+1 ( 0 ) x k ( 0 ) )C 0 t Φ α,α ( tτ )B( u k+1 ( τ ) u k ( τ ) )dτ = e k ( t )C Φ α,1 ( t )B L d e k ( 0 )C 0 t Φ α,α ( tτ )B L p e k ( τ )dτ C 0 t Φ α,α ( tτ )B L d D 0 C t α e k ( τ )dτ C 0 t Φ α,α ( tτ )B L i I 0 t 1α e k ( τ )dτ . (5)

根据引理2和引理3,得

0 t Φ α,α ( tτ )B L d D 0 C t α e k ( τ )dτ = 0 t D τ C t 1α Φ α,1 ( tτ )B L d D 0 C t α e k ( τ )dτ

= Φ α,1 ( tτ )B L d e k ( τ )| 0 t 0 t d dτ ( Φ α,1 ( tτ ) )B L d e k ( τ )dτ =B L d e k ( t ) Φ α,1 ( t )B L d e k ( 0 )+ 0 t Φ α,α ( tτ )AB L d e k ( τ )dτ (6)

另外,公式(5)中

C 0 t Φ α,α ( tτ )B L i I 0 t 1α e k ( τ )dτ =C 0 t D τ T 1α ( Φ α,1 ( tτ ) )B L i I 0 t 1α e k ( τ )dτ =C 0 t Φ α,1 ( tτ )B L i D τ T 1α I 0 t 1α e k ( τ )dτ =C 0 t Φ α,1 ( tτ )B L i e k ( τ )dτ (7)

将(6)和(7)两个式子代入式(5),我们可以得到

e k+1 ( t )= e k ( t )C Φ α,1 ( t )B L d e k ( 0 )CB L d e k ( t )+C Φ α,1 ( t )B L d e k ( 0 ) C 0 t Φ α,α ( tτ )AB L d e k ( τ )dτ C 0 t Φ α,α ( tτ )B L i I 0 t 1α e k ( τ )dτ =( ICB L d ) e k ( t )C 0 t Φ α,α ( tτ )( B L p +AB L d ) e k ( τ )dτ C 0 t Φ α,1 ( tτ )B L i e k ( τ )dτ (8)

t=0 代入式(8)可得

e k+1 ( 0 )=( ICB L d ) e k ( 0 ) (9)

将式(9)两边同时取范数,得到

e k+1 ( 0 ) ( ICB L d ) e k ( 0 )

根据定理1中假设 ICB L d <1 lim k e k+1 ( 0 ) =0

对式(8)两边同时取范数,整理可得

e k+1 ( t ) ICB L d e k ( t ) +a 0 t e k ( τ ) dτ (10)

其中 a= a 1 + a 2 ,根据引理4,上式中 a 1 = c 1 ξ C B L p +AB L d a 2 = c 0 ξ C B L i

将式(8)两端同乘以 e λt ,并计算上确界,整理可得

e k+1 ( t ) λ ICB L d e k ( t ) λ +a sup 0tT 0 t e λ( tλ ) e λτ e k ( τ ) dτ ICB L d e k ( t ) λ + a λ e k ( t ) λ =β e k ( t ) λ (11)

其中, β= ICB L d + a λ

根据定理1中条件可知,存在一个足够大的 λ ,使得 β<1 。故 lim k e k ( t ) λ =0 。证明完毕。

5. 数值仿真

考虑如下分数阶线性时不变系统:

{ D 0 C x t α k ( t )=[ 0.1 0.4 0.2 0.1 ] x k ( t )+[ 0 0.2 ] u k ( t ) y k ( t )=[ 0.2 1 ] x k ( t )

其中,系统的运行区间为 t[ 0,1 ]

给定期望输出为

y d ( t )=12 t 2 ( 1t ),t[ 0,1 ]

为了验证控制算法对初值学习的敏感性,首次迭代时的初值 x( 0 ) 是利用rand函数在区间 [ 0,1 ] 上随机选择的,不等于期望初值 x d ( 0 )

在初值学习的开环PI1−αDα型控制算法中, L p , L i , L d 的参数分别设置为0.9、1.8、0.8, α=0.5 ,通过计算可得 ICB L d =0.84<1 ,满足定理1中收敛性条件。

Figure 1. The iterative outputs of the PI1−αDα-type control algorithm at the 7th and 16th iterations

1. PI1−αDα型控制算法在第7、16次的迭代输出

Figure 2. The iterative outputs of the PI1−αDα-type control algorithm with initial value learning at the 7th and 16th iterations

2. 初值学习的PI1−αDα型控制算法在第7、16次的迭代输出

当传统的PI1−αDα型控制算法被应用于分数阶线性时不变系统时,给定的期望输出 y d ( t ) 以及第5、10、60次迭代学习时的实际输出如图1所示,系统的跟踪误差变化曲线趋势如图2所示。可以看出,针对具有任意初始状态的系统,在传统的PI1−αDα型控制算法作用下,系统输出与期望输出之间存在偏差;随着迭代次数的增加,在Lebesgue-2范数的意义下,系统在整个区间 [ 0,1 ] 上的跟踪误差都收敛有界。

将提出的PI1−αDα型控制算法(3)应用于具有任意初值的系统时,系统第5、10、60次输出如图3所示,跟踪误差曲线如图4所示。显然,系统的输出可以完全跟踪期望输出,在Lebesgue-2范数的意义下,跟踪误差随着迭代次数的增加收敛到零。以上结果表明,对于具有系统任意初值的系统,所提出的算法是可行和有效的。

Figure 3. The tracking error variation trend of the PI1−αDα-type control algorithm

3. PI1−αDα型控制算法的跟踪误差变化趋势

Figure 4. The tracking error variation trend of the PI1−αDα-type control algorithm with initial value learning

4. 初值学习的PI1−αDα型控制算法的跟踪误差变化趋势

由仿真效果可知,较之传统的PI1−αDα型控制算法,论文基于初值学习的PI1−αDα控制算法改进效果十分明显,消除了随机初值对系统的不良影响,跟踪效果更好。

参考文献

[1] Gao, S., Song, Q., Jiang, H., Shen, D. and Lv, Y. (2025) Decentralized Learning Control for High-Speed Trains with Unknown Time-Varying Speed Delays. Applied Mathematical Modelling, 137, Article 115711.
https://doi.org/10.1016/j.apm.2024.115711
[2] Xiao, T. (2024) Iterative Dissipativity of Partial Difference Equation Dynamics in Open-Loop Iterative Learning Control Mode. Mathematics, 12, Article 3128.
https://doi.org/10.3390/math12193128
[3] Jin, X. (2016) Iterative Learning Control for Non-Repetitive Trajectory Tracking of Robot Manipulators with Joint Position Constraints and Actuator Faults. International Journal of Adaptive Control and Signal Processing, 31, 859-875.
https://doi.org/10.1002/acs.2734
[4] Kim, D.I. and Kim, S. (1996) An Iterative Learning Control Method with Application for CNC Machine Tools. IEEE Transactions on Industry Applications, 32, 66-72.
[5] Yang, D.R., Lee, K.S., Ahn, H.J., et al. (2003) Experimental Application of a Quadratic Optimal Iterative Learning Control Method for Control of Wafer Temperature Uniformity in Rapid Thermal Processing. IEEE Transactions on Semiconductor Manufacturing, 16, 36-44.
https://doi.org/10.1109/tsm.2002.807740
[6] Gorinevsky, D. (2002) Loop Shaping for Iterative Control of Batch Processes. IEEE Control Systems, 22, 55-65.
[7] Mezghani, M., Roux, G., Cabassud, M., Le Lann, M.V., Dahhou, B. and Casamatta, G. (2002) Application of Iterative Learning Control to an Exothermic Semibatch Chemical Reactor. IEEE Transactions on Control Systems Technology, 10, 822-834.
https://doi.org/10.1109/tcst.2002.804117
[8] Saab, S.S. (2004) A Stochastic Iterative Learning Control Algorithm with Application to an Induction Motor. International Journal of Control, 77, 144-163.
https://doi.org/10.1080/00207170310001646282
[9] Trubin, M.V. and Yurkevich, V.D. (2024) Iterative Learning Control for an Electrohydraulic Strength Test Bench. Optoelectronics, Instrumentation and Data Processing, 60, 420-425.
https://doi.org/10.3103/s8756699024700481
[10] He, Z. and Chen, Z. (2024) Iterative Learning Control for Non-Normal and Biased Measured Targets. Proceedings of the Institution of Mechanical Engineers, Part I: Journal of Systems and Control Engineering, 238, 1450-1458.
https://doi.org/10.1177/09596518241236977
[11] Zhang, C., Li, S. and Zhang, Z. (2024) Industrial Robot Arm Dynamic Modeling Simulation and Variable-Gain Iterative Learning Control Strategy Design. Journal of Mechanical Science and Technology, 38, 3729-3739.
https://doi.org/10.1007/s12206-024-0644-5
[12] Zhao, H., Wang, F., Zhang, Y., Zheng, Z., Ma, J. and Fu, C. (2024) Fractional-Order Modeling and Stochastic Dynamics Analysis of a Nonlinear Rubbing Overhung Rotor System. Fractal and Fractional, 8, Article 643.
https://doi.org/10.3390/fractalfract8110643
[13] Bensidhoum, T. and Bouakrif, F. (2020) Adaptive P-Type Iterative Learning Radial Basis Function Control for Robot Manipulators with Unknown Varying Disturbances and Unknown Input Dead Zone. International Journal of Robust and Nonlinear Control, 30, 4075-4094.
https://doi.org/10.1002/rnc.4988
[14] Zhang, X., Xu, W.B. and Lu, W.R. (2021) The Control Strategy of Manipulator Based on Fractional-Order Iterative Learning. Automatic Control and Computer Sciences, 55, 368-376.
https://doi.org/10.3103/s014641162104009x
[15] Wu, H., Huang, J., Wu, K., Lopes, A.M. and Chen, L. (2024) Precise Tracking Control via Iterative Learning for One-Sided Lipschitz Caputo Fractional-Order Systems. Mathematical Biosciences and Engineering, 21, 3095-3109.
https://doi.org/10.3934/mbe.2024137
[16] Luo, D., Wang, J. and Shen, D. (2020) Iterative Learning Control for Locally Lipschitz Nonlinear Fractional-Order Multi-Agent Systems. Journal of the Franklin Institute, 357, 6671-6693.
https://doi.org/10.1016/j.jfranklin.2020.04.032
[17] Luo, Y., Liu, C. and Liu, G. (2019) Consensus Tracking by Iterative Learning Control for Linear Heterogeneous Multiagent Systems Based on Fractional-Power Error Signals. Algorithms, 12, 185.
https://doi.org/10.3390/a12090185
[18] Ghasemi, I., Ranjbar Noei, A. and Sadati, J. (2016) Sliding Mode Based Fractional-Order Iterative Learning Control for a Nonlinear Robot Manipulator with Bounded Disturbance. Transactions of the Institute of Measurement and Control, 40, 49-60.
https://doi.org/10.1177/0142331216659130
[19] Chen, Y.Q. and Moore, K.L. (2001) On D α-Type Iterative Learning Control. Proceedings of the 40th IEEE Conference on Decision and Control, Orlando, 4-7 December 2001, 4451-4456.
[20] Zhao, Y., Li, Y. and Liu, H. (2022) Fractional-Order Iterative Learning Control with Nonuniform Trial Lengths. International Journal of Control, Automation and Systems, 20, 3167-3176.
https://doi.org/10.1007/s12555-021-0536-8
[21] Aydin, M. and Mahmudov, N.I. (2022) Iterative Learning Control for Impulsive Fractional Order Time‐Delay Systems with Nonpermutable Constant Coefficient Matrices. International Journal of Adaptive Control and Signal Processing, 36, 1419-1438.
https://doi.org/10.1002/acs.3401
[22] Wang, G.X., Wang, R., Yi, D.H., et al. (2024) Iterative Learning Formation Control via Input Sharing for Fractional-Order Singular Multi-Agent Systems with Local Lipschitz Nonlinearity. Fractal and Fractional, 8, 347.
https://doi.org/10.3390/fractalfract8060347
[23] Gao, Y.H., Lou, W.D., Lu, H.L., et al. (2020) Consensus Control of Multi-Agent Robot System with State Delay Based on Fractional-Order Iterative Learning Control Algorithm. Journal Européen des Systèmes Automatisés, 53, 771-779.
https://doi.org/10.18280/jesa.530603
[24] Liu, Q. and Tian, S.P. (2022) Iterative Learning Control Analysis for Linear Fractional-Order Singular Systems. International Journal of Control, Automation and Systems, 20, 3951-3959.
https://doi.org/10.1007/s12555-021-0682-z
[25] Wang, Y.Q., Zhou, F.G., Yin, L., et al. (2021) Iterative Learning Control for Fractional Order Linear Systems with Time Delay Based on Frequency Analysis. International Journal of Control, Automation and Systems, 19, 1588-1596.
https://doi.org/10.1007/s12555-019-0295-y
[26] Lan, Y.H. and Cui, Z.M. (2018) ILC with Initial State Learning for Fractional Order Linear Distributed Parameter Systems. Algorithms, 11, Article 85.
https://doi.org/10.3390/a11060085
[27] Kumar, S., Yadav, P. and Singh, K.V. (2024) Product Integration Techniques for Fractional Integro-Differential Equations. Mathematical Methods in the Applied Sciences, 48, 2833-2858.
https://doi.org/10.1002/mma.10464
[28] Gao, Z.Y., Liu, S.D. and Wang, J.R. (2017) PID-Type Iterative Learning Control for Impulsive Ordinary Differential Equations. Journal of Applied Mathematics and Computing, 54, 41-55.
https://doi.org/10.1007/s12190-016-0995-x