Adaptive robust multivariable control of noninvertible memoryless systems with bounded disturbances: a generalization

The article is dedicated to robust adaptive control in discrete time for certain classes of undefined memoryless multivariate systems with unmeasurable bounded disturbances, the bounds of which are assumed to be known. The systems considered have a number of control inputs that does not exceed the n...

Full description

Saved in:
Bibliographic Details
Published in:Проблемы управления и информатики
Date:2022
Main Authors: Zhiteckii, L.S., Solovchuk, K.Yu.
Format: Article
Language:English
Published: Інститут кібернетики ім. В.М. Глушкова НАН України 2022
Subjects:
Online Access:https://nasplib.isofts.kiev.ua/handle/123456789/210873
Tags: Add Tag
No Tags, Be the first to tag this record!
Journal Title:Digital Library of Periodicals of National Academy of Sciences of Ukraine
Cite this:Adaptive robust multivariable control of noninvertible memoryless systems with bounded disturbances: a generalization / S. ZhiteckiiL, K.Yu. Solovchuk // Проблеми керування та інформатики. — 2022. — № 2. — С. 22-38. — Бібліогр.: 29 назв. — англ.

Institution

Digital Library of Periodicals of National Academy of Sciences of Ukraine
_version_ 1859821176591745024
author Zhiteckii, L.S.
Solovchuk, K.Yu.
author_facet Zhiteckii, L.S.
Solovchuk, K.Yu.
citation_txt Adaptive robust multivariable control of noninvertible memoryless systems with bounded disturbances: a generalization / S. ZhiteckiiL, K.Yu. Solovchuk // Проблеми керування та інформатики. — 2022. — № 2. — С. 22-38. — Бібліогр.: 29 назв. — англ.
collection DSpace DC
container_title Проблемы управления и информатики
description The article is dedicated to robust adaptive control in discrete time for certain classes of undefined memoryless multivariate systems with unmeasurable bounded disturbances, the bounds of which are assumed to be known. The systems considered have a number of control inputs that does not exceed the number of outputs. Стаття присвячена робастному адаптивному керуванню в дискретному часі деякими класами невизначених багатовимірних систем без пам’яті за наявності невимірних обмежених збурень, межі яких припускаються відомими. Розглянуто системи, у яких число керувальних входів не перевищує числа виходів. Основна особливість об’єктів, що підлягають керуванню, полягає у тому, що матриці коефіцієнтів підсилення таких об’єктів не можуть бути обернені. Введено припущення, що елементи цих матриць апріорі невідомі, але є інформація про можливі межі інтервалів, до яких вони належать. Поставлено і вирішено задачу, що полягає в тому, аби у контурі зворотного зв’язку побудувати регулятор, здатний впоратися з матрицями коефіцієнтів підсилення, які не можуть бути обернені, а також з параметричною невизначеністю для пригнічування зовнішніх збурень і забезпечення обмеженості всіх керувань та вихідних сигналів. Для вирішення вищезазначеної задачі використовується робастний адаптивний підхід у поєднанні з так званою концепцією псевдооберненої чи оберненої моделей. Вивчено три різні випадки. У першому випадку побудований робастний адаптивний регулятор, який може застосовуватися до невизначеного об’єкта з виродженою квадратною матрицею коефіцієнтів підсилення. У другому випадку на основі псевдообернених моделей, параметри яких оцінюються за допомогою стандартної рекурентної процедури адаптації, запропоновано робастний метод керування об’єктами з невідомими прямокутними матрицями коефіцієнтів підсилення повного рангу. Запропонований у першому випадку підхід поширюється на третій випадок керування невідомими об’єктами, матриці коефіцієнтів підсилення яких є прямокутними матрицями неповного рангу. Встановлено асимптотичні властивості запропонованих у статті робастно-адаптивних регуляторів. Для підтримки теоретичного дослідження наведено результати числових прикладів.
first_indexed 2026-03-16T12:21:29Z
format Article
fulltext © L.S. ZHITECKII, K.YU. SOLOVCHUK, 2022 22 ISSN 1028-0979 АДАПТИВНЕ КЕРУВАННЯ ТА МЕТОДИ ІДЕНТИФІКАЦІЇ UDC 681.5 L.S. Zhiteckii, K.Yu. Solovchuk ADAPTIVE ROBUST MULTIVARIABLE CONTROL OF NONINVERTIBLE MEMORYLESS SYSTEMS WITH BOUNDED DISTURBANCES: A GENERALIZATION Leonid Zhiteckii International Research and Training Center for Information Technologies and Systems of the National Academy of Sciences of Ukraine and Ministry of Education and Science of Ukraine, Kyiv, leonid_zhiteckii@i.ua Klavdiia Solovchuk Poltava Scientific Research Forensic Center of the MIA of Ukraine, solovchuk_ok@ukr.net The discrete-time robust adaptive control for some classes of the uncertain multivariable memoryless (static) systems in the presence of unmeasurable bounded disturbances, whose bounds are assumed to be known is addressed. The systems where the number of the con- trol inputs do not exceed the number of their outputs are considered. The main feature of plants to be controlled is that their gain matrices are noninvertible. The assumption that the elements of these matrices are unknown a priori but there is information about possible bounds on these elements. The problem stated and solvesd here is to design a feedback con- troller to be capable to cope with the noninvertibility of the gain matrices and also with the parametric uncertainty in order to reject the external disturbances and to ensure the boun- dedness of all the control and output system signals. To solve the problem above mentioned, the robust adaptive approach together with the so-called pseudoinverse or inverse model- based concept is used. Three different cases are studied. In the first case, the robust adaptive controller applicable to the uncertain plant with the square singular gain matrix is designed. The robust method employing the pseudoinverse model-based controllers whose parameters are estimated via a standard recursive adaptation procedure is proposed in the second case to deal with the unknown nonsquare gain matrices having the full rank. The approach pro- posed in first case is extended to the third case dealing with the control of the unknown plants the gain matrices of which represent the nonsquare matrices of not full rank. Asymp- totic properties of the robustly-adaptive controllers proposed in this paper are established. Results of numerical examples given to support the theoretic study. Keywords: discrete time, multivariable memoryless plant, noninvertibility, pseudoinverse model-based concept, uncertainty, estimation algorithm, robust adaptive control. Introduction A long-standing in [1] problem of the optimal controller design for multivariable system in the presence of unmeasurable disturbances remains an important problem from both theoretical and practical points of view. Within the framework of this actual problem, new approaches have been proposed by many researches. The latest results in this scientific area have been reported in numerous papers including [2] and generalized in several recent books [3–6] dealing with advanced multivariable control systems. mailto:leonid_zhiteckii@i.ua mailto:solovchuk_ok@ukr.net Міжнародний науково-технічний журнал «Проблеми керування та інформатики», 2022, № 2 23 Among other methods advanced in the modern control theory, the inverse model- based method that is an extension of the well-known internal model principle seems to be perspective in order to cope with arbitrary unmeasurable disturbances and to opti- mize some classes of multivariable control systems. It turned out that this method first intuitively devised in [7] makes it possible to optimize the closed-loop control system containing the multivariable static (memoryless) plants whose gain matrices are square and nonsingular. Since the beginning of the 21st century, a significant progress has been achieved utilizing the inverse model-based approach [8, 9]. However, this approach is quite unacceptable if the gain matrices are either square but singular or nonsquare be- cause they are noninvertible. To optimize the closed-loop control system containing an arbitrary multivariable static plant, the so-called pseudoinverse (generalized inverse) model-based concept has been proposed and substantiated in [10] dealing with the possible noninvertibility of gain matrices whose elements are assumed to be known. This fruitful concept was ex- tended in [11, 12] to robust control of a noninvertible and uncertain plant with unmeas- urable bounded disturbances. Unfortunately, the robust control theory may not be em- ployed if the initial parametric uncertainty is «wide» enough. Meanwhile, the adaptive approach gives some universal tool to deal with such a type of uncertainty. Foundations of this approach have been extended and generalized in the books [13–19]. In the recent works [20, 21], the different adaptive control ideas are advanced to cope with the noninvertible multivariable memoryless system in the presence of para- metric uncertainties. In particular, the adaptive pseudoinverse model-based control idea using the standard identification method is exploited in [20, 22] to reject unmeasurable bounded disturbances acting on the uncertain plants with the nonsquare gain matrices of full rank. Novel approaches to deal with the multivariable memoryless plants having unknown square and nonsquare gain matrices of not full rank are reported in [21–23]. The purpose of this paper is to generalize the results achieved in [20–23] and relat- ed to the case where there are no nonparametric uncertainties. 1. Problem formulation Let 1n n ny Bu v−= + (1) be the vector-valued difference equation of a static (memoryless) plant represent- ing some linear multivariable discrete-time system to be stabilized. In this equa- tion, ,m r n ny R u R  and m nv R are the m dimensional measured output, con- trol input and unmeasured external disturbance vectors, respectively, at the n th time instant ( 1, 2, )n =  defined by (1) ( ) T (1) ( ) T.[ , , ] ,., .[ , ]m r n n n n n ny y y u u u=  = and (1) ( ) T...[ , , ] ,m n n nv v v= (11) (1 ) ( 1) ( ) r m mr b b B b b     =       (2) is an arbitrary time-invariant m r gain matrix. Consider the case when the number of the control inputs (1) ( ).., ., r n nu u is not less than two but does not exceed the number of the outputs (1) ( ), , m n ny y meaning that 2 .r m  (3) 24 ISSN 1028-0979 Next, suppose that the rank of B satisfies the inequality rank B r (4) implying that B may be not a full rank matrix. Note that the rank of 𝐵 satisfying (4) together with (3) give that B becomes a noninvertible matrix if either r m= but rank B r or r m irrespective of rank .B The following basic assumptions with respect to the gain matrix B and the se- quences ( ) ( )( ) 0 1{ } , , ( 1, , ) i ii nv v v i m=  =  are made. A1) all the elements of 𝐵 are all unknown. However, there are some interval esti- mates defined as ( ) ( ) ( ) , 1, , , 1, , ,ij ij ijb b b i m j r  =  =  (5) where the upper and lower bounds ( )ijb and ( ) ,ijb respectively, on ( )ijb are assumed to be known. A2) ( )s ( =1, ..., )i nv i m are all the arbitrary scalar sequences bounded in modulus according to ( ) ( )     0,1, 2,  ...,i i nv n     = where ( )i s are constant. A3) The upper bounds, ( )i s are known a priori. Let 0 0(1) 0( ) T[ , , ]my y y=  be a desired output vector 0( ) c( onstiy i  = 1, , m=  . Suppose that 0(1) 0( ) 0my y++  implying that, at least, one 0( )iy of 0(1) 0( ), ..., my y is nonzero. Define the output error vector 0 ,n ne y y= − (6) with the components ( ) 0( ) ( ) ,i i i n ne y y= − i.e., (1) ( ) T, ..., ] .[ m n n ne e e= The problem is to design the feedback controller guaranteeing the ultimate bound- edness of the sequence 1 2{ } , , ,ne e e=  in the form lim sup n n e →   (7) provided lim sup .n n u →   (8) Remark. The requirement (8) is here introduced additionally since it may not be satisfied even if (7) takes place. 2. Preliminaries Assume, for the time being, that B is a known noninvertible matrix. In this case, the so-called pseudoinverse control 1n n nu u B e+−= + (9) proposed in several works (see, e.g., [24]) ensures the minimum of the upper bound on the Euclidean norm 2|| ||ne of the output error vector of the closed-loop control sys- tem (1), (6), (9) with any bounded sequence 1 2{ } , ,nv v v=  Here the notation P+ of any pseudoinverse matrix introduced in [24, Theorem 3.4] and defined as Міжнародний науково-технічний журнал «Проблеми керування та інформатики», 2022, № 2 25 T 1 T 0 ,lim ( )rP P P I P+ − → = +  where rI denotes the identity r r matrix is used. Recall that if rank ,P r= for some ( ) ,m rP R  then the expression of P+ is simplified [25, item 7.46]; we have T 1 T.( )P P P P+ −= (10) The closed-loop control system (1), (6), (9) is designed as shown in Fig. 1. In this control system, the variable 1:n n nu u u −= − produced by the pseudoinverse model represents the increment of the control action during one step determined as :n nu B e+= whereas the signal 𝑢𝑛 is formed as the sum 1 . n n k k u u = =  Fig. 1 3. Robustly-adaptive control of plants with square singular gain matrices Let 𝐵 be an unknown square singular r r matrix, i.e., det  0.B = (11) Basic idea to deal with a matrix B satisfying the requirement (11) is to replace adaptive identification of the true plant having the singular gain matrix B to the adaptive identification of a so-called fictitious plant with the nonsingular gain matrix B of the form 0 ,rBB I= +  (12) where rI denotes the identity r r matrix and 0 is a fixed quantity [26]. Although �̃� as well as 𝐵 remain unknown, the requirement det  0B  (13) can always be satisfied by the suitable choice of 0 in the expression (12). In fact, each ith eigenvalue ( )i B of B lies in one of the 𝑟 closed regions of the complex z-plane consist- ing of all the Gerŝgorin discs [27, p. 146]: ( ) ( ) 1 , 1, , . rii ij j j i z b b i r =  −  =  (14) Since, at least, one of the eigenvalues ( )i B is equal to zero (due to the singularity of 𝐵), by virtue of (12) there are an integer 𝑖 (1 ≤ 𝑖 ≤ 𝑟) and the numbers ( ) ( ) ( ) ( ) ( ) 1 ( ) : 1 ,   :i j j j i r ri i ij i i j i ii jb b b b = =   = −  −=  (15) 26 ISSN 1028-0979 such that if ( ) ( )i i  (16) then either ( ) 0i  but ( ) 0i  or ( ) 0i  but ( ) 0.i  These numbers are defined as the intersection of the i th Gerŝgorin disc with the real axis of the complex z-plane as shown in Fig. 2, a and 3, a, respectively. In both cases, ( ) ( ) 0i i  if the inequa- lity (16) is satisfied because ( )i and ( )i cannot have the same sign. Denoting (1) ( ) (1) ( ): min{ , ..., }, : max{ ..., },,r r =    =   (17) consider the following two cases: (i) | | | |;   (ii) | | | |   (The case when | | | | =  can be combined with any of two cases.) In order to go to the gain matrix B of the ficti- tious plant having the form (12) in the case (i), it is sufficient to shift the Gerŝgorin discs (14) right taking 0 | |   (18) as shown in Fig. 2b. In the case (ii), the discs (14) need to be shifted left according to 0 | |;  −  (19) see Fig. 3, b. In both cases, the nonsingularity of B is guaranteed. Nevertheless, the conditions (18) and (19) cannot be satisfied, as yet. In fact, the numbers  and  given by the expressions (17) depend on ( )i s and ( )i s defined by (15). But they are un- known because ( )ijb s are all unknown. а б Fig. 2 а б Fig. 3 Im Z-plane Im Z-plane Re Re Im Z-plane Im Z-plane Re Re Міжнародний науково-технічний журнал «Проблеми керування та інформатики», 2022, № 2 27 The following operations are proposed to choose a number 0 satisfying the reqiurement (13). Introduce ( ) ( ) ( ) ( ) min 1, ( )( ) ( ) ( ) max 1, : max{| |, | |}, : max{| |, | |} r i ii ij ij j j i r iji ii ij j j i b b b b b b =  =   = −  = +   (20) minimizing and maximizing in ( )( ) ( )[ , ] ijij ijb b b the right-hand sides of (15) for ( )i  and ( )i respectively. Further, the number 0 is found to satisfy the conditions 0 max 0 max maxmin min min if | | | |, if | | | |  −          (21) where maxmin ,   represent some quantities defined as follows: (1) ( )(1) ( ) mix n minmin min min ma, max{ }.: min{ , ..., } : , ..., rr    =   = (22) It can be clarified that if the conditions (21) together with (20) and (22) are satis- fied then the condition (13) will without fail be ensured. After determining the quantity 0 we can proceed to the consideration of the ficti- tious plant. Since the input variables (1) ( ), ..., r n nu u and the disturbances (1) ( ), ..., r n nv v of both true plant and fictitious plant are the same, this feature makes it possible to des- cribe our fictitious plant by the equation 1 ,n n ny Bu v−= + (23) similar to the equation (1), where (1) ( ) T[ , ..., ] ,m n n ny y y= denotes the output vector of the fictitious plant. It is interesting that the components of ny can be measured while the components of nv in the equation (23) remain unmeasurable. In fact, substituting the expression (12) into (23), due to (1) we produce 0 1.n n ny y u −= + (24) It is seen from the equation (24) that ny can always be found indirectly having nu and ny to be measured. Now, our problem reduces to the known problem of adaptive control applicable to the fictitious plant (23) with the unknown gain matrix B in the presence of arbitrary bounded disturbances (1) ( ), ..., .r n nv v Its solving follows the steps of the section above. Namely, the adaptive control law is designed in the form 1 1 ,n n n nu u B e− −= + (25) 28 ISSN 1028-0979 in which, instead of the current estimate nB of B, another nB is exploited, and the error vector ne defined in (6) is replaced by 0 n ne y y= − (26) with ny given by the expression (24). The adaptive identification algorithm used to determine the estimates nB may be taken as ( ) *( ) 0 1 ( ) *( ) *( ) ( ) ( ) 11 2 1 2 1, , if | | , sign otherwise, || || i i n in i i in i i n i n n nn n b e b e e b u u i r − −− −      =  −  +  =     (27) which is similar to that in [22]. In this algorithm, 0 i and i are given by 0 , 1, ,2i i i i r = =   (28) ( )T*( ) ( ) 11 ii i n n nne y b u −−= −  (29) represent the i th component of the identification error * ne given as * 1 1,n n n ne y B u− −= −  (30) where ( )( ) ( ) 1: , ii i n n ny y y − = − and the notation ( )T ( 1) ( ): [ , , ]...i i ir n n nb b b= of the i th row of nB is introduced. The coefficients ( )i n s are chosen from the intervals ( )0 2i n        (31) to satisfy det 0.nB  (32) The feedback adaptive robust control system described in the (1), (25), (27) is designed as depicted in Fig. 4. In this figure, the notation * 1 1:n n ny B u− − =  is in- troduced. The asymptotic properties of the adaptive control system are established in the theorem below. Theorem 1. Determine 0 using the formula (21) together with the expres- sions (20) and (22), and choose an arbitrary initial 0 0 0 rB B I= +  with ( ) 0 0( ) ij B b= whose elements satisfy the conditions ( )( ) ( ) 0 . ijij ijb b b  Subject to assumptions A1–A3, the adaptive controller described in the equations (25) and (27) together with (24) and (26) when applied to the plant (1) leads to (7) and (8). Proof. See [26]. □ Міжнародний науково-технічний журнал «Проблеми керування та інформатики», 2022, № 2 29 Fig. 4 4. Robustly-adaptive control of plants with nonsquare gain matrices having full rank Let (1) be the equation of a nonsquare multivariable memoryless system ( 2 r m  ) whose gain matrix, B, as full rank: rank B r= (33) Suppose that ( )i s are known. For controling this system under condition of the parametric uncertainty given by (5), the adaptive controller will be designed here as the adaptive pseudoinverse model-based controller described by 1n n n nu u B e+−= + (34) with nB+ defined by (10). Noting that the requirement rank nB r= (35) will be satisfied to calculate nB+ by (10), we will update the elements of ( )( ),ij n nB b= exploiting the following standard adaptive estimation algorithm proposed in [28, sect. 4.2]: ( ) ( ) ( ) 0( ) 1 ( ) ( ) ( ) ( ) ( ) ( ) 11 2 1 2 1, , if | | , sign { } otherw )ise ( || || i i i i nn i i i i n i i n n n nn n b e b i e e P b u u m − −− −     =  −  −    =  (36) 30 ISSN 1028-0979 In this algorithm, ( ) ( 1) ( ) T...: [ , , ]i i ir n n nb b b= is the i th row of ,nB ( ) { }iP w  repre- sents the projection of w onto the i th set ( 1) ( )( ) ( 1) ( )[ , ] [ , ] i iri i irb b b b =   and ( ) ( ) 0( ) ( )2 , .i i i i =     The coefficients ( )i n s in (36) are chosen from (31) to satisfy (35). The recursive procedure (36) is the on-line identification algorithm need to imple- ment the adaptive pseudoinverse model-based controller (34). The adaptive closed-loop control system containing this controller is shown in Fig. 5. Fig. 5 The convergence and robustness properties of the adaptive pseudoinverse model- based controller (34), (36) are given below. Proposition. Provided that Assumptions A1) to A3) are valid, and the conditi- on (33) is satisfied, the adaptive control algorithm defined in (34), (36) and applied to the plant (1) gives: (i) the estimate sequence { }nB converges in the sense of ;n n B B→ ⎯⎯⎯→ (ii) the ultimate boundedness given in the expressions (7), (8) is achieved. Proof. Due to space limitation, the proof of Proposition is omitted. □ 5. Robustly-adaptive control of plants with nonsquare gain matrices having not full rank Let B be a nonsquare m r matrix of the form (2) with unknown rank satis- fying (4). Define the so-called submatrices 1[ [ ], , [ ] |1, , ] r r rB i k i k r R  [43, part I, subsect. 2.2] whose rows represent the rows of B with the numbers 1[ ], , [ ]ri k i k ( 11 [ ] [ ]ri k i k m    ). The quantity of these matrices is equal to . m N r   =     Deno- ting by [ ]B k the submatrix which corresponds to a k th subset 1{ [ ], , [ ]},ri k i k write the equations of some k plants as: Міжнародний науково-технічний журнал «Проблеми керування та інформатики», 2022, № 2 31 1[ ] [ ] [ ], 1, , ,n n ny k B k u v k k N−= + = (37) where 1( [ ]) ( [ ]) [ ] [ , , ]ri k i k r n n ny y Rk y=  and 1( [ ]) ( [ ]) [ ] [ , , ] .ri k i k r n n nv Rk v v=  In accordance with the approach proposed in the previous section, pass from the equation (37) to the equations of the fictitious plants described by 1[ ] [ ] [ ], 1, , ,n n ny k B k u v k k N−= + = (38) with the same 1nu − and [ ].nv k In these equations, [ ]ny k denotes the r-dimensional output vector related to the k th fictitious plant whose gain matrix [ ]B k is defined as follows: 0[ ] [ ] [ ] ,rB k B k k I= +  (39) where 0[ ]k is a fixed quantity depending on k. This quantity is calculated for each 1, ,k N= using the technique described in the previous section. Namely, with the constraints (5) in mind, 0[ ]k can always be found to satisfy the conditions det [ ] 0 1, ,B k k N  = (40) similar to (32). It follows from the equations (37) to (39) that 0 1[ ] [ ] [ ] .n n ny k y k k u −= + (41) This expression shows that although [ ]B k as [ ]B k remain unknown, however, the components of all N the vectors [ ]ny k can indirectly be «measured» after measuring the components of ny and 1,nu − and it is essential. If the conditions (40) are satisfied, then the problem of the adaptive stabilization of the true plant (1) can be reduced to the problem of simultaneous adaptive stabilization of all N fictitious plants (38) with unknown but nonsingular r r gain matrices [ ]B k ( 1, ,k N= ) via forming at each n th time instant a set of N different «potentially» possible controls [1], , [ ]n nu u N and selecting one of them in accordance with cer- tain choice rule [23] given below. Following [23], the adaptive control law to be applicable to any fictitious plant is designed in the form 1 1[ ] [ ] [ ],n n n nu k u B k e k− −= + 1, , ,k N= (42) where 0[ ] [ ] [ ]n ne k y k y k= − with 10( [ ]) 0( [ ])0 T[ ] [ , ..., ]ri k i k y k y y= defines the output error vector related to the k th fictitious plant at the n th time instant, and [ ] r r nB k R  is the current estimate of unknown r r matrix [ ]B k at the same time instant satisfying det [ ] 0 1, , .nB k k N  = (43) As the adaptation algorithms, the standard recursive procedures for the adaptive identification of each k th fictitious plant (37) described by 32 ISSN 1028-0979 ( ) *( ) 0 1 *( ) *( ) ( )( ) ( ) 11 2 1 2 n 1, , , 1, , [ ] if | [ ]| , [ ] sig [ ] [ ] [ ] otherwise, || || i i n in i i ii i n i n n n nn n b k e k e k e k b k u i r N b k u k − −− −      −  = +      =  =   (44) are proposed. In these algorithms, ( )[ ]i nb k denotes the r-dimensional estimate vector ob- tained by transposing the i th row of [ ],nB k and ( ) ( )T*( ) ( ) 11 1[ ] [ ] [ ] [ ] i ii i n n nn ne k y k y k b k u −− −= − −  (45) represents the scalar variable making sense of the i th component of *( )[ ]i r ne k R that is the identification error vector related to the k th fictitious plant. The coefficients ( )i n s are chosen from the ranges (31) to satisfy the requirement (43). Next, add the adaptation algorithms described in the formulas (44) together with (45) by an algorithm for estimating unknown B defined as follows: ( ) *( ) 0 1 ( ) *( ) *( ) ( ) ( ) 11 2 1 2 if | | , sign otherwis 1, , ,e, || || i i n in i i i n i i n i n n nn n b e b i e e b u m u − −− −     =  −  +    =  (46) where ( )Ti nb represents the i th row of the estimate matrix nB and ( ) ( )T*( ) ( ) 11 1 i ii i n n nn ne y y b u −− −= − −  (47) is the i th component of the identification error vector * 1 1 1n n n n ne y y B u− − −= − −  ( i and 0 i are given by the conditions (28)). The estimation procedure defined in the algorithm (46) together with the equation (47) makes it possible to estimate the m predicted output errors ( ) 1[ ] i ne k+ ( 1, ,i m=  ) for each i th output of true plant (1) at any n using the formula ( ) 0( ) ( )T ( ) .| [ ] | | [ ] | , 1, ,i i i i n n ne k y b u k i m= − + =  (48) The synthesis of the adaptive controller is finished by the choice of the control nu from the set { [1], , [ ]}n nu u N with [ ]nu k given by (42). This choice is implemented by the rule giving the minimum of the 1-norm of (1) ( ) T 1 1 1[ ] [ [ ], , [ ]] m n n ne k e k e k+ + += according to ( ) [ ] 1 arg min | [ ] |, n m i n n u k i u e k = =  (49) where ( ) 1 [ ] i n e k + s are specified by (48). The asymptotic properties of the adaptive controller described in this section are given in theorem below. Theorem 2. Consider the feedback control system containing the plant (1) in which 𝑟 < 𝑚, and the adaptive controller defined in the equations (44), (49) to- Міжнародний науково-технічний журнал «Проблеми керування та інформатики», 2022, № 2 33 gether with (41), (48) and (43). Using the constraints (5), determine 0[1], 0, [ ]N to satisfy the requirement (40). Let assumptions A1)–A3) be valid. Then, this controller applied to the plant (1) guarantees that the control objecti - ves (7) and (8) will be achieved. Proof. See [39]. □ Note that Theorem 2 does not guarantee that the ultimate error lim sup || ||n ne→ will become as in the nonadaptive case when there is no parametric uncertainty and the pseudoinverse model-based controller proposed in [24] can by applied. 6. Simulation To demonstrate the behavior of the robust adaptive closed-loop control system de- signed in Section 3–5, several simulation experiments were conducted. Simulation experiment 1. In this experiment, the adaptive closed-loop control system (1), (6), (25) to (32) was simulated. The elements of B were given as: (11) 4,b = (12) 2,b = (21) 2,b = (22) 1b = (det 0).B = The interval estimates of these elements were chosen as follows: (11)1 5,b  (12)0 2,b  (21)0 2,b  (22)1 2.b  By the formulas (20) and (22), it was found: (1) min 1, = − (2) min 1, = − (1) max 7, = (2) max 4, = min 1, = − max 7. = It turned out that maxmin | | | | .   Therefore it is required that 0 1  to satisfy the inequalities (21). Namely, 0 1,1 = was put. From the conditions (11) 0 [1, 5],b  (12) 0 [0, 2],b  (21) 0 [0, 2],b  (22) 0 [1, 2]b  the following initial estimates of nB were taken: (11) 0 1,b = (12) 0 1,b = (21) 0 0,b = (22) 0 1,9.b = Then (11) 0 2,1,b = (12) 0 1,b = (21) 0 0,b = (22) 0 3.b = In this simulation experiment, the sequences ( ){ } ( 1, 2)i nv i = were generated as i.i.d. random variables [29, p.40] belonging to [ 1, 1].− It was put: 0 T[1, 3] .y = The performance of the adaptive estimation algorithm is shown in Fig. 6. Fig. 6 Fig. 7 shows simulation results illustrating a successful behavior of the robustly- adaptive control system when this experiment was conducted. 2 1 0 1 2 3 4 5 n =0 n =6 n =13 n =106 n =116 n=314 n =1 n =2 0 1 2 1 2 n =1 n =2 n =9 n =105 n =166 n =534 34 ISSN 1028-0979 Fig. 7 Simulation experiment 2. To verify how the adaptive controller proposed in Sec- tion 4 performs, we give an illustrative example. In this example, the true B and the ini- tial 0B were chosen as: 0 0,2 1,4 50 20 0,8 2,4 , 30 40 1,1 0.5 10 10 B B         = =            to ensure 0rank rank 2.B B= = The desired output vector was taken as 0 T[2, 7, 3] .y = ( ){ }i nv were generated as pseudorandom independently identically dis- tributed (i.i.d.) sequences satisfying to (1) [ 0,1; 0,1],nv  − (2) [ 0,2; 0,2]nv  − and (3) [ 0,08; 0,08].nv  − Computer simulations have been carried out to evaluate the behavior of the adap- tive control system (1), (34), (36). This behavior is presented in Fig. 8. It demonstrates that the closed-loop control systems containing the adaptive pseudoinverse model-based controller (34), (36) is successful enough. Fig. 8 b a n 2,8 2,6 2,4 20 3,0 40 b a n 60 80 2 1 100 0 20 3 40 60 80 100 0 0 2nu n 20 40 60 80 100 0 2nu 2ny 10 0 20 0 2 y 4 n 20 40 60 80 100 0 2ny 3,5 3,0 4,0 3,2 2,5 2,0 1,5 Міжнародний науково-технічний журнал «Проблеми керування та інформатики», 2022, № 2 35 Simulation experiment 3. This simulation experiment was conducted to illustrate the performance of the adaptive control proposed in Section 5 for the case when 2,r = 3.m = As the gain matrix, 4 2 2 1 3 1,5 B     =       of not full rank ( rank 1B = ) was taken. Since 3,N = it produces the following three submatrices: 4 2 4 2 2 1 [1] , [2] and [3] 2 1 3 1,5 3 1,5 B B B       = = =            Further, the three vectors (1) (2) T[1] [ , ] ,n n ny y y= (1) (3) T[2] [ , ]n n ny y y= and (2) (3) T[3] [ , ]n n ny y y= were introduced to describe the plants (37) having the gain matrices [1], [2] and [3]B B B respectively. The quantities 0 0[1] 1,1, [2] 1,2 =  = and 0[3] 1,3 = were taken to satisfy the conditions (40) guaranteeing [ ]B k to be nonsingu- lar were derived from the equation (5). The initial 0 0 0[1], [2] and [3]B B B were chosen as 0 0 0[ ] [ ] [ ] rB k B k k I= +  with the initial elements of 0[ ]B k which were selected from B inside the corresponding ranges ( ) ( )[ , ] ij ijb b specified as follows: (11) [1, 5],b  (12) [0, 2],b  (21) [0, 2],b  (22) [1, 2],b  (31) [1, 4],b  (32) [0, 5].b  Namely, we set (11) 0 1,b = (12) 0 1,b = (21) 0 0,b = (22) 0 1,9,b = (31) 0 2,b = (32) 0 2,1.b = The desired output vector was given as 0 T[1, 3, 7] .y = The performance of the simulated adaptive closed-loop control system with the disturbance sequences ( ) ( )( ) 0 1{ } , , i ii nv v v= generated as some pseudorandom i.i.d. vari- ables taken from (1) (2) (3)0,1 0,1; 0,2 0,2; 0,08 0,08n n nv v v−   −   −   is presented in Fig. 9. Fig. 9 a n 20 40 60 80 100 0 1nB B− 3 4 2 0 1 1 [ ] [ ]nB k B k− 1nB B− 1 [3] [3]nB B− 1 [2] [2]nB B− 1 [1] [1]nB B− b n 20 40 60 80 100 0 3 k 2 0 1 36 ISSN 1028-0979 Continuation Fig. 9 Fig. 9, a–d demonstrate that the performance of the proposed adaptive controller applied to the static multivariable plant having some nonsquare gain matrix with not full rank is successful enough. Conclusion The adaptive control concept together with the pseudoinverse (generalized in- verse) model-based approach is the suitable tool to deal with discrete-time uncer- tain multivariable systems whose gain matrices are noninvertible. It has been shown that a successful behavior of the robustly-adaptive closed-loop control sys- tem containing the square singular or nonsquare plant can be achieved regardless of the rank of its gain matrix. Л.С. Житецький, К.Ю. Соловчук АДАПТИВНЕ РОБАСТНЕ КЕРУВАННЯ БАГАТОВИМІРНИМИ СИСТЕМАМИ БЕЗ ПАМ’ЯТІ, ЩО НЕ МОЖУТЬ БУТИ ОБЕРНЕНИМИ, З ОБМЕЖЕНИМИ ЗБУРЕННЯМИ: УЗАГАЛЬНЕННЯ Житецький Леонід Сергійович Міжнародний науково-навчальний центр інформаційних технологій та систем НАН України і МОН України, м. Київ, leonid_zhiteckii@i.ua Соловчук Клавдія Юріївна Полтавський науково-дослідний експертно-криміналістичний центр МВС України, solovchuk_ok@ukr.net Стаття присвячена робастному адаптивному керуванню в дискретному часі де- якими класами невизначених багатовимірних систем без пам’яті за наявності не- вимірних обмежених збурень, межі яких припускаються відомими. Розглянуто системи, у яких число керувальних входів не перевищує числа виходів. Основна особливість об’єктів, що підлягають керуванню, полягає у тому, що матриці кое- фіцієнтів підсилення таких об’єктів не можуть бути обернені. Введено припу- щення, що елементи цих матриць апріорі невідомі, але є інформація про можливі межі інтервалів, до яких вони належать. Поставлено і вирішено задачу, що поля- c n 20 40 60 80 100 0 1nu 3 4 2 0 1 1ne d n 20 40 60 80 100 0 11 10 8 9 12 mailto:leonid_zhiteckii@i.ua mailto:solovchuk_ok@ukr.net Міжнародний науково-технічний журнал «Проблеми керування та інформатики», 2022, № 2 37 гає в тому, аби у контурі зворотного зв’язку побудувати регулятор, здатний впо- ратися з матрицями коефіцієнтів підсилення, які не можуть бути обернені, а та- кож з параметричною невизначеністю для пригнічування зовнішніх збурень і за- безпечення обмеженості всіх керувань та вихідних сигналів. Для вирішення ви- щезазначеної задачі використовується робастний адаптивний підхід у поєднанні з так званою концепцією псевдооберненої чи оберненої моделей. Вивчено три різні випадки. У першому випадку побудований робастний адаптивний регулятор, який може застосовуватися до невизначеного об’єкта з виродженою квадратною мат- рицею коефіцієнтів підсилення. У другому випадку на основі псевдообернених моделей, параметри яких оцінюються за допомогою стандартної рекурентної процедури адаптації, запропоновано робастний метод керування об’єктами з не- відомими прямокутними матрицями коефіцієнтів підсилення повного рангу. За- пропонований у першому випадку підхід поширюється на третій випадок керу- вання невідомими об’єктами, матриці коефіцієнтів підсилення яких є прямокут- ними матрицями неповного рангу. Встановлено асимптотичні властивості запропонованих у статті робастно-адаптивних регуляторів. Для підтримки теоре- тичного дослідження наведено результати числових прикладів. Ключові слова: дискретний час, багатовимірний об’єкт без пам’яті, нездатність до обернення, концепція псевдооберненої моделі, невизначеність, алгоритм оці- нювання, робастне адаптивне керування. REFERENCES 1. Davison E. The output control of linear time-invariant multivariable systems with un- measurable arbitrary disturbances’. IEEE Trans. Autom. Contr. Oct. 1972. AC-17, N 5. P. 621–631. 2. Gubarev V.F., Mishchenko M.D., Snizhko B.M. Model predictive control for discrete MIMO linear systems. In: Y.P. Kondratenko, A.A. Chikrii, V.F. Gubarev, J. Kacprzyk (eds.). Advanced control techniques in complex engineering systems: Theory and applica- tions. Dedicated to Prof. V.M. Kuntsevich. Studies in systems, decision and control, Spring- er Nature Switzerland AG, Cham, 2019. 203. P. 63–81. DOI: https://doi.org/10.1007/978-3- 030-21927-7_10. 3. Skogestad S., Postlethwaite I. Multivariable feedback control. Wiley, Chichester, 1996. 4. Albertos P., Sala A. Multivariable control systems: an engineering approach. Springer, London, 2006. 5. Glad T., Ljung L. Control theory: multivariable and nonlinear methods. Taylor & Francis, New York, 2000. 6. Tan L. A generalized framework of linear multivariable control. Elsevier, Oxford, 2017. 7. Pukhov G.E., Zhuk K.D. Synthesis of interconnected control systems via inverse operator meth- od. Nauk. dumka, Kiev, 1966 (in Russian). 8. Lyubchyk L.M. Disturbance rejection in linear discrete multivariable systems: inverse model ap- proach. In: Proc. 18th IFAC World Congress. Milano, Italy, 2011. P. 7921–7926. 9. Lyubchyk L.M. Inverse model approach to disturbance rejection problem, In: Kondratenko Y.P., Kuntsevich V.M., Chikrii A.A., Gubarev V.F. (Eds), Advanced control systems: theory and ap- plications. River Publishers, Gistrup. 2021. P. 129–166. 10. Skurikhin V.I., Zhiteckii L.S., Solovchuk K.Yu. Control of interconnected plants with singular and ill-conditioned transfer matrices based on pseudo-inverse operator method. Upravlyayush- chye sistemy i mashiny. 2013. 29, N 3. P. 14−20. 11. Zhiteckii L.S., Solovchuk K.Yu. Pseudoinversion in the problems of robust stabilizing multivari- able discrete-time control systems of linear and nonlinear static objects under bounded disturb- ances. Journal of Automation and Information Sciences. 2017. 49, N 5. P. 35–48. 12. Skurikhin V.I., Zhiteckii L.S., Solovchuk K.Yu. Stabilization of a nonlinear multivariable dis- crete-time time-invariant plant with uncertainty on a linear pseudoinverse model. Journal of Computer and Systems Sciences International. 2017. 56, N 5. P. 759–773. 38 ISSN 1028-0979 13. Kuntsevich V.M. Control under uncertainty: guaranteed results in control and identification prob- lems. Nauk. dumka, Kiev, 2006 (in Russian). 14. Fomin V.N., Fradkov A.L., Yakubovich V.A. Adaptive control of dynamic plants. Nauka, Mos- cow, 1981 (in Russian). 15. Goodwin G.C., Sin K.S. Adaptive filtering, prediction and control. Prentice-Hall, Engewood Cliffs, 1984. 16. Tao G. Adaptive control design and analysis. John Wiley and Sons, New York, 2003. 17. Narendra K.S., Annaswamy A.M. Stable adaptive systems. Dover Publications, New York, 2012. 18. Ioannou P., Sun J. Robust adaptive control. Dover Publications, New York, 2013. 19. Ǻström K.J. and B. Wittenmark. Adaptive control: 2nd Edition, Dover Publications, New York, 2014. 20. Zhiteckii L.S., Solovchuk K.Yu. Adaptive stabilization of some multivariable systems with nonsquare gain matrices of full rank. Cybernetics and Computer Engineering. 2018. 192, N 2. P. 44–60. 21. Zhiteckii L.S., Solovchuk K.Yu. Solving a problem of adaptive stabilization for some static mimo systems. Cybernetics and Computer Engineering. 2019. 197, N 3. P. 33–50. 22. Zhiteckii L.S., Solovchuk K.Yu. Robust adaptive controls for a class of nonsquare memor- yless systems. In: Kondratenko, Y.P., Kuntsevich, V.M., Chikrii, A.A., Gubarev, V.F. (Eds), Advanced Control Systems: Theory and Applications. River Publishers, Gistrup, 2021. P. 203–226. 23. Zhiteckii L.S., Azarskov V.N., Solovchuk K.Yu. Adaptive robust control of inter-connected static plants with nonsquare gain matrixes. In: Proc. 13th All-Russian Control Problems Council (VSPU-2019). IPU, Moscow, 2019. P. 713–718. 24. Albert A. Regression and the Moore-Penrose pseudoinverse. Academic Press, New York, 1972. 25. Voevodin V.V., Kuznetsov Yu.A. Matrices and calculations. Nauka, Moscow, 1984 (in Rus- sian). 26. Azarskov V.N., Zhiteckii L.S., Solovchuk K.Yu. Parametric identification of the interconnected static closed-loop system: a special case. In: Proc. 12th All-Russian Control Problems Council (VSPU-2014). IPU, Moscow, 2014. P. 2764–2776. 27. Marcus M., Minc H. A survey of matrix theory and matrix inequalities. Aliyn and Bacon, Boston, 1964. 28. Zhiteckii L.S., Skurikhin V.I. Adaptive control systems with parametric and nonparametric uncer- tainties. Nauk. dumka, Kiev, 2010 (in Russian). 29. Walter E., Pronzato L. Identification of parametric models. Springer, Paris/Milan/Barcelona, 1997. Submitted 01.02.2022
id nasplib_isofts_kiev_ua-123456789-210873
institution Digital Library of Periodicals of National Academy of Sciences of Ukraine
issn 0572-2691
language English
last_indexed 2026-03-16T12:21:29Z
publishDate 2022
publisher Інститут кібернетики ім. В.М. Глушкова НАН України
record_format dspace
spelling Zhiteckii, L.S.
Solovchuk, K.Yu.
2025-12-19T17:14:22Z
2022
Adaptive robust multivariable control of noninvertible memoryless systems with bounded disturbances: a generalization / S. ZhiteckiiL, K.Yu. Solovchuk // Проблеми керування та інформатики. — 2022. — № 2. — С. 22-38. — Бібліогр.: 29 назв. — англ.
0572-2691
https://nasplib.isofts.kiev.ua/handle/123456789/210873
681.5
10.34229/2786-6505-2022-2-2
The article is dedicated to robust adaptive control in discrete time for certain classes of undefined memoryless multivariate systems with unmeasurable bounded disturbances, the bounds of which are assumed to be known. The systems considered have a number of control inputs that does not exceed the number of outputs.
Стаття присвячена робастному адаптивному керуванню в дискретному часі деякими класами невизначених багатовимірних систем без пам’яті за наявності невимірних обмежених збурень, межі яких припускаються відомими. Розглянуто системи, у яких число керувальних входів не перевищує числа виходів. Основна особливість об’єктів, що підлягають керуванню, полягає у тому, що матриці коефіцієнтів підсилення таких об’єктів не можуть бути обернені. Введено припущення, що елементи цих матриць апріорі невідомі, але є інформація про можливі межі інтервалів, до яких вони належать. Поставлено і вирішено задачу, що полягає в тому, аби у контурі зворотного зв’язку побудувати регулятор, здатний впоратися з матрицями коефіцієнтів підсилення, які не можуть бути обернені, а також з параметричною невизначеністю для пригнічування зовнішніх збурень і забезпечення обмеженості всіх керувань та вихідних сигналів. Для вирішення вищезазначеної задачі використовується робастний адаптивний підхід у поєднанні з так званою концепцією псевдооберненої чи оберненої моделей. Вивчено три різні випадки. У першому випадку побудований робастний адаптивний регулятор, який може застосовуватися до невизначеного об’єкта з виродженою квадратною матрицею коефіцієнтів підсилення. У другому випадку на основі псевдообернених моделей, параметри яких оцінюються за допомогою стандартної рекурентної процедури адаптації, запропоновано робастний метод керування об’єктами з невідомими прямокутними матрицями коефіцієнтів підсилення повного рангу. Запропонований у першому випадку підхід поширюється на третій випадок керування невідомими об’єктами, матриці коефіцієнтів підсилення яких є прямокутними матрицями неповного рангу. Встановлено асимптотичні властивості запропонованих у статті робастно-адаптивних регуляторів. Для підтримки теоретичного дослідження наведено результати числових прикладів.
en
Інститут кібернетики ім. В.М. Глушкова НАН України
Проблемы управления и информатики
Адаптивне керування та методи ідентифікації
Adaptive robust multivariable control of noninvertible memoryless systems with bounded disturbances: a generalization
Адаптивне робастне керування багатовимірними системами без пам'яті, що не здатні бути оберненими, з обмеженими збуреннями: узагальнення
Article
published earlier
spellingShingle Adaptive robust multivariable control of noninvertible memoryless systems with bounded disturbances: a generalization
Zhiteckii, L.S.
Solovchuk, K.Yu.
Адаптивне керування та методи ідентифікації
title Adaptive robust multivariable control of noninvertible memoryless systems with bounded disturbances: a generalization
title_alt Адаптивне робастне керування багатовимірними системами без пам'яті, що не здатні бути оберненими, з обмеженими збуреннями: узагальнення
title_full Adaptive robust multivariable control of noninvertible memoryless systems with bounded disturbances: a generalization
title_fullStr Adaptive robust multivariable control of noninvertible memoryless systems with bounded disturbances: a generalization
title_full_unstemmed Adaptive robust multivariable control of noninvertible memoryless systems with bounded disturbances: a generalization
title_short Adaptive robust multivariable control of noninvertible memoryless systems with bounded disturbances: a generalization
title_sort adaptive robust multivariable control of noninvertible memoryless systems with bounded disturbances: a generalization
topic Адаптивне керування та методи ідентифікації
topic_facet Адаптивне керування та методи ідентифікації
url https://nasplib.isofts.kiev.ua/handle/123456789/210873
work_keys_str_mv AT zhiteckiils adaptiverobustmultivariablecontrolofnoninvertiblememorylesssystemswithboundeddisturbancesageneralization
AT solovchukkyu adaptiverobustmultivariablecontrolofnoninvertiblememorylesssystemswithboundeddisturbancesageneralization
AT zhiteckiils adaptivnerobastnekeruvannâbagatovimírnimisistemamibezpamâtíŝonezdatníbutiobernenimizobmeženimizburennâmiuzagalʹnennâ
AT solovchukkyu adaptivnerobastnekeruvannâbagatovimírnimisistemamibezpamâtíŝonezdatníbutiobernenimizobmeženimizburennâmiuzagalʹnennâ