Muhammad al-Xorazmiy nomidagi TATU Farg'ona filiali "Al-Farg'oniy avlodlari" elektron ilmiy jurnali ISSN 2181-4252 Tom: 1 | Son: 4 | 2023-yil
"Descendants of Al-Farghani" electronic scientific journal of Fergana branch of TATU named after Muhammad al-Khorazmi. ISSN 2181-4252 Vol: 1 | Iss: 4 | 2023 year
Электронный научный журнал "Потомки Аль-Фаргани" Ферганского филиала ТАТУ имени Мухаммада аль-Хоразми ISSN 2181-4252 Том: 1 | Выпуск: 4 | 2023 год
ALGORITHMS FOR SYNTHESIS OF ADAPTIVE CONTROL SYSTEMS WITH IMPLICIT REFERENCE MODELS BASED ON THE SPEED GRADIENT METHOD
Sevinov Jasur Usmonovich
Doctor of technical sciences, prof.
Tashkent State Technical University named after
Islom Karimov E-mail: sevinovj asur@gmail .com,
Boborayimov Okhunjon Khushmurod ogli,
PhD student of Tashkent State Technical University named
after Islom Karimov E-mail:[email protected].
Abstract: Algorithms for the synthesis of adaptive control are considered, in which the reference model acts as a certain reference equation. To obtain precise statements about the properties of a generalized custom object, some auxiliary conditions such as smoothness and regularity are introduced. The use of such systems has made it possible to reduce the requirements for the structure of the main circuit and for the completeness of the measured information, which is advisable to use in the presence of strong unmeasured disturbances, as well as in problems of controlling high-order multidimensional objects, when the implementation of a reference model in the system is impossible or difficult.
I Key words: adaptive control, speed gradient algorithm, control object, reference model, unmeasured disturbance, control action.
Introduction. The current stage of development of control theory and technology is characterized by increasing requirements for control systems, increasing complexity of controlled objects, and high rates of design and commissioning of systems. These circumstances lead, as a rule, to the fact that the available initial information is insufficient to build systems with high quality indicators and it is necessary to replenish the information during the operation of the system. Such systems, in which the information necessary to improve functioning is collected during operation, immediately processed and used for control, are called adaptive. Adaptive systems are currently finding increasing use for managing objects and processes under conditions of uncertainty. An adaptive system uses a control law with variable coefficients. The coefficients are changed by a special algorithm (adaptation algorithm) based on current information about the state of the process obtained during normal operation of the installation. The adaptation algorithm is constructed in such a way as to adapt to a specific situation and ensure the achievement
of the control goal for any possible value of the unknown parameters of the object [1-7].
The gradient principle is widely used to synthesize the adaptation algorithm. In this case, the evaluation function, the gradient of which determines the direction of change in the adjustable parameters, may coincide with the evaluation function that sets the control goal, or may differ from it. If the control goal is given to the designer of the adaptive system from the outside, then the adaptation goal is set by the designer himself when synthesizing the adaptation algorithm. The separation of control and adaptation goals, generally speaking, expands the designer's capabilities. However, this makes it more difficult to justify the system's performance, since achieving the adaptation goal may not directly result from achieving the original control goal. The above considerations explain the origin of speed gradient algorithms. However, our task is to establish the general properties of algorithms that allow us to judge the stability and quality of specific adaptive systems [3,8-14,21].
172
Muhammad al-Xorazmiy nomidagi TATU Farg'ona filiali "Al-Farg'oniy avlodlari" elektron ilmiy jurnali ISSN 2181-4252 Tom: 1 | Son: 4 | 2023-yil
"Descendants of Al-Farghani" electronic scientific journal of Fergana branch of TATU named after Muhammad al-Khorazmi. ISSN 2181-4252 Vol: 1 | Iss: 4 | 2023 year
Электронный научный журнал "Потомки Аль-Фаргани" Ферганского филиала ТАТУ имени Мухаммада аль-Хоразми ISSN 2181-4252 Том: 1 | Выпуск: 4 | 2023 год
Formulation of the problem
Let the equation of a generalized custom object be given:
dx / dt = F (x, c, t,h)
x = {x (i) x(n)} where x {x x } - is the state vector of
the generalized customizable object (GCO);
c = (c (1) c (N)}
c {c ,•••,c } - vector of adjustable parameters;
h = f/(i) p(q)\
h \h ,---,h } - vector of unknown object parameters and external influences. Let us assume that the adaptation goal is specified using the evaluation
functional Qt, and approaching the goal corresponds to a decrease in the values. We will consider two cases
Qt [2,3,10]:
a) the estimated functional Qt is a non-negative function of the phase coordinates of GCO:
Qt = Q(x(t),t). We will call such a functional Qt local.
b) Qt is the integral functional:
R( x, c, t )
Qt =| R(x(s), c(s), s)ds
0 , where "v-'-'-v - is some
non-negative function..
In each of these cases, it is possible to calculate
function Qt - the rate of change of functional Qt by virtue of equation (1) for a fixed c. Obviously, in case a)
Qt = dQ(x(t), t)/ dt + F(x(t), c(t), t, h)T VxQ(x(t), t),
and in case b), Qt = R( x(t ^ c(t ^t) i.e. in both cases:
Qt = y(x(t), c(t), t),
where ^t) - is some function that we will assume is continuously differentiable with respect to the components of vector c.
Let's call the following adaptation algorithm the speed gradient algorithm:
dc / dt = - rVciy( x, c, t )
(2)
where r =r > 0 - is a positive definite N x N - is the matrix of gain factors.
It should be noted that the right-hand side in relation (2) may turn out to depend on unknown
parameters h or on phase coordinates of the GCO that are inaccessible to measurement, and then algorithm (2) will be unrealizable. The question of the feasibility of the algorithm and the class E of adaptability of the system (1), (2) must be solved separately in each specific problem.
To obtain precise statements about the properties of the system (1), (2), it is necessary to impose some auxiliary conditions such as smoothness and regularity on the right-hand sides of the system and
the functional Qt, excluding "pathological" cases [1,2,15,16]. We will assume that the right-hand sides of (1), (2) are locally bounded uniformly in t > 0, i.e.,
for any P > 0 the following inequality holds:
||F(z, t,h)|| + IVc¥(z, t,)|| <Kp<»
(3)
INI ^p, t > о
at \r\\~ , where z = {x, c} is the state
vector of the system (1), (2). Note that condition (3) does not exclude the possibility of discontinuities along
t in the right-hand sides of (1), (2). Function Q(^t) in the case of local functional Qt will be considered
x t
uniformly continuous in x, t in any domain of the form
{x'11 'Ixt > 0}. In addition, we will require sufficient smoothness of the functions
Q(x,t), F(x,c,th), R(x,c,t) so that all their derivatives arising in the formulations of the statements exist and are continuous in x, c .
Let us assume that the adaptation goal to be achieved by algorithm (2) is given by the relation: lim Qt = 0
(4)
(in the case of local functional Qt ) or the
relation:
lim R(x(t), c(t), t) = 0
(5)
(in the case of integral functional Qt ). We are interested in the conditions under which goals (4), (5) in systems (1), (2) are achieved for any initial values
c(0). The first of these conditions is the convexity
of the c function X'C't ), i.e., the fulfillment of the inequality [21]:
173
Muhammad al-Xorazmiy nomidagi TATU Farg'ona filiali "Al-Farg'oniy avlodlari" elektron ilmiy jurnali ISSN 2181-4252 Tom: 1 | Son: 4 | 2023-yil
"Descendants of Al-Farghani" electronic scientific journal of Fergana branch of TATU named after Muhammad al-Khorazmi. ISSN 2181-4252 Vol: 1 | Iss: 4 | 2023 year
Электронный научный журнал "Потомки Аль-Фаргани" Ферганского филиала ТАТУ имени Мухаммада аль-Хоразми ISSN 2181-4252 Том: 1 | Выпуск: 4 | 2023 год
y/(x, C, t) - y/(x, C', t) > (c - C') Vcy(x, c', t) (6)
for any c, c , x, t. Condition (6) is satisfied, for example, in the frequently encountered case when the right side of the GCO equation (1) linearly depends on the adjustable parameters. The second main condition is the fundamental achievability of the goal, i.e. the
existence of an "ideal" vector c* (depending, perhaps,
on %), such that in system (1) at c = c* goal (4) or (5) is achieved. More precisely, by the condition of attainability we will understand the fulfillment of the inequality [21]:
y/(x, c*, t) <-ßQt + d(t),
œ
ß> 0, ¡(t) > 0, Jd(t)
(7)
< œ
where
Solution of the task
It can be shown that from relation (7) it follows that goals (4), (5) are achieved (in the case of goal (4)
with the additional requirement $ > 0 ).
Let the functional Qt be local and the convexity
condition (6), the reachability condition (7) for $ > 0, and the growth condition are satisfied [2,10]: inf Q(x, t) ^ œ
x
■ œ
at ""'' ' . (8)
Then all trajectories of system (1), (2) are bounded and satisfy (4) [21].
Let the control object be described by the equations:
dx / dt = Ax + bu, y = L x, u = cTy ^
where x e R", u e r1, y e R, - vector of measured outputs; c - vector of adjustable parameters. It is required to find an algorithm for adjusting
vector c(t) so as to ensure achievement of the goal: x(t) ^ 0, c(t) ^ const ^o)
Let us choose the evaluation function
Q = 0,5x'^x, where H = HT > 0, velocity gradient principle.
Qt = xTH(Ax + bcTy), VcQt = xTHby
and use the We have
Since the
value XHb should depend only on the measured
quantities, we arrive at condition ~ , and if it is satisfied, we write the speed gradient algorithm in the form:
dc / dt = - g уГу
(11)
where r = r > 0.
The only condition (8) that needs special verification is the solvability condition (7). It will be
satisfied if there is a vector c* such that x HA*x < 0,
where A* = A + bc*L . Therefore, to check (7) we need
conditions for the existence of a matrix H = H > 0
and a vector c* such that
HA + Al H < 0, Hb = Lg, A = A + bcl LT (12)
For the existence of a matrix H = H > 0 and a vector c* satisfying (12), it is necessary and sufficient
that the polynomial = g Wbe a Hurwitz polynomial of degree n -1 with positive coefficients,
where
W(Ä) = LT (ÄIn - A)-1 b
- is the transfer vector
function of the object, = A) - is the
common denominator of its components.
As is known [17, 18], systems in which the numerator of the transfer function is a Hurwitz polynomial are called minimum-phase. Minimumphase systems, the numerator of the transfer function of which has the maximum possible (equal to n-1) degree and positive coefficients, will be called strictly minimum-phase. The transfer function of a strictly minimum-phase system will also be called strictly minimum-phase. Taking into account the introduced ones, (12) can be formulated as follows: for the
existence of a matrix H = H > 0 and a vector c* satisfying (12), it is necessary and sufficient that the
function g Wbe strictly minimal-phase.
From (12) and (8) it follows that for a given vector, the adaptation goal (10) in the system (9), (11) is achieved for any object (9) such that function
gTW(Ä) •
is strictly minimal-phase. If adaptability class
174
0
Muhammad al-Xorazmiy nomidagi TATU Farg'ona filiali "Al-Farg'oniy avlodlari" elektron ilmiy jurnali ISSN 2181-4252 Tom: 1 | Son: 4 | 2023-yil
"Descendants of Al-Farghani" electronic scientific journal of Fergana branch of TATU named after Muhammad al-Khorazmi. ISSN 2181-4252 Vol: 1 | Iss: 4 | 2023 year
Электронный научный журнал "Потомки Аль-Фаргани" Ферганского филиала ТАТУ имени Мухаммада аль-Хоразми ISSN 2181-4252 Том: 1 | Выпуск: 4 | 2023 год
E is specified, then the synthesis of an adaptive stabilization system is reduced to finding a vector g that provides for any object from class E strict minimum
phase function g W(A).
Let's consider a common special case when the object is described by equation
A(P)l = B(P)u (13)
A(A) = An + an-1An-1 + ••• + a0, B(A) = bmAm +... + b0
and the output variable 1 along with its l -1 derivatives is available for observation, i.e.
y = [f,fj,...,f(l
. The controller will be described by
(-i
C (A) = £ ciAi equation u =C(p)l, where i=0 . The
r = vI
adaptation algorithm (at l ) is written by the
equations:
c. =-vG(p)n.n{i \ i = 0,1,•••, l -1
in which the coefficients of the polynomial
G(p correspond to the components of the vector g from (11).
The transfer vector function of the object has the form:
w (A) = ba&A'";*-1} A(A) ,
where
gTW(A) = G(A)B(A)/A(A), S(A) = A(A) Therefore, the strict minimum phase property of function
gTW (A)
means that the polynomials
G(A), B(A)
are
Hurwitz [1, 19], the signs of their coefficients coincide m +l = n. Consequently, by choosing a Hurwitz polynomial G(A) of degree n - m -1 with coefficients of the same sign as those of B^AA, we can guarantee
the achievement of the adaptation goal ^ 0 for any minimum-phase object of the form (13). To implement an adaptive controller, an n - m -1 derivative of the controlled object coordinate is required [13,20]. The results of modeling such systems for various special cases show that the adaptation process in them proceeds several times faster than the
transition process according to x(t), even in the case of an unstable object (13). In this case, the value
G (after adaptation becomes close to zero, which
allows us to interpret equation G(pll =0 as a reference characterizing the quality of the object's processes after adaptation is completed.
The described approach can be extended to
systems for monitoring the reference influence r (t). In this case, Qt =[x x0(t)] H[x x°] should be taken
evaluation
function,
as
where
an
x°it) - is the equilibrium state of the system with "ideal" regulator coefficients, calculated under the assumption that the setting influence is established at
the level of r (t).
For a special case of object (13), the adaptive controller will have the form:
u = C(p)q- Cl(p)r, dci /dt = -y'1S{t)v(l\ i = 0,...,n -m -\, dcu /dt = -/"S(t)r(i), i = 0,...,k, S(t) = G(p)n~G1(p)r(t),
(14)
where the degree of polynomial
С (A)
is equal
to n m 1; the power of C A is equal to the number
of k available derivatives of r(t); C(AA - Hurwitz
polynomial; Vi,Vi> 0.
Equation G(p)lM = Gi(p)r can be considered
as a reference and polynomials C(AA, C1(AA can be selected based on the desired quality of the system after adaptation. It can be shown, using (8), that
l(t) -Vm (t) ^ 0 if the setting influence r(t) is established more accurately, if
]\r (i\t )
dt <œ, i = 1„..,( +1
Conclusion
Thus, the use of such systems makes it possible to reduce the requirements for the structure of the main circuit and for the completeness of the measured information. It is advisable to use them in the presence of strong unmeasured disturbances, as well as in control problems of high-order multidimensional
175
о
Muhammad al-Xorazmiy nomidagi TATU Farg'ona filiali "Al-Farg'oniy avlodlari" elektron ilmiy jurnali ISSN 2181-4252 Tom: 1 I Son: 4 | 2023-yil
"Descendants of Al-Farghani" electronic scientific journal of Fergana branch of TATU named after Muhammad al-Khorazmi. ISSN 2181-4252 Vol: 1 | Iss: 4 | 2023 year
Электронный научный журнал "Потомки Аль-Фаргани" Ферганского филиала ТАТУ имени Мухаммада аль-Хоразми ISSN 2181-4252 Том: 1 | Выпуск: 4 | 2023 год
objects, when implementation of a reference model in the system is impossible or difficult.
Literature
1. Miroshnik I.V., Nikiforov V.O., Fradkov A.L. Nonlinear and adaptive control of complex dynamic systems. -SPb.: Nauka, 2000. - 549 p.
2. Fradkov A.L. Adaptive control in complex systems: non-search methods. - Moscow: Nauka, 1990. -S. 296.
3. Andrievsky B.R., Fradkov B.R. Speed gradient method and its applications // Automation and Technology, 2021, No. 9. -WITH. 3-72.
4. Igamberdiyev X.Z., Sevinov J.U., Zaripov O.O. Regulyarniye metodi i algoritmi sinteza adaptivnix system upravleniya s nastraivayemimi modelyami. -T.: TashGTU, 2014. -160 p.
5. Igamberdiev Kh.Z., Yusupbekov A.N., Zaripov O.O. Regular methods for assessing and managing dynamic objects under conditions of uncertainty. - T.: Tashkent State Technical University, 2012. - 320 s.
6. Furtat IB., Gushchin P.A., Nguyen B., Kolesnik N.S. Adaptive control with a guarantee of a given quality of regulation // Management of large systems, No. 102, 2023. - pp. 44-57.
7. Fradkov A.L., Grigoriev G.K., Decentralized adaptive control of synchronization of networks of dynamic systems under limited disturbances // Automation and Technology, 2013, No. 5. pp. 137155.
8. Alisher Mallayev, Jasur Sevinov, Suban Xusanov, Okhunjon Boboraimov. Algorithms for the synthesis of gradient controllers in a nonlinear control system / Proceedings of the II International Conference on Advances in Materials, Systems and Technologies: (Camstech-Ii 2021). Krasnoyarsk, July 29-31, 2021. -pp. 51-53
9. Pechen A.N. On the speed gradient method for generating unitary quantum operations in closed quantum systems // Uspekhi Matematicheskikh Nauk, 2016, T 71, No. 3(429). -C. 205-206.
10. Sevinov J.U., Boborayimov O.Kh. Synthesis of Management Systems for Dynamic Objects Based on Speed Gradient Algorithms // International scientific and technical journal "Chemical technology. Control and management." Tashkent. 2022. No. 3. -pp. 61-63.
11. Boborayimov O.Kh., Okyay K.M. Synthesis of Control Systems With Multilayer Neural Networks Based on Velocity Gradient Methods // International scientific and technical journal "Chemical technology. Control and management." Tashkent. 2023. No. 3. -pp. 34-39.
12. Sevinov J.U., Boboraimov O.Kh., Algorithms for Synthesis of Adaptive Decentralized Control of Interconnected Systems by the Speed Gradient Method // Central Asian Journal of Theoretical and Applied Sciences. Volume: 04 Issue: 10, Oct 2023, pp. 129137.
13. Andrievsky B.R., Guzenko P.Yu., Fradkov A.L., Control of nonlinear oscillations of mechanical systems by the speed gradient method // Automation and Technology, 1996, No. 4. - P. 4-17.
14. Andrievsky B.R., Stotsky A.A., Fradkov A.L., Speed gradient algorithms in control and adaptation problems // Automation and Technology, 1988, No. 12. - P. 3-39.
15. Bakushinsky A.B., Kokurin M.Yu. Iterative methods for solving irregular equations. - M.: Lenand, 2006. - 214 p.
16. Mamirov U.F. Regular algorithms for parametric synthesis of adaptive control systems for uncertain objects / Prospects of Development of Science and Education Proceedings of 9th Conference December 25, 2020, -PP. 44-46.
17. Voronov A, A. Stability, controllability, observability. - M.: Nauka, 1979. - 335 p.
18. Oleg Gasparyan. Multidimensional discrete automatic control systems: Method of characteristic transfer functions. LAP LAMBERT Academic Publishing, 2016. - P. 312.
19. Andrievsky B.R. Global stabilization of an unstable pendulum with flywheel control // Management of large systems, 2009, No. 24. - pp. 258280.
20. Fradkov A.L. On the application of cybernetic methods in physics // Uspekhi Fizicheskikh Nauk, 2005, T.175, No. 2. - pp. 113-138.
21. Sharshenaliyev J. Synthesis of algorithms for adaptive control of dynamic systems // News of the National Academy of Sciences of the Kyrgyz Republic. 2009. No. 2.-P. 5-14.
176