Professional Documents
Culture Documents
net/publication/270175807
CITATIONS READS
7 103
2 authors:
All content following this page was uploaded by Andrzej Dzieliński on 06 July 2016.
Fixed final time and free final state optimal control problem
for fractional dynamic systems
– linear quadratic discrete-time case
A. DZIELIŃSKI1 and P.M. CZYRONIS2∗
1
Electrical Engineering Institute of Control and Industrial Electronics, Warsaw University of Technology,
75 Koszykowa St., 00-662 Warszawa, Poland
2
Bumar Elektronika S.A., 30 Poligonowa St., 04-451 Warszawa, Poland
Abstract. The optimization problem for fractional discrete-time systems with a quadratic performance index has been formulated and
solved. The case of fixed final time and a free final state has been considered. A method for numerical computation of optimization problems
has been presented. The presented method is a generalization of the well-known method for discrete-time systems of integer order. The
efficiency of the method has been demonstrated on numerical examples and illustrated by graphs. Graphs also show the differences between
the fractional and classical (standard) systems theory. Results for other cases of the fractional system order (coefficient α) and not illustrated
with numerical examples have been obtained through a computer algorithm written for this purpose.
Key words: fractional order systems, discrete-time systems, optimal control, linear quadratic performance index.
681
Unauthenticated
Download Date | 7/6/16 12:41 PM
A. Dzieliński and P.M. Czyronis
2. Problem formulation and general solution We define the scalar functions Hk , called the Hamiltonians,
which are defined as follows:
The fractional continuous-time system is described by the T
equations [31] X k
RL α
Hk = xTk Qxk +uTk Ruk + dj xk−j + Buk λk+1 . (4)
a Dt x(t) = Ax(t) + Bu(t), j=0
where x ∈ ℜn , u ∈ ℜm are respectively the state and control Using (4) and (2), we define a new performance index ex-
vectors, A ∈ ℜn×n , B ∈ ℜn×m and RL a Dt x(t) is Riemanna-
α
pressed by Hamiltonians, of the form
Liouville fractional derivative. N −1
The Grunwald-Letnikov (shifted) approximation of frac-
X
J(u) = xTN SxN + (Hk − xTk+1 λk+1 ). (5)
tional order derivative [17] is given as k=i
and
X
+ dj xk−j + Buk − xk+1 λk+1 .
j=0 λN = (S + S T )xN , λ0 ∈ ℜ. (9d)
Unauthenticated
Download Date | 7/6/16 12:41 PM
Fixed final time and free final state optimal control problem for fractional dynamic systems...
The conditions (9) for α = 1 are equivalent to the conditions Substituting (9d) to (15) we obtain
for continuous-time systems of integral order (not fractional)
after the discretization. Substituting α = 1 to (9) yields λ1 ΨTN −1
. ..
. = S + S T xN
uk = −[R + RT ]−1 B T λk+1 , (10a) . .
T λN ΨT0
λk = [Q + Q ]xk + ATd λk+1 , (10b)
xk+1 = Ad xk + Buk , (10c)
0 ΨT0 ··· ΨTN −2
where Ad = A + In .
.. .. .. ..
. . . .
+ (16)
0 0 ··· ΨT0
3. Problem solution in case of free final state
0 0 ··· 0
and fixed final time
In the case of free final state xN the variation dxN is not equal x0
to zero. Therefore, we consider Eqs. (9d) and (9a)–(9c). We T ..
· [Q + Q ] .
.
assume that the number of discrete time points N is given.
The initial conditions and final time are given as xN −1
x(k = 0) = x0 , N ∈ Z+ . (11) From the above equation it follows that this time range
of the vector
λk does not depend from x0 . Taking
as T1 =
Taking into account initial conditions (11) and applying the
R + RT , T2 = Q + QT and T3 = S + S T the above
z-transform to Eqs. (9a) and (9c) we obtain the equations in
relationship is written in the form
z domain. Applying then the inverse z-transform we obtain
the solution as
λ1
k−1 λ
X 2
xk = Ψk x0 − Ψk−i−1 B[R + RT ]−1 B T λi+1 , (12)
..
=
.
i=0
where
λN −1
k−1
X λN
Ψ0 = In , Ψk = dj Ψk−j−1 . (13)
ΨT0 T2 ΨTN −3 T2 ΨTN −2 T2 ΨTN −1 T3
j=0 ···
Equation (12) in matrix form gives .. .. .. .. ..
.
. . . .
x1 Ψ1 0 ··· ΨT0 T2 ΨT1 T2 ΨTN −2 T3 (17)
. .
. = . x0 ΨT0 T2 ΨT1 T3
0 ··· 0
. .
xN ΨN 0 ··· 0 0 ΨT0 T3
(14)
x1
Ψ0 ··· 0 λ1
.. . .. . x2
. T −1 T . .
− . . . B[R + R ] B .
..
· .
.
ΨN −1 · · · Ψ0 λN
xN −1
Using the well-known matrix operations it can be easily shown
that the solution of (9b) has the form xN
λ1
ΨTN −1
Substituting (17) to (14) we obtain
. .. 0 0 0 0
. =
W11 W12 ··· W1,N W1,N
λN
. . 0 0 0
−1
0
λN ΨT0
W21 W22 ··· W2,N −1 W2,N
.. .. .. .. ..
. . . . .
0 ΨT0 ··· ΨTN −2
WN0 −1,1 WN0 −1,2 ··· WN0 −1,N −1 WN0 −1,N
.. .. ..
..
. . . . 0
WN,1 0
WN,2 ··· 0
WN,N 0
WN,N
+ (15) −1
0 0 ··· ΨT0
(18)
0 0 ··· 0 x1 Ψ1
x2
Ψ2
.. ..
x0
· = x0 ,
. .
..
T
· [Q + Q ]
.
.
xN −1
ΨN −1
xN −1 xN ΨN
Unauthenticated
Download Date | 7/6/16 12:41 PM
A. Dzieliński and P.M. Czyronis
Unauthenticated
Download Date | 7/6/16 12:41 PM
Fixed final time and free final state optimal control problem for fractional dynamic systems...
We assume N = 5 and α = 0.7. Using the foregoing consid- The minimum values of performance index Jmin are given
erations, we obtain as follows
" # " #
0.6 −0.4377
h i h i
x0 = , x1 = , J0 = 6.1269 , J1 = 0.6609 ,
0.8 0.5011 h i h i
J2 = 0.1886 , J3 = 0.0818 ,
" # " #
−0.2301 −0.1526
x2 = , x3 = ,
0.2257 0.1454
h i h i
" # " # J4 = 0.0364 , J5 = 0.0133 .
−0.1075 −0.069
x4 = , x5 = ,
0.1064 0.09 Figures 1–3 show the above considerations for the sys-
" # " # tem (1) with matrices (22) and the performance index (2) with
7.7728 −0.3464 matrices (23) for four different values of α = 0.5, 0.7, 0.9, 1.0
λ0 = , λ1 = ,
9.4877 2.1705 and N = 5. Individual results were obtained with the help of
" # " # a dedicated computer program implementing the above is-
−0.3372 −0.2434 sues.
λ2 = , λ3 = ,
0.9682 0.5947 From Fig. 3, we conclude that the performance index
takes smaller values for fractional discrete-time systems for
" # " #
−0.1632 −0.0961
λ4 = , λ5 = , α = 0.5, 0.7, 0.9 than for discrete-time system of integer or-
0.3852 0.2221
h i h i der α = 1.0. In terms of physical and technical properties
u0 = −0.7389 , u1 = −0.1469 , performance indexes present such values as energy consump-
h i h i tion, fuel consumption, cost of production, profit, time, accu-
u2 = −0.0539 , u3 = −0.0294 , racy, etc. Consideration of the optimization problems in a case
h i h i of fractional systems can bring greater benefits than for the
u4 = −0.0149 , u5 = 0 . systems of integer order.
Fig. 1. Optimal trajectory and its zoom of end points forα = 0.5, 0.7, 0.9, 1.0 and N = 5
Unauthenticated
Download Date | 7/6/16 12:41 PM
A. Dzieliński and P.M. Czyronis
Fig. 2. Optimal control and its zoom of end points for α = 0.5, 0.7, 0.9, 1.0 and N = 5
Fig. 3. The minimum values of the performance index and its zoom of end points for α = 0.5, 0.7, 0.9, 1.0 and N = 5
Unauthenticated
Download Date | 7/6/16 12:41 PM
Fixed final time and free final state optimal control problem for fractional dynamic systems...
Example 2. Consider a discrete-time fractional system (1) The minimum values of performance index Jmin are given
with matrices (22) and performance index (4) with matri- as follows
ces (23). The initial conditions are given as (24). We assume h i h i h i
N = 10 and α = 0.5. Using the foregoing considerations, we J0 = 5.7746 , J1 = 0.4429 , J2 = 0.0859 ,
obtain optimal control in form h i h i h i
J3 = 0.0495 , J4 = 0.0303 , J5 = 0.0198 ,
h i h i
u0 = −0.6417 , u1 = −0.1469 ,
h i h i h i
J6 = 0.0133 , J7 = 0.0088 , J8 = 0.0056 ,
h i h i
u2 = −0.0518 , u3 = −0.0315 ,
h i h i
J9 = 0.0032 , J10 = 0.0014 .
h i h i
u4 = −0.0216 , u5 = −0.0161 , (25)
h i h i
u6 = −0.0127 , u7 = −0.0103 , Now we assume that elements of matrices (22) of fraction-
h i h i h i al discrete-time system (1) have been changed. We consider
u8 = −0.0086 , u9 = −0.054 , u10 = 0 . two cases, i.e.
Unauthenticated
Download Date | 7/6/16 12:41 PM
A. Dzieliński and P.M. Czyronis
Fig. 5. The minimum values of the performance index for α = 0.5 and N = 10
Fig. 6. The minimum values of the performance index for α = 1.0 and N = 10
Unauthenticated
Download Date | 7/6/16 12:41 PM
Fixed final time and free final state optimal control problem for fractional dynamic systems...
Unauthenticated
Download Date | 7/6/16 12:41 PM
A. Dzieliński and P.M. Czyronis
[21] O.P. Agrawal, “Fractional variational calculus in terms of Riesz [28] D. Sierociuk, I. Tejado, and B.M. Vinagre, “Improved frac-
fractional derivatives”, J. Physics. Math. Theor. 40 (24), 6287– tional Kalman filter and its application to estimation over lossy
6303 (2007). networks”, Signal Processing 91 (3), 542–552 (2011).
[22] O.P. Agrawal, “A general finite element formulation for frac- [29] T. Kaczorek, “New stability tests of positive standard and frac-
tional variational problems”, J. Mathemat. Analysis and Appl. tional linear systems”, Circuits and Systems 2, 261–268 (2011).
337 (1), 1–12 (2008). [30] T. Kaczorek, “Positive linear systems consisting of n subsys-
[23] G.S.F. Frederico and D.F.M. Torres, “Fractional conservation tems with different fractional orders”, IEEE Trans. on Circuits
laws in optimal control theory”, Nonlinear Dynamics 53 (3), and Systems 58, 1203-1210 (2011).
215–222 (2008). [31] T. Kaczorek, Selected Problems of Fractional Systems Theo-
[24] Z. D. Jelicic and N. Petrovacki, “Optimality conditions and ry, Lecture Notes in Control and Information Sciences 411,
a solution scheme for fractional optimal control problems”, J Springer, Berlin, 2011.
Struct Multidisciplinary Optimization 38 (6), 571–581 (2008). [32] T. Kaczorek, “Positive fractional continuous-time linear sys-
[25] R.K. Biswas and S. Sen, “Fractional optimal control problems tems with singular pencils”, Bull. Pol. Ac.: Tech. 60 (1), 9–12
with specified final time”, ASME J. Comput. Nonlinear Dyn. (2012).
6, 021009.1–021009.6 (2011). [33] A. Dzieliński and P. M. Czyronis, “Fixed final time opti-
[26] R.K. Biswas and S. Sen, “Fractional optimal control problems: mal control problem for fractional dynamic systems – linear
a pseudo-state-space approach”, J. Vib. Control 17, 1034–1041 quadratic discrete-time case”, Advances in Control Theory and
(2010). Automation 1, 71–80 (2012).
[27] D. Sierociuk and A. Dzieliński, “Fractional Kalman filter algo- [34] A. Dzieliński and P. M. Czyronis, “Computer algorithms for
rithm for the states, parameters and order of fractional system solving optimization problems for discrete-time fractional sys-
estimation”, Int. J. Appl. Math. Comput. Sci. 16 (1), 129–140 tems”, Eur. Control Conf. ECC 1, CD-ROM (2013).
(2006).
Unauthenticated
Download Date | 7/6/16 12:41 PM
View publication stats