Anda di halaman 1dari 4

2007 4th International Conference on Electrical and Electronics Engineering (ICEEE 2007)

IJu J

Decentralized Fuzzy Gain Scheduling Control for an Open Irrigation Canal


Prototype
0. Begovichl. E. Martinez', V. M. Ruiz2
'Unidad Guadalajara, CINVESTAV-IPN, Jalisco, Mexico
2IMTA, Juitepec, Morelos, Mexico
Phone (33) 37703700 Fax (33) 37703709 E-mail: obegoviggdl.cinvestav.mx
the three LQG, as in classical Gain Scheduling Control [9],
Abstract A decentralized fuzzy gain scheduling control fuzzy logic is used. The proposed fuzzy sets associated with
is designed and tested in simulation on a multipool open the premise variable (inflow signal), allows to switch the
irrigation canal prototype. The main objective of the controller three controllers in soft way to attain stability and good
is to regulate the downstream level of each canal's pool in spite performance. For the canal considered in this paper: the
of large inflow disturbances. Each local controller, of the
decentralized control, is a gain scheduling controller composed model to design the controller is an Input-Output (I/0)
by a bank of three Linear Quadratic Gaussian (LQG) model obtained by identification; the canal operation is
regulators switched by fuzzy logic. The adequate closed-loop constant level downstream and the control method used is
performance obtained suggests the evaluation of the developed upstream control [10]. The control variables are the opening
scheme on field applications. of the slide gate structures. In order to evaluate the level
regulation performance, large inflow variations are
Keywords Gain Scheduling, irrigation canals, fuzzy introduced at the canal head.
logic, control applications.
This paper is organized as follows. In section 2 the
I. INTRODUCTION characteristics of the laboratory canal are described and the
linear model used to design the controller is presented. In
Adequate water administration and distribution in Section 3, some preliminaries about LQG controllers and,
agriculture requires particular attention, since agriculture is fuzzy logic are given. Section 4 describes the methodology
the biggest water consumer activity. Distribution of water to obtain the controller. In section 5, the closed-loop results
usually requires an extensive canal network to transport are presented. Finally, conclusions and future work are
water from storage reservoirs to farmers. Currently, canals stated.
are operated manually, with a large staff, in order to obtain
reliable irrigation service. To improve canal operation and II. MODEL
irrigation service, automatic control offers an attractive
solution. Canal description. The laboratory canal used in this study
is a zero slop rectangular canal of 64 cm wide, 50 m long
There are several works on automatic canal regulation. and 1 m high available at IMTA (Mexican Institute of Water
These works range from PI [1], LQ regulator [2], nonlinear Technology), see Fig. 1. The control structures are slide
control [3] to partial differential equation controllers [4]. gates and they divide the canal in four pools. The inflow is
However, nowadays, there is not a panacea controller, and adjusted with a servo-valve. At the downstream end of the
all researches provide a data base useful to find the canal, the level is regulated by a manual overshot gate. Each
controller able to satisfy the multiple specifications required gate is equipped with a linear actuator and two pressure
for a particular canal. In particular, an adequate water level sensors (upstream and downstream). All gates operate in
controller must be able to regulate the level in each pool of submerged condition. There are not lateral gates.
canal in spite of flow disturbances caused by weather, water
offtakes and/or variation in feeding sources. Nonlinear Model. In an open canal, the flow and level
dynamics are described by two nonlinear partial differential
Inspired in the nice results obtained in [5] and [6], in equations called the Saint-Venant equations [11]. This
this work, a decentralized fuzzy gain scheduling control is model is used to study the flow behavior, but in general, it is
designed to this purpose. A decentralized control for an not used for control design due to its complexity. Then a
irrigation canal consists of local controllers, one for each simple model must be found for this aim.
canal's pool and driven only by the respective pool signals.
In our design, each local controller is a Gain Scheduling Linear Model. In order to design the controller, Input-
controller composed by a bank of three LQG controllers [7] Output (I/0) linear models are derived in this work. This
which are switched by fuzzy logic [8]. Each LQG controller procedure consists in the identification of three transfer
is designed to regulate the level at its reference value, but matrices, one for each set point signaled in Table 1.
considering different inflow conditions. In this way each
local controller resulting will be able to handle bigger flow
disturbances. Instead of use standard logic to switch among

262 IEEE Catalog Number: 07EX1762C


1-4244-1166-1/07/$25.00 ©2007 IEEE. ISBN: 1-4244-1166-1
Library of Congress: 2007923398
2007 4th International Conference on Electrical and Electronics Engineering (ICEEE 2007)

IJu J

Slide U1 U2 U3
3.5 ~~~2.5 18

y]
Overshot
gate (cm)|0 3000 0 3000 0
K
3000
3-
18

13m P11 12m Pl 12m Ml 13m 0 3000 0 3000

14 B5m
Y3
Figure 1. Hydraulic canal prototype scheme. (cm)
Time (s) 0 3000
To find the transfer functions forming a transfer matrix, a Figure 2. Identification and validation when the inflow is 801/s
standard identification procedure is used [12]: The first step
is to determine the input and output variables. The opening 0.85 0.625 4.2
deviations, from the kth set point, of the gates located at the 48s + 1 168s + 1 3890S2 + 399s + 1
downstream end of the pools are the input variables. They 0.70 4.675
are denoted by: u k ( 1, 2, 3), where j denote the jth-pool
H 1 (s) = 0
149s + 1 365s + 1
(1 a)
and k (k=1,2,3) the set point. The downstream water level 0 0
5.12
deviations from the set point are the output variables. They 326s + 1
are denoted by yik (i =1, 2, 3), where i denotes the ith-pool.
1.21 0.862 6.806
During the second phase, the variations in the water level yi 72s + 1 245s + 1 5420s2 + 552s + 1
(i = 1, 2, 3) are registered when a step signal is applied at 0.93 7.064
each gate. In Figure 2, the levels yi" (i = 1, 2, 3) are depicted. H2 (s) = 0
216s+1 514s+1
(lb)
Similar plots are obtained for the other two set points. Note 0 0
7.42
in Fig. 2 that these levels can be approached by responses of 462s+1
linear systems. From information of each level curve a 2.04 1.762 10.609
transfer function can be derived using an identification 84s+1 348s +1 7150S2 + 725s + 1
toolbox (e.g. Matlab System Identification Toolbox). When H3(s)= 0
1.81 10.739
all transfer functions for the set point k are estimated, the 360s + 1 691s+1 (1 c)
transfer matrix for this set point can be completed (for more 10.83
0 0

detail about this procedure the reader can be addressed to 653s + 1


[13]). The transfer matrices Hf(s) obtained are shown in (1). Then, to design the local controllers we use only the
Validation between the estimated linear system H'(s) and diagonal entries of each transfer matrix in (1). For example
canal responses are also shown in Figure 2. to design the first local gain scheduling (i.e. the controller
As it is discussed in [14] and [6], systems with triangular for the first pool) we use.
transfer matrices, as in our case, present low interaction
among local subsystems, and then it is possible to get a (s)= 0.85 1(s)=1.21 2.04 (2a)
48s±1 I 72s±1 I 84s±1
decentralized controller with acceptable performance. to design the second local gain scheduling
TABLE 1. SET POINTS
0.70 0.93 1.81 (b
Set Inflow u l (cm), u 3 Ykol Yko2 yko3
22(S)
149s I;
=
216s + 1
+ 360s + I
Point (I/s) uk2 (cm) (cm)
and to design the third local gain scheduling
(cm) (cm) (cm)
I; h3,3(s)4627.42-+I;
5.12 2 10.83 c
h3 (s) 5. -

h33(s) 653s+
1 80 20 20 70.7 63.5 53.5
2 65 14.7 15.6 70.7 63.5 57.3
11.6 III. PRELIMINARIES
3 50 10.9 70.7 63.5 57.5
A. LQG control
Let a minimal state representation of a linear plant

;x(t) = Ax(t) + Bu(t) + v(t)


y(t) = Cx(t) + w(t)

263 IEEE Catalog Number: 07EX1762C


1-4244-1166-1/07/$25.00 ©2007 IEEE. ISBN: 1-4244-1166-1
Library of Congress: 2007923398
2007 4th International Conference on Electrical and Electronics Engineering (ICEEE 2007)

IJu J

where x is the state, u the input, y the output, v and w are inflow to each pool i (the first is measured and the other two
noises with spectrums Qf and Rf respectively . Under flows are estimated). According to its values, the fuzzy sets
habitual assumptions, the LQG signal u that minimizes depicted in Fig. 3, are proposed. The following rules
J =E{xT (t)Q6X(t) + uT (t)R6u(t)} indicate the way to blend the three controllers (See [13], [5])
is given by: 1
If Q. is M1 then u' -LQGi(s)y,i
S
u(t) = -Kk(t)
where Q, and R, are LQ weighting matrices, K is the LQ s
gain [7] and x(t) is the Kalman estimated [7]. The Kalman If Qiis M2then u, -

S
LQGi()~
estimated state is obtained from a Kalman filter [7], given
by If Q, is M3 then u - LQG, (s)y,1
x(t) = Ax(t) + Bu(t) + L(Cx(t) - y(t)) s

where L is the Kalman gain, and A, B, C are the matrices of


the system. The LQG control is a simple and modem Ui =
Xuk'A k(Qi)
k=l
technique, well known in control theory that in many control
problems offers very attractive solutions. The interested where i is the pool (i=1,2,3), Qi is inflow to i-th pool,
reader can be referred to [7].
Yni= yi- yoi and LQGk (s) 4 (sI - Ak BkK7 CCk )-l Kk
B. Functional Fuzzy Systems The values of gains Kik and Lik obtained are found in [ 13].
In a functional fuzzy system, the i-th Rule has the form [8] The scheme for a local controller is shown in Fig. 4.
If z1 is Mil,. . . ,zj is Mij,... ,zg is Mig Then bi=f(-)
where zj (=1,...,g) are the premise variables; 1=1.r are V. RESULTS
the fuzzy rules; Mij are the fuzzy sets and f(o) is a function
To test the designed controller, the inflow variations showed
in the argument (.). The premise of this rule is defined as in in Fig. 5 are applied to canal. In Fig. 6 the level responses
a standard fuzzy system. The consequent, instead of a for these inflow variations are shown when the gain
linguistic term with an associated membership function, is a scheduling control designed is used and when a single LQG
function. The argument of each f can be the premise is also used. In this figure can be seen that the best level
variables, but other variables may also be used. The choice regulation is obtained from our design. Control actions,
of these functions depends on the application being satisfying physical constrains (opening limits and rates) are
considered. Defuzzification may be obtained by obtained and their curves can be found in [13]
r r
b ZbQf,(z)
i=l
ZI(z)
i=l
where
r I
E fi' (z) # H Hij (zj )
°; * * f (z) = j=l
i=l

and tij(zj) is the membership function of zj and


z [Z1 Zg]1 ls)
Figure 3. Fuzzy sets
IV. CONTROL DESIGN

The goal of the controller designed is to regulate the


downstream level at the end of the first three pools in face of
large inflow disturbances. To do that, each local Gain
Scheduling Controller (i.e the controller of each pool) is
composed by 3 LQG, each of them designed for a different
inflow condition. To reject flow disturbances and the small
existing interactions between pools, an integral action is
added to each transfer of (2). Then a LQGki is designed
using a state space representation (Aik1Bik; Cik) of each
transfer function in (2) augmented with the integral action.
Instead of use standard logic to switch among the three
LQGki k-1,2,3 as in classical Gain Scheduling Control, Figure 4. Proposed local controller scheme
fuzzy logic is used. To do that, the premise variables are the

264 IEEE Catalog Number: 07EX1762C


1-4244-1166-1/07/$25.00 ©2007 IEEE. ISBN: 1-4244-1166-1
Library of Congress: 2007923398
2007 4th International Conference on Electrical and Electronics Engineering (ICEEE 2007)

IJu J

80F
VI. CONCLUSION
The decentralized fuzzy Gain Scheduling Control designed
has presented good regulation in face of large inflow
I-N
disturbances, then this controller can be a good option to
cn
1-11 control a real canal working in a similar situation.
8
It-0
As future work, this controller will be tested in real-time in
the prototype under study.

ACKNOWLEDGMENT
40
0 10 30 50 70 90 110 130 Author thank to the LAFMAA project for financial
Time (min)
support.
Figure 5. Inflow variations
REFERENCES
100

80 - Gain Scheduling [1] Mareels I., E. Weyer, Su Ki Ooi, M. Cantoni, Y. Li, G. Nair,
60 "Systems engineering for irrigation systems: successes and
40 _ challenges", 16 th IFAC World Congress, Prague, Czech
20_
Republic, 2005.
[2] Balogun O., M. Hubbard, J. Devries, "Automatic control of canal
0
Single LQG flow using Linear Quadratic Regulator theory", Journal of
80 Hydraulic Engineering 114(1):75-102, 1988.
60 [3] Dulhoste, J-F., Georges, D. and Besan9on, G. "Non-linear
40 _
control of open-channel water flow based on a collocation
control model". ASCE Journal of Hydraulic Engineering.
20_
130(3):254-256, 2004.
v
0 10 30 50 70 90 110 130
[4] Chen M.L. and D. Georges, "Nonlinear robust state feedback
Time (min) control of an open-channel hydraulic system", European Control
Conference Ecc'01. Porto, Portugal, 2001.
(a) [5] Begovich O., V. M. Ruiz, D. Georges, G. Besan9on, "Real-time
inn application of a fuzzy gain scheduling control scheme to a multi-
80
pool open irrigation canal prototype", Journal of Intelligent &
Gain Scheduling Fuzzy Systems, Vol. 16, No 3, pp. 189-199, 2005.
60 .
[6] Begovich O., J. C. Felipe, V. M. Ruiz, "Real-Time
40 implementation of a decentralized control for an open irrigation
20 canal", Asian Journal of Control, Vol. 9, No 2, pp. 170-179, Jun
0
2007.
[7] Anderson B. D. and J. B. Moore, Optimal Control: Linear
80 Single LQG Quadratic Methods, New Jersey, Prentice Hall, 1990.
60 [8] Passino K.V., S. Yurkovich, Fuzzy Control, Addison-Wesley,
40 1998.
20
[9] Leith D. J., "Survey of gain-Scheduling analysis & design",
International Journal ofControl, 73, pp. 100 1-1025, 2000.
0
0 10 30 50 70 90 110 130 [10] Buyalski C.P., D.G. Ehler, H.T. Flavey, D.C. Rogers and E.A.
Time (min) Serfozo, Canal Systems Automation Manual. Vol. 1. US
(b) Department of the Interior, Bureau of Reclamation, Denver
EUA, 1991.
100 [11] Chow V. T., Open-channel hydraulics, McGraw-Hill, 1988.
80 [12] Ljung, Lennart, System Identification Theory for the User,
Gain Scheduling Prentice Hall, 1987.
60
[13] Martinez-Maldonado E., "Decentralized Control using fuzzy
40
logic for an irrigation canal prototype", (in spanish) MSc. Thesis,
20 CINVESTAV-Gdl, April 2007.
I.-I
0 _ [14] Skogestad S., I. Postlethwaite, Multivariable Feedback Control:
80
Analysis and Design, John Wiley and Sons, 1996.
Single LQG
60

40

20

n1 00 3 5 70
0 10 30 50 70 90 110 130
Time (min)
(c)
Figure 6. Level responses: (a) pool 1, (b) pool 2
(c) pool 3.

265 IEEE Catalog Number: 07EX1762C


1-4244-1166-1/07/$25.00 ©2007 IEEE. ISBN: 1-4244-1166-1
Library of Congress: 2007923398

Anda mungkin juga menyukai