Documentos de Académico
Documentos de Profesional
Documentos de Cultura
irrigation canals
J. Mohan Reddy
Department
of Agricultural
Engineering,
University
of Wyoming,
Laramie,
WY,
USA
A lumped-parameter
model was used for simulating the dynamics of an irrigation canal using the
concepts of optimal control theory. Using the state-space approach presented by Corriga et al., the
problem was formulated as a discrete optimal control problem, and the solutions for gate opening in
the presence of known variations in water withdrawal rates were obtained by solving the algebraic
Riccati equation. An example problem was considered, and the variations in the depth offlow obtained
by using the optimal control theory were compared with the results obtained from an unsteady openchannel flow model. For small and slow changes in the j7ow rates, the difference between the two
models in predicting the changes in water surface elevation was not significant. However, as the
variations in withdrawal rates increased, the accuracy of the optimal control model (which is based
upon small perturbations around the uniform flow) in predicting the water surface elevations decreased.
Kevwords: automation
mode1, state-space
of canals, constant-volume
model
Introduction
A significant benefit of a demand- or user-oriented
system is that such a system should be easier to manage
than other systems. By definition, such a system should
automatically
respond to new turnout discharges,
thereby decreasing the need for calculations and adjustments by canal operators to ensure smooth canal
operation. In social conditions in which a limited demand system is more desirable because of the likelihood of farmers stealing water or some other reason,
the farmers themselves might not be allowed to turn
their water supply on or off. An automatic downstream-controlled system makes operation by the canal
personnel much simpler than conventional methods.
Although adoption of a user-oriented water delivery
schedule is desirable, the existing solutions usually involve very detailed hydraulic modelling using a distributed parameter model. 2,3 To make the conversion
from upstream oriented to downstream oriented, there
is a need to develop relatively simple mathematical
models for improved operation of the canal systems
using downstream control. Burt developed a simple
control logic based upon regression analysis of data
obtained from extensive simulations of the unsteady
open-channel flow equations. However, the rules on
amount of gate opening appear a bit arbitrary.4
Lumped-parameter
models are much simpler than
distributed-parameter
models. Corriga et al. presented a continuous-time lumped parameter model for
optimal operation of irrigation canals based upon constant-volume control. However, the control was suboptimal in nature, since the effect of disturbances was
neglected in deriving the control law. This resulted in
a steady-state error in the volume of water maintained
in the reaches.
Rapid developments in digital computing technology coupled with sign&ant
reductions in their cost
make real-time data processing and control more attractive. This calls for discrete-time control rather than
continuous-time. Therefore the objective of this paper
is to present a discrete-time control technique by incorporating the effect of known external disturbances
(changes in withdrawal rates) acting on the system and
compare the performance of the linearized lumpedparameter model with the results obtained using an
unsteady nonlinear open-channel flow model. The case
of random (unknown) disturbances will be dealt with
in a separate paper.
Mathematical
model
by the Saint-
_
Address
cultural
USA.
Received
450
reprint requests
to Dr. Reddy at the Department
of AgriEngineering,
University
of Wyoming,
Laramie,
WY 82071,
Q
;It+B$=o
a6
vav
--$+--+*=&_G
3 January
1989; accepted
20 March
1990
gdx
gat
(continuity)
(1)
(momentum)
(2)
0 1990 Butterworth-Heinemann
Optima,
%(hb(i- 1)
bi
hai)
(3)
qu(i+
qbi
the transformed
.___
J. M. Reddy
z=o
$0
dQ
z+BsH=O
(7)
and
q=o
h=O
(4)
qci
*r
constant-volume
(cg -
u$$ - (2~
dH
+ y)- s(s + P)H = 0
dx
(8)
(5)
Q(x,s)=
and
-Bs
C,-+c,-ffl
&w
(9)
a2
>
S*h
3 + 2&l
The boundary
conditions
Solution of equations
C,; = 2;:::
Derivation
(11)
2(cS - u;,
iaZF
forx; = 0
(12)
fOrXi = Li
(13)
(14)
Let us consider the ith canal reach belonging to a Ncascaded reach open-channel system as shown in Figwe 1, where the reaches are joined by controlled gates.
451
Optimal constant-volume
J. M. Reddy
GATE i+l
GATE i
-LiGATE 1
Vi(s) = B
GATE N
GATE 2
REACH 1
hb(i-l)
hai
(Clieall~ + CIiea2f) dx
hbi ha(i+l)
R?cH
BCli(elL~ -
1) + BC2i(eaz~Lr - 1)
qb(i-l)
qoi
I
%(i-1)
(15)
ff2i
ali
qbi
-1
qa(i+l)
HA;
rf
HBi = 0
0,
and
qci
HAi = 0,
HgifO
(17)
Application
of the superposition
principle results in
(18)
which expresses the volume variation in the ith reach
as a function of the level variations at the upper and
lower ends of the reach. It can be used in designing
the feedback control law that operates on the upperend control gate in order to keep a constant stored
volume in the reach (Figure 1). Corriga et al. used a
proportional (P) control law (the expression for the gate
opening) of the following form in the analysis:
Vi(s) = - G,i(s)Vi(s)
= -Bs(%
+ 2)
= Q,z,;
(22)
(19)
Wzi(S)
=
E
BI
G,i(s) = &
(20)
Appl.
Math.
Modelling,
1990, Vol.
14, September
(25)
(21)
452
Qi(L;y
S) =
-Bs
eW,Li
+ Czj -
Cli ali
a2i
>
= QB;
(26)
and
QB~(s) = WX(S)HA~(S) + W4i(s)HBi(s)
(27)
Optimal
(2%
W&(S) = gg
Al
BI
where Poi is the initial point, corresponding to the uniform flow conditions.
To derive the state equations, equations (25) and
(27) are written in the following form, after using a
low-order approximation for W(S):
h!Aj(s) = T
QA;(s) + +
(I -
Zl;s)QBi(S)
constant-volume
control
for irrigation:
J. M. Reddy
(2%
and the equation for the flow through the upstream end
control gate of the ith reach which, in the time domain,
is given by equation (3). In equation (29), Qcj is the
variation in the water withdrawal rate in the ith reach.
Equation (3) can be linearized and then Laplace-transformed, leading to
(32)
= $
(1 -
TziS)QA;(S) + y
QBi(s)
(33)
Kri = -Kzi
K3i = - Kdi
and the constants Tj; are the time lags in the propagation of an infinitesimal perturbation in the upstream
and downstream directions. By taking the inverse Ltransforms of equations (29), (30), (32), and (33) and
expressing the volume variation in a given reach by
f
u(t) =
(34)
(qaib) - qbi(T)) dT
(35)
Since the objective is to derive a discrete optimal control law, the state equation (equation (35)) must be
converted to a discrete-time version of the following
form (after setting x = v):
B: m
1,
I:mxm,
v:m X 1,
u:l x I
(36)
= Wk)
(37)
(38)
453
Optimal constant-volume
J. M. Reddy
stant error between the actual and desired operating
condition of the canal. Depending upon the magnitude
of the disturbance, the error may become unacceptable. However, the effect of these disturbances can often
be eliminated by using the so-called integral feedback
(Figure 2). This is achieved by appending additional
state variables of the formlo
qua
4
C
(44)
Schematic
GK) are. The more negative these values are, the larger
will be the value of K and therefore the higher the
required signal energy and the associated cost.9
The above discussion suggests that we should try
to make a tradeoff between the rate of decay (response
of the system) of x(k) to its initial value and the energy
of the input. In the quadratic regulator problem this is
done by choosing u(k) to minimize the cost function
2 (x(k)TQx(k)
J =
+ u(k)=Ru(k))
(39)
k=l
(40)
feedback
F[Pk+l
(41)
pk+ ,G(GTpk+
,G +
WGTPk+
x F + Q
,)I
(42)
with
(43)
454
Appl.
Math. Modelling,
WV = [x(k) w(k)lT
F=
OT
[ TD
I 1
G = [G
GJT
C = [C
O]T
(46)
u(k) = [K,
(47)
or
K,]
The second term in equation (47) accounts for the disturbances. In the absence of any disturbances the outer
feedback loop in Figure 2 is not required (i.e., K, =
0) to bring the system to equilibrium.
Evaluation
- Q)
PN = H
in which
(45)
of the constant-volume
method
Optimal constant-volume
whether it is a constant-volume control or a constantlevel control, we are always interested in knowing the
expected deviations in the depth of flow at the downstream end of the reach where the lateral canals are
usually located. This is helpful in deciding the appropriate structures to discharge the required quantity of
water into the lateral canals. To accomplish the above,
the variations in flow rates and depths of flow predicted
by the above model were added to the initial flow rates
and depths of flow, respectively, in the channel. An
unsteady open-channel flow model developed by Hromadka et al. ** was used to simulate the canal dynamics
using the same data. This model needs, as boundary
K12 = 2.85 x 10P5m-*s
K3* = 7.51
z-1, = 1739s
T,* = 2475 s
T,, = 917 s
Tz2 = 1334 s
10-5mP2s
/3; = Om*/s
y; = 3.76m2/s
r;
a; = 0.61 m/s
p; = 5.08 m*/s
= - 2.99 m*/s
~4 = 0 m2/s
[ 0.9955
0.0038
and
0.9933
0.0024
J. M. Reddy
The gain matrix K was obtained by solving the steadystate Riccati equation:
K =
-5.160
1.525 x lop3
x lop4
1O-4
-5.875
5.002 x lop3
[ - 1.570
1.943 x 10-4
IO-3
G=
-8.817
33.700
- 7.402
1.870 1
-- 27.905
30.200
The weighting matrices are diagonal in nature, and the
elements of the matrices are usually found by trial and
error. The selection is based upon the desired closedloop dynamics of the given system. The elements of
the weighting matrices were as follows:
0.9955
F = [ 0.0038
9.9773
0.0191
0.0024
0.9933
0.0121
9.9663
0 0
0 0
1 0
0 1
and
30.203 1
G = I - 7.4020
151.1451
1 -37.1480
- 27.9052
33.7322
- 139.7006
168.9403
Figure 3.
lation
Geometry
of the irrigation
0.0290
[ 0.0014
0.0192
0.0268
0.0011
-0.0002
0.0005
0.0010
These values were used in simulating the system dynamics by using equation (45).
Initially, a 10% change (increase) in the farmers
withdrawal rate (qc(l> = 0.295 m3/s and qc(2) = 0.25
Appl.
Math.
Modelling,
1990,
Vol.
14,
September
455
Optimal constant-volume
J. M. Reddy
m3/s) was introduced in both the reaches, and the variation of the gate openings, changes in the volume of
water (state of the system), and changes in the depth
of flow and the flow rates at the upstream and downstream ends of both the reaches were simulated. Simulations were also done by increasing the changes in
the water withdrawal rates to 20% of the original withdrawal rates.
Results and discussion
First the volume variations in both the reaches were
simulated for a 10% change in the withdrawal rates.
Figure 4 presents the volume variations obtained by
using the proportional (P) and proportional-plus-integral (PI) control laws. The predicted variation in volumes using the PI control was less than 3 m3 in contrast
to 80 m3 obtained by using Corriga et al.s5 method.
This shows the advantage of the PI controller over the
proportional (P) controller in keeping the volume variation at almost zero. But it should be pointed out that
the percent variation in the volume of water was still
negligible in both cases for a 10% change in withdrawal
rate.
The gate openings predicted by the discrete-time
model with the PI controller (Figure 5) were compared
with the values predicted by the continuous-time model
with the P controller (Figure 6). The gate openings
predicted by the PI controller (in the presence of disturbances) were slightly higher than the values predicted by the P controller. This resulted in negligible
volume variation in the case of the PI control. Also,
the PI control has some overshoot at the beginning and
end of the disturbance period. This can be smoothed
out by increasing the value of R in the cost function
(equation (39)).
The variations in flow rates at the upstream and
downstream ends of the reaches are depicted in Figure
7. These variations in flow rates in both the reaches
were less than the required values. The total change
required in flow rate in the first reach was 0.55 m3/s
(0.298 m3/s and 0.250 m3/s). But the actual variation
obtained by using the linear optimal control model was
approximately 0.40 m3/s. This resulted in a negative
change in the volume of water stored in both the reaches.
In Figure 8 the predicted variations in the depth of
flow at the upstream (hai) and downstream (hbi) ends
of the reaches are presented. The change in the depth
of flow was positive at the upstream ends of the reaches
because of the increase in flow rates into the reaches.
The maximum increase in the upstream depth of flow
was 0.027 m and occurred in the first reach. Conversely, the change in the depth of flow at the downstream end of the reaches was negative with a maximum deviation of - 0.025 m. This also occurred in the
first reach. These predicted variations in depth are
negligible.
Later on, the variations in the volume due to a 20%
change in withdrawal rate were simulated. Again, the
volume variations in the first and second reaches were
found to be negligible (Figure 9). The variations in gate
456
Appl.
Math. Modelling,
-1001
0
500
1000
duration
1500
2000
of disturbance.
see
3000
2500
2
duration
Figure 5. Variation
mand: discrete-time
-0.1
I
0
3
of simulation.
4
set
5
(thousands)
I
duration
3
of disturbance.
sec(thousands)
Figure 6. Variation in gate openings for a 10% change in demand: continuous time model with P control
Optimal constant-volume
2
duration
4
5
sec(thousands)
3
of stmulatmn.
i
a
n
9
e
c------,
0
2
duration
-7,
3
of simulatmn,
J. M. Reddy
-0.04
I
4
5
s&thousands)
l-
m
e
2
duration
3
of simulation,
4
5
secfthousands)
.__
a
r
-1
i
0
n
0.2
-2
:
a
reach 2
react! I
L
m
0.1
:
e
-3
-4A
O-0.1
f
0
2
duration
3
of simulatmn.
4
5
sec(thousands)
0
w
-0.2
-0.3
-0.4
7
h
-0.5
-0.6
L
0
r
1
2
duration
3
of simulation.
4
5
secfthousands)
Appl.
Math. Modelling,
457
Optimal constant-volume
J. M. Reddy
Appl.
References
1
458
(which is based upon the linearized Saint-Venant equations) in predicting the deviations in depth of flow decreased. By using the distributed-parameter
(full hydrodynamic) model, the range in which the performance
of the optimal control model is acceptable can be identified for any given system. Since the computational
effort involved in the optimal control model is much
less, it can be used for operation of irrigation canals.
Math. Modelling,
9
IO
11
12
13
14