Environmental Modelling & Software 51 (2014) 207e220

Contents lists available at ScienceDirect

Environmental Modelling & Software

journal homepage: www.elsevier.com/locate/envsoft

Mathematical model for robust control of an irrigation main canal

pool
R. Rivas-Perez a, *, V. Feliu-Batlle b, F.J. Castillo-Garcia b, A. Linares-Saez c
a
Department of Automatica and Computer Science, Havana Polytechnic University, Calle 114 No 11901, CUJAE, Marianao,
Ciudad de la Habana C.P. 19390, Cuba
b
Escuela Tcnica Superior de Ingenieros Industriales, Universidad de Castilla-La Mancha, Campus Universitario s/n, Ciudad Real C.P. 13005, Spain
c
ABEIMA, Avda la Buharia, 2, Sevilla 41018, Spain

a r t i c l e i n f o

a b s t r a c t

Article history:
Received 3 October 2013
Accepted 3 October 2013
Available online 31 October 2013

This paper describes the formulation and development of a mathematical model for high-performance
robust controller design techniques, based on a complete identication for control procedure, of an irrigation main canal pool (true plant), which is characterized by the exhibition of large variations in its
dynamic parameters when the discharge regime changes in the operating range [Qmin, Qmax]. Real-time
eld data has been used. Four basic steps of the proposed procedure have been dened in which all the
stages, from the design of the experiments to the model validation, are considered. This procedure not
only delivers a nominal model of the true plant, but also a reliable estimate of its model uncertainty
region bounded by the true plant models under minimum and maximum operating discharge regimes
(limit operating models). The model uncertainty set, dened by the nominal model and its uncertainty
region, is characterized by its being as tight as possible to the true irrigation main canal pool. The obtained results are very promising since this kind of models facilitates the design of robust controllers,
which allow improving the operability of irrigation main canal pools and also substantially reduce water
losses.

2013 Elsevier Ltd. All rights reserved.

Keywords:
Systems identication
Irrigation canal
Time-varying parameters
Model uncertainty set
Robust controller
Water efcient use

1. Introduction
A signicant part of the control system design of irrigation main
canals is devoted to obtaining their mathematical models. These
mathematical models should provide an accurate description of the
relevant irrigation main canal pool dynamics. The physical dynamics of an irrigation main canal pool (plant) are usually modeled
and simulated by using the SainteVenant equations, owing to their
capacity to represent the nonlinear hydraulic characteristics of real
interest (Chaudhry, 1993). These equations are not easy to use
directly as a model for control system design (Kovalenko, 1983;
Litrico and Fromion, 2009; Rivas-Perez et al., 2007). Linearization
or simplications of the SainteVenant equations are therefore
recurrently used by the irrigation canal control research community. Linear and rational models open up the possibility to apply
well-known control system design techniques, which are relatively
easy to implement.

* Corresponding author. Tel.: 53 72611631, 53 72663285.

E-mail addresses: raul_rivas_perez@yahoo.es, rivas@electrica.cujae.edu.cu
(R. Rivas-Perez), Feliu.Batlle@uclm.es (V. Feliu-Batlle).
1364-8152/$ e see front matter
2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.envsoft.2013.10.002

Water demand for irrigation varies with time as a result of the

users variable water needs. In order to satisfy water demand, the
irrigation main canals need to be operated under different
discharge regimes in the range [Qmin, Qmax], in order to maintain
water levels and supply the desired discharge rates at specic locations. Experiments developed by certain authors (Deltour and
Sanlippo, 1998; Litrico et al., 2006; Rivas-Perez et al., 2008a;
Schuurmans et al., 1999) conrm that when the discharge regimes
change in the operating range [Qmin, Qmax] and/or other hydraulic
parameters change, the irrigation main canal pools may exhibit
large variations in their dynamic parameters. The mathematical
models to be obtained must therefore consider these parameter
variations. Indeed, control system design methods are usually
based on a nominal model, whereas the dynamics of irrigation
main canal pools vary with the alteration in operating hydraulic
conditions, thus causing uncertainties in the nominal model (FeliuBatlle et al., 2011; Litrico et al., 2006; Rivas-Perez et al., 2011).
Identication for control is an area in which a renewed interest
has been shown since the beginning of the 1990s and which still
attracts a growing number of researchers (Gevers, 2005). One of its
main objectives is to estimate mathematical models that are suitable for high performance robust control design techniques

208

R. Rivas-Perez et al. / Environmental Modelling & Software 51 (2014) 207e220

(control-oriented models), i.e. estimate the plant nominal models

and its uncertainties regions (Chen and Gu, 2000).
Various works concerning the design of robust controllers to
control water distribution in irrigation main canal pools, which are
characterized by large time-varying dynamic parameters, have
been reported (Calderon-Valdez et al., 2009; Feliu-Batlle et al.,
2005, 2011; Litrico and Fromion, 2009; Rivas-Perez et al., 2002).
These controllers should guarantee a specied minimum level of
performance for the whole range of variation of canal pool
dynamical parameters (model uncertainty set). One of the main
problems in the design of robust controllers is that of systematically
obtaining the model uncertainty set when the irrigation canal is
operating under different discharge regimes (Litrico and Fromion,
2009; Kovalenko et al., 1993; Rivas-Perez, 1984). In this paper, an
identication for control procedure is used to obtain a reliable
model uncertainty set of a main irrigation canal pool when the
design of a robust controller is demanded.
The paper makes an intensive use of standard model structures
and algorithms found in the literature, for example in Data-Based
Mechanistic (DBM) modeling of hydrological and other environmental systems (see, e.g. Andrews et al., 2011; Camacho and Lees,
1999; Jakeman et al., 2006; Lekkas et al., 2001; McIntyre et al.,
2011; Ochieng and Otieno, 2009; Ooi and Weyer, 2008; Price
et al., 1999; Romanowicz et al., 2006; Young, 1998, 2011; Young
and Garnier, 2006).
The main contributions of this paper are: 1) to the best of our
knowledge it is the rst time that a complete algorithmic procedure
has been formulated and developed (using real-time eld data) for
the identication for robust control of an irrigation main canal pool,
which is characterized by the exhibition of large variations in its
dynamic parameters when the discharge regime changes in the
operating range; 2) the estimation, through the use of real-time
eld data and the Prediction Error Framework, of a nominal
model of the true plant and its explicit uncertainty region, bounded
by the true plant models under minimum (lower) and maximum
(upper) operating discharge regimes (limit operating models); 3)
the derivation of a true plant model uncertainty set, dened by the
nominal model and its uncertainty region, which is characterized
by its being as tight as possible to the true plant; 4) the design of a
robust controller for an irrigation main canal pool based in this
model uncertainties set. The approaches used in this paper for
obtaining the mathematical model are within the indirect techniques based directly on the experimental data (Garnier and Wang,
2008). The obtained plant mathematical model has been developed
using the software platform of the System Identication Toolbox of
Matlab.
The paper is organized as follows. In Section 2 the algorithmic
procedure is presented in four steps. The main results are given in
Section 3. Section 4 provides some comments and conclusions.

Remarks: a) the model parameters of an irrigation main canal

pool change with variations in its operating discharge regime (Q(t)).
The identication experiments should therefore be developed so
that they represent the typical discharge regimes in which the true
plant is operated; b) according to robust control theory it is desirable to reduce the model uncertainty region as little as possible, i.e.,
to make the nominal model represent the true plant dynamics as
well as possible, or to make the model uncertainty set as tight as
possible; and c) in irrigation main canal pool practice the limit
operating discharge regimes are obtained when Q(t) Qmin (minimum (lower) limit of Q(t)) and Q(t) Qmax (maximum (upper)
limit of Q(t)) with Q(t) [Qmin, Qmax]. The subscripts min, nom, max
represent respectively the minimum, nominal and maximum
discharge regimes.
This paper focuses on the application of this four basic steps
algorithm in order to obtain a mathematical model for highperformance robust control of an irrigation main canal pool.
3. Results
In this Section we develop the four basic steps algorithm in the
identication for control procedure proposed in Section 2 for the
rst pool of the Aragon Imperial Main Canal (AIMC).
3.1. Irrigation main canal description
The irrigation main canal considered in this paper is the Aragon
Imperial Main Canal, which pertains to the Ebro Hydrographical
Confederation in Zaragoza, Spain. It obtains its water from the Ebro
River thanks to the elevation of the Pignatelli dam. The water
passes through the Gate House, which controls the designed
discharge at its origin, although this value may sometimes be superior as a result of a high ow in the Ebro River. The Ebro River
represents the Spanish most important catchment of the Iberian
Peninsula (Pedregal et al., 2009). The AIMC is a 108 km long crossstructure canal. It has a trapezoidal cross-section and ten pools of
different lengths which are separated by undershoot ow gates.
Several of the pools in the AIMC are characterized by timevarying dynamics when the discharge regimes change in the
operating range [Qmin, Qmax] and/or other hydraulic parameters
change (Rivas-Perez et al., 2011). The controllers which are
currently installed do not, therefore, guarantee an effective water
distribution control. In order to improve the control and management of the whole AIMC, the implementation of decentralized
robust control systems of water distribution has been considered
for canal pools that are characterized by time-varying dynamics.
These systems will be based on independently tuned robust controllers to which additional control elements will be added in order
to minimize pool interactions. The rst step in this study is,
therefore, to obtain the control-oriented models of the canal pools.

2. Methodology
3.2. Design of the experiments (step 1 of the procedure)
We propose the following four basic steps algorithm for the
identication for control procedure of a true irrigation main canal
pool from N eld sampled measurements of the input and output
Z N futk ; ytk gN
k1 .
Step 1: Design of the experiments;
Step 2: Data collection, parameter estimation and validation of
the linear nominal model of the true plant under nominal operating
discharge regime (Q(t) Qnom);
Step 3: Data collection, parameter estimation and validation of
the linear models of the true plant under limit operating discharge
regimes (model uncertainty region);
Step 4: Delivery of the true plant model uncertainty set,
comprised by the nominal model and its uncertainty region.

The eld data and results reported in this paper were obtained
from the rst pool in the AIMC, which is known as the Bocal. It is a
lined trapezoidal canal pool of 8 km in length, with a variable depth
of between 3.5 and 4.15 m, a variable width of between 21 and
26.9 m, and a design maximum discharge of 30 m3/s, in its entire
extension. Fig. 1 shows an upper view of the Bocal in which it is
possible to observe the Ebro River, the Pignatelli dam, and the Gate
House.
This canal pool is operated in a downstream end regulation
mode (Kovalenko, 1983). The downstream end water level is
controlled by means of 10 upstream undershoot gates located in
the Gate House on the side of the canal. The measurements available

R. Rivas-Perez et al. / Environmental Modelling & Software 51 (2014) 207e220

209

Experiments reported in previous works (Litrico and Fromion,

2009; Rivas-Perez et al., 2008b; Schuurmans et al., 1999) show
that canal pools similar to the Bocal experience large variations in
their model parameters when the operating discharge regimes
change across their upstream control gates.
The Bocal is operated under different discharge regimes through
its upstream control gates which change randomly. For control
oriented modeling the main operating discharge regimes are: the
nominal operating discharge regime (Q(t) Qnom(t) 22 m3/s) and
the limit operation discharge regimes: the minimum (lower)
operating discharge regime (Q(t) Qmin(t) 15 m3/s), and the
maximum (upper) operating discharge regime (Q(t) Qmax(t)
30 m3/s). The variation range of the operating discharge regimes of
this canal pool is therefore:
Fig. 1. An upper view of the Bocal.

are the upstream (yup(t)), downstream (ydw(t)) and downstream

end (ydwe(t)) water levels and the position of the control gates
(u(t)). Fig. 2 shows an equivalent diagram of the Bocal in which the
10 control gates are represented by means of an equivalent control
gate.
The opening magnitude of each upstream control gate depends
on the discharge that it is necessary to deliver and which is derived
from the Ebro River. For example, if it were necessary to deliver a
discharge that required an opening magnitude of 150 cm in the
upstream control gates, 5 control gates would simultaneously open
with an opening magnitude of 30 cm, which is equivalent to a total
gate opening magnitude of 150 cm. The 10 control gates can
therefore be represented by means of a single equivalent control
gate, which has an opening range of 0e700 cm. The discharge
through the equivalent control gate is estimated in real time by the
computer (PC) of the canal pool using the respective sensor signals
and the following equation (Chaudhry, 1993):

Q t Cd L

q
p
2g ut yup t  ydw t;

(1)

where Cd is the gate discharge coefcient, L is the gate width, and g

is the gravity acceleration. This information is very important for
accurate operation of the canal pool and for effective management
of irrigation water. The canal pool PC also estimates the variations
of the downstream end water level and of the equivalent control
gate by the expressions: Dydwe(t) ydwe_f(t)  ydwe_ini(t), Du(t)
uf(t)  uini(t) where the subscripts f, ini denote respectively the nal
and initial values.

Qmin ; Qmax  fQ tj15  Q t  30g m3 =s;

with Qnom Qmin ; Qmax :

(2)

The usual operating discharge regime of the Bocal through its

upstream control gates is the nominal regime. The linear model
obtained under the most common operating discharge regime
constitutes the nominal model. When the Bocal is operated under
other discharge regimes, different from the nominal, the dynamic
behavior of the nominal plant undergoes variations, and in this case
the model parametric uncertainties are obtained. These uncertainties are bounded (model uncertainty region) by the limit
operation discharge regimes.
Bearing in mind that the Bocal is operated in a downstream end
regulation mode, the control oriented model will thus consider the
downstream end water level variation Dydwe(t) as the output variable, and the equivalent control gate position variation Du(t) as the
input variable. The fundamental disturbance variable v(t) is the
unknown offtake discharge q(t), which is shown in Fig. 2, together
with the effects of pool interactions (upstream and downstream). It
is not necessary to know the water level variations for the entire
extension of the pool in order to control the water distribution in
the Bocal. It is sufcient to measure them at certain specic points
which will depend on the canal operating method to be used
(Rivas-Perez, 1984). In this case, since the water distribution is
carried out by gravity offtakes, a good distribution is obtained by
maintaining a constant water level at the offtake. Therefore, a linear
model with lumped parameters and a time delay can adequately
characterize the dynamic behavior of an irrigation main canal pool
at the measurement points.

Fig. 2. An equivalent diagram of the Bocal.

210

R. Rivas-Perez et al. / Environmental Modelling & Software 51 (2014) 207e220

Fig. 3. Step response under nominal operating regime of the Bocal.

3.3. Data collection, parameter estimation and validation of the

linear nominal model of the true plant under nominal operating
discharge regime (step 2 of the procedure)

water level variation (cm)

This step has the following two stages: a) nonparametric identication, and b) parametric identication (Ljung, 1999). The
nonparametric identication is very often a rst step in obtaining
experimental information on the dynamic properties of a plant. In
this case a preliminary experiment, such as a step response, is
performed to gain primary knowledge (by visual inspection) about
plant dynamic characteristics. Generally, this provides good insights into important properties of the plant, as e.g. the presence
and length of time delays, possible model order, static gain and
time constants (Van den Hof and Bombois, 2004). The information
obtained from nonparametric identication is then used in parametric identication to determine the plant model based on a more

informative experiment applying a binary sequence as persistent

excitation signal, e.g. a pseudo random binary sequence (PRBS).
As was pointed out in the previous Subsection, the linear
nominal model of the real plant is obtained when the operating
discharge regime through the upstream control gates corresponds
to the nominal discharge regime (Q(t) Qnom(t) 22 m3/s). This
operating discharge regime is usually achieved with a total opening
magnitude of the equivalent control gate of 140 cm.
a) Nonparametric identication: Experiment based on the
response to a step like input was carried out in order to obtain
primary knowledge about the model order and time delay. The
experiment consists of maintaining the downstream gate in a xed
position, and then applying a step command of 120 cm to the
equivalent control gate (unom_ini(t) 20 cm). Therefore, a total of 5
upstream control gates received a simultaneous increment in their
opening magnitudes of 24 cm, in such a way that a total equivalent

83.0

166.0

249.0

332.0

415.0

249.0

332.0

415.0

time (min)

gate opening (cm)

170

120

70

20

83.0

166.0
time (min)

Fig. 4. Experimental eld data of the Bocal obtained with a PRBS under nominal operating discharge regime.

R. Rivas-Perez et al. / Environmental Modelling & Software 51 (2014) 207e220

control gate opening magnitude of 140 cm (unom_f(t)) was reached.

The data of the upstream (yup_nom(t)), downstream (ydw_nom(t)) and
downstream end (ydwe_nom(t)) water levels, along with those of the
total opening magnitude of equivalent control gate unom_f(t), were
registered and stored in the canal pool PC. The sampling device
installed in the Bocal, samples data every second. The water levels
and gate position are given in cm and were uniformly sampled with
a period of 1 min. The variations Dydwe_nom(t) and Dunom(t) were
estimated by the canal pool PC. For simplicity we denote
Dydwe_nom(t) Dynom(t). The plant response under the nominal
operating discharge regime to a step command is presented in
Fig. 3. Based in this response it is assumed that the nominal dynamic behavior of the rst pool in the AIMC can be described by the
following general second order plus time delay transfer function:

Gnom s

Dynom s
Knom

esnom s ;
Dunom s
T1nom s 1T2nom s 1

(3)

where Knom is the static gain, T1nom, T2nom are the time constants,
and snom is the time delay. We consider that T1nom is the dominant
time constant (the largest of those associated with the dynamics of
the canal pool), while T2nom is the smallest time constant that
represents the motors equivalent control gate dynamics secondary canal dynamics. T2nom is usually much smaller than T1nom
(Rivas-Perez et al., 2007).
The transfer function (3) is similar to that obtained by other
authors (e.g. Litrico and Fromion, 2009). It is a standard model of
the type used in DBM modeling of hydrological and other environmental systems (Young, 2011, 1998). The second order assumed
dynamics for this model needs to be conrmed in the next stage
(parametric identication and model validation) of this step using
model structures with different orders, delays and sampling period
(Garnier and Wang, 2008). The approximated nominal values of the
parameters estimated from Fig. 3 are Knom z0.044 cm/cm, T1nom
z46.95 min, T2nom z0.85 min, snom z6 min, tss_nom z78.33 min
(settling time). The model evaluation (Bennett et al., 2013), i.e. the
comparison of the step test and the prediction given by the linear
model (3) with the parameters estimated nominal values, is shown
in Fig. 3 also.
b) Parametric identication: Experiment based on the response
to a pseudo random binary sequence (PRBS) as input was also
carried out in order to collect data containing the maximum information with regard to the dynamic behavior of the nominal
plant. The PRBS are persistent excitation signals, which contain
frequency spectrums that are sufciently wide to represent the real
plant dynamics (Ljung, 1999). This command sequence was
designed in such a way that signicant, although not very large,
downstream end water level variations were obtained. According
to the linear system theory (Dorf and Bishop, 2005), in order to
capture the main dynamic behavior of a real plant, its input should
be excited around the frequency at which its Bode diagram presents
bends, i.e. around the plant cutoff frequency. This frequency may be
computed by means of the time constants of the nominal plant
obtained in the experiment based on the step command.
It was determined that the PRBS should change the upstream
control gate opening magnitude at intervals that were multiples of
10 min with a maximum variation interval of 50 min. The sampling
period was 1 min. In this case, the downstream gate was again kept
in a xed position and the equivalent control gate received an
increment in its opening magnitude of 120 cm in such a way that
a total equivalent control gate opening magnitude of 140 cm was
achieved. These experiments lasted 415 min (approximately 7 h).
The experimental eld data collected was stored in a computer and
they are shown in Fig. 4. An additional procedure was that of

211

splitting the collected data into data for estimation and data for
validation (left and right of the vertical red line, respectively).
The data collected from these experiments were analyzed with
the purpose of verifying their suitability for the parameter estimation procedure. Their adequacy was veried and these data
were, therefore, used directly in this step of the proposed identication for control procedure. Different model structures such as
ARX, OE, BoxeJenkins and ARMAX were tested to determine which
of them best described the dynamic behavior of the nominal plant.
These structures are the most used in the design of control systems
and are represented by means of the following expressions (Ljung,
1999):

Aq b
y ARX
b
y OE

b
y BJ

nom t

Bq nk
q
unom t xOE
Fq

Bq nk
Cq
x
q
unom t
Fq
Dq BJ

nom t

nom t

Bqqnk unom t xARX

Aq b
y ARMAX

nom t

nom t;

(4)

nom t;

(5)

nom t;

Bqqnk unom t CqxARMAX

(6)

nom t;

(7)

where b
y ARX nom t; b
y OE nom t; b
y BJ nom t; b
y ARMAX nom t are the
nominal model output signals (the estimated downstream end
water level) of the respective structures ARX, OE, BoxeJenkins and
ARMAX (j structures), A(q), B(q), C(q), D(q) and F(q) are polynomials
dened as: A(q) 1 a1q1.anaqna; B(q) b1q1.bnbqnb;
C(q) 1 c1q1.cncqnc; D(q) 1 d1q1.dndqnd;
F(q) 1 f1q1.fnfqnf, na, nb, nc, nd, nf are the orders of the
respective polynomials, ai, bi, ci, di, f are the parameters of the
polynomials to be estimated, nk is the plant time delay, and xARx_nom(t), xOE_nom(t), xBJ_nom(t), xARMAx_nom(t) are uncorrelated random
white noise sequences with zero mean of the nominal model
structures ARX, OE, BoxeJenkins and ARMAX.
The parameter estimation procedure for the selected model
structures was developed using the data located to the left of the
vertical line in Fig. 4 and the Matlab System Identication Toolbox.
q N nom j of each of
The estimation of the nominal parameter vector b
the selected model structures j (from a nominal data set) was carried out on the basis of the Prediction Error framework using a least
mean square criterion to minimize the prediction error. This
parameter vector was computed by means of the following
expression (Ljung, 1999):

b
q

N nom j

arg min
qnom

arg min
qnom

N


1 X
2 t; qnom j
N t1 j
N 

2
1 X
y nomj t; qnom j ;
ynom t  b
N t1

(8)

where j(t,qnomj), b
y nom j t; qnom j are the prediction error and the
nominal model output signal of the selected model structure j, and
N is the total number of the nominal eld data used in the
parameter estimation (N 300). The nominal parameter vector
b
q N nom j was estimated for different model orders, time delays and
sampling periods in order to obtain the nominal model that best
reproduces the eld data in each of the selected model structures j.
Once the parameter estimation phase has been concluded, it is
necessary to decide whether the nominal models obtained are
sufciently accurate for the plants dynamic behavior under the
nominal operating discharge regime. The procedure used to evaluate the quality of the nominal models obtained is known as model

212

R. Rivas-Perez et al. / Environmental Modelling & Software 51 (2014) 207e220

Table 1
Validation results of the plant nominal models.
Model
structure

Model order Performance Variance of the

residual error (cm)
(na, nb, nc, index (FIT)
nd, nf, nk)

ARX
OE
BoxeJenkins
ARMAX
ARX
OE
BoxeJenkins
ARMAX

(1,
(1,
(1,
(1,
(2,
(2,
(2,
(2,

1,
1,
1,
1,
2,
2,
2,
2,

5)
5)
1, 1, 5)
1, 6)
6)
5)
2, 2, 7)
2, 6)

72.55%
72.86%
72.41%
72.55%
76.28%
83.90%
80.26%
87.08%

0.72 < e(t) < 0.74

1.0 < e(t) < 1.25
0.65 < e(t) < 0.64
0.7 < e(t) < 0.7
0.7 < e(t) < 0.58
0.53 < e(t) < 0.55
0.58 < e(t) < 0.6
0.4 < e(t) < 0.4

FPE

AIC

0.0708
0.1676
0.0684
0.0717
0.0499
0.0781
0.0567
0.0355

2.648
1.786
2.682
2.636
2.998
2.549
2.868
3.337

validation (Ljung, 1999). This procedure is then applied based on the

cross-validation method (Bennett et al., 2013) and by using the
experimental data located to the right of the vertical red line
(N 120) shown in Fig. 4, which were reserved for this purpose.
This procedure gives a good indication of whether or not the
identied model captures the dominating dynamics of the true
plant. The cross-validation results of the estimated nominal models
showed that the best dynamic behavior in each of the selected
model structures j was obtained with second order models. These
results are presented in Table 1. These second order nominal
models were adjusted with different time delays and sampling
period.
The nominal model that would best reproduce the dynamic
behavior of the nominal plant was selected by quantifying the
degree of accuracy of each of the nominal models with regard to the
validation data. This was done by using a performance index (FIT),
which constituted a quantitative measure of the model quality
which could be obtained from the norm of the residual errors. The
variance of the residual error was also used, which in turn is equal
to the average value of the residual error spectrum (Ljung, 1999).
The residual error is the part of the data that the model could not
reproduce, and it therefore bears information about the quality of
the model. Performance indexes and variances of the residual errors calculated from the second order nominal models obtained are
shown in Table 1. For comparison purposes, this table also shows
the performance indexes and the variances of the residual errors
obtained from the rst order models of each of the selected model
structures j. Table 1 shows that the best performance index (87.08%)
was obtained by the second order nominal model with the ARMAX
structure. This index is approximately 10.8%, 3.2% and 6.8% higher
than the respective indexes obtained with the ARX, the OE and the
BoxeJenkins model structures. This table also shows that the
smallest variance of the residual errors was obtained with the same
ARMAX structure nominal model.
The Akaikes Final Prediction Error (FPE) criterion was used to
evaluate the model quality. Its closely related Information Theoretic
Criterion (AIC) (Ljung, 1999) was also used. According to Akaikes
theory, in a collection of different models the most accurate is the
one with the smallest FPE (or AIC). Table 1 shows that the smallest
FPE (or AIC) was obtained by the second order nominal model with
the ARMAX structure.
Table 2
Performance indexes of the nominal models using different Ts.
Model
Model order Performance Performance Performance Performance
structure (na, nb, nc, index (FIT) index (FIT) index (FIT) index (FIT)
Ts 2 min Ts 1.5 min Ts 1 min Ts 0.5 min
nd, nf, nk)
ARX
OE
Boxe
Jenkins
ARMAX

(2, 2, 6)
74.21%
(2, 2, 5)
80.32%
(2, 2, 2, 2, 7) 76.60%

75.42%
81.19%
78.48%

76.87%
83.20%
80.26%

76.28%
82.90%
79.32%

(2, 2, 2, 6)

85.29%

87.08%

86.41%

85.29%

It is well known that the choice of the sampling period for

obtaining eld data is one of the important options in system
identication (Ljung, 1999). Table 2 shows the effects of the sampling period in the nominal model accuracy with different model
structures. From this table it is observed that the best performance
index (87.08%) was obtained by the second order nominal model
with the ARMAX structure with a sampling period of 1 min. This
model can be explicitly represented by means of the following
expression:

b
y ARMAX

nom t

1:4022b
y ARMAX

nom t

 0:435 b
y ARMAX

 1

nom t

 2

0:0007786unom t  5
0:0005905unom t  6 xARMAX
 0:9736xARMAX

nom t

0:1508xARMAX

nom t  2:

nom t

 1
(9)

The cross-validation results of the nominal model (9) are presented in Fig. 5. This gure shows that the obtained nominal model
adequately reproduces the eld plant data, even when considering
data that were not used in the parameter estimation.
In the residual analysis of the model (9), the auto-correlation
function of the residuals and the cross-correlation function between the input and the residuals do not go outside of the 99%
condence regions, as shown in Fig. 6. The residuals are, therefore,
white and totally uncorrelated with the input signal. We can thus
be satised with the accuracy of the model (9). The true plant
nominal model selected (9) can therefore be represented in the
Laplace domain by means of the following transfer function:

h
b nom s G
bu
G

nom s

bv
H

nom s

(10)

where:

bu
G

nom s

Dynom s
0:0417
e6s ;

Dunom s 15:84s 11:3s 1

(11)

bv
H

nom s

Dynom s
4:24s 10:9s 1
:

Dvnom s
15:84s 11:3s 1

(12)

Expression (11) shows that the parameter values of the nominal

model of the Bocal are different from those obtained from the step
response because the parametric identication procedure is more
rigorous and is not based on graphical results.
3.4. Data collection, parameter estimation and validation of the
linear models of the true plant under limit operating discharge
regimes (model uncertainty region) (step 3 of the procedure)
For the design of a high performance robust controller it is
extremely important to know the variation ranges of the dynamic
parameters of the plant nominal model, i.e. DG(s) (Litrico and Fromion,
2009), since the robust control paradigm pursues the design of one
linear controller to stabilize a set of linear models in DG(s) and achieve
the best worst-case performance over DG(s) (Chen and Gu, 2000;
Gevers, 2005). Therefore, it is necessary to identify the dynamic
behavior of the true plant under the limit operation discharge regimes.
3.4.1. Linear model of the true plant under minimum operating
discharge regime
The minimum operating discharge regime through the upstream
control gates (Q(t) Qmin 15 m3/s) is obtained when the users water

R. Rivas-Perez et al. / Environmental Modelling & Software 51 (2014) 207e220

213

Fig. 5. The cross-validation results of the estimated nominal model (9).

Correlation function of residuals. Output yARMAXnom

1
0.5
0
0.5

10

lag

15

20

25

Cross corr. function between input unom and residuals from output yARMAXnom
0.5

0.5
25

20

15

10

0
lag

10

15

20

25

Fig. 6. Correlation test of the estimated nominal model (9).

demand (water withdrawal) decreases in relation to the nominal

regime as a result of a reduction in the water needed for irrigation. In
this case, a minimum volume of water ows through the upstream
control gates, the celerity coefcient decreases, and the canal pool
dynamic parameters therefore undergo considerable changes (Litrico
and Fromion, 2009; Rivas-Perez et al., 2011). This operating discharge
regime is usually achieved with a total opening magnitude of the
equivalent control gate of 120 cm. Therefore, the equivalent control
gate received an increment in its opening magnitude of 100 cm in

such a way that a total gate opening magnitude of 120 cm was achieved. The experimental eld data collected under this operating
discharge regime were stored in a computer and ltered. These
experimental eld data are shown in Fig. 7.
The model that best represents the dynamic behavior of the true
plant under this operating discharge regime was estimated by
following the same procedure as that shown in Subsection 3.3. As a
result, the following second order model with the ARMAX structure
and time delay was obtained:

Table 3
Validation results of the plant ARMAX models under limit operating discharge regimes.
Limit operating
discharge regimes

Model order
(na, nb, nc, nk)

Performance
index (FIT)

Variance of the
residual error (cm)

FPE

AIC

Minimum operating
discharge regime
Maximum operating
discharge regime

(2, 2, 2, 7)

85.6%

0.46 < e(t) < 0.37

0.0389

3.245

(2, 2, 2, 5)

80.04%

0.27 < e(t) < 0.42

0.0644

2.739

214

Fig. 8. The cross-validation results of the true plant ARMAX models under minimum and maximum operating discharge regimes.

R. Rivas-Perez et al. / Environmental Modelling & Software 51 (2014) 207e220

Fig. 7. Experimental eld data of the Bocal obtained with a PRBS under minimum and maximum operating discharge regimes.

R. Rivas-Perez et al. / Environmental Modelling & Software 51 (2014) 207e220

Table 4
Model uncertainty set of the true plant.
Model uncertainty set of the true plant
Model
parameters

Nominal model
b nom s)
(G

b
Bounds of uncertainty region (D Gs)
Maximum discharge
regime model

Minimum discharge
regime model

K (cm/cm)
T1 (min)
T2 (min)
s (min)

Knom 0.0417
T1nom 15.84
T2nom 1.3
snom 6

Kmin 0.0303
T1min 13.33
T2min 1.11
smin 5

Kmax 0.0736
T1max 25.18
T2max 1.37
smax 7

215

operating discharge regime is usually achieved with a total opening

magnitude of the equivalent control gate of 180 cm. The data
collected under this operating discharge regime are shown in Fig. 7.
Similarly, the model that best represents the dynamic behavior of
the true plant under this operating discharge regime was estimated. As a result, the following second order model with the
ARMAX structure and time delay was obtained:

b
y ARMAX

max t

1:334 b
y ARMAX

max t

 0:3768 b
y ARMAX

 1

max t

 2

0:0007545umax t  4
0:0005458umax t  5 xARMAX
b
y ARMAX

min t

1:443 b
y ARMAX

min t

 0:4632 b
y ARMAX

 1

min t

 2

 1:0151xARMAX

max t

0:2312xARMAX

max t  2;

max t

 1
(17)

0:0008367umin t  6
 0:0006479umin t  7 xARMAX
 0:9903xARMAX

min t  1

0:1406xARMAX

min t

min t

 2:
(13)

The cross-validation results of the obtained model (13) on the

validation data set are shown in Fig. 8, and also in Table 3. This table
also presents the variance of the residual error, the FPE and the AIC.
The model (13) can thus be represented in the Laplace domain
by means of the following transfer function:

i
b
b
b
G
min s G u min s G v min s ;

The cross-validation results of the obtained model (17) on the

validation data set are shown in Fig. 8 and also in Table 3. This table
also shows the variance of the residual error, the FPE and the AIC.
The model (17) has been represented in the Laplace domain by
means of the following transfer function:

h
b max s G
bu
G

b
G
u

min s

Dymin s
0:0736
e7s ;

Dumin s
25:18s 11:37s 1

(15)

b
G
v

min s

Dymin s
5:18s 10:89s 1
:

Dvmin s
25:18s 11:37s 1

(16)

3.4.2. Linear model of the true plant under maximum operating

discharge regime
The maximum operating discharge regime through the upstream control gates is obtained when Q(t) Qmax 30 m3/s. This

bv
G

max s

(18)

where:

bu
G

max s

Dymax s
0:0303
e5s ;

Dumax s 13:33s 11:11s 1

(19)

bv
G

max s

Dymax s
2:85s 11:03s 1
:

Dvmax s
13:33s 11:11s 1

(20)

(14)

where:

max s

3.5. The model uncertainty set of the true plant (step 4 of the
procedure)
It is well known that the hydraulic parameters (the friction coefcient, the pool geometry, the downstream water elevation, the
main velocity, etc.) and/or the discharge regime Q(t) of main irrigation canal pools may change randomly within the operating
range, thus originating uncertainties in the nominal model
dynamical parameters. Therefore, the exact dynamic behavior of
these canal pools is unknown on multiple occasions. This has been
reported by various authors (see e.g. Deltour and Sanlippo, 1998;

Fig. 9. Step responses of the true plant model uncertainty set obtained.

216

R. Rivas-Perez et al. / Environmental Modelling & Software 51 (2014) 207e220

Fig. 10. Bode plots of the true plant model uncertainty set obtained.

water level variation (cm)

6
Mnom
= 4.99%
p
4

tnom
= 20.65 min
s

20

40

60

80

100
time (min)

120

140

160

180

200

Fig. 11. Nominal time response of the Bocal with PI controller (25).

Litrico and Fromion, 2009; Rivas-Perez et al., 2007; Schuurmans

et al., 1999). These parametric uncertainties are bounded (model
uncertainty region DG(s)) by the limit operation discharge regimes
models in the following ranges:

uncertainty set can therefore be represented by the expression

(Gevers, 2005):

D
Kmin  Kt  Kmax ; T1min  T1 t  T1max ; T2min  T2 t  T2max ;

smin  st  smax :
(21)
Consequently, the model uncertainty set of the true plant is
comprised of its nominal model and all the models (family of
models) that can be found in its uncertainty region, which is
bounded by the true plant limit operating models (minimum and
maximum discharge regime models). The true plant model

n
GD




Bocal sGD Bocal s

o
b nom s D Gs
b
G
;

(22)

b
where D Gs
is the model uncertainty region, bounded by the true
plant limit operating discharge regime models. Table 4 shows the
model uncertainty set of our true plant bounded by the parameters
of the limit models, which were derived using the results yielded in
Subsections 3.3 and 3.4.
The time-domain responses and the Bode plots of the resulting
true plant model uncertainty set when the discharge regime
changes in the operating range [Qmin, Qmax] are presented in Figs. 9
and 10. For time domain responses was used a total equivalent

b
Fig. 12. PI controller: settling time and overshoot of the time responses when operating discharge regime changes (D Gs).

R. Rivas-Perez et al. / Environmental Modelling & Software 51 (2014) 207e220

217

(gain crossover frequency uc and phase margin fm), the following

designing equation of the controller (23) has been found (FeliuBatlle et al., 2011):

Kp <x; Ki uc Jx; being x

cosfm  jsinfm
;
b u nom juc
G
(24)

Fig. 13. Block diagram of the control system with SP based HN robust controller.

control gate opening magnitude variation of 100 cm (1 m). These

gures show both the response of the true plant nominal model
and its uncertainty region bounded by the responses of the true
plant models under minimum and maximum operating discharge
regimes (limit models).
It should be noted that such plants are highly nonlinear and
therefore the obtaining of approximate linear models for the design
of robust controller is very complicated.
3.6. Application case of the obtained model: robust controller
design
In this Subsection we develop the design of a robust controller
for our irrigation main canal pool using the model uncertainty set
(22), whose ranges of parameters variations are shown in Table 4.
Let us assume the following required control objectives: a) nominal
b nom s): settling time t nom z20 min, overshoot M nom  5%,
plant ( G
s
p
steady state error enom
0; b) entire range of operating discharge
ss
b
regimes (D Gs):
settling time tsr  45 min, overshoot Mpr z30%,
steady state error erss 0.
In order to illustrate how hard these required control objectives
are, a PI controller embedded in a conventional feedback control
scheme is rstly designed in order to fulll the nominal operating
discharge regime requirements. Once this controller is tuned, the
requirements for the entire range of variation of the operating
discharge regimes will be checked. The frequency-domain based
design procedure of the PI controller will be used considering the
b nom s. The transfer function of the conplant nominal model G
ventional PI controller is represented as:

RPI s Kp

Ki
;
s

(23)

where Kp and Ki denote the proportional and integral gains

respectively. By means of the conventional frequency requirements

where <() and J() represent real and imaginary parts of a complex
number respectively. The gain crossover frequency uc is related to
the settling time tsnom of the response to a step command, while the
phase margin fm is related to the time response overshoot Mpnom .
Using the second order transfer function relation between the
settling time and the gain crossover frequency, and between the
overshoot and the phase margin as a rst approximation (Dorf and
Bishop, 2005), the following PI controller is easily obtained:

RPI s 33:38

1:682
;
s

(25)

which fullls the frequency design specications uc 0.0812 rad/s

and fm 62.1. Note that owing to the integral action of the PI
controller, the steady state error enom
of the response of the
ss
controlled system to a step command must be zero. The time
response of the nominal plant with the PI controller (25) is shown
in Fig. 11.
These results show that the PI controller (25) fulll the time
domain requirements for the nominal operating discharge regime
of our canal pool (control objective (a)). Nevertheless, if the settling
time and the overshoot of the time responses for all discharge regimes are obtained, we can check that these requirements are not
b
fullled at all for the entire range of variation D Gs
(control
objective (b)). Fig. 12 presents the variation of the time response
settling time and overshoot when the operating discharge regime
changes from the maximum discharge regime to its minimum one
b
(D Gs).
This gure also represents the region that denes the
fulllment of the control objective (b), and shows that, although the
PI controller (25) guarantees the nominal requirements, the
resulting settling time and overshoot are often higher than their
allowed values when the operating discharge regime varies.
Objective b) can be ensured by controller (25) only near the nominal operating discharge regime.
Then a Smith predictor (SP) based HN robust controller is
designed to fulll control objectives (a) and (b) simultaneously, as
an illustrative application of the identication for control procedure
developed in the previous Subsections. The SP based control

water level variation (cm)

nom

Mp

= 0.107%

2
tnom = 20.67 min
s

20

40

60

80

100
time (min)

120

140

Fig. 14. SP based HN robust controller: nominal time response.

160

180

200

218

R. Rivas-Perez et al. / Environmental Modelling & Software 51 (2014) 207e220

b
Fig. 15. SP based HN robust controller: settling time and overshoot of the time responses when the operating discharge regime changes (D Gs).

scheme is frequently used in the control of irrigation main canal

pools with time delay (see e.g. Castillo-Garcia et al., 2011; Deltour
and Sanlippo, 1998; Guin et al., 1989; Rivas-Perez, 1984).
Fig. 13 shows the block diagram of the proposed robust control
system, where Y*(s), U(s) and Y(s) are the signals: reference, control,
and system output respectively, RN(s) is the HN controller. The true
plant G(s) and the nominal plant model Gnom(s) have been represented as G(s)G0 (s)ess and Gnom s G0nom sesnom s being G0 (s)
and G0nom s the rational part of G(s) and Gnom(s) respectively.
Assuming as rst instance a perfect tuning of the SP control
system, i.e., Gnom(s) G(s), the HN controller can be designed for
G0nom s using the frequency responses of the obtained model unb
certainty set (D Gs),
which has been shown in Fig. 10. Using the
proposed method in (Farhati et al., 2012; Morari and Zariou, 1989),
we can select the weighting functions of the HN controller as:

W1

s
SN

us

s S0 us

(26)

being SN the maximum peak magnitude of S (sensitivity function

S 1/(1 G0 (s)), S0 the maximum steady-state error allowed and us
the minimum bandwidth frequency at 3 dB, W2 is usually chosen
as a constant. Choosing us 0.21 rad/s (see Fig. 10), a nearly zero

state error S0 0.001, SN 1 and W2 0.001, the resulting HN

robust controller is obtained as:

RN s

s3

7032s2 5854s 341:5

:
21:98s2 72:37s 0:015

(27)

Fig. 14 shows the nominal time response provided by the HN

robust controller (27). Note that this controller provides the same
settling time than the PI controller, but with a nearly zero overshoot
value. We shall mention that this nominal time response cannot be
provided by a PI controller embedded in a unity feedback control
scheme (time response that is shown in Fig. 11 is the best one that
can be provided by the PI controller for the nominal operating
discharge regime).
If the settling time and the overshoot of the time responses for
b
all operating discharge regimes (D Gs)
are obtained, we can check
that these requirements are now fullled for the entire range of
variation of operating discharge regimes (control objective (b)).
Fig. 15 shows the variation of the time response settling time and
the overshoot when the operating discharge regimes change from
b
the maximum discharge regime to its minimum one (D Gs).
This
gure also represents the regions where the control objective (b) is
fullled.

Fig. 16. PI controller vs. SP based HN robust controller: time responses for different discharge regimes.

R. Rivas-Perez et al. / Environmental Modelling & Software 51 (2014) 207e220

Finally, Fig. 16 compares the time responses that both controllers (25) and (27) provide for the minimum, nominal and maximum
operating discharge regimes. From this gure it is observed that the
quality of the time response of the Bocal control system has been
signicantly improved throughout the range of variation of the
b
operating discharge regimes (D Gs)
if the designed robust
controller (27) were used.
4. Comments and conclusions
This paper develops a mathematical model for the design of a
robust control system of the most important pool of the AIMC,
known as the Bocal, in Spain. This research is in fact a rst step
towards the implementation of a high-performance robust control
system in the whole AIMC which will be based on this kind of
models. These control systems have a special relevance in irrigation
main canal pools whose dynamic parameters change drastically
with the discharge regime variations in the operating range [Qmin,
Qmax], and disturbances from different sources are present.
The approach proposed in this paper is based on a complete
algorithmic procedure of identication for control, which enables
one to deliver the nominal model Gnom(s) of the true plant when the
irrigation main canal pool is operated under a nominal discharge
regime (Q(t) Qnom) and also delivers an explicit model uncertainty
region DG(s) of the true plant when the Bocal is operated under
other discharge regimes, different from the nominal one. This uncertainty region is bounded by the limit operating models obtained
when the irrigation main canal pool is operated under minimum
(Q(t) Qmin) or maximum (Q(t) Qmax) discharge regimes. The
proposed algorithmic procedure thus delivers the true plant model
uncertainty set (22), which is dened by the true plant nominal
model and its uncertainty region.
The model has been developed based on the software platform
of system identication toolbox of Matlab. The wide range of applications in which MATLAB is the working framework shows that
it is a powerful, comprehensive and easy-to-use real-time environment for implementing different technologies, among which
are those related to identication and control of environmental
systems.
The proposed algorithmic procedure of identication for control
has been applied to deliver the model uncertainty set (22) with
parameter variation ranges shown in Table 4. This model has been
used later to design an HN controller (27) robust in all the above set.
Comparative simulations show that the time responses of the Bocal
obtained with this controller e which takes into account the canal
pool parameter variations provided by our identied model e
signicantly outperform the responses provided by a standard PI
controller designed only from information on the nominal plant
dynamics.
The results are very encouraging since control-oriented models
facilitate the design of highly efcient robust controllers, which
allow the operability of the irrigation main canal pools to be
increased and the service to the farmers to be improved. The next
objective of this research will be to extend the proposed procedure
to obtain a control-oriented model of the whole AIMC considering
pool interactions and wave propagations.
Acknowledgments
The authors wish to acknowledge the help received from the
Ebro Hydrographical Confederation authorities in carrying out this
research, the fruitful discussions concerning the modeling and
operation of irrigation main canals, and their unconditional support. The authors would also like to thank the journal editor, the
associated editor and the anonymous reviewers for their detailed

219

reviews, invaluable comments and recommendations, which have

made the improvements to the original paper possible.
References
Andrews, F.T., Croke, B.F.W., Jakeman, A.J., 2011. An open software environment for
hydrological model assessment and development. Environ. Model. Softw. 26
(10), 1171e1185.
Bennett, N.D., Croke, B.F.W., Guariso, G., Guillaume, J.H.A., Hamilton, S.H.,
Jakeman, A.J., Marsili-Libelli, S., Newhama, L.T.H., Norton, J.P., Perrin, C.,
Pierce, S.A., Robson, B., Seppelt, R., Voinov, A.A., Fath, B.D., Andreassian, V., 2013.
Characterising performance of environmental models. Environ. Model. Softw.
40, 1e20.
Calderon-Valdez, S.N., Rivas-Perez, R., Ruiz-Torija, M.A., Feliu-Batlle, V., 2009.
Fractional PI controller design with optimized robustness to time delay changes
in main irrigation canals. In: 14th IEEE Conference on Emerging Technologies
and Factory Automation, September 22e25, Palma de Mayorca, Spain,
pp. 1411e1417.
Camacho, L.A., Lees, M.J., 1999. Multilinear discrete lag-cascade model for channel
routing. J. Hydrol. 226, 30e47.
Castillo-Garcia, F.J., Feliu-Batlle, V., Rivas-Perez, R., Sanchez-Rodriguez, L., 2011. Time
domain tuning of a fractional order PIa controller combined with a Smith predictor for automation of water distribution in irrigation main channel pools. In:
IFAC Proceedings Volumes (IFAC-PapersOnline) 18 (Part 1), pp. 15049e15054.
Chaudhry, M.H., 1993. Open-channel Flow. Prentice-Hall, Englewood Cliffs, N.J.
Chen, J., Gu, G., 2000. Control Oriented System Identication. Wiley Interscience,
N.Y.
Deltour, J.L., Sanlippo, F., 1998. Introduction of Smith predictor into dynamic
regulation. J. Irrig. Drainage Eng. ASCE 124 (1), 47e52.
Dorf, R.C., Bishop, R.H., 2005. Modern Control Systems. Pearson Education, Madrid.
Farhati, S., Bel Hadj Ali Naoui, S., Abdelkrim, M.N., 2012. Robust control of time
delay systems. Comput. Techno. Appl. 3, 69e75.
Feliu-Batlle, V., Rivas-Perez, R., Castillo-Garcia, F.J., Sanchez-Rodriguez, L., Linares
Saez, A., 2011. Robust fractional order controller for irrigation main canal
pools with time-varying dynamical parameters. Comput. Electron. Agric. 76,
205e217.
Feliu-Batlle, V., Rivas-Perez, R., Castillo-Garcia, F.J., 2005. Fractional robust control to
delay changes in main irrigation canals. In: IFAC Proceedings Volumes (IFACPapersOnline) 16 (Part 1), pp. 28e33.
Garnier, H., Wang, L., 2008. Identication of Continuous-time Models from Sampled
Data. Springer-Verlag, London.
Gevers, M., 2005. Identication for control: from the early achievements to the
revival of experiment design. Eur. J. Control 11, 1e18.
Guin, K.T., Nguen, Mi, Rivas-Perez, R., 1989. Adaptive control of a time-delay plant
using a searchless model-reference self-tuning system. Autom. Remote Control
49 (12, Part. 2), 1620e1627.
Jakeman, A.J., Letcher, R.A., Norton, J.P., 2006. Ten iterative steps in development
and evaluation of environmental models. Environ. Model. Softw. 21 (5), 602e
614.
Kovalenko, P.I., Matzeliuk, E.M., Rivas-Perez, R., Lebedev, V.I., 1993. Automatic
Control System of Water Level in a Canal. Patent No SU 1837263, International
Patent Classication G05D 9/12, Newsletter No 32.
Kovalenko, P.I., 1983. Automation of Land Reclamation Systems. Kolos, Moscow.
Lekkas, D.F., Imrie, C.E., Lees, M.J., 2001. Improved non-linear transfer function and
neural network methods of ow routing for real-time forecasting.
J. Hydroinformatics 3, 153e164.
Litrico, X., Fromion, V., 2009. Modeling and Control of Hydrosystems. Springer,
London.
Litrico, X., Fromion, V., Baume, J.-P., 2006. Tuning of robust distant downstream PI
controllers for an irrigation canal pool: (II) implementation issues. J. Irrig.
Drainage Eng. ASCE 132 (4), 369e379.
Ljung, L., 1999. System Identication. Theory for the User, Second ed. Prentice Hall,
Englewood Cliffs, N.J.
McIntyre, N., Young, P., Orellana, B., Marshall, M., Reynolds, B., Wheater, H., 2011.
Identication of nonlinearity in rainfall-ow response using data-based
mechanistic modeling. Water Resour. Res. 47 (3), W03515.
Morari, M., Zariou, E., 1989. Robust Process Control. Prentice Hall, Englewood
Cliffs, N.J.
Ochieng, G., Otieno, F., 2009. Data-based mechanistic modelling of stochastic
rainfall-ow processes by state dependent parameter estimation. Environ.
Model. Softw. 24, 279e284.
Ooi, S., Weyer, E., 2008. Control design of an irrigation channel from physical data.
Control Eng. Pract. 16, 1132e1150.
Pedregal, D.J., Rivas-Perez, R., Feliu, V., Sanchez, L., Linares, A., 2009. A non-linear
forecasting system for the Ebro River at Zaragoza, Spain. Environ. Model. Softw.
24, 502e509.
Price, L., Young, P.C., Berckmans, D., Janssens, K., Taylor, J., 1999. Data-based
mechanistic modelling and control of mass and energy transfer in agricultural
buildings. Annu. Rev. Control 23, 71e82.
Rivas-Perez, R., Feliu-Batlle, V., Castillo-Garcia, F.J., Sanchez-Rodriguez, L., LinaresSaez, A., 2011. Control oriented model of a complex irrigation main canal
pool. In: IFAC Proceedings Volumes (IFAC-PapersOnline) 18 (Part 1),
pp. 2919e2924.

220

R. Rivas-Perez et al. / Environmental Modelling & Software 51 (2014) 207e220

Rivas-Perez, R., Feliu-Batlle, V., Castillo-Garcia, F., Linarez-Saez, A., 2008a. System
identication for control of a main irrigation canal pool. In: IFAC Proceedings
Volumes (IFAC-PapersOnline) 17 (Part 1), pp. 9649e9654.
Rivas-Perez, R., Feliu-Batlle, V., Sanchez-Rodrguez, L., Pedregal-Tercero, D.J.,
Linares- Saez, A., Aguilar-Mariosa, J.V., Langarita-Garca, P., 2008b. Identication of the rst pool of the Imperial de Aragon main irrigation canal. Hydraul
Eng. Mexico 23 (1), 71e87.
Rivas-Perez, R., Feliu-Batlle, V., Sanchez-Rodriguez, L., 2007. Robust system identication of an irrigation main canal. Adv. Water Resour. 30, 1785e1796.
Rivas-Perez, R., Prada Moraga, C., Peran-Gonzalez, J.R., Kovalenko, P.I., 2002. Robust
adaptive predictive control of water distribution in irrigation canals. In: IFAC
Proceedings Volumes (IFAC-PapersOnline) 15 (Part 1), pp. 97e102.
Rivas-Perez, R., 1984. Technological Process Control in Main Canals of Irrigation
Systems, with Application to Irrigation Systems of Cuba. Ph.D. thesis. Scientic
Research Institute on Land Reclamation and Hydraulic Engineering of Ukrainian
Academy of Agricultural Sciences (UkrNIIGIM), Kiev, Ukraine.

Romanowicz, R.J., Young, P.C., Beven, K.J., 2006. Data assimilation and adaptive
forecasting of water levels in the River Severn catchment. Water Resour. Res. 42
(6), W06407.
Schuurmans, J., Hof, A., Dijkstra, S., Bosgra, O.H., Brouwer, R., 1999. Simple water
level controller for irrigation and drainage canals. J. Irrig. Drainage Eng. ASCE
125 (4), 189e195.
Van den Hof, P.M.J., Bombois, X., 2004. System Identication for Control. Lecture
Notes DISC Course. Delft Center for Systems and Control and Delft University of
Technology, Nederlands.
Young, P.C., 2011. Recursive Estimation and Time-series Analysis: an Introduction
for the Student and Practitioner. Springer-Verlag, Berlin.
Young, P.C., Garnier, H., 2006. Identication and estimation of continuous-time,
data-based mechanistic models for environmental systems. Environ. Model.
Softw. 21, 1055e1072.
Young, P.C., 1998. Data-based mechanistic modeling of environmental, ecological,
economic and engineering systems. Environ. Model. Softw. 13, 105e122.