Model predictive control of a Venlo-type greenhouse system considering electrical

energy, water and carbon dioxide consumption

Dong Lin∗, Lijun Zhang, Xiaohua Xia

Department of Electrical, Electronic and Computer Engineering, University of Pretoria, Pretoria 0002, South Africa

Abstract
Greenhouse systems consume lots of energy, water and carbon dioxide (CO2 ) to provide a suitable growth
environment for crops. Due to the problems of operation mode, some greenhouse systems are inefficient and
need to be optimized. In this paper, four optimization strategies for improving the operation efficiency of
greenhouse systems are studied. Strategy 1 minimizes the energy consumed for greenhouse heating, cooling,
ventilation and irrigation. Strategy 2 minimizes the water consumed for irrigation. Strategy 3 minimizes the
CO2 consumed for greenhouse CO2 enrichment. Strategy 4 minimizes the total cost of energy, water and CO2
consumed. These optimization strategies are based on a multi-input multi-output (MIMO) climate model
and a modified evapotranspiration model. Moreover, a sensitivity analysis is conducted to study the influence
of electricity price, water price, CO2 price and the range of system constraints on the optimization results.
Finally, a model predictive controller (MPC) is designed to reject system disturbances and address model
plant mismatch. The MPC controller is compared with a commonly used open loop controller. A performance
index relative average deviation (RAD) is introduced to evaluate the tracking performance of the proposed
MPC and the compared open loop control. Simulation results show that Strategy 4 reduce the total cost by
66.60 %, 92.68 % and 68.83% compared with Strategy 1, Strategy 2 and Strategy 3 respectively. Changes
in electricity price have a greater impact on optimization results than changes in water price and CO2 price.
Both temperature constraints and relative humidity constraints have a great influence on the optimization
results. The controller designed is verified to be effective.
Keywords: Greenhouse, optimization strategy, operation efficiency, sensitivity analysis, model predictive
control

1. Introduction and the South African economy [7]. Moreover, South

Africa is one of the 30 driest countries in the world
In recent years, the problem of energy and water [8]. Forty-nine percent of people in South Africa live
shortage has become more and more serious [1], [2], in water-scarce areas. South Africa’s average rainfall
[3]. Research in [4] shows that about 1.2 billion people is about 40 % less than the world’s annual average
in the least developed countries have no access to elec- rainfall. Research shows that the agricultural sector
tricity. Moreover, about 4 billion people in the world consumes the most water, followed by the municipal
are facing serious water shortage [5]. The electrical sector and the industrial sector [9] .
energy shortage is particularly serious in South Africa The global population will reach 10.1 billion by
[6]. In 2019, South Africa experienced its most seri- 2050. Due to the limited arable land, the traditional
ous energy crisis, and for the first time started load cultivation mode is facing many challenges in meet-
shedding stage 6 to protect the power system from ing the increasing food demand [10]. For example,
total blackouts. The energy shortage problem has a about 80% of the land in South Africa is used for a-
profound negative impact on both people’s daily life griculture, but only about 11% of the land is suitable
for cultivation. Moreover, due to the construction
∗
Corresponding author of urbanization and the overuse of land, arable land
Email address: [email protected] (Dong Lin) is gradually decreasing. The food shortage in some

Preprint submitted to Elsevier March 26, 2021

countries and regions is getting worse [11]. fogging, not the water consumed for irrigation. In
Greenhouse cultivation can effectively solve the [24], the optimization of the irrigation amount for
above problems. A greenhouse is a kind of agri- muskmelon in a plastic greenhouse is studied. The
cultural building that can provide a suitable growth results show that different irrigation amounts have
environment for crops [12]. Greenhouses are wide- significant effects on plant growth, fruit yield and
ly used all over the world. Research in [13] shows quality. In [25], the effects of four different levels
that there are approximately 3.64 million hectares of of drip irrigation on crop growth are studied. The
greenhouses worldwide. Greenhouse cultivation mod- results show that the optimal water requirement is
e can obtain higher yield than outdoor cultivation about 75% of crop evapotranspiration for the Troy
mode [14]. Therefore, it is an effective way to alle- 489 variety of tomato. Two optimal control strate-
viate food shortage. To maintain the environmen- gies of CO2 enrichment in greenhouse tomato crops
t required for the growth of crops, a lot of energy are studied in [26]. In [27], the leakage rate, CO2
and water are consumed [15]. Research in [16] shows supply amount, and CO2 concentration are estimat-
that the greenhouse energy consumption is the largest ed and used to study the CO2 enrichment efficiency
part of total agricultural energy consumption. In ad- in a greenhouse without ventilation.
dition, carbon dioxide (CO2 ) needs to be supplied Due to the complexity of the greenhouse environ-
to the greenhouse to increase crop yields. Howev- ment, greenhouse modeling is challenging. Some re-
er, some traditional greenhouse operation modes are search studied different modelling methods for green-
inefficient and cause a lot of waste. Therefore, it is house systems. For example, a nonlinear robust i-
necessary to optimize the operation of the greenhouse dentification method of greenhouse model based on a
to improve the efficiency of energy, water and CO2 u- multi-objective evolutionary algorithm is studied in
tilization. [28]. Modelling and control of greenhouse tempera-
Some optimization methods are proposed to re- ture and humidity are studied in [29]. In [30], the
duce greenhouse energy consumption. In [17], an op- applicability of extended Kalman filter in automatic,
timal control algorithm for greenhouse tomato pro- on-line and adaptive parameter estimation in a physi-
duction is proposed and compared with a conven- cal based greenhouse model is investigated. However,
tional proportional-integral (PI) control. The results most studies only consider the control of greenhouse
show that compared with the PI control, the energy temperature, relative humidity and CO2 concentra-
efficiency of the proposed optimal control is increased tion by heating and cooling, ventilation and CO2 sup-
by 8.5%. In [18], the control of the LED array fil- ply. Relatively few studies considered the influence of
l light for greenhouse cultivation based on parallel supplemental light on the greenhouse climate [31]. In
particle swarm optimization is studied. The results [32], an irrigation modelling method for greenhouse
show that compared with fluorescent lamps, the en- hot pepper grown based on soil water balance is s-
ergy saving is about 82.6%, and compared with in- tudied. A modified crop evapotranspiration model
candescent lamps, the energy saving is about 54.2%. is used to calculate irrigation water requirement in
A model optimization forecasting method to predic- the greenhouse and good predictive performance is
t the greenhouse energy demand for better accuracy obtained.
and cost time performance is proposed in [19]. A In our previous work, we studied how to reduce
hierarchical control strategy to optimize greenhouse the total cost of greenhouse energy consumption, ven-
operation is proposed in [20]. In [21], a model pre- tilation and CO2 supply [33]. However, the cost of
dictive control strategy is proposed to optimize the greenhouse water consumption was not considered.
efficiency of greenhouse heating system. Few studies have analyzed optimization strategies that
Some research studied different methods to reduce consider energy, water and CO2 consumption of green-
greenhouse water and CO2 consumption. In [22], a house system operation. In addition, solar radia-
greenhouse multi-objective optimization strategy is tion control has been neglected in many studies on
studied to maximize profit, fruit quality and water greenhouse control. Therefore, there is a research
use efficiency. In [23], a model-based predictive con- gap in the optimization of greenhouse system oper-
trol strategy is proposed to reduce energy consump- ation that considers energy consumption, water con-
tion and water consumption. However, this study sumption and CO2 consumption. These problems are
only considered the water consumed for greenhouse solved in this study. In this paper, four optimization

2
strategies for a greenhouse operation under South than the commonly used open loop controller.
Africa climate are studied. Strategy 1 minimizes en- The remainder of this paper is organized as fol-
ergy consumption, Strategy 2 minimizes water con- lows: In Section 2, the system model is presented.
sumption, Strategy 3 minimizes CO2 consumption, In Section 3, the optimization problem is formulat-
and Strategy 4 minimizes the total cost of energy ed. In Section 4, the controller design is conducted.
consumption, water consumption and CO2 consump- Simulation results are shown in Section 5. Section 6
tion. Strategy 1, Strategy 2 and Strategy 3 are three concludes this paper.
commonly used greenhouse optimization strategies in
previous studies and are compared with Strategy 4 in 2. System model
this study. The greenhouse climate model proposed
in [34], [35] and a modified evapotranspiration model Greenhouse systems generally include a heating
presented in [32] are used. Moreover, a sensitivity and cooling system, a ventilation system, a CO2 sup-
analysis is conducted to study the impact of elec- ply system and a lighting system, etc. People adjust
tricity price, water price, CO2 price and the range the greenhouse temperature, relative humidity and
of constraints on the optimization results. Finally, CO2 concentration by controlling greenhouse system-
to address system disturbances and model mismatch, s to provide a suitable environment for crop growth.
an MPC controller is designed and compared with an Figure 1 is the schematic diagram of a greenhouse
open loop controller. system.
The main contributions of this paper include: 1)
For greenhouse operation optimization, most research
focus on how to reduce energy consumption, few stud- Farmer
ies consider water consumption and CO2 consump- System settings
tion. In this paper, the proposed optimization strate-
Electricity price
gy takes into account energy consumption, water con-
sumption and CO2 consumption. 2) For the green-
Control center
house control, most studies only consider the control Weather data

of greenhouse heating, cooling, ventilation and CO2 Greenhouse climate

Control signal
supply, but not the control of solar radiation which data

has a great impact on both greenhouse climate and Power grid

irrigation water demand. In this paper, the control of

Electricity supply
greenhouse heating, cooling, ventilation, CO2 supply Weather station

and solar radiation are considered. 3) Most studies Water

W at supply

adopt the method of changing crop planting mode or CO2 supply

Well
Greenhouse
irrigation mode to save water. These studies focus
Pump
CO2 generator
on how to improve water use efficiency rather than
how to reduce water demand. In this paper, a mod-
el for balancing crop irrigation water needs based on Figure 1: Schematic diagram of a greenhouse control system
soil moisture is established. The influence of green-
house climate and solar radiation power on irrigation In this paper, we studied greenhouse climate con-
water demand is analyzed. The proposed strategy trol and irrigation control. The corresponding green-
improves water use efficiency and reduces water de- house climate model and water demand model used
mand. 4) The influence of the changes of electricity are based on energy and mass balance. Figure 2
price, water price, CO2 price, temperature constraint shows the energy, water and CO2 flow of a green-
and relative humidity constraint on the optimization house control system.
results is studied through sensitivity analysis. The
sensitivity analysis performed can provide a deeper 2.1. Greenhouse climate model
insight into the greenhouse optimization problem. 5) In this paper, the greenhouse model presented in
To improve the control accuracy of the greenhouse [34] and [35] is used and given below.
system, an MPC controller is designed and a better
reference trajectory tracking performance is obtained

3
 is the absolute water vapour concentration at crop
level. Hair is the absolute water vapour concentration
of the greenhouse air.
ge is obtained using:
Energy loss
Solar radiation
2LAI
ge = , (5)
Lighting (1 + )rb + rs

Transpiration Photosynthesis
where LAI is the leaf area index,  is the ratio of
Ventilation fan
latent to sensible heat content of saturated air. rb
is the boundary layer resistance, rs is the stomatal
Heating Cooling
CO2 supply
Water pump
resistance.
Energy flow Water flow CO2 flow
Hcrop can be calculated by:

rb Rn
Figure 2: Greenhouse energy, water and CO2 flow Hcrop = Hair,sat +  , (6)
2LAI L
where Hair,sat is the saturated vapour concentration.
2.1.1. Temperature
According to [37], Hair,sat can be approximated by:
Greenhouse temperature is an important factor
affecting crop growth. It is determined by greenhouse Hair,sat = 5.5638e0.0572Tair . (7)
heating, cooling, ventilation, lighting, solar radiation,
etc. The temperature is governed by:  and rs can be obtained by:
dTair 1  = 0.7584e0.0518Tair , (8)
= (Qsun + Qlamp − Qcov
dt Ccap (1)
Rn
−Qtrans − Qvent + Qc ), rs = (82 + 570e−γ LAI )(1 + 0.023(Tair − 20)2 ), (9)
where Tair is the greenhouse temperature, Ccap is the where γ is a crop parameter, Rn is the net radiation
greenhouse heat capacity, Qsun is the incoming radi- at crop level.
ation from the sun, Qlamp is the lamp heating power.
Qcov is the heat transfer through the cover, Qtrans is Rn = 0.86(1 − e−0.7LAI )(Qsun + PE ), (10)
the energy absorption of crop transpiration. Qvent is
the energy change caused by ventilation. Qc is the where PE is the power of lighting.
heating or cooling power.
Qsun can be calculated by: Qlamp = ηPE , (11)

Qsun = α1 (1 − sr )Irad , (2) where η is the conversion coefficient of lamp power

into heating power.
where α1 is the transmission coefficient of the cover
material, sr is the shading rate and is adjusted by the Qvent = gv ρair Cp,air (Tair − Tout ), (12)
greenhouse shading system, Irad is the solar radiation
power. where gv denotes the specific ventilation rate, ρair is
According to [36], Qcov can be described by: the density of the air, Cp,air is the heat capacity of
the air.
Qcov = α2 (Tair − Tout ), (3) Please note that this paper gives a brief introduc-
tion to the model used. For further details, such as
where α2 is the heat transfer coefficient of the cover,
the model parameters and the physical meaning of
Tout is the outside temperature.
the variables, please refer to [38, 39].
Qtrans can be obtained by:

Qtrans = ge L(Hcrop − Hair ), (4)

where ge is the transpiration conductance, L is the

energy needed to evaporate water from a leaf. Hcrop
4
2.1.2. Relative humidity where Cair is the CO2 concentration inside the green-
The relative humidity of the greenhouse has a house, Cinj is the CO2 injection rate, Cass is the CO2
great influence on the growth of crops. The rela- assimilation, Cvent is the changes in CO2 concentra-
tive humidity is affected by crop transpiration, water tion due to ventilation.
vapor condensation and ventilation. The relative hu- Cass and Cvent can be obtained by:
midity RHair can be obtained using: 1
Cass = 2.2 × 10−3 (1 − e−0.003(Qsun +PE ) ),
RHair = Hair /Hair,sat , (13) 1 + C0.42
air
(20)
where Hair is the vapour concentration of the green-
house air. Hair can be calculated by: Cvent = gv (Cair − Cout ). (21)

dHair 1 2.2. Irrigation water demand model

= (Htrans − Hcov − Hvent ), (14)
dt h The irrigation water demand model used in this
where Htrans is the vapour produced by plant transpi- paper is based on the soil water balance. The schemat-
ration, Hcov is the vapour condensation to the cover, ic diagram of soil water balance is shown in Figure 3.
Hvent is the vapour flux caused by ventilation. h is Crop water demand is affected by precipitation, irri-
the average height of the greenhouse. gation, evapotranspiration, surface runoff, etc.
Htrans is influenced by Hcrop and Hair , and it can
be described by:

Htrans = ge (Hcrop − Hair ). (15)

precipitation evapotranspiration
Hcov can be obtained by:

Hcov = gc 0.2522e0.0485Tair (Tair − Tout )


 (16)
−(Hair,sat − Hair ) ,
Irrigation Surface runoff
where gc is the condensation. gc can be obtained by:
 Deep percolation
0 if Tair ≤ Tout Storage change
gc = 1/3 (17)
pgc (Tair − Tcov ) if Tair > Tout , Water table

where pgc is related to the properties of the conden-

sation surface. Figure 3: Schematic diagram of soil water balance
Hvent is influenced by the ventilation and the hu-
midity both inside and outside the greenhouse. The The water balance can be given as:
value of Hvent can be obtained by:
P + I + W = ET + R + D + ∆S, (22)
Hvent = gv (Hair − Hout ), (18)
where P is the precipitation, I is the irrigation, W is
where gv is the ventilation rate. In this paper, gv is the water from the water table, ET is the crop evap-
controlled by changing the power of the ventilation otranspiration, R is the surface runoff, D is the deep
fan. percolation, ∆S is the soil water content change. Due
to the greenhouse is a closed environment, there is no
2.1.3. CO2 concentration precipitation, P = 0. W can be ignored because the
CO2 concentration is also an important climatic water table is more than 25 meters deep. R is negligi-
factor affecting greenhouse crop growth. CO2 con- ble because the greenhouse is flat and there is no loss
centration can affect crop photosynthesis [40]. People of irrigation water. D can be ignored according to
can use CO2 enrichment to improve crop yield. The the research in [41]. Therefore, the soil water balance
CO2 concentration model is as follows. can be changed to:
dCair 1
= (Cinj − Cass − Cvent ), (19) I = ET + ∆S. (23)
dt h
5
 The dynamic model of soil water content can be calculate the evapotranspiration in the greenhouse,
expressed as: there will be a large error [47].
According to [32], a modified Penman-Monteith
dS equation for greenhouse evapotranspiration calcula-
= I − ET. (24)
dt tion is introduced and given below.
In this paper, the real-time irrigation mode is 1713
used. However, it is difficult to measure soil water 0.408∆(Rn − G) + γ Tair +273 (es − ea )
ETo = . (31)
content quickly and accurately without causing dam- ∆ + 1.64γ
age to crops. Therefore, the control of irrigation in
Compared with Rn , the soil heat density G is rel-
many studies is to keep the soil water content con-
atively small. According to [48], G can be approxi-
stant. The greenhouse irrigation water demand is
mately zero when the ground is covered with vegeta-
equal to the greenhouse water consumption. The soil
tion.
water content measurement is not needed. Related
research can be found in [42], [43]. The water de- 2.3. Model analysis
mand for irrigation can be obtained by:
The proposed greenhouse climate model had been
I = ET. (25) validated in [34], [35]. The proposed greenhouse crop
reference evapotranspiration model had been validat-
ET can be calculated by: ed in [32]. It should be pointed out that in most cas-
es, the developed model can accurately predict the
ET = ETo × Kc , (26) actual value, but when the temperature outside the
greenhouse is low (below 0 ◦ C), the prediction error
where ETo is the reference evapotranspiration, Kc is is large. Similar results can be found in [49]. Howev-
the crop coefficient and depends on the crop type and er, the average temperature in South Africa is high.
growth stage. For example, in Pretoria, the administrative capital
According to [44], the Penman-Monteith equation of South Africa, the average temperature of the win-
can be used to calculate ETo . ter is above 10 ◦ C. Therefore, the proposed model
is suitable for greenhouse climate prediction in this
0.408∆(Rn − G) + γ Tair900
+273 u2 (es − ea ) study.
ETo = ,
∆ + γ(1 + 0.34u2 )
(27)
3
(MJ/hour/m2 )

0.8
where ∆ is the slope of the vapor pressure curve, G is 2 0.6
the soil heat density, γ is the psychometric constant,
1
es is the saturation vapour pressure, ea is the average 0.4
sun

0
daily actual vapour pressure, u2 is the wind speed at
Q

1 0.2
0.8 30
2 meter height. 0.6
0.4 20
25
0.2 15
RHair Tair (°C)
17.27 × Tair
es = 0.6108 × exp( ), (28)
Tair + 237.3 Figure 4: Crop reference evapotranspiration (mm/hour)

ea = es × RHair , (29)
According to (31), ETo is related to the temper-
4098 × es ature, relative humidity and radiation power. As
∆= . (30)
(Tair + 237.3)2 shown in Figure 4, ETo increases with the increase of
Please note that the Penman-Monteith equation temperature and radiation power but decreases with
in (27) is for the calculation of outdoor evapotranspi- the increase of relative humidity. Therefore, the fol-
ration. Related research can be found in [45], [46]. lowing methods can be used to reduce greenhouse wa-
However, it is not suitable for the greenhouse evapo- ter demand: reducing the temperature of the green-
transpiration calculation. The reason is that the wind house, reducing the radiation power, and increasing
speed in the greenhouse is very low. If (27) is used to the relative humidity in the greenhouse.

6
3. Optimization to the ventilation power Qv and is determined by the
type of ventilation fan.
This paper aims to study different optimization Z tf
strategies for greenhouse operation to reduce ener-
E3 = Qw dt, (36)
gy, water and CO2 consumption while keeping green- ti
house climate within the required range to provide
1
a suitable environment for crop growth. The corre- Qw = ρw Vw ghw , (37)
η
sponding optimization problems are formulated in the
following sections. where Qw is the pumping power, η is the energy effi-
ciency of the water pump system, ρw is the water den-
3.1. Decision variables sity, Vw is the volume of pumped water, g is the accel-
In this paper, the system studied is a multiple- eration of gravity, hw is the height of water pumping.
input multiple-output (MIMO) system. There are 3.2.2. Strategy 2
four inputs and three outputs. The inputs (decision
The objective of Strategy 2 is to minimize green-
variables) include the controlled heating or cooling
house water consumption. In this paper, the wa-
power Qc , the ventilation rate gv , the CO2 injection
ter consumed for greenhouse irrigation is considered.
rate Cinj and the controlled shading rate sr . The
The objective function is as follows:
outputs are greenhouse temperature Tair , relative hu-
Z tf
midity RHair , CO2 concentration Cair . People use
J2 = I(t)dt. (38)
the greenhouse heating and cooling system, ventila- ti
tion system and CO2 supply system to control the
temperature, relative humidity and CO2 concentra- 3.2.3. Strategy 3
tion. The CO2 used for CO2 enrichment is very expen-
sive and should be used effectively. The objective
3.2. Objective functions of Strategy 3 is to minimize greenhouse CO2 con-
3.2.1. Strategy 1 sumption. Therefore, the objective function can be
Greenhouse operation consumes lots of energy, e- obtained by:
specially in winter when the temperature is low. The Z tf
objective of Strategy 1 is to minimize greenhouse en- J3 = Cinj (t)dt. (39)
ti
ergy consumption. Most of the previous studies on
energy optimization only considered the energy con- 3.2.4. Strategy 4
sumption of greenhouse heating and cooling. In this The objective of Strategy 4 is to reduce the cost
paper, the energy consumed for heating, cooling, ven- of greenhouse energy consumption, water consump-
tilation and irrigation is considered. The objective tion and CO2 consumption. Therefore, the objective
function of Strategy 1 is given by: function can be given by:

J1 = E1 + E2 + E3 , (32) J4 = ω1 J1 + ω2 J2 + ω3 J3 , (40)

where E1 is the energy consumed for heating and where ω1 , ω2 and ω3 are prices of energy, water and
cooling, E2 is the energy consumed for ventilation, E3 CO2 respectively. The time-of-use (TOU) electricity
is the energy consumed for irrigation water pumping. tariff in South Africa is used for energy cost calcula-
tion and given by:
Z tf

E1 = |Qc (t)| dt, (33)  ωo t ∈ [0, 6] ∪ [22, 24]
ti ω1 (t) = ωs t ∈ [9, 17] ∪ [19, 22], (41)
ωp t ∈ [6, 9] ∪ [17, 19]

Z tf
E2 = Qv dt, (34)
ti where ωo , ωs , ωp are the off-peak, standard, peak tar-
Qv = λgv , (35) iff in R/kWh. R is the South Africa Currency, Rand.
In this study, the groundwater is used for greenhouse
where ti is the initial time, tf is the final time for irrigation. Due to the groundwater is free, ω2 = 0.
optimization. λ is the conversion coefficient from gv ω3 = R1000/ton.
7
 min max
3.3. System constraints Cair ≤ Cair ≤ Cair , (48)
3.3.1. Input constraints where min
Tair
and max
Tair
are the lower and upper limit
In this study, inputs include the heating/cooling of greenhouse temperature. RHair min and RH max are
air
power Qc , ventilation rate gv , CO2 injection rate Cinj the lower and upper limit of greenhouse relative hu-
and shading rate sr . The corresponding constraints min and C max are the lower and upper limit
midity. Cair air
are as follows: of greenhouse CO2 concentration.

Qmin
c ≤ Qc ≤ Qmax
c , (42) min
Irad ≤ Qsun , (49)

gvmin ≤ gv ≤ gvmax , (43) The radiation power after shading control (Qsun )
min to provide
should be greater than the limit value Irad
min max
Cinj ≤ Cinj ≤ Cinj , (44)
sufficient light for crop growth.
where Qmin
c and Qmax
c are the lower and upper limit
of the heating or cooling power. gvmin and gvmax are 4. Controller design
the lower and upper limit of ventilation rate. Cinjmin
max
and Cinj are the lower and upper limit of CO2 in- Figure 5 shows the hierarchical structure of green-
jection rate. The range of input constraints is deter- house control. Hierarchical control can decompose
mined by the characteristics of the greenhouse system complex problems into different subproblems, thus
studied. effectively reducing the computational complexity of
 min
complex problems [54], [55]. It can be seen that the
0 ≤ sr ≤ 1, if Irad ≥ Irad greenhouse control includes two layers. On the op-
min (45)
sr = 0, if Irad < Irad timization layer, reference points are generated by
min is the lower limit of solar radiation pow- greenhouse optimization. On the control layer, a cli-
where Irad
mate controller is designed to track the reference tra-
er for shading control. To provide sufficient light for
jectories obtained from the optimization layer.
crop growth, the shading control should be imple-
mented only when the radiation power is greater than Control layer Optimization layer
min . Moreover, the controlled shading rate s varies
Irad Meteorological data
r Disturbances

between 0 and 1. Qc Tair

Setpoints gv
Climate
RH air
Cinj Greenhouse Optimization
3.3.2. State constraints controller

sr Cair
Greenhouse climate factors such as temperature,
relative humidity, CO2 concentration and light in-
Electricity price CO2 price
State feedback
tensity affect crop growth and yield. Therefore, the
greenhouse climate factors should be in a suitable
range to provide the necessary environment for crop Figure 5: Greenhouse hierarchical control structure
growth. The too high or low temperature will cause
crops to wither or even die [50]. Too high relative
4.1. Open loop controller design
humidity will cause the outbreak of some diseases in
The discrete state-space model is as follows:
crops [51]. Too low or too high CO2 concentration
will affect crop photosynthesis and thus affect crop x(k + 1) = f (x(k), u(k)), (50)
yield [52]. Too low lighting power will reduce crop
yield [53]. Please note that different types of crop- where x(k), u(k) are the state vector, input vector
s have different state constraints at different growth at time kTo . x(k) = [Tair (k), RHair (k), Cair (k)]T ,
stages. The range of state constraints is set by farm- u(k) = [Qc (k), gv (k), Cinj (k), sr (k)]T . k = 0, 1, 2, · · · ,
ers based on their experiences (weigh the expected No − 1. No = Tt /To . Tt is the total simulation time.
yield and costs) or obtained from the optimization of To is the sampling period.
crop growth. The state constraints are as follows: The open loop controller solves the optimization
problem:
min max
Tair ≤ Tair ≤ Tair , (46) u∗ = arg min J, (51)
u
min max
RHair ≤ RHair ≤ RHair , (47) subject to the constraints (42)−(50). J ∈ [J1 , J2 , J3 , J4 ].
8
4.2. MPC controller design Algorithm 1: The proposed MPC algorithm
The MPC sampling interval Tm is smaller than Initialization: Given initial state value x(0),
the open loop control sampling interval To . Tm = u(0) and set k = 0;
To /Nm , where Nm is a positive integer. For the time while k ≤ Nm − Np do
tm ∈ [k1 To +k2 Tm , k1 To +(k2 +1)Tm ], k1 = 0, 1, 2, · · · , Compute the solution U of the optimal
No − 1, k2 = 0, 1, 2, · · · , Nm − 1, the MPC take the problem formulated in (55);
value u(k1 + 1) that obtained from the open loop op- Apply the first value of the solution U and
timization (51) as the inputs reference uref (k1 + 1) discard the rest of the solution;
to track the corresponding state variables reference Calculate the state of next interval x(k + 1)
trajectories xref (k1 + 1). by x(k + 1) = f (x(k), u(k));
In the sequel, a commensurate quantification as- k = k + 1;
sumption is made: all variables are quantized in the end
two sampling schemes, they are represented by start- Np = Np − 1;
ing values and remained in the same sampling in- while k < Nm do
terval. This assumption ensures that the MPC can Compute the solution U of the optimal
reach the steady state obtained from the open loop problem formulated in (55);
optimization. Apply the first value of the solution U and
The MPC objective function can be given by: discard the rest of the solution;
Calculate the state of next interval x(k + 1)
Np
X by x(k + 1) = f (x(k), u(k));
Jm = (∆x(k + i|k))T Q(∆x(k + i|k)) k = k + 1; Np = Np − 1;
i=1
(52) end
c −1
NX
+ (∆u(k + i|k))T R(∆u(k + i|k)),
i=0
5. Simulation
where Np and Nc are optimization horizon and con-
trol horizon respectively. |k means that the predicted 5.1. Simulation data
value is based on the information up to time k. ∆x In this research, we studied four optimization s-
is the tracking error. ∆u is the control effort. Q and trategies for the operation of a Venlo-type greenhouse
R are the weighting matrices that penalize the future under South Africa climate. The meteorological data
tracking and control efforts respectively. The con- of a winter day (July 1, 2016, 0:00 to 23:59) is used.
trol effort in the objective function is to avoid abrupt The data come from a weather station at the Uni-
0 0
changes in the control action [56]. ∆x(k + i|k) and versity of Pretoria (25◦ 75 S, 28◦ 23 E) and is shown
∆u(k + i|k) are given by: in Figure 6. The greenhouse model parameters are
from [34] and [35] and are shown in Table 1. The
∆x(k + i|k) = x(k + i|k) − xref (k + i). (53) system constraints are shown in Table 2.
 Lamps are installed to provide artificial lighting.
 u(k + i|k) − u(k − 1), i = 0 Air-water heat exchangers are installed for heating
∆u(k + i|k) = u(k + i|k) − u(k + i − 1|k), (54) and cooling. For CO2 enrichment, the OCAP (or-
i = 1, 2, · · · , Nc − 1

ganic CO2 for assimilation by plants) network sup-
Denote U = [u(k|k), u(k+1|k), u(k+2|k), · · · , u(k+ plies the organic CO2 crops needed. For greenhouse
Nc − 1)|k]T . The MPC controller solves the nonlinear supplemental lighting, the artificial lighting power is
optimization problem: set to zero for day time(07:00 to 18:00) and 110W/m2
for night time (19:00 to 06:00).
U ∗ (k) = arg min Jm (k), (55)
U 5.2. Optimization results
subject to the constraints (42)−(50). The optimization problem is solved by the ’fmin-
The MPC algorithm is as follows. con’ code of the MATLAB Optimization Toolbox.
The interior point algorithm is selected as the opti-
mization algorithm. The optimization results of pro-
9
 Table 1: Greenhouse model parameters
Parameter Value Unit Parameter Value Unit
α1 0.7 − pgc 1.8 × 10−3 m◦ C−1/3 s−1
α2 10 Wm−2◦ C−1 ωo 0.5157 R/kWh
γ 0.008 − ωs 0.9446 R/kWh
LAI 2.6 − ωp 3.1047 R/kWh
Ccap 30000 J/m2 ◦ C λ 0.06 W/m3
h 7 m η 0.75 −
s 40709 m2 g 9.8 m/s2
L 2450 J/g hw 7 m
rb 150 s/m ω3 1000 R/ton
ρair 1.225 kg/m3 Kc 0.7 −
Cp,air 1003 J/kg◦ C
RHout (%) Qsun (W/m ) Tout ( C)

30
greenhouse heating system should work to maintain
°

20
10
0
the greenhouse temperature.
0 5 10 15 20
In addition, we can find that greenhouse venti-
2

1000
500 lation is mainly at noontime. That is because the
0
0 5 10 15 20
outdoor temperature is high and ventilation will not
100
lose a lot of energy during this period.
50
0
0 5 10 15 20
Time (hour)

Figure 6: Meteorological data for July 1, 2016

Table 2: Greenhouse system constraints

Variable Value Unit
min ◦
Tair 14 C
max ◦
Tair 26 C
min
RHair 0 %
max
RHair 90 %
min
Cair 400 ppm
max
Cair 2000 ppm
Qmin
c -200 W/m2 Figure 7: Optimization results of Strategy 1
Qmax
c 200 W/m2
gvmin 0 m/s
gvmax 0.02 m/s
min
Cinj 0 g/m2 s 5.2.2. Strategy 2
max
Cinj 0.02 g/m2 s The optimization result of Strategy 2 is shown in
Figure 8. The energy consumption of Strategy 1 is
mainly used for greenhouse heating, while the energy
posed four strategies are shown in the following sec- consumption of Strategy 2 is used for both green-
tions. house heating and greenhouse cooling. Moreover, we
can find that the greenhouse temperature is low but
5.2.1. Strategy 1 the relative humidity is high. Moreover, the shading
The optimization result of Strategy 1 is shown in rate of Strategy 2 is greater than that of Strategy 1,
Figure 7. We can find that the energy consumed is and the radiation power of Strategy 2 is smaller than
mainly used for heating in the morning. The reason that of Strategy 1. That is because low temperature,
is that the temperature in the greenhouse has gradu- high relative humidity and low radiation power help
ally dropped to the set lower temperature limit after reduce crop evapotranspiration and thus reduce wa-
a cold night. There is very little energy from solar ra- ter consumption. It should be pointed out that the
diation available in the early morning. Therefore, the energy consumption of Strategy 2 is much higher than

10
that of Strategy 1.

Figure 10: Optimization results of Strategy 4

Figure 8: Optimization results of Strategy 2

12000.00
10300.00
10000.00
5.2.3. Strategy 3 7970.00
8000.00
The optimization results of Strategy 3 is shown in
Figure 9. It can be seen that the CO2 injection rate 6000.00

is small. The CO2 concentration is low. Strategy 3 4000.00 3320.00

2180.00 2570.00
consumes less CO2 than other strategies. Moreover, 2000.00
787.30 792.60
1120.00 743.68
748.71
130.54 67.32
we can find that the ventilation rate is high. The 0.00
Off peak Standard Peak Off peak Standard Peak
reason is that ventilation can send the outdoor CO2
Energy consumption (kWh) Energy cost (Rand)
into the greenhouse to keep the greenhouse CO2 con- Strategy 1 Strategy 4
centration within the required range. It should be
noted that ventilation will cause energy loss in the
greenhouse and increase the energy consumption of Figure 11: Comparison of Strategy 1 and Strategy 4 under
the greenhouse. TOU tariff

0.6
Strategy 1
0.5 Strategy 2
Strategy 3
Strategy 4
0.4
sr

0.3

0.2

0.1

0
0 5 10 15 20
Time (hour)

Figure 12: Greenhouse shading rate sr

Figure 9: Optimization results of Strategy 3

Table 3 shows the energy consumption, water con-
sumption, CO2 consumption and total cost of the
5.2.4. Strategy 4 proposed four strategies. We can find that Strate-
The optimization result of Strategy 4 is shown in gy 1 has the least energy consumption, Strategy 2
Figure 10. We can find that the energy consump- has the least water consumption, Strategy 3 has the
tion and CO2 consumption of Strategy 4 are very least CO2 consumption, and Strategy 4 has the lowest
small. That is because energy cost and CO2 cost are total cost. Figure 11 shows the comparison between
the main part of the greenhouse total cost. Reduc- Strategy 1 and Strategy 4 under the TOU tariff. It
ing energy consumption and CO2 consumption will can be seen that the energy consumption of Strategy
effectively reduce the total cost.
11
 Table 3: Comparison of four strategies
Strategy 1 Strategy 2 Strategy 3 Strategy 4
Energy consumption (kWh) 4243.20 106656.73 31299.40 5539.21
Water consumption (ton) 86.61 68.16 94.50 86.92
CO2 consumption (ton) 21.18 21.08 0.65 0.95
Cost (Rand) 32307 147410 34624 10791

4 is higher than that of Strategy 1, but the energy

cost of Strategy 4 is lower than that of Strategy 1. 140000 125610
The reason is that Strategy 4 consumes less energy 120000
than Strategy 1 in the peak time with a high price. 100000
The control variable shading rate sr is shown in

Cost (Rand)
80000
Figure 12. It can be seen that the shading rate of S-
60000
trategy 2 is bigger than that of other strategies. The 31150
40000 21077
reason is that the increase of shading rate can reduce 10197
8917
2826
21176
20000 934 1.09 2.77
the solar radiation, and thus reduce the evapotran- 757 920 0.87 1.08
645 953

spiration and water consumption. Although reduc- 0

Heating/cooling Ventilation Irrigation CO₂ injection
ing solar radiation power can reduce water consump- Strategy 1 Strategy 2 Strategy 3 Strategy 4
tion, the energy consumed to maintain the green-
house temperature within the required range increas-
es accordingly. Therefore, the control of solar radia- Figure 14: Cost composition of four strategies
tion power should be in a reasonable range.
Figure 13 shows the energy consumption compo-
sition of four strategies. It can be seen that most en- 5.3. Sensitivity analysis
ergy consumed for heating and cooling, followed by Sensitivity analysis can provide insights into the
ventilation and the least for irrigation. Among the influence of model parameter uncertainties on the
four strategies, Strategy 1 has the least total energy performance of the optimal controller [36]. In this
consumption and Strategy 2 has the most energy con- paper, the initial electricity price is R 0.5157/kWh
sumption. The cost composition of four strategies is for the off peak period, R 0.9446/kWh for the stan-
shown in Figure 14. We can find that Strategy 2 has dard period, and R 3.1047/kWh for the peak period.
the highest total cost and Strategy 4 has the lowest The initial water price is zero. The initial CO2 price
cost. The reason is that Strategy 2 consumes much is R 1000/ton. The initial upper limit of temperature
more energy for heating and cooling than the other is 26 ◦ C, the lower limit of temperature is 14 ◦ C, and
three strategies. the upper limit of relative humidity is 90%. However,
the electricity price and CO2 price may change over
time. The water price should not be set to zero if
120000
the impact of water consumption on the sustainable
105960
development of society is considered. The state con-
Energy consumption (kWh)

100000
straints of different types of crops at different growth
80000
stages should be different values. Therefore, it is nec-
60000
essary to analyze how the changes of these parameters
40000 28748 and constraints affect the optimization results.
20000
3334 907 2.20
695 2549 4634 903 In this paper, we take Strategy 4 as an example
1.73 2.40 2.21
0 to analyze the influence of the changes of electricity
Strategy 1 Strategy 2 Strategy 3 Strategy 4
price, water price, CO2 price, temperature constraint
Heating/cooling Ventilation Irrigation
and relative humidity constraint on the optimization
results. The change is in increments of 5%. The max-
imum change of electricity price, CO2 price and con-
Figure 13: Energy consumption composition of four strategies
straints range is 15% of the initial value. The maxi-
mum change of water price is 15% of South Africa’s
12
water tariff (R 14.27 per kiloliter).
25000

5.3.1. Influence of prices change 20000

The sensitivity analysis of electricity price, water

Cost (Rand)
15000
price and CO2 price is shown in Figure 15. It can be
seen that the cost increases with the increase of elec- 10000
tricity price, water price and CO2 price. Moreover,
5000
compared with the change of water price and CO2
price, the change of electricity price has a greater im- 0
-15% -10% -5% 0 5% 10% 15%
pact on the cost. When the price of electricity in-
Tmax Tmin RHmax
creased by 15%, the cost increased by 13.68%. How-
ever, when the price of water and CO2 increased by
15%, the cost only increased by 1.9% and 1.33% re-
Figure 16: Greenhouse operation cost under the constraints of
spectively. different percentage changes

13000
In addition, the cost decreases with the increase of
the upper limit of relative humidity (RHmax ). There-
12000
fore, the greenhouse operating cost can be effectively
Cost (Rand)

reduced by setting appropriate temperature and rel-

11000
ative humidity constraints. It can be concluded that
10000
cost is sensitive to changes in greenhouse temperature
and relative humidity. Therefore, it is important to
9000 carefully set the temperature and relative humidity
-15% -10% -5% 0 5% 10% 15%
constraints for the optimization.
Electricity price Water price CO₂ price

5.4. MPC
The optimization results shown in Figure 10 are
Figure 15: Sensitivity analysis of electricity price, water price
and CO2 price taken as the reference trajectories. The MPC param-
eter settings are as follows: the predictive horizon
Np = 10, the control horizon Nc = Np , the sam-
5.3.2. Influence of constraints change pling interval Tsm = 60 s, the total simulation time
The optimization results of strategy 4 under the Tm = 24 h. In this paper, Q = diag(100, 100, 100),
constraints of different percentage changes are shown R = diag(1, 1, 1).
in Figure 16. We can find that both increasing the The comparison results between the open loop
upper limit of temperature (Tmax ) and reducing the control (51) and the proposed MPC (55) under 2%
lower limit of temperature (Tmin ) will reduce the cost system disturbances are shown in Figure 17. To com-
of greenhouse operation. The difference is that re- pare the tracking performance of open loop control
ducing the lower limit of temperature can significant- and MPC, the tracking performance index relative
ly reduce the cost while increasing the upper limit average deviation (RAD) is introduced.
of temperature cannot. When the lower limit of the Denote the value of actual measurement as xmeas ,
temperature constraint is reduced by 15% (from 100% the relative deviation (RD) of x is defined by:
to 85%), the cost is reduced by 43.31% (from R 10791
xmeas (i) − xref (i)
to R 6123). However, when the upper bound is in- RD(i) = . (56)
creased by 15% (from 100% to 115%), the cost is only xref (i)
reduced by 2.90% (from R 10791 to R 10490). More- The RAD can be obtained by:
over, when the upper limit of temperature reaches
110%, increasing the upper limit will no longer affect N
1 X
the optimization results. RAD = RD(i). (57)
N
i=1

13
 Reference trajectory Open loop control MPC
40
Tair (°C)

20

0
0 5 10 15 20
100
RHair (%)

95

50 90

85
1 1.2 1.4 1.6 1.8 2
0
0 5 10 15 20
Cair (ppm)

600

400

200
0 5 10 15 20
Time (hour)

Figure 17: Comparison of open loop control and MPC under 2% system disturbances

where N is the total sampling times. For the open 6. Conclusion

loop control, N = 288. For the MPC, N = 1440.
The comparison of RAD between open loop con- Four optimization strategies to improve the oper-
trol and MPC is shown in Figure 18. It can be seen ating efficiency of a Venlo-type greenhouse are stud-
that, compared with the open loop control, the MPC ied. These strategies are to minimize greenhouse en-
can reduce 81.22% temperature RAD (from 6.23% to ergy consumption (Strategy 1), irrigation water con-
1.17%), 76.41% relative humidity RAD (from 7.46% sumption (Strategy 2), CO2 consumption (Strategy
to 1.76%), and 69.51% CO2 concentration RAD (from 3) and operation cost (Strategy 4) while keeping green-
3.28% to 1%). Compared with the open loop control, house climatic factors include the temperature, rela-
the proposed MPC can effectively reduce the tracking tive humidity, and CO2 concentration within the re-
error. quired range. A multi-input multi-output greenhouse
(MIMO) climate model and a modified evapotran-
spiration model are adopted. Moreover, a sensitiv-
8.00% 7.46% ity analysis is conducted to study the influence of
7.00% 6.23% electricity price, water price, CO2 price and system
6.00% constraints on optimization results. Finally, a mod-
5.00% el predictive controller (MPC) is designed to address
4.00% 3.28%
system disturbances. A performance index relative
3.00%

2.00%
1.76% average deviation (RAD) is introduced to evaluate
1.17% 1.00%
1.00%
the tracking performance.
0.00% Simulation results show that the cost of Strate-
Temperature Relative humidity CO₂ concentration
gy 4 (R 10791) is reduced by 66.60%, 92.68% and
Open loop control MPC
68.83% compared with Strategy 1 (R 32308), Strate-
gy 2 (R 147440) and Strategy 3 (R 34624) respective-
ly. Changes in electricity price have a greater impact
Figure 18: Comparison of RAD between open loop control and
MPC
on optimization results than changes in water price
and CO2 price. Both temperature constraints and
relative humidity constraints have a great influence
on the optimization results. The MPC controller de-
signed is verified to be effective.
In future research, we will focus on the following
14
aspects. 1) Use a hybrid energy system composed greenhouse climate control problem, IEEE Transaction-
of PV panel, wind generator, power grid, diesel gen- s on Automation Science and Engineering 13 (2) (2015)
772–788.
erators and battery bank to power the greenhouse
[15] D. Xu, S. Du, G. van Willigenburg, Adaptive two time-
system. 2) The greenhouse optimization process con- scale receding horizon optimal control for greenhouse let-
siders some long-term objectives such as crop yields tuce cultivation, Computers and Electronics in Agricul-
and greenhouse production profits. 3) Distributed ture 146 (2018) 93–103.
[16] A. Nadal, P. Llorach-Massana, E. Cuerva, E. López-Capel,
control of large-scale greenhouse systems. 4) To veri-
J. I. Montero, A. Josa, J. Rieradevall, M. Royapoor,
fy the effectiveness of the proposed strategies through Building-integrated rooftop greenhouses: An energy and
relevant experimental studies. environmental assessment in the mediterranean context,
Applied Energy 187 (2017) 338–351.
[17] F. Tap, Economics-based optimal control of greenhouse
References tomato crop production, Ph.D. thesis, Wageningen Agri-
cultural University, Wageningen, The Netherlands (2000).
[1] Y. Zhang, S. Hou, J. Liu, H. Zheng, J. Fang, J. Wang,
[18] F. He, L. Zeng, D. Li, Z. Ren, Study of led array fil-
The impacts of interprovincial electricity transmission on
l light based on parallel particle swarm optimization in
China’s water crisis: Mitigate or aggravate, Journal of
greenhouse planting, Information Processing in Agricul-
Cleaner Production (2020) 121696.
ture 6 (1) (2019) 73–80.
[2] L. Zhang, X. Xia, J. Zhang, Improving energy efficiency
[19] J. Chen, J. Yang, J. Zhao, F. Xu, Z. Shen, L. Zhang, En-
of cyclone circuits in coal beneficiation plants by pump-
ergy demand forecasting of the greenhouses using nonlin-
storage systems, Applied Energy 119 (2014) 306–313.
ear models based on model optimized prediction method,
[3] E. Cuce, D. Harjunowibowo, P. M. Cuce, Renewable and
Neurocomputing 174 (2016) 1087–1100.
sustainable energy saving strategies for greenhouse sys-
[20] M. C. Bozchalui, C. A. Cañizares, K. Bhattacharya, Op-
tems: A comprehensive review, Renewable and Sustain-
timal energy management of greenhouses in smart grids,
able Energy Reviews 64 (2016) 34–59.
IEEE Transactions on Smart Grid 6 (2) (2014) 827–835.
[4] R. Poudyal, P. Loskot, R. Nepal, R. Parajuli, S. K. Khad-
[21] A. Ramı́rez-Arias, F. Rodrı́guez, J. L. Guzmán, M. R.
ka, Mitigating the current energy crisis in Nepal with re-
Arahal, M. Berenguel, J. C. López, Improving efficiency
newable energy sources, Renewable and Sustainable Ener-
of greenhouse heating systems using model predictive con-
gy Reviews 116 (2019) 109388.
trol, IFAC Proceedings Volumes 38 (1) (2005) 40–45.
[5] M. M. Mekonnen, A. Y. Hoekstra, Four billion people fac-
[22] A. Ramı́rez-Arias, F. Rodrı́guez, J. L. Guzmán,
ing severe water scarcity, Science Advances 2 (2) (2016)
M. Berenguel, Multiobjective hierarchical control archi-
e1500323.
tecture for greenhouse crop growth, Automatica 48 (3)
[6] M. Kohler, Differential electricity pricing and energy effi-
(2012) 490–498.
ciency in South Africa, Energy 64 (2014) 524–532.
[23] X. Blasco, M. Martı́nez, J. M. Herrero, C. Ramos, J. San-
[7] I. Pretorius, S. Piketh, R. Burger, The impact of the South
chis, Model-based predictive control of greenhouse climate
African energy crisis on emissions, Transactions on Ecol-
for reducing energy and water consumption, Computers
ogy and the Environment 4 (2015) 255–264.
and Electronics in Agriculture 55 (1) (2007) 49–70.
[8] S. Hedden, J. Cilliers, Parched prospects-the emerging wa-
[24] C.-Z. Zeng, Z.-L. Bie, B.-Z. Yuan, Determination of opti-
ter crisis in South Africa, Institute for Security Studies
mum irrigation water amount for drip-irrigated muskmel-
Papers 2014 (11) (2014) 16–16.
on (Cucumis melo L.) in plastic greenhouse, Agricultural
[9] Z. Donnenfeld, C. Crookes, S. Hedden, A delicate balance:
Water Management 96 (4) (2009) 595–602.
Water scarcity in South Africa, ISS Southern Africa Re-
[25] V. Salokhe, M. Babel, H. Tantau, et al., Water require-
port 2018 (13) (2018) 1–24.
ment of drip irrigated tomatoes grown in greenhouse in
[10] A. Costantino, L. Comba, G. Sicardi, M. Bariani, E. Fab-
tropical environment, Agricultural Water Management
rizio, Energy performance and climate control in mechan-
71 (3) (2005) 225–242.
ically ventilated greenhouses: A dynamic modelling-based
[26] Z. Chalabi, A. Biro, B. Bailey, D. Aikman, K. Cockshull,
assessment and investigation, Applied Energy 288 (2021)
Optimal control strategies for carbon dioxide enrichment
116583.
in greenhouse tomato Crops. Part 1: Using pure carbon
[11] Y. Guo, H. Zhao, S. Zhang, Y. Wang, D. Chow, Model-
dioxide, Biosystems Engineering 81 (4) (2002) 421–431.
ing and optimization of environment in agricultural green-
[27] T. Kuroyanagi, K.-i. Yasuba, T. Higashide, Y. Iwasaki,
houses for improving cleaner and sustainable crop produc-
M. Takaichi, Efficiency of carbon dioxide enrichment in
tion, Journal of Cleaner Production (2020) 124843.
an unventilated greenhouse, Biosystems Engineering 119
[12] C. Bersani, A. Ouammi, R. Sacile, E. Zero, Model predic-
(2014) 58–68.
tive control of smart greenhouses as the path towards near
[28] J. Herrero, X. Blasco, M. Martı́nez, C. Ramos, J. Sanchis,
zero energy consumption, Energies 13 (14) (2020) 3647.
Non-linear robust identification of a greenhouse model us-
[13] S. Zhang, Y. Guo, H. Zhao, Y. Wang, D. Chow, Y. Fang,
ing multi-objective evolutionary algorithms, Biosystems
Methodologies of control strategies for improving energy
Engineering 98 (3) (2007) 335–346.
efficiency in agricultural greenhouses, Journal of Cleaner
[29] N. Bennis, J. Duplaix, G. Enéa, M. Haloua, H. Youlal,
Production (2020) 122695.
Greenhouse climate modelling and robust control, Com-
[14] Y. Su, L. Xu, D. Li, Adaptive fuzzy control of a class
puters and Electronics in Agriculture 61 (2) (2008) 96–107.
of MIMO nonlinear system with actuator saturation for
[30] S. Speetjens, J. Stigter, G. Van Straten, Towards an adap-

15
 tive model for greenhouse control, Computers and Elec- [46] J. Zhang, J. Liu, P. E. Campana, R. Zhang, J. Yan,
tronics in Agriculture 67 (1-2) (2009) 1–8. X. Gao, Model of evapotranspiration and groundwater lev-
[31] D. Katzin, S. van Mourik, F. Kempkes, E. J. van Hen- el based on photovoltaic water pumping system, Applied
ten, Greenlight–an open source model for greenhouses Energy 136 (2014) 1132–1137.
with supplemental lighting: Evaluation of heat require- [47] R. Qiu, J. Song, T. Du, S. Kang, L. Tong, R. Chen, L. Wu,
ments under led and hps lamps, Biosystems Engineering Response of evapotranspiration and yield to planting den-
194 (2020) 61–81. sity of solar greenhouse grown tomato in northwest China,
[32] R. Qiu, S. Kang, F. Li, T. Du, L. Tong, F. Wang, R. Chen, Agricultural Water Management 130 (2013) 44–51.
J. Liu, S. Li, Energy partitioning and evapotranspiration [48] K. A. Czyzyk, S. T. Bement, W. F. Dawson, K. Mehta,
of hot pepper grown in greenhouse with furrow and drip Quantifying water savings with greenhouse farming,
irrigation methods, Scientia Horticulturae 129 (4) (2011) in: IEEE Global Humanitarian Technology Conference
790–797. (GHTC 2014), IEEE, 2014, pp. 325–332.
[33] D. Lin, L. Zhang, X. Xia, Hierarchical model predictive [49] Y. Su, L. Xu, E. D. Goodman, Nearly dynamic program-
control of Venlo-type greenhouse climate for improving ming NN-approximation–based optimal control for green-
energy efficiency and reducing operating cost, Journal of house climate: A simulation study, Optimal Control Ap-
Cleaner Production (2020) 121513. plications and Methods 39 (2) (2018) 638–662.
[34] P. Van Beveren, J. Bontsema, G. Van Straten, E. Van Hen- [50] H. Hu, L. Xu, B. Zhu, R. Wei, A compatible control algo-
ten, Optimal control of greenhouse climate using minimal rithm for greenhouse environment control based on MOCC
energy and grower defined bounds, Applied Energy 159 strategy, Sensors 11 (3) (2011) 3281–3302.
(2015) 509–519. [51] C. Stanghellini, T. de Jong, A model of humidity and
[35] P. Van Beveren, J. Bontsema, G. Van Straten, E. Van Hen- its applications in a greenhouse, Agricultural and Forest
ten, Minimal heating and cooling in a modern rose green- Meteorology 76 (2) (1995) 129–148.
house, Applied Energy 137 (2015) 97–109. [52] H.-P. Kläring, C. Hauschild, A. Heißner, B. Bar-Yosef,
[36] E. Van Henten, Sensitivity analysis of an optimal control Model-based control of CO2 concentration in greenhouses
problem in greenhouse climate management, Biosystems at ambient levels increases cucumber yield, Agricultural
Engineering 85 (3) (2003) 355–364. and Forest Meteorology 143 (3-4) (2007) 208–216.
[37] J. Bontsema, J. Hemming, C. Stanghellini, P. De Viss- [53] D. Xu, S. Du, G. van Willigenburg, Double closed-loop
er, E. van Henten, J. Budding, T. Rieswijk, S. Nieboer, optimal control of greenhouse cultivation, Control Engi-
On-line estimation of the transpiration in greenhouse hor- neering Practice 85 (2019) 90–99.
ticulture, in: Proceedings Agricontrol 2007. 2nd IFAC In- [54] J. Mei, X. Xia, M. Song, An autonomous hierarchical con-
ternational Conference on Modeling and Design of Control trol for improving indoor comfort and energy efficiency of
Systems in Agriculture, Osijek, Croatia, 2008, pp. 29–34. a direct expansion air conditioning system, Applied Ener-
[38] E. Van Henten, Greenhouse climate management: an op- gy 221 (2018) 450–463.
timal control approach, Ph.D. thesis, Wageningen Agri- [55] M. S. Masaki, L. Zhang, X. Xia, A hierarchical predictive
cultural University, Wageningen, The Netherlands (1994). control for supercapacitor-retrofitted grid-connected hy-
[39] R. JC, Optimal control design for a solar greenhouse, brid renewable systems, Applied Energy 242 (2019) 393–
Ph.D. thesis, Wageningen Agricultural University, Wa- 402.
geningen, The Netherlands (2007). [56] J. Coelho, P. de Moura Oliveira, J. B. Cunha, Greenhouse
[40] R. Linker, P. Gutman, I. Seginer, Robust controllers for air temperature predictive control using the particle swar-
simultaneous control of temperature and CO2 concentra- m optimisation algorithm, Computers and Electronics in
tion in greenhouses, Control Engineering Practice 7 (7) Agriculture 49 (3) (2005) 330–344.
(1999) 851–862.
[41] A. Bianchi, D. Masseroni, A. Facchi, Modelling water re-
quirements of greenhouse spinach for irrigation manage-
ment purposes, Hydrology Research 48 (3) (2017) 776–
788.
[42] P. E. Campana, H. Li, J. Yan, Techno-economic feasibility
of the irrigation system for the grassland and farmland
conservation in China: Photovoltaic vs. wind power water
pumping, Energy Conversion and Management 103 (2015)
311–320.
[43] P. E. Campana, H. Li, J. Zhang, R. Zhang, J. Liu, J. Yan,
Economic optimization of photovoltaic water pumping
systems for irrigation, Energy Conversion and Manage-
ment 95 (2015) 32–41.
[44] R. G. Allen, L. S. Pereira, D. Raes, M. Smith, et al., Crop
evapotranspiration-guidelines for computing crop water
requirements-FAO irrigation and drainage paper 56, Fao,
Rome 300 (9) (1998) D05109.
[45] P. E. Campana, H. Li, J. Yan, Dynamic modelling of a
PV pumping system with special consideration on water
demand, Applied Energy 112 (2013) 635–645.

16