2017 American Control Conference

Sheraton Seattle Hotel

May 24–26, 2017, Seattle, USA

Climate Control in Greenhouse using Intelligent Control Algorithms

Revathi S, Radhakrishnan T K, and Sivakumaran N

Abstract— Greenhouse climate control problem has received nonlinear fuzzy immune PID algorithm is tested on the
considerable attention in agriculture engineering research. The interacting multivariable greenhouse climate control.
greater part of accomplishing ensured farming within the For the purpose of comparison, four controller
greenhouse environment is achieved by controlling the algorithms namely Proportional-Integral-Derivative (PID)
temperature and relative humidity. As the result of process control, Linear Quadratic Regulator control, Rule based
dead times and extreme interdependence of these parameters,
Fuzzy Logic Control and Fuzzy Immune PID control are
the control problem is classified to be non-linear and multi-
variable. With the advances in intelligent control systems, more studied. In “non-intelligent” technique, PID and LQR
and more decisions involved in greenhouse control can be approach that is more classical is considered. In “Intelligent”
automated. Thus, more emphasis can be placed on emulating technique, Fuzzy PID and Fuzzy immune PID are
the abilities of the expert operator. In this paper, intelligent and considered. Even though fuzzy logic is widely used in
non-intelligent control techniques for addressing the problem control, fuzzy immune control is new and it is not
of automated climate control in a greenhouse are investigated. prevalently used in the control arena. Numerical simulation
These include proportional-integral-derivative (PID) and is performed and the advantages of intelligent controllers are
Linear-Quadratic regulator (LQR) as a ‘non-intelligent’ compared against non-intelligent controller with system
technique and fuzzy PID and fuzzy immune PID as ‘intelligent’
responses to set-point changes.
technique. The new study is made for implementing the
nonlinear fuzzy immune PID controller for greenhouse climate
control. This controller has a simple structure and its II. GREENHOUSE CLIMATE MODELLING AND DECOUPLING
parameters can be conveniently adjusted. It consists of a PID Based on mass and energy balance inside the
controller and a basic immune proportional controller in greenhouse, a dynamic model describing the latent and
cascaded connection, the nonlinear function of the immune
sensible heat and the water vapor balance is developed [2].
proportional controller is realized using fuzzy reasoning. Thus,
controller parameters are adjusted online by the rules of This is used as a multi-season model. This simplified model
immune feedback controller and fuzzy controller. The considers only the significant disturbances such as outside
simulation results are compared for the effectiveness and temperature and humidity, and the solar radiation. The
applicability to greenhouse environmental problem. differential equations for coupled energy and water vapor
mass balances are mentioned in (1a) and (1b) respectively.
I. INTRODUCTION
𝑑𝑇𝑖𝑛(𝑡) 1
Greenhouses allow producers to grow plants at a = [𝑞ℎ𝑒𝑎𝑡𝑒𝑟 (𝑡) + 𝑆𝑖 (𝑡) − 𝜆𝑞𝑓𝑜𝑔 (𝑡)] −
𝑑𝑡 𝜌𝐶𝑝 𝑉
time when it would be impossible to grow outside because ̇
𝑉 (𝑡) 𝑈𝐴
of adverse climate, pests and diseases [1]. In order to [𝑇 (𝑡) − 𝑇𝑜𝑢𝑡 (𝑡)] − [𝑇 (𝑡) − 𝑇𝑜𝑢𝑡 (𝑡)](1a)
𝑉(𝑡) 𝑖𝑛 𝜌𝐶 𝑉 𝑖𝑛
𝑝
achieve maximum returns from greenhouse cultivation, it is
important to maintain an environment that minimizes the 𝑑𝜔𝑖𝑛 (𝑡) 1 1 𝑉̇ (𝑡)
energy consumption. The greenhouse climate is a nonlinear = 𝑞𝑓𝑜𝑔 (𝑡) + 𝐸(𝑆𝑖 (𝑡), 𝜔𝑖𝑛 (𝑡)) − [𝜔𝑖𝑛 (𝑡) −
𝑑𝑡 𝜌𝑉 𝜌𝑉 𝑉(𝑡)
Multiple input Multiple Output (MIMO) system 𝜔𝑜𝑢𝑡 (𝑡)] (1b)
characterized by high coupling, with strong variation of
parameters due to crop development stages, and it is highly where Tin (t) , Tout (t) are the greenhouse and ambient
influenced by disturbances (outside temperature and temperature respectively (°C), V is the greenhouse volume
humidity, wind speed, solar radiation). The main state (m3 ), ρ is the air density (1.2 kg/m3 ), Cp is the specific heat
variables for greenhouse climate control are air humidity and of air (1006 J(KgK)−1 ), qheater is the heat provided by the
temperature. The interdependence of these parameters makes greenhouse heater (W), Si is the intercepted solar radiant
it a very challenging control problem. The efficacy of
energy (W), qfog is the water capacity of the fog system (g
H2 O/s), λ is the latent heat of vaporization (2257 J/g), V̇ is
the ventilation rate (m3 /s), ωin and ωout are the interior and
Revathi S is with Department of Instrumentation and Control in the exterior absolute humidity (g H2 O/m3 ), E(Si , ωin ) is the
National Institute of Technology, Tiruchirapalli, Tamil Nadu, India- evapotranspiration rate of the plants (g H2 O/s). In this
620015.( phone: 9488611107; e-mail: [email protected]).
Radhakrishnan T K is with the Chemical Engineering Department, paper, just summer operations are considered and hence
National Institute of Technology, Tiruchirapalli, Tamil Nadu, India-620015. heater element is not used, i.e. qheater = 0.
(e-mail: [email protected]). 𝑆 (𝑡)
Sivakumaran N, is with Department of Instrumentation and Control,
𝐸(𝑆𝑖 (𝑡), 𝜔𝑖𝑛 (𝑡)) = 𝛼 𝑖 − 𝛽𝑇 𝜔𝑖𝑛 (𝑡) (2)
𝜆
National Institute of Technology, Tiruchirapalli, Tamil Nadu, India-620015.
(e-mail: [email protected]).

978-1-5090-5992-8/$31.00 ©2017 AACC 887

where α is the coefficient to account for shading and leaf
area index, and βT is the coefficient to account for
thermodynamic constants and other factors affecting
evapotranspiration. By using (2), the nonlinear model can be
expressed as (3a) and (3b).

𝑈𝐴 1 𝜆
𝑥̇1 (𝑡) = − 𝑥 (𝑡) − 𝑥1 (𝑡)𝑢1 (𝑡) − 𝑢 (𝑡) +
𝜌𝐶𝑝 𝑉 1 𝑉 𝜌𝐶𝑝 𝑉 2
1 𝑈𝐴 1
𝑣 (𝑡) + 𝑣 (𝑡) + 𝑢1 (𝑡)𝑣2 (𝑡) (3a)
𝜌𝐶 𝑉 1
𝑝 𝜌𝐶 𝑉 2 𝑝 𝑉 Figure 1. Block diagram of General PID Controller
𝛽𝑇 1 1 1
𝑥̇ 2 (𝑡) = 𝑥2 (𝑡) + 𝑢2 (𝑡) + 𝑣1 (𝑡) − 𝑥2 (𝑡)𝑢1 (𝑡) +
𝜌𝑉 𝜌𝑉 𝜆𝑝 𝑉 𝜌𝑉 The proportional part is responsible for following
1
𝑢1 (𝑡)𝑣3 (𝑡) (3b) the desired set point, while the integral and derivative part
𝜌𝑉
account for the accumulation of past errors and the rate of
change of error in the process respectively. The PID tuning
The state variables 𝑥1 and 𝑥2 are the greenhouse inside
rules based on direct synthesis method is in Table I.
temperature and the interior absolute humidity. The input
control variables 𝑢1 and 𝑢2 are the fan ventilation rate and TABLE I. PID TUNING RULES BASED ON DIRECT-SYNTHESIS DESIGN
the water capacity of the fog system. The external
disturbances 𝑣1 , 𝑣2 and 𝑣3 correspond to the intercepted 𝑲𝒑 𝑻𝒊 𝑻𝒅
solar radiation, outside temperature and outside absolute 2𝜆 + 1.5𝐿 2 𝜆+1.5L 𝐿𝜆 + 0.5𝐿2
humidity. By using combined scheme of feedback with 𝐾(𝜆2 + 𝑠𝜆𝐿 + 0.5𝐿2 ) 2𝜆 + 1.5𝐿
simultaneous feedforward linearization, the temperature
changes are modeled as a self-regulating first-order plus b. LQR Controller Design
dead time (FOPDT) model while interior humidity changes The LQR theory is a powerful method for the
are modeled as an integrator plus dead-time model as shown control of linear systems. It generates controllers with
in (4a) and (4b). guaranteed closed-loop stability and robustness property
𝐾 𝑒 −𝑑𝑇 𝑠 even in the face of uncertain gain and phase variation at the
𝑋1 (𝑠) = 𝑇 ̃𝑇 (𝑠)
𝑈 (4a)
𝜏𝑇 𝑠+1 plant input/output [10]. In addition, the LQR-based
𝐾𝜔 𝑒 −𝑑𝜔 𝑠
𝑋2 (𝑠) = ̃𝜔 (𝑠)
𝑈 (4b) controllers provide reliable closed-loop system performance
𝑠
𝜌𝐶𝑝 𝑉
̃𝑇 , 𝜏 𝑇 = 𝜌𝐶𝑝 𝑉 𝑉
̃𝜔 and despite the presence of stochastic plant disturbances. It
where 𝐾𝑇 = 𝐾 , 𝐾𝜔 = 𝐾 minimizes the following quadratic objective function shown
𝑈𝐴 𝑈𝐴 𝛽𝑇
𝑑 𝑇 = 𝑑𝜔 = 30𝑠𝑒𝑐 in (6).
∞
III. CONTROL STRATEGIES J(k) = ∫0 ( x T (t)Qx(t) + uT (t)Ru(t)) dt (6)
where ‘Q’ is a nonnegative-definite matrix that penalizes the
Control of greenhouse systems has become more departure of system states from the equilibrium and ‘R’ is a
important in the recent years. Traditionally, both non- positive-definite matrix that penalizes the control input is
intelligent and intelligent techniques have been deployed in used. The solution to the LQR problem is obtained via
literature. Lagrange multiplier-based optimization technique and is
A. Non-Intelligent control technique given by (7).
K = 𝑅−1 𝐵𝑇 P (7)
a. PID Controller Design
The PID controller is the most widely used B. Intelligent control technique
controller in industry today and it possesses the simple The parameters of the PID controllers K p , K 𝐼 , K 𝑑
robust and stable properties. PID controller with the transfer are tuned adaptively to the changes in error and rate of
function is shown in (5) change of error using fuzzy intelligent technique. This leads
1 to the optimal values of the controller parameters.
𝐶(𝑠) = K p (1 + + TD s) (5)
Ti s
where K p , Ti , TD are the proportional gain, integral time and a. Design of Fuzzy PID Controller
derivative time respectively. A wide range of methods are A fuzzy expert system is an expert system that uses
available in literature for tuning the parameters of PID a collection of fuzzy membership functions and rules,
controller to processes modeled by integrator plus dead time instead of Boolean logic, to reason about data. A fuzzy PID
models viz., empirical formulae, analytical methods and controller takes the conventional PID controller as the
frequency-domain approaches. The best values are used in foundation and uses the fuzzy reasoning and variable
the simulation [3] [4] [5]. The general PID control structure universe of discourse to regulate the PID parameters [6]. The
is in Fig. 1. structure of the controller is in Fig. 2.

888
 This is an immune feedback control. The P controller is:
u(k) = k p e(k) (11)
Where we α(k) assume as the system deviation e(k) , S(k)
as the control input u(k), then equation (11) is rewritten as
u(k) = K (1 − ηf(u(k), Δu(k))) e(k) = K p1 e(k) (12)
where K p1 = K (1 − ηf(u(k), Δu(k))), K = k1 is control
reaction rate, if k1 increases, it can improve the response
speed. η = k 2 /k1 is control-stability effect, and if η
Figure 2. Fuzzy PID Control structure
increases, it can reduce the overshoot and improve the
stability. The block diagram representation is in Fig. 4.
The characteristics of the fuzzy system (such as
robustness and adaptability) incorporated into the controlling
method for better tuning of PID parameters. The self-tuning
characteristics of this controller to tune the controlling
parameters on-line automatically, gives the most suitable
values of PID parameters that results in optimization of the
process output [7].

b. Design of Fuzzy Immune PID Controller

Biological immune system is a complex system,
which includes lymphocytes and antibody molecules [8] [9].
T cells and B cells form lymphocytes. Further, the T cells
differentiate into TH cells (named helper T-cells) and TS cells
(named suppressor T-cells) by antigen stimulus. TH cells Figure 4. Block diagram of the fuzzy immune PID control system
could promote the production of B cells, TS cells could
inhibit the production of B cells. Both the TH cells and TS From (11) and (12), the selection of f(∗) is crucially
cells stimulate B cells to produce antibodies to destroy the important. In this paper, the principle of fuzzy control
antigen. T cells and B cells work together to maintain the system employed to approximate the nonlinear function f(∗).
balance of the body immune system. The basic configuration The motivation behind this selection is the relative easiness
diagram of a biological immune system is in Fig. 3. of designing nonlinearity in the fuzzy logic paradigm.
IV. SIMULATION RESULTS
For the simulation study, a greenhouse of surface
area 1000𝑚2 and height of 4m is considered. The maximum
water capacity of the fog system is 26 (g H2 O/s)and
maximum ventilation rate corresponds to 23 (m3 /s). The
desired temperature set point of 20℃ and desired humidity
ratio set point of 24 g/Kg is applied. The outside weather
conditions are 𝑇𝑜𝑢𝑡 = 35℃, 𝜔𝑜𝑢𝑡 = 4 𝑔/𝑘g and 𝑆𝑖 =
300 𝑊. 𝑚−2 .
Optimal PID parameters are obtained by
substituting 𝜆 = 𝐿 in the Table I. are 𝐾𝑝 = .03, 𝐾𝑖 =
.00028 and 𝐾𝑑 = 0.384. Proportional term 𝐾𝑝 , integral term
Figure 3. Illustration of immune feedback model in biological system
𝐾𝑝
𝐾𝑖 = and derivative term K D = 𝐾𝑝 TD applied to the
Ti
𝑡ℎ
Suppose the number of the 𝑘 generation antigen is α (k), PID controller in the MATLAB/Simulink environment for
the number of TH cells is T𝐻 (k), then, the greenhouse climate control.
TH (k) = k1 α(k) (8) The design of a stabilizing LQR controller is
where k1 is the positive factor. The TS cells could affect the straightforward and provides some means of tuning the
production of B cells and assuming this effect of TS cells to closed-loop such that the amplitude of the control input
B cells is TS (k), then, minimized. Numerical tuning gave R=8 × 105 for the
TS (k) = k 2 f(∆s(k))α(k) (9) required operating conditions of greenhouse environment.
where k 2 is an inhibitory factor and its symbol is positive. The servo response of non-intelligent controllers for a step
f(∗) is a nonlinear function, which represents the inhibition change in temperature and humidity is in Fig. 5. Unlike PID
amount of TS (k) cells. The output of f(∗) is limited to [0, 1]. controller, LQR controller makes a smooth set point tracking
By (8) and (9), the total stimulation received by B cells is: without an overshoot or undershoots but it produces an
S(k) = TH (k) − TS (k) = (k1 − k 2 f(∆s(k)))α(k) (10) offset.

889
 40
Temperature step change response curves for Non-Intelligent Controllers
N NB NB N O P
Temperature ( degree celcius)
35 O N N O P P
Desired Temperature
30
PID P N O P P PB
LQR

25
PB O P P PB PB
20

15
TABLE IV. FUZZY RULES FOR K D
500 600 700 800 900 1000 1100 1200 1300 1400 1500
Time (sec) CE NB N O P PB
Humidity step change response curves for Non-Intelligent Controllers
20

18 E
16 NB N N NB NB P
Humidity ratio (g/Kg)

14
PID
12
LQR N N N N N O
10 Desired Humidity level

8
O O O O O O
6

2
P PB O P P PB
0
500 600 700 800 900 1000 1100 1200 1300 1400 1500
Time (sec) PB PB P P P PB
Figure 5. Servo Response for non-intelligent controllers

In fuzzy PID design, two inputs to the fuzzy logic In fuzzy Immune PID controller, the proportion
controller are error ‘E’ and change in error ‘CE’ and the factor ′𝐾𝑝 ′ is the output of immune fuzzy adjusting
output are the PID parameters. The fuzzy membership and 𝐾𝐼 , 𝐾𝐷 are adjusted online according to fuzzy PI
functions are ‘NB’, ‘N’, ‘O’, ‘P’, ‘PB’ which means controller. In the fuzzy immune adjusting, double inputs and
negative big, negative, zero, positive, positive big single output are used. The inputs are u (k) and ∆u (k), and
respectively for all the membership variables. The fuzzy the output is f (u (k), ∆u (k)). The input variables are
control rule for K p , K I , K D are listed in the Table II, III, IV fuzzified by two fuzzy sets, which are named “Positive” (P)
respectively. and “Negative” (N), and the output variable is fuzzified by
three fuzzy sets, which are named “Positive” (P), “Negative”
TABLE II. FUZZY RULES FOR K P
(N) and “Zero” (Z). Fuzzy inference rules are derived by
CE NB N O P PB Lyapunov synthesis method to ensure f(∗) on the stability of
the system. The fuzzy rules are in Table V.
E
NB PB PB P P O TABLE V. FUZZY RULES FOR IMMUNE FUZZY TUNED K P

∆U P N
N PB PB P O N
U
O P P O N N P N Z

P P O N N NB N Z P

PB O N N NB NB
Using fuzzy logic Zadeh AND operation, Mamdani
algorithm and centroid defuzzification, the f(∗) can be fitted
precisely. The combined response with implementation of
TABLE III. FUZZY RULES FOR K 𝐼
intelligent controllers for the variation in step change in
CE NB N O P PB temperature and humidity in the greenhouse is in Fig. 6. It is
inferred that fuzzy immune PID has less settling time
E compared to fuzzy PID control.
NB NB NB N N O

890
 Temperature step change response curves for Intelligent Controllers Response of temperature during unknown disturbance
40 32

Temperature (Degree celcius) 30

35
28 Desired Temperature
Fuzzy immune PID
30 26 Fuzzy PID

Temperature (Degree
24 Celsius)
25 Fuzzy Immune PID
Fuzzy PID 22
Desired temperature

20 20

18
15
500 600 700 800 900 1000 1100 1200 1300 1400 1500
16
Time (sec) 500 1000 1500 2000
Humidity step change response curves for Intelligent Controllers Response of Humidity during unknown disturbace Time (sec)
22

20 Response of Humidity during unknown disturbance

34
18
Humidity ratio (g/Kg)

32
16 Fuzzy Immune PID
30 Fuzzy PID
14 Fuzzy immune PID Desired Humidity
Fuzzy PID 28
12 Desired humidity

Humidity (g/Kg)
26
10
24
8
22
6
800 900 1000 1100 1200 1300 1400 1500 20
Time (sec)
18
Figure 6. Servo response for intelligent controllers
16

14
600 800 1000 1200 1400 1600 1800 2000

TABLE VI. COMPARISON OF PERFORMANCE CRITERIA FOR Time (sec)

GREENHOUSE Figure 7. Regulatory response of intelligent controllers

Fuzzy From Fig. 7, it is inferred that the fuzzy immune

Fuzzy PID controller responds more quickly and achieves good
Controller PID LQR Immune
PID performances than fuzzy PID. Thereby self-adaptability of
PID
fuzzy immune PID controller makes it more applicable to a
Temperature
41.96 101.6 36.71 36.13 complex greenhouse environment.
IAE
Humidity
35.80 83.34 29.38 28.75 V. CONCLUSION
Temperature For complex greenhouse environment with
22.03 35.13 22.62 21.97
measurable external disturbance, using feedback-
ISE
Humidity feedforward linearization and decoupling a FOPDT and
9.41 16.58 7.81 7.678
Integrator with dead time model for changes in interior
Temperature temperature and humidity is developed. To this equivalent
0.58 38.62 0.65 0.271
model, two classes of controllers are implemented in
∫ ∆𝑢2
Humidity simulation and responses for the set point step changes in
0.16 11.09 0.67 0.68
humidity and temperature are observed. From the previous
section, it is obvious that in order to get better control
From Table VI, it is inferred that the integral performance than that of the non-intelligent controller, the
absolute error and integral square error is high for LQR and control parameters should be adjusted immediately in a
PID compared to intelligent controllers. Moreover, the complicated time varying and time delay system by the
intelligent controller uses less time to return to the desired
intelligent control algorithm. Moreover, the simulation
set point value and the offset is relatively zero. In addition,
results with a fuzzy immune PID control scheme for the
the controller effort of fuzzy PID and fuzzy immune PID are
greenhouse climate control shows a satisfactory performance
closer to the controller effort of PID and relatively lesser
than LQR. Thus with minimal controller output the desired even in the presence of interactions between the process
temperature and humidity is achieved by implementation of variables and the external meteorological conditions in
intelligent control algorithms. comparison to a fuzzy PID controller as evident from the
lowest performance criteria values. This is possible because
In addition, to investigate the regulatory response of the controller performs the dual role of fuzzy and biological
intelligent controllers, a unit disturbance is given at 700th immune feedback. This provides strong robustness and
instance and the response of fuzzy PID and fuzzy immune better dynamic performance and thereby ensuring optimum
PID are studied. growing conditions for the plants inside the greenhouse.

891
 REFERENCES
[1] Bennis N, Duplaix J, Enéa G, Haloua M, Youlal H, “Greenhouse
climate modeling and robust control”, Computers and Electronics in
Agriculture, vol. 61,no. 2, pp. 96–107, May 2008.
[2] L.D. Albright, R.S. Gates, K.G. Arvanitis, and A. Drysdale,
“Environmental control for plants on earth and in space,” IEEE
Control System Magazine, vol. 21, no. 5, pp. 28-47, Oct. 2001.
[3] M. Chidambaram and R. Padma Sree, “A simple method of tuning
PID controllers for integrator/dead-time processes,” Computers and
Chemical Engineering, vol. 27, no. 2, pp. 211-215, Feb. 2003.
[4] A. Seshagiri Rao, V.S.R. Rao, M. Chidambaram, “Direct synthesis
based controller design for integrating processes with time delay,”
Journal of the Franklin Institute, vol. 346, no. 1, pp. 38-56., Feb.
2009.
[5] L. Wang and W.R. Cluett, “Tuning PID controllers for integrating
processes,” IEE Proceedings - Control Theory and Applications, vol.
144, no. 5, pp. 385-392, Sep. 1997.
[6] Ma Yin-hui, Ding Jian-she and Xu De-song, “Design of PID
parameter self-adjusting fuzzy controller based on MATLAB and its
simulated applications,” Power Equipment, no.2, 112-124, 2005.
[7] G. Cao, C. Lou and D. “An Application of fuzzy self-tuning PID
control to air conditioning system,” Heat Ventilating and Air
Conditioning, vol. 34, no.10, 106-109, October 2004.
[8] Wang, W., Gao, X. Z., Wang, C. H., “Fuzzy Immune PID Controller
in Material-level Control of Preheating Cylinder,” International
Conference on Informatics and Control Technologies, pp. 52-55,
2006.
[9] Peng Daogang, Yang Ping, Wang Zhiping, et al., “Fuzzy immune
PID control and its application to main steam temperature control
system,” Computer Automated Measurement & Control, vol. 13,
no.3, 250-251, 2005.
[10] William S. Levine, The Control Handbook. CRC press, 1995.

892