Evolutionary Split Range Controller for a Refrigeration System

Gerardo José Amador Soto and Jesús-Antonio Hernandez-Riveros

Universidad Nacional de Colombia, Facultad de Minas, Cra. 80 #65-223, Medellín, Colombia

Keywords: Refrigeration, Adaptive and Learning Systems, Controller Design, Evolutionary Algorithms, Split Range

Control, Multiobjective Control.

Abstract: Every year more than 80 million units of domestic refrigerators are produced worldwide. Hundreds of millions

are used continuously so the impact on electricity is significant. Typical initiatives in energy efficiency for

refrigeration systems are aimed at: a) redesigns b) new materials and c) good use practices. A different

approach in energy efficiency for Vapour Compression Refrigeration Systems (VCRS) is the implementation

of control strategies that directly reduce energy consumption while guaranteeing operating conditions. The

difficulty lies in the multiple energy domains of the system (electric/mechanical/hydraulic/thermal), high

coupling, multiplicity of variables, strong non-linearity and non-stationary regime. This paper focuses on

increasing the energy efficiency of a VCRS with the implementation of an optimal split range and multi-

objective evolutionary control. The evolutionary control is expanded to variable speed compressors and

electronic expansion valves. The effectiveness of the evolutionary method was demonstrated through the

Benchmarking of the IFAC. Now, in the multi-domain model of the VCRS, the MAGO algorithm is applied

to optimally tune a split range controller to achieve a more precise behaviour and multi-objective to save

energy. The cases studied go from traditional control to multivariable and multivariable-multi-objective

control, the results in energy saving are outstanding.

1 INTRODUCTION

Heating, ventilation and air conditioning (HVAC) has

become a technological option that provides many

ways to contribute to humanity, the conservation of

meals through the control of indoor air quality, etc.

Approximately 30% of the total energy in the world

is consumed in HVAC processes, as well as in

refrigerators and water heaters (Jahangeer, Tay and

Raisul Islam, 2011). It is expected that the world's

energy consumption rates will increase by 33% from

2010 to 2030 (Khan, Ryan and Abebe, 2017). In

industry, refrigeration systems consume large

amounts of electricity, where refrigeration can be

responsible for up to 85% of total energy

consumption (depending on the industry sector). To

address this problem, the aim is to improve the

efficiency of the systems (HVAC).

The IIR (Industrial Info Resources) estimates the

number of refrigeration systems in operation

worldwide (Coulomb, Dupon and Pichard, 2015), as

summarized in Figure 1. The total number of HVAC

in operation worldwide is approximately 3 billion,

including 1.5 billion domestic refrigerators.

Annually, more than 80 million units of domestic

refrigerators are produced worldwide (Corte et al.,

2014). Today hundreds of millions are used, therefore

the global impact of the electric power consumption

of these systems is significant. In some countries, the

use of refrigeration systems has become an increasing

need allowing the development of technologies and

equipment with high efficiencies to fulfil this type of

tasks, in addition to the concern to reduce the

environmental impact.

Figure 1: Refrigeration systems in operation worldwide

(Coulomb, Dupon and Pichard, 2015).

Soto, G. and Hernandez-Riveros, J.

Evolutionary Split Range Controller for a Refrigeration System.

DOI: 10.5220/0007930803410351

In Proceedings of the 16th International Conference on Informatics in Control, Automation and Robotics (ICINCO 2019), pages 341-351

ISBN: 978-989-758-380-3

341

This paper presents a proposal to improve the energy

performance of a vapour compression refrigeration

system (VCRS) by means of implementing an

evolutionary control structure. To do this, two case

studies are presented. The first case seeks to improve

the energy performance of the system by controlling

the superheat temperature therefore improving the

thermal cycle of the system. For the second case, the

aim is to directly reduce the energy consumption on

the external power source by implementing different

evolutionary control strategies. In both cases the

temperature behaviour inside the cold chamber is the

direct control goal to satisfy. In the second case,

saving energy from the external power source is

added as a new objective.

This paper is organized as follows. In section 2 the

basics on VCRS is presented. In section 3 the

evolutionary control method is described together

with the optimizing process. In Section 4, the

effectiveness of the evolutionary control method is

verified with two case studies. Conclusions are

presented in section 5

2 VAPOUR COMPRESSION

REFRIGERATION SYSTEMS

The VCRS are the most used among all refrigeration

systems. These systems belong to the general class of

vapour cycles, where the working fluid (refrigerant)

presents phase changes at least during one

compression process. In a VCRS, cooling is obtained

by extracting thermal energy from an insulated space

to reduce its temperature. The input to the system is

in the form of mechanical energy required to run the

compressor. Hence, these systems are also called as

mechanical refrigeration systems. VCRS are

available to suit almost all applications with

refrigeration capacities ranging from few watts to few

megawatts. A wide variety of refrigerants can be used

to suit different applications, capacities etc. The

actual vapour compression cycle is based on Evans-

Perkins cycle, which is also called as reverse Rankine

cycle. In principle, all vapour compression

refrigeration systems are used to remove heat from

one location and transfer it to another by means of

mechanical power (compression).

A VCRS has four main components: a

compressor, a condenser, an expansion device and an

evaporator. In a cycling process, a circulating

refrigerant enters the compressor as saturated vapour

and it is compressed to a higher pressure, resulting in

a higher temperature as a superheated vapour. This

hot compressed vapour is condensed to liquid by

cooling air flowing across a coil carrying away heat

from the system. This high-pressure, high

temperature liquid leaving the condenser when

passing through an expansion valve is cooled and

reduced in pressure. In the evaporator, this low

pressure, low temperature liquid is converted to

vapour, absorbing heat from the refrigerated space

and keeping it cool, to then return to the compressor

and repeat the process. (see Figure 2).

Figure 2: A vapour compression refrigeration cycle scheme

(Bejarano, 2017).

3 EVOLUTIONARY CONTROL

STRATEGIES FOR A VCRS

VCRS refrigeration systems are closed cycles whose

components are connected through several pipes and

valves. This implies a difficulty in controlling these

processes due to some of the characteristics of the

system, such as multiple energy domains (electric /

mechanical / hydraulic / thermal), high coupling and

multiplicity of variables; operating under conditions

of strong non-linearity and in a non-stationary

regime. The inherent complexity of these systems

combined with the implementation of new control

strategies is not a trivial task. To overcome these

complexities, an optimization problem arises, since

the main objective is to find a combination of

controller parameters whose objective function

through maximization or minimization guarantees the

search for solutions that allow to improve the indexes

of efficient performance of the process. Given its

nature as a global optimizer of problems in different

areas of science, engineering and other branches of

knowledge (Fleming and Purshouse, 2002) one way

to address the problem of energy efficiency in VCRS

is through evolutionary algorithms (EA). The EA will

calculate the optimum points of operation of the

process based not only on the achievement of the

ICINCO 2019 - 16th International Conference on Informatics in Control, Automation and Robotics

342

desired production objectives, but also on the

reduction of energy consumption; both objectives are

to be met in this work.

The traditional control system for VCRS is an

on/off strategy applied on the expansion valve. The

highest efficiency of the evaporator is achieved if the

refrigerant at the outlet of the evaporator is saturated

vapour. Through the development of new

technologies such as variable speed compressors or

electronic expansion valves, it is possible to operate

the cycle with a certain degree of superheating of the

refrigerant at the outlet of the evaporator. This

approach requires a multivariable control, which is

very demanding technique in the system modelling

and in the tuning of the controllers. This paper

describes the strategy to tune a PID controller applied

to two actuating elements on the system (compressor

speed and expansion valve) to satisfy the expected

cooling demand and maximizing the energy

efficiency of the VCRS. The applied strategy is based

on the split range control (Smith, 2010). This control

technique is used when a single controller is applied

to manage two final control elements. The PID

controller strives to keep a temperature behaviour in

the cooling chamber manipulating simultaneously the

compressor speed and the expansion valve aperture.

In strong coupled systems, as VCRS, the split range

control could oscillate. The common practice is

configuring the sequencing of the final control

elements, in complementary, exclusive or progressive

modes, defining their range of operation, and

establishing a dead band between the two ranges. But,

here, the split range controller is complete free.

Besides this, the split range controller is tuned for a

multiobjective system, saving energy in the power

source and simultaneously fulfilling the desired

behaviour.

3.1 Multidynamics Algorithm for

Global Optimization (MAGO)

The MAGO (Hernández-Riveros and Ospina, 2010)

is an auto-organized EA that has only two parameters:

number of generations and population size. MAGO

uses statistical operators instead of genetic operators

and through the covariance matrix of the population

in each generation considers the relationships among

variables from the problem. MAGO is a real-value

EA that has shown its capacity solving engineering

problems (Hernández-Riveros Jesús-Antonio and

Cindy, 2018), (Balarezo-Gallardo and Hernández-

Riveros, 2017), (Bejarano et al., 2018). MAGO has

three different autonomous dynamics for evolving the

population, this way getting a larger exploration-

exploitation balance and less likelihood to

convergence to a local optimum are guaranteed.

In each generation, MAGO partitions the

population in three subgroups, each one with its own

evolutionary dynamics. To determine the number of

individuals for each dynamic, the actual population is

observed as in a normal distribution. The average of

the current generation, really a virtual individual, is

calculated on purpose. The number of elements

within one standard deviation of the actual population

conforms the cardinality of the Emergent Dynamics.

The cardinality of the Crowd Dynamics corresponds

to the difference between the first and second

deviation. The number of remaining elements is the

cardinality of the Accidental Dynamics. These

cardinalities change in each generation. Once the

number of individuals within each dynamic is

determined, MAGO proceeds to create individuals

who will make up the new population and so

continuing with the evaluation of new solutions.

From the fitness function evaluation of each

individual, the actual population is reorganized from

the best to the worst individual. The first N1

individuals within one standard deviation of the

actual population compose the Emergent Dynamics.

This N1 individuals obtaining the best values in their

objective function mutate applying the Nelder-Mead

method of numerical derivation, Equation 1.





()

=



()

+

()

×(



(



)

−



(



)

(1)

Where 



(



)

is the best individual of generation j

and 



()

is a randomly selected individual, usually the

same test individual. 

()

is a matrix that includes

information about the covariance of the problem

variables, Equation 2.



()



()

‖



()

‖

(2)

With 

()

is the sample covariance matrix of the

individual population in generation j.

Emergent Dynamics is improved elite making

faster convergence of the algorithm but keeping an

equilibrium between exploitation-exploration among

the best individuals.

The Crowd Dynamics keeps the memory of the

evolution process and is a sampling from a uniform

distribution determined by the upper and lower limits

of the second dispersion and the mean of the current

population, on the hyper-rectangle [LB

(j)

, UB

(j)

Equations3 and 4 are vectors with the diagonal of the

population dispersion matrix of the generation j,

described by Equation 5.

Evolutionary Split Range Controller for a Refrigeration System

343



()

=



()

−



(

()

)

(3)



()

=



()



(

()

)

(4)



(



)



(



)

=



(



)





(



)

… 



(



)





(5)

The Accidental Dynamics are samples from a

uniform distribution throughout the searching space,

similarly as in the initial population. It is smaller in

magnitude but has two basic functions: maintaining

the diversity of the population, and ensuring

numerical stability of the algorithm. Following is the

MAGO pseudo code:

1: j:= 0; Random initial population with a uniform

distribution.

2: Repeat

3: Evaluate each individual with the fitness function.

4: Calculate the population covariance matrix and the

first, second and third dispersion of the population.

5: Calculate cardinalities N1, N2 and N3 of the 3

dynamics.

6: Select the N1 best individuals, move toward the

best of all according to equation 1, make compete

with their parents, and choose the best of them to the

next generation j + 1.

7: Sample N2 individuals from a uniform distribution

in the hyper rectangle [LB(j), UB(j)], and pass to the

next generation j + 1.

8: Sample N3 individuals with a uniform distribution

over the entire search space. Pass to the next

generation j + 1.

9: j = j + 1

10: Until to satisfy a stopping criterion.

4 SPLIT RANGE CONTROL

STUDY CASES

Next, two cases of application of the MAGO

algorithm in different control structures for a VCRS

to improve its efficiency and control the temperature

inside the cold chamber are presented. The first study

case corresponds to the IFAC benchmark which only

represents the thermodynamic behaviour of the

VCRC. This case is presented to demonstrate the

effectiveness of the evolutionary control strategy

using the MAGO. The second case study corresponds

to a complete VCRS, whose entire model includes

multiple energy domains (electric, mechanical,

hydraulic and thermal). Three multivariable and

multi-objective control strategies are proposed.

Besides to the temperature control inside the cooling

chamber, an additional objective is added, that is, the

energy saving on the power source of the VCRS.

4.1 Case 1: Benchmark IFAC (Only

Thermal Model)

The model for the Benchmark (Bejarano et al., 2018)

is a refrigeration cycle of one compression stage and

one load. This model cannot be modified. The main

features of the model are:

1) Relatively low complexity, while

faithfully capturing the dynamics of the essential

plant and its non-linearities in a wide range of

operation.

2) Oriented to control because the

manipulated variables, the controlled variables and

the significant disturbances are shown.

3) The model is realistic since restrictions are

considered in the manipulated variables.

Figure 3: Discrete decentralized controllers included by

default in the refrigeration control benchmarking (Bejarano

et al., 2018).

Table 1: Benchmark control strategy.

Control Objectives

1. Reach a desire value of outlet temperature of

the evaporator secondary flux by manipulating

the opening of the expansion device.

2. Reach a desire value of superheat temperature

by manipulating the compressors speed.

Control Variables

Output variables

y1: outlet temperature of the evaporator

secondary flux (

,,

)

y2: superheat temperature (



)

Controlled variables

u1: Expansion device opening (





)

u2: Compressors speed ()

Elements

System: VCRC

Actuator 1: Expansion Valve.

Actuator 2: Compressor.

Mono Objective Optimization Problem

Minimization of the controlled variables error

ICINCO 2019 - 16th International Conference on Informatics in Control, Automation and Robotics

344

The multivariable control structure included by

default in the PID Benchmark 2018 consist of a

discrete decentralized control scheme (Figure 3) in

which the variables to control are the outlet

temperature of the evaporator secondary flux and the

degree of superheating (Table 1). The control system

is designed to obtain these two variables, tracking

their references as efficiently as possible, in the

presence of disturbances. The coefficient of

performance  is used as an indicator of steady-

state quality.

Tuning Procedure of the Discrete Controllers and

Optimization of the IFAC Benchmark’s VCRS.

As can be seen in Figure 4 the VCRS is composed of

two controllers; the first one is a discrete transfer

function that acts on the aperture of the expansion

valve. The second is a discrete PI controller that

corresponds to the compressor’s speed.

Figure 4: Coupled evolutionary tuning procedure for the

Benchmark IFAC 2018.

Table 2 shows the input data for the MAGO algorithm.

This mono-objective optimization approach seeks to

minimize the error between the reference and output signals

of the control variables.

Table 2: Input data for MAGO.

Data Values

Individuals 20

Generations 10

Upper bound [-1 0 1 -1.9 1 2.7 2.7]

Lower bound [-1.3 -0.6 0.7 -2 0.9 0.4 0.5]

Table 3 presents the obtained parameters for each

controller for the benchmarking applying MAGO.

Table 3: Parameters of each controller applying MAGO.

Expansion Valve

(Controller1)

Compressor

(Controller 2)

Benchmark

−1,0136 − 0,0626 0,9988

1 − 1,9853 0,9853

P: 0,4200

I: 0.9524

MAGO

1,1039 − 0,2901 0,8961

1 − 1,9185 0,9184

P: 1,2829

I: 1,6916

The results obtained for the temperature of the

secondary flux in the evaporator, Te sec out, and the

temperature of superheating, Tsh, are presented in

Figure 5 and Figure 6. The C1 controller corresponds

to the Benchmark and C2 to the results with MAGO.

The MAGO was implemented with a decentralized

MIMO control structure consisting of two discrete

controllers (a transfer function and a PI). The MAGO,

independent of the structure and domain of the

controller, finds the parameters in the function of

achieving both control objectives. As can be seen in

Figure 5 the optimal tuning method of controllers

applying the MAGO evolutionary strategy manages

to reach the reference values, achieving a behaviour

like the Benchmark's default strategy but improving

in the handling of temperature variation. The facility

to implement evolutionary tuning through the MAGO

algorithm in a complex system is highlighted, whose

model was not available to adapt it to the control

strategy that was to be applied.

Figure 5: Controlled variables.

Figure 6: Manipulated variables.

4.2 Case 2: Complete VCRS

(Multi-domain Energy based

Model)

The multi-domain energy-based model for a complete

VCRS is presented in Figure 9. The bond-graph (BG)

method is a graphical modelling approach in which

energy ports are connected by bonds that specify the

transfer of energy between system components.

Power, the rate of energy transport between

Evolutionary Split Range Controller for a Refrigeration System

345

components, is the universal currency of physical

systems (Gawthrop and Bevan, 2007). The main

advantage of the BG technique lies in its ability to

determine the energy consumption of a system, in

general, as well as its components, which is of vital

relevance for this second case study. Another

advantage is that the mono-domain models generated

from the BG have a modular coupling facility (unified

base) which can result in a coupled multi-domain

model.

Figure 7: BG base model of a VCRC (thermal behaviour).

Figure 8: Base model of a VCRS. Adapted from (Schné,

Jaskó and Simon, 2015).

The final model arises from the modular union of

the BG base model Figure 7 that corresponds only to

the thermal dynamics of the VCRC whose approach

was developed by (Schné, Jaskó and Simon, 2015)

Figure 8. The construction of the step-by-step model

is detailed in (Amador Soto, 2019).

From the base model described above

representing the dynamics of a VCRS, there is a need

to detail with greater precision the action of the

compressor (simplified in the base model by its

output temperature Tcomp). The reason for this is that

the impact of the compression action on the cooling

system is usually known, but it is not its direct energy

consumption seen from the power source. That is why

once having the BG representation of the thermal

cycle it is necessary to have also the representation

for the compressor and finally joint both models,

which does not imply a problem in the use of the BG

technique due to its modular nature (see Figure 9).

The system of coupled differential equations (6-11)

emerges from of the BG model. The state variables

are 



: Armature current, 



: Motor rotational

speed, 



: Hydraulic flow, 



: Temperature in

the condenser, 



: Temperature in the evaporator,





: Temperature inside the cold chamber.

4.2.1 On-off Control Baseline

To control the temperature inside the refrigerator

there is thermostat, whose sensor is connected to the

evaporator, Figure 10. The round knob inside a

refrigerator compartment can do the thermostat

setting. When the set temperature is reached inside

the refrigerator, the thermostat stops the electric

supply to the compressor and the compressor stops.

When the temperature falls below certain level it

restarts the supply to the compressor.

Figure 9: BG representation for the complete energy-based

model of the VCRS.







=−





∗



−



+



∗









(6)







=−





∗



−



∗



+



∗









(7)











∗



−



∗













− 



∗













(8)









=−













−



∗













+



















(9)













−









−







+













(10)







=−





−













−













(11)

ICINCO 2019 - 16th International Conference on Informatics in Control, Automation and Robotics

346

Figure 10: Conventional ON-OFF VCRS basic scheme.

While conventional on-off control commonly has

a fluctuating behaviour (Figure 11), other control

options have arisen due to the growth of the

electronics field. Variable speed compressors and

electronic expansion valves have gradually replaced

older single speed compressors and thermostatic

expansion valves, respectively. Such new

components allow the development of smarter control

strategies, not only to save energy but also to reduce

fluctuations in the controlled variables and therefore

achieve a more accurate control (Bejarano et al.,

2018). In this sense, in this document continuous

controllers are implemented for temperature control

and simultaneously reduce energy consumption.

Figure 11: Typical thermal behaviour for a conventional

VCRS by ON-OFF control.

4.2.2 Continuous Control Strategies

A multivariable SISO and MISO control structures

are presented, both consisting on a continuous

centralized control scheme in which the variable to

control is the outlet temperature of the cold chamber

while at the same time reducing the energy

consumption from the power source.

Tuning Procedure of the Energy-based Complete

VCRS Model.

Three different control strategies are applied.

• First Control Strategy:

Centralized control SISO (manipulated Variable:

aperture of the expansion valve). See Table 5, Table

6, Figure 12, Figure 13, and Figure 14.

• Second Control Strategy:

Centralized control SISO (manipulated Variable:

compressor speed). See Table 7, Table 8, Figure 15,

Figure 16, and Figure 17.

• Third Control Strategy:

Centralized control MISO (manipulated Variables:

compressor speed + aperture of the expansion valve).

See Table 9, Table 10, Figure 18, Figure 19, Figure

20.

In all cases the MAGO runs with the same set of

parameters of the algorithm, Table 4. The operation

of the actuator opening the expansion valve goes from

0% to 100%, and the compressor speed goes from 0

to 1200 RPM. For the evolutionary tuning of the split

range controller, there is not sequencing, restrictions

on the range nor dead band among the two actuators.

Table 4: Input data for MAGO.

Data Values

Individuals 50

Generations 25

Upper bound [10 10 10]

Lower bound [-10 -10 -10]

Table 5: First, control strategy (by expansion valve

aperture).

Control objective

A set of stable temperature inside the cold chamber

(2, 1, 0 -1 ℃)

Energy objetive

Reduce the energy consumption of the power

source of the system.

Control variables (SISO) Elements

Output variable (y):

Temperature inside the cold

chamber (



)

Controlled variable (u):

Expansion device opening

(





)

System:

CompleteVCRC

Actuator:

Expansion Valve.

Multi objective optimization problem

1. Minimization of the controlled variable error

2. Minimization power source energy

Evolutionary Split Range Controller for a Refrigeration System

347

Table 6: Obtained controller parameters applying MAGO.

Expansion Valve (PID

Controller)

Kp: 0.0959

Ti: 1.8297

Td: 4.3237

Figure 12: VCRC intervention applying the first control

strategy for temperature control and power source’s energy

reduction.

Figure 13: Controlled variable for the first control strategy.

Figure 14: Manipulated variable for the first control

strategy.

Table 7: Second control strategy (by compressor speed).

Control Objective

A set of stable temperature inside the cold chamber

(2, 1, 0 -1 ℃).

Energy Objective

Reduce the energy consumption on the power

source of the system.

Control Variables (SISO) Elements

Output variable (y):

Temperature inside the cold

chamber (



)

Controlled variable (u):

Compressors speed (



)

System:

CompleteVCRC

Actuador:

Compressor.

Multi Objective Optimization Problem

1. Minimization of the controlled variable error

2. Minimization power source energy

Table 8: Obtained controller parameters applying MAGO.

Compressor

(PID Controller)

Kp: 0.0965

Ti: 2.993

Td: 1.5037

Figure 15: VCRC intervention with the second control

strategy for temperature control and power source’s energy

reduction.

Figure 16: Controlled variable for the second control

strategy.

ICINCO 2019 - 16th International Conference on Informatics in Control, Automation and Robotics

348

Figure 17: Manipulated variable for the second control

strategy.

Finally, the energy consumption corresponding to

each control strategy applied to the system is

presented in Table 11.

Table 9: Third control strategy (by expansion device

aperture + compressor speed).

Control Objective

A set of stable temperature inside the cold chamber

(2, 1, 0 -1 ℃).

Energy Objective

Reduce the energy consumption of the power

source of the system.

Control Variables (MISO) Elements

Output variable (y):

Temperature inside the cold

chamber (



)

Controlled variable (u1):

Expansion device opening

(



)

Controlled variable (u2):

Compressors speed (



)

System:

CompleteVCRC

Actuador1:

Expansion device

Actuador2:

Compressor.

Multi Objective Optimization Problem

1. Minimization of the controlled variable error

2. Minimization Power Source energy

Table 10: Obtained controller parameters applying MAGO.

Expansion Valve

(PID Controller 1)

Compressor

(PID Controller 2)

Kp: 0.7096

Ti: 4.295

Td: -1.7924

Figure 18: VCRC intervention with the third control

strategy for temperature control and power source’s energy

reduction.

Table 11: Summary energy consumption of control

strategies applying MAGO.

Figure 19: Controlled variable for the third control strategy.

Figure 20: Manipulated variables for the third control

strategy.

Comparing results between the three proposed

control strategies, it was found that, by controlling the

speed of the compressor, applying a voltage profile, it

is possible to obtain better results in reducing energy

consumption. On the other hand, manipulating the

expansion device improves the behaviour of the

controlled variable (temperature), reaching the

reference values in less time. However, changing the

Evolutionary Split Range Controller for a Refrigeration System

349

opening value of the expansion device alters the

demands on the compressor, increasing the motor

current while applying a fixed voltage. This is

reflected in an increase in energy consumption. For

its part, the third strategy combines the benefits of

both previous strategies, so it is possible to achieve

good results in reducing energy consumption and

good behaviour of the controlled variable, quickly

reaching the reference values without increasing the

demand of the compressor.

4 CONCLUSIONS

The proposed evolutionary control approach was

applied for two different developments in VCRS

(case study 1 - only the thermal part, and case study 2

- including the electrical, rotational, hydraulic and

thermal parts) under conditions of multivariability,

high coupling, non-linearity and restrictions, among

others. The results obtained for both study cases show

that intervention inside the refrigeration system by

means of applying control structures can achieved

energy savings for the thermal circuit and for the

whole system. The MAGO algorithm achieves

remarkable results for the different control strategies,

independently of both the structure and the domain of

the controller to be tuned.

For the first study case, we use a predefined model

formulated for control purpose (by transfer function)

that tries to reach the temperature behaviour

improving indirectly the energy performance of the

system. On the other hand, case study 2 illustrates the

use of a single unified energy based model (by

differential equations of the whole system) to reduce

directly the source's energy consumption and at the

same time achieving the desired temperature

behaviour for the system using three different control

strategies. Evolutionary tuning was applied to the two

different systems without additional procedures. The

split range controller was expanded to multivariable

and after to multiobjective purposes. This

evolutionary control method can be implemented

without any inconvenience in developments for

control of cooling systems of multiple loads and

stages.

Savings and control opportunities were identified

according to the strategy (MISO or MIMO). The

more variables in the process are controlled, the

greater energy savings are obtained.

Future work is to combine the advantages of the

manipulation of the compressor speed and the

opening of the expansion valve with two independent

but coupled controllers. A greater range of solutions

is foreseen to improve the savings in energy

consumption while satisfying the expected cooling

demand.

REFERENCES

Amador Soto, G. J. (2019) Control de Procesos

Industriales con Minimización de Consumo Energetico.

Universidad Nacional de Colombia. Available at:

http://bdigital.unal.edu.co/71599/.

Balarezo-Gallardo, S.-F. and Hernández-Riveros, J. A.

(2017) “Evolutionary Parameter Estimation of Coupled

Non-linear Oscillators,” in: Solano, A. and Ordoñez, H.

(eds.) Advances in Computing. Cham: Springer

International Publishing, pp. 457–471.

Bejarano, G. (2017) Optimization and multivariable control

of refrigeration systems. Universidad de Sevilla.

Available at: https://idus.us.es/xmlui/handle/11441/

63847 (Accessed: February 12, 2019).

Bejarano, G. et al. (2018) “BENCHMARK PID 2018

Benchmark for PID control of refrigeration systems

based on vapour compression,” pp. 1–20. Available at:

http://servidor.dia.uned.es/~fmorilla/benchmarkPID20

18/.

Corte, E. et al. (2014) “Sistemas de refrigeración doméstica

- estado del arte de las mejoras en la eficiencia

energética,” Revista de la Facultad de Ciencias

Químicas de la Universidad de Cuenca Ecuador, 9, pp.

19–40. Available at: https://publicaciones.ucuenca.

edu.ec/ojs/ index.php/quimica/article/view/253.

Coulomb, D., Dupon, J. L. and Pichard, A. (2015) “The

Role of Refrigeration in the Global Economy,” The

29th Informatory Note on Refrigeration Technologies,

29(November, 2015), p. 16.

Fleming, P. and Purshouse, R. (2002) “Evolutionary

algorithms in control systems engineering: a survey,”

Control Engineering Practice, 10(11), pp. 1223–1241.

doi: 10.1016/S0967-0661(02)00081-3.

Gawthrop, P. J. and Bevan, G. P. (2007) “Bond-Graph

Modeling: A tutorial introduction for control

engineers,” IEEE Control Systems, 27(2), pp. 24–45.

doi: 10.1109/MCS.2007.338279.

Hernández-Riveros, J. and Ospina, J. D. (2010) “A multi

dynamics algorithm for global optimization,”

Mathematical and Computer Modelling, 52(7–8), pp.

1271–1278. doi: 10.1016/j.mcm.2010.03.024.

Hernández-Riveros Jesús-Antonio, Carmona Cindy, Urrea

Q. Jorge. (2018) “Evolutionary Control Systems,” (in:

Incorporating Nature-Inspired Paradigms). doi:

10.4018/978-1-5225-5020-4.

Jahangeer, K. A., Tay, A. A. O. and Raisul Islam, M. (2011)

“Numerical investigation of transfer coefficients of an

evaporatively-cooled condenser,” Applied Thermal

Engineering. Elsevier Ltd, 31(10), pp. 1655–1663. doi:

10.1016/j.applthermaleng.2011.02.007.

Khan, K. H., Ryan, C. and Abebe, E. (2017) “Optimizing

HVAC Energy Usage in Industrial Processes by

Scheduling Based on Weather Data,” IEEE Access, 5,

ICINCO 2019 - 16th International Conference on Informatics in Control, Automation and Robotics

350

pp. 11228–11235. doi: 10.1109/ACCESS.2017.2715

239.

Schné, T., Jaskó, S. and Simon, G. (2015) “Dynamic

models of a home refrigerator,” MACRo 2015- 5th

International Conference on Recent Achievements in

Mechatronics, Automation, Computer Science and

Robotics, pp. 105–114. doi: 10.1515/macro-2015-0010.

Smith, C. (2010) Advanced Process Control: Beyond Single

Loop Control. Wiley.

Evolutionary Split Range Controller for a Refrigeration System

351