Searching the Optimal Combination of Fire Risks Reducing
Measures at Oil and Gas Processing Facilities with the use of Genetic
Algorithm
Sergey Gudin
1
, Renat Khabibulin
2
and Denis Shikhalev
2
1
Ghent University, 9000, Ghent, Belgium
2
The State Fire Academy of EMERCOM of Russia, 129366, B. Galushkina 4, Moscow, Russia
Keywords: Genetic Algorithm, Oil and Gas Processing Facilities, Quantitative Risk Assessment, Optimisation.
Abstract: The search for the combination of fire risk-reducing measures at oil and gas processing facilities is a
complicated task. There may be a large number of measures to reduce fire risks which need to be optimized,
both technically and economically. The analysis of the existing programs for risk assessment has been
conducted. The structure of database with the values of risk-reducing measures has been worked out. To
reduce the time required for this task, a genetic algorithm approach has been proposed.
1 INTRODUCTION
At present, there are a lot of quantitative risks
assessment systems, which can qualitatively
determine explosion and fire dangerous factors in the
territory of oil and gas processing facilities. As a rule,
after risks assessment procedures, risk values are
inappropriate. In these cases, some measures for
reducing risk values are required. There can be a lot
of measures for reducing risk values (installation of
alarm system, automatic fire extinguishing system, a
decrease in stored material, etc.). In most cases, one
measure is not enough. It is necessary to find a set of
measures that maximally reduce fire risk values and
do not require a lot of expenses. Fire risks values may
be different in each case of a combination of risk
values reducing measures. Each situation requires
risk assessment procedures, but risk assessment
procedure requires a lot of operations and time. So,
the number of the procedures of risks assessment will
grow in geometric progression with the amount of
objects on the territory (figure 1). Special algorithm
for optimization of combinations of measures for
reducing risk values has been developed. In the paper
presented, the risk acceptance criteria approach when
using genetic algorithms for searching optimal
combination of fire risk-reducing measures at oil and
gas processing facilities.
Figure 1: Number of combinations of measures to ensure fire safety.
1
100000000
1E+16
1E+24
1E+32
1E+40
1E+48
1E+56
12345678910
Number of combinations
of measures to ensure fire
safety
Number of measures to ensure fire safety
1 object on the territory 5 objects on the territory
10 objects on the territory 20 objects on the territory
Gudin S., Khabibulin R. and Shikhalev D.
Searching the Optimal Combination of Fire Risks Reducing Measures at Oil and Gas Processing Facilities with the use of Genetic Algorithm.
DOI: 10.5220/0006188904890496
In Proceedings of the 9th International Conference on Agents and Artificial Intelligence (ICAART 2017), pages 489-496
ISBN: 978-989-758-220-2
Copyright
c
2017 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
489
2 RELATED WORKS
Risk management has become a vital topic both in
academia and practice during the past several
decades(Desheng, Shu-Heng, David, 2014). When
evaluating safety of projects, it is common to use risk
acceptance criteria to support decision-making
(Abrahamsen and Aven, 2008). However, the
analysis of the existing fire risk estimation software
(Gudin and Khabibulin, 2015) showed that the
systems of fire risks assessment do not contain special
methods and algorithms, which make it possible to
find a combination of measures for reducing fire risk
values.
One of the existing methods for solving
optimization problems is genetic algorithms used to
solve a lot of decision-making problems with great
amount of information (Xuancai et al., 2016). Many
scientists came to the conclusion that genetic
algorithms can be used in solving complex tasks
(Martorell et al., 2005; Ramirez et al., 2009). Besides,
their effectiveness as to the optimization were
highlighted in many studies (Panov and Shary, 2011;
Schaefer, 2012; Sergienko, 2009).
One of the methods of using genetic algorithms
has been already presented (Caputo et al., 2011). This
method is used in searching for the economic
optimum risk level. It is based on the minimization of
total expenses. After the experiments, the authors
show that he can solve that problem quite well.
Reduction in expenses is an important task, but the
first task is to provide fire safety, because it directly
influences the safety plant workers and people living
in residential areas.
Upon the review of the related works, it was
identified that genetic algorithm can solve the
problem of searching for the best combination of
measures for reducing fire risk values with risk
acceptance criteria.
3 GENETIC ALGORITHM FOR
FINDING COMBINATIONS IN
ORDER TO REDUCE RISK
VALUES
For the purposes of analysis of most of the events
stored in the database, comprehensive assessment of
fire risks should be conducted when analyzing each
event, except for specific associated cases. For
example, with the time that people spend in the
building, when the values of individual fire risk
would change only by factoring the probability of the
people being there. In such cases, comprehensive risk
estimation is not required, and such exceptions should
be included into the program code separately.
In general, the genetic algorithm model proposed
by John Holland (Holland, 1975) was used. For
crossing of chromosomes a method with a single
point of exchange was used. Mutation procedure was
slightly modified for searching the combination of
fire risk-reducing measures.
First of all, after crossing, there may be identical
gens in the chromosome. It is necessary to have only
unique gens in the chromosome. So, all identical gens
are going to mutate, except for one.
To add more, a special method was invented to
form the first population. It consists of 2 steps:
1) Evaluating the effectiveness of measures;
2) Generation chromosomes by using the roulette
wheel method.
To create combinations with different number of
measures added the operation for accidentally
deleting of chromosome.
3.1 Objective Function
One of the obligatory criteria of the genetic algorithm
use is an objective function presenting quantification
of the efficiency of computed solutions. The
suggested objective function consists of three
parameters:
1. The amount of fire risks calculated values
in the territory of an enterprise, which does not
exceed acceptable values (Q).
2. The parameter of average deviation of
infeasible calculated values of fire risks in the
territory of oil and gas facility and the adjacent
residential zone from the acceptable values (D).
3. Reduced cost of measures (P).
The following formula is used for the calculation
of the amount of acceptable fire risks values in the
territory of oil and gas facility:
=
+
(
)
+ ,
(1)
Where:
=
1, ≤
0, >
;
(2)
(
)
=
1, ≤
0, >
;
(3)
ICAART 2017 - 9th International Conference on Agents and Artificial Intelligence
490
(
)
=
1, ≤
0, >
;
(4)
R – Final value of an individual fire risk for the
workers of the enterprise;
J – Amount of workers on the enterprise;
I – Final value of individual fire risk for people
located in housing, social, and business or recreation
areas;
M – Amount of facilities with people in residential
zone;
S – Final value of social fire risks quantity for people
located in housing, social, and business or recreation
areas;
R
a
– Acceptable value of individual fire risk quantity
for the workers of enterprise;
I
a
– Acceptable value of individual fire risk quantity
for people located in housing, social, and business or
recreation areas;
S
a
– Final value of social fire risk quantity for people
located in housing, social, and business or recreation
areas;
α – Acceptable criterion of value as to individual fire
risk quantity for the workers of the enterprise;
β – Acceptable criterion of value of individual fire
risk quantity for people located in housing, social, and
business or recreation areas;
γ – Acceptable criterion of value of social fire risk
quantity for people located in housing, social, and
business or recreation areas;
Q – Amount of fire risks quantity acceptable within
the framework of the given case.
Total costs of procedures implementation are
calculated according to the following formula:
=
⋅
+С
(5)
where P
i
– total costs of i-th procedure, eur/year;
K
i
– capital cost for the purposes of the procedure
implementation, eur/year;
С
ei
– exploitation costs of i-th procedure.
Adduction of this set cost parameters to the
current period is performed by multiplying them by
the coefficient of the relative cost effectiveness of
additional capital investments. (E
n
).
Parameter (D) reflects average deviation of
infeasible calculated fire risks quantities at the
protected facility and the adjacent residential area
from feasible values. The current parameter takes on
the value from 0 to 1, and it is used in cases, where
not of all the values of fire risks are feasible, and also
used in order to define points in the territory of the
facility, where all values will be most designated to
feasible values:
=
++
++
,
(6)
Where
=
∑
/
;
(7)
=
∑
/
;
(8)
=
,>
0,≤
;
(9)
А – Dimensionless parameter of the average
infeasible individual fire risks quantities in the
territory of the facility (less than R
a
) from feasible
values (R
a
);
R
z
– Value of the quantity of infeasible individual
risks in the territory of gas distributing plant;
Z – Amount of infeasible values of individual fire
risks quantity in the territory of the enterprise;
B – Dimensionless parameter of the average
infeasible values of the individual fire risks in the
territory of the facility (lesser I
a
) from feasible values
(I
a
);
I
y
– Infeasible values of the individual fire risks in
residential area;
Y – Amount of infeasible values of individual risks
quantity in the residential area;
C – Dimensionless parameter of deviation of social
risk in the residential area from feasible value.
The current criteria were correlated in a unified
object function. Because of the primary objective
within the set of procedures on the optimization of
fire risks control is safety, the highest priority goes to
parameter K, the next criterion according to priority
is economic component (P). Against the backdrop of
couple combinations of procedure parameters K and
P are equal, and the required values of fire risks are
non-subnormal, Q becomes the key parameter that
provides rectangular distribution of the risk zones in
the territory of gas and oil facilities. Therefore, the
object function in the system of fire risks
management is presented in the following way:
=
(
max
(
)
,min
(
P
)
,max
(
))
(10)
3.2 Generation of the First Population
The logical sequence of actions as to the evaluation
of the effectiveness of risk value-reducing measures
can be represented as follows:
1. Choosing a fire risk-reducing value
measure from the database.
Searching the Optimal Combination of Fire Risks Reducing Measures at Oil and Gas Processing Facilities with the use of Genetic Algorithm
491
2. Analysis of the application with regard to
the objects in the territory.
3. Calculation of fire risks values for each
suitable object with the selected measure.
4. Save results to the specified massive.
5. If all measures are considered, go to the
end, or go to step 1.
The first population is generated by using the
approach that called the “roulette wheel” (Gen and
Cheng, 1997). The value of the effectiveness of each
measure is expressed by fitness function ((
)).
The next step after the assessment of the effectiveness
of measures is to calculate the overall function of all
measures:
= (
)−
(
)
(11)
3. Calculate the probability of selection (
) for each
measures
:
=
(
)
−min
(
)
(12)
k =1,2,…,number_of_measures
4. Calculate the total probability
for each measure
(
):
=
(13)
Each measure is one gen in the chromosome, so
chromosome consists of many measures.
=(
[
1
]
,
[
2
]
,
[
3
]
,…[])
(14)
where M1, M2, M3, Mk – measures for reducing
fire risks values.
A chromosome consists of k measures. The
selection process begins with rotation of a wheel for
k times; each time, one chromosome’s gen is selected
by the following algorithm:
1. Generate a random number r from the interval
[0, 1].
2. If ≤
, then select the first measure
;
otherwise, go to the k-th measure (2≤ ≤
count_of_mesuares) such as
≤ ≤
.
3. If the selected chromosome has already been
chosen, go to step 1 or choose this measure.
The above procedures are to be repeated while
counting the population not equal to the required
number.
3.3 Fitness Function
Evaluation function matching of chromosomes is
performed in two steps:
1. Implementation of all measures contained in the
chromosome.
2. Calculate the parameters of the goal function.
Fitness function plays the role of environment and
evaluates chromosomes according to their capability
to perform optimization criterion.
3.4 Selection
For the selection, the roulette wheel approach was
chosen. The roulette wheel can be constructed as
follows:
1. Calculate the value of the function of
compliance (
) for each chromosome
.
2. Calculate the overall function of the population
concerned:
= (
) −
−
=1,
(15)
k =1,2,…,pop_size
3. Calculate the probability of selection (
) for each
chromosome
:
=
(
)−
=1,
(16)
k =1,2,…,pop_size
4. Calculate the total probability (
) for each
chromosome (
):
=
,=1,2,…,_
(17)
The selection process begins with the rotation of
a wheel pop_size times; each time one chromosome
is selected by the following algorithm:
1. Generate a random number r from the interval
[0, 1].
2. If ≤
, then select the first chromosome
;
otherwise choose the k-th chromosome (2≤ ≤
_) such that
≤ ≤
.
3.5 Crossing and Mutation
For crossing of chromosomes, a method with a single
ICAART 2017 - 9th International Conference on Agents and Artificial Intelligence
492
point of exchange is used. In accordance with this
method, one point of exchange is randomly selected,
with respect to which parts of chromosomes are
swapped-parents. For this purpose, it generates the
integer in the interval [1, count_of_chromosomes],
which is the point of gene exchange.
Mutation consists of the change in one or more
genes with mutation probability equal ratio. If we
suggest different number of measures in
combinations, there is a little probability of removing
one gen from a chromosome.
4 ANALYSIS
The analysis of the received model was performed on
the basis of estimating the fire risk of a standard
facility in gas and oil industry (gas distribution plant),
in the territory of which propane-butane fraction was
major circulated substance.
In order to create the model of enterprise
optimization combination according to the reduction
of fire risks calculated values on the territory of the
current oil and gas enterprise, the list of suggested
procedures to reduce fire risks with provisional
capital and exploitation total costs from base value
has been formed (table 1).
Efficiency analysis as to the model has been
conducted in several stages, with the use of the
presented object function. During the first stage, Q
and D parameters of the object function, along with
the use of each procedure separately, were estimated
(table 2).
On the next stage the search of procedures
combinations with the help of suggested model was
conducted in order to analyze and select the best
parameters. As search stopping criterion was chosen
situation when combination includes only one
procedure. After the range of computing experiments
the following optimal parameters where defined:
– crossing percentage: 80;
– mutation percentage: 30;
– possibility of gene deletion from the
chromosome (procedures from the set): 75.
In order to create high variability of procedures
combination high crossing (80%) and mutation (30%)
percentage was chosen (figure 2, 3). After the range
of experiments it was discovered that possibility of
gene deletion from the chromosome (procedures from
the set) significantly influence the time of selection
and in this case the quality of the selected
combinations do not change up to the particular
moment (figure 4). Therefore the possibility of gene
deletion from the chromosome was defined equal to
75%.
Mutation of 30% of specimens was chosen,
because in this value, the variability of the suggested
combination of the procedures significantly increases
and causes the increase in the quality of the algorithm
work results. The use of value under 30% causes the
decrease in the observed combinations variability and
the quality of results. According to the increase in the
variability, the decrease in the algorithm quality is
also observed (figure 3).
Table 1: Procedures of value of fire risks quantity reduction
in the territory of gas distribution plant.
№ Procedure of
reduction of value of
quantity of fire risk
K, run
С
ei
,
eur/year
P,
eur/year
1 Reduce filling
degree on 15 %.
0 X X
2 Reduce the
probability of
unstable unit
presence by 20%
0 0.3X 0.3X
3 Install automatic fire
alarm unit
0.3X 0.1X 0.16X
4 Install automatic fire
extinguishing unit or
water spray unit
under the control by
independent
organization
(irrespective of the
type of the fire
extinguishing unit)
X 0.3X 0.5X
5 Install automated
automatic fire
extinguishing unit
(water or foam) or
waterspray unit
without control of
performance
capability by
independent
organization
0.6X 0.15X 0.27X
6 Install other types of
automatic fire
extinguishing unit
without the control
of performance
capability by
independent
organization
0.5X 0.1X 0.2X
7 Install flanging 30
m
2
0.15X 0.01X 0.04X
Searching the Optimal Combination of Fire Risks Reducing Measures at Oil and Gas Processing Facilities with the use of Genetic Algorithm
493
Table 2: List of possible procedures and parameters of object function using Q and D parameters.
№ Procedure Object Q D
1
Reduce possibility of the object’s presence by 20 %
Road tank 16 781561535748
2 Railway tank 16 779786139413
3
Reduce filling degree by 15 %.
Road tank 16 779307110031
4 Railway tank 16 779307110031
5 Separator 16 779307110031
6 Tank 100 m
3
16 779307110031
7 Tank 50 m
3
16 779307110031
8 Tank 100 м
3
(group 2) 16 779307110031
9
Install automated automatic fire extinguishing unit (water or
foam) or water spray unit without the control of performance
capability by independent organization
Road tank 16 779307110031
10 Railway tank 16 779307110031
11 Separator 16 779307110031
12 Tank 100 m
3
16 779307110031
13 Tank 50 m
3
16 779307110031
14 Tank 100 m
3
(group 2) 16 779307110031
15
Install automatic fire alarm unit
Road tank 16 788404842408
16 Railway tank 16 781226809106
17 Separator 16 790786801193
18 Tank100 m
3
17 870910212155
19 Tank 50 m
3
17 953271699395
20 Tank 100 m
3
(group 2) 17 974221794362
21
Install automatic fire extinguishing unit or water spray unit under
the control by independent organization (irrespective of the type of fire
extinguishing unit)
Road tank 16 779307110031
22 Railway tank 16 779307110031
23 Separator 16 779307110031
24 Tank100 m
3
16 779307110031
25 Tank 50 m
3
16 779307110031
26 Tank 100 m
3
(group 2) 16 779307110031
27
Install other types of automatic fire extinguishing unit without the
control of performance capability by independent organization
Road tank 16 779307110031
28 Railway tank 16 779307110031
29 Separator 16 779307110031
30 Tank100 m
3
16 779307110031
31 Tank 50 m
3
16 779307110031
32 Tank 100 m
3
(group 2) 16 779307110031
33
Install flanging 30 m
2
Road tank 16 779307110031
34 Railway tank 16 779307110031
35 Separator 16 779307110031
Figure 2: Correspondence of Q parameter average value with mutation possibility.
15,50
16,00
16,50
17,00
17,50
18,00
18,50
1 102030405060708090100
Average value of Q parameter
Possibility of mutation, %
The maximum value
of the goal function
ICAART 2017 - 9th International Conference on Agents and Artificial Intelligence
494
Figure 3: Correspondence of Q parameter average value with crossover possibility.
Figure 4: Correspondence of procedures combinations and number of feasible risks on the territory of the enterprise (Q)
selection with possibility of their chromosome gene deletion.
After selecting the procedures as to the reduction
of fire risks of the calculated values with the use of
the suggested model, the list of possible combinations
was defined. A table was formed with the lists of the
most efficient combinations with various quantity of
procedures (table 3).
Procedures selection time amounted a little bit
more than 21 min. The suggested model was always
finding procedures combination with high value of
objective function, though during the process of
combination selection options are possible with
greater amount of procedures, but with lower value of
objective function. For example: when procedures
No. 3, 4, 5, 6, 1, 16, 21, 24, 10, 11 were combined,
then the parameters of the objective function were
equal to: Q=16, D=0,78, P=3,2X. Despite the fact that
in table 4 with various quantity of procedures only
one option of a procedure set is presented, the
program can output multiple alternate options of
combinations with high value of objective function
according to the required quantity of procedures in
one combination.
Table 3: Rating of procedures combinations as to the
decrease in fire risks calculation values at various quantities
of solutions in them.
Quantity
of
solutions
Numbers of solutions
according to table 3
Q D P
10
7, 8, 1, 2, 15, 18, 19,
26, 27, 33
18 0 2,42X
9
7, 8, 1, 2, 15, 18, 19,
26, 27
18 0 2,38X
8
5, 1, 15, 17, 18, 19, 22,
24
18 0 2,24X
7 7, 1, 2, 15, 18, 19, 26 18 0 1,88X
6 5, 1, 15, 18, 19, 22 18 0 1,58X
5 5, 16, 18, 20, 10 18 0 1,05X
4 1, 15, 16, 19 18 0 1,03X
3 5, 15, 16 18 0 0,62X
2 7, 15 17 0,98 0,46X
15
16
17
18
19
0 102030405060708090100
Average value of Q
parameter
Possibility of crossover, %
15
15,5
16
16,5
17
17,5
18
18,5
0
10
20
30
40
50
60
70
25 50 75 85 95
Value of Q and D parameters
Time of procedure
selection, min
Possibility of gene deletion, %
Time of procedure selection, min Value of objective funtion
Maximum variability with
the maximum value of the goal
function
The maximum value of the
goal function with the
minimum time
Searching the Optimal Combination of Fire Risks Reducing Measures at Oil and Gas Processing Facilities with the use of Genetic Algorithm
495
5 CONCLUSIONS AND FUTURE
WORK
The model of procedures for the management of fire
risks at oil and gas facilities was presented with the
use of genetic algorithms with modifications for
solving the assigned task:
1. Instead of the use of binary row, the
chromosome is used, the genes of which serve as
identifier of the procedures.
2. The primary population is generated
according to a specific algorithm.
3. In order to create the set from various quantity
of procedures, a changed mutation operation is used
consisting of random deletion of one of the
chromosome-s gene.
The efficiency of the model obtained was tested
in information system “FireRisks”. As a result, it was
concluded that one of the main advantages of the
suggested approach is the significant decrease in
calculation operations, which, in turn, solves the issue
of optimizing fire risks management procedures at the
facilities with the use of modern information systems.
The offered model also possesses high variability
of suggested variants, except for the significant
reduction in the required time for conducting the
variable combination of procedures on fire risk
calculated values decrease.
At present, unification is conducted of the created
models into a single system of intellectual support of
decision-making in the field of fire risks management
at oil and gas complex facilities.
The way forward is to create algorithms using
CMA Evolution Strategy, Differential evolution and
Simulated Annealing to compare the effectiveness of
the obtained models in the management of fire risks
in the oil refining facilities.
REFERENCES
Desheng Dash Wu, Shu-Heng Chen, David L. Olson,
Business intelligence in risk management: Some recent
progresses, Information Sciences, Volume 256, 20
January 2014, Pages 1-7, ISSN 0020-0255,
http://dx.doi.org/10.1016/j.ins.2013.10.008.
Abrahamsen, E. B., & Aven, T. (2008). On the compliance
of risk acceptance criteria with normative theories for
decision-making. Reliability Engineering & System
Safety, 93(12), 1906-1910. doi:http://dx.doi.org/
10.1016/j.ress.2008.03.021.
Gudin, S. V., Khabibulin, R. Sh., Rubtsov, D. N. Problems
of decision-making in fire risks management in the
territories of oil processing facilities using modern
software products // Fire and Explosion Safety. —
2015. — № 6 (62). — pp. 40-45. DOI:
10.18322/PVB.2015.24.12.40-45.
Xuancai Zhao, Qiuzhen Lin, Jianyong Chen, Xiaomin
Wang, Jianping Yu, Zhong Ming, Optimizing security
and quality of service in a Real-time database system
using Multi-objective genetic algorithm, Expert
Systems with Applications, Volume 64, 1 December
2016, Pages 11-23, ISSN 0957-4174,
http://dx.doi.org/10.1016/j.eswa.2016.07.023.
Martorell, S., Villanueva, J. F., Carlos, S., Nebot, Y.,
Sánchez, A., Pitarch, J. L., & Serradell, V. (2005).
RAMS+ C informed decision-making with the
application to multi-objective optimization of technical
specifications and maintenance using genetic
algorithms. Reliability Engineering & System Safety,
87(1), 65-75.
Ramirez, A. J., Knoester, D. B., Cheng, B. H., & McKinley,
P. K. (2009). Applying genetic algorithms to decision-
making in autonomous computing systems. The paper
presented during the Proceedings of the 6th
international conference on Autonomous computing.
Panov N.V., Shary S.P. (2011). Interval evolutionary
algorithm search the global optimum - News of Altai
State University, №. 12.
Schaefer I.A. (2012). Investigation on the efficiency of
genetic algorithm constrained optimization - Youth and
Science: Proceedings of the VIII All-Russian scientific
and technical conference of students, graduate students
and young scientists dedicated.
Sergienko R. B. (2009) Efficacy co-evolutionary genetic
algorithm constrained optimization. Herald of Siberian
State Aerospace University. Academician M.F.
Reshetnev № 3.
Caputo, A. C., Pelagagge, P. M., & Palumbo, M. (2011).
Economic optimization of industrial safety measures
using genetic algorithms. Journal of Loss Prevention in
the Process Industries, 24(5), 541-551.
Holland J.N (1975). Adaptation in Natural and Artificial
Systems. Ann Arbor, Michigan: Univ. Michigan Press.
Gen, M., & Cheng, R. (2000). Genetic algorithms and
engineering optimization (Vol. 7). John Wiley & Sons.
ICAART 2017 - 9th International Conference on Agents and Artificial Intelligence
496
