See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/260549157

An Improved Genetic Algorithm for Resource Constrained Project Scheduling

Problem

Article in International Journal of Computer Applications · September 2013

DOI: 10.5120/13520-1302

CITATIONS READS

10 527

3 authors:

s. Diana Ganapathy L.
NITIE-National Institute of Industrial Engineering NITIE-National Institute of Industrial Engineering
2 PUBLICATIONS 10 CITATIONS 113 PUBLICATIONS 315 CITATIONS

SEE PROFILE SEE PROFILE

Ashok Pundir
NITIE-National Institute of Industrial Engineering
135 PUBLICATIONS 321 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Economic Corridors: Collaborating & Converging with the Global Opportunities View project

Journal of Industrial Engineering International View project

All content following this page was uploaded by s. Diana on 02 April 2018.

The user has requested enhancement of the downloaded file.

 International Journal of Computer Applications (0975 – 8887)
Volume 78 – No.9, September 2013

An Improved Genetic Algorithm for Resource

Constrained Project Scheduling Problem

S Diana L Ganapathy Ashok K Pundir

Research Scholar Professor, Professor & Dean (Student
National Institute of Industrial National Institute of Industrial Affairs & Placement)
Engineering, Mumbai Engineering, Mumbai National Institute of Industrial
Engineering, Mumbai

ABSTRACT details of the proposed Genetic Algorithm Section 5 presents

Project scheduling with limited resources is a challenging the computational results. Summary and Conclusion is
management problem that is of immense importance to both presented in Section 6.
practitioners and researchers. This problem is known to be
NP-hard even under the simplifying assumptions of single
2. PROBLEM FORMULATION
renewable resource constraint, its constant availability over The RCPSP consists of n+2 activities J = (0, 1, 2, 3…n+1)
time and minimization of makespan as objective. This paper where j = 0 and j = n+1 are dummy activities which represent
presents an improved Genetic Algorithm (GA) based the project start and project end respectively. Renewable
approach for the single mode resource constrained project resources are of several types k where k = 1, 2… K and each
scheduling problem (RCPSP) with makespan minimization as are available in limited quantity Rk at any point in time. Each
objective. The proposed approach uses binary string based activity j (j = 1, 2… n) requires a processing time or duration
representations and operators for chromosomes. The approach denoted by dj and a certain amount, rj,k of resources type k for
was tested on some difficult instances with high optimality completion.
gap in the J120 data set of PSPLIB. It was found that the
The activities in the project are interrelated by two kinds of
proposed approaches gave better results as compared to
constraints namely Precedence constraint and Resource
activity list based representations that are commonly used.
availability constraint. Precedence constraints dictates that an
General Terms activity j cannot be started until all its immediate predecessors
Genetic Algorithm, Project Scheduling, Resources Allocation Pj are completed; and the resource availability constraints
require that sum of resources needed on any day by all
Keywords activities scheduled on that day should not exceed availability
Resource Constrained Project Scheduling Problem, RCPSP, of any resource type.
Genetic Algorithm, Project Makespan, PSPLIB
As an illustration, R2 = 5, it means that the availability of
1. INTRODUCTION resource type 2 is limited to 5 units every day, and if r 3,2 = 4, it
Due to increasing complexity of managing modern businesses means that job 3 requires 4 units of resource type 2 every day
that depend on short product life cycles, practitioners have for its completion.
recognized the need to efficiently allocate scarce resources in
industrial projects to reduce their completion time. This The parameters dj, Rk, and rj,k are assumed to be integer and
problem is widely known as the Resource Constrained Project non negative as is typically assumed in the literature on
Scheduling Problem (RCPSP). Briefly stated, the RCPSP is RCPSP. The decision variables to be determined are the finish
concerned with scheduling of project activities considering times, Fj for each activity j. Thus, the finish time of the last
both the precedence and resource availability constraints. A job, n+1, is given by Fn+1. The finish time of the last job is
typical objective in solving this problem is one of minimizing also the makespan (total duration) of the project. Thus, the
the project makespan, or the time required to complete all the objective of RCPSP is to find precedence and resource
activities in the project. Over the time, various variations of feasible completion times of all activities such that the
RCPSP have been proposed by researchers with different makespan (total duration) of the project, given by the finish
objective functions, resource requirement and resource time of the last job, Fn+1, is minimized.
availability patterns. The RCPSP is known to be NP-hard
even under the simplifying assumptions of single renewable The conceptual model of RCPSP described by Christofides et
resource constraint, its constant availability over time and al. (1987) is as follows:
minimization of makespan as objective. Due to NP hard status
of the problem, one cannot solve for optimality when the Min Fn1 (1)
number of activities increases. Various approximation
methods are available to solve RCPSP. In this paper, an Subject to:
improved version of Genetic Algorithm (GA) is proposed and
tested on some of the worse instances of J120 Data set. Fi  F j  d j i  Pj (2)

The paper is organized as follows. Section 2 presents a

problem formulation. Section 3 presents some literature
related to Genetic Algorithms for RCPSP. Section 4 gives

34
 International Journal of Computer Applications (0975 – 8887)
Volume 78 – No.9, September 2013

r
jA( t )
j ,k  Rk k K ; t  0 (3)
is employed to generate a precedence and resource feasible
active schedule using one and two point crossover techniques.

Fj  0 j 1,2,3,.....n  1 (4)
GA proposed by Kohlmorgen et al. (1999) uses the concept of
Island Model and studied the variants of parallel Genetic
algorithm. A random key representation was employed which
uses a real values between 0 and 1 for each task. A two point
The objective function (1) minimizes the total duration
standard crossover was applied to the representation.
(makespan) of the project by minimizing the finish time of
n+1th activity. Constraint (2) shows the precedence
Lee and Kim (1996) tested three different algorithms namely
relationship which doesn’t allow activity J to start until all its
Simulated Annealing (SA), Tabu Search (TS) procedure and a
predecessors, i, are finished. Constraints (3) limit the resource
GA based on the random key representation. A parallel SGS
demand imposed by the activities being processed at time t to
was used to develop an active schedule. A different variant of
the available capacity. Constraints (4) shows the finish time to
pair wise interchange for SA and TS was applied and for GA
be non – negative.
one point crossover was employed.
The difficulty with this formulation is that the conceptual
Alcaraz and Maroto (2001) developed a GA which used a
constraint (3) cannot be explicit because it depends on the
serial SGS and activity list representation. A schedule is
unknown jobs scheduled at time t. Various time indexed
generated using an additional gene which decided whether a
formulations are available to work around this difficulty, but
forward or backward scheduling needs to be employed.
they exponentially increase the size of the formulation.
During the crossover stage child’s activity can be generated
by using either forward or backward scheduling. Alcaraz et al.
3. REVIEW OF GENETIC ALGORITHM (2004) (c.f. Kolisch and Hartmann, 2006) extended the
LITERATURE FOR RCPSP genetic algorithm of Alcaraz and Maroto (2001). SGS is
Genetic algorithm (GA) was developed by Goldberg (1989) as generated based on an additional gene (Hartmann, 2002) and
a computational approach to solve hard problems. It mimics the forward backward improvement of Tormos and Lova
the principles of biological evolution to solve hard (2001).
optimization problems. It provides an environment where the
solutions in a population continuously crossbreed, mutate and Coelho and Tavares (2003) introduced a crossover operator
compete with each other in a survival of the fittest strategy, called Late join Function Crossover and they employed Serial
until they evolve into the best solutions. The search for better SGS and activity list representation. The late join crossover
solutions in the GA based approach is largely context creates an offspring by copying the father chromosomes and
independent and hence they can be readily applied across a then swapping the adjacent pair from the mother in reverse
variety of situations. order.

GA considers a population of solutions instead of one Hindi et al. (2002) proposed a crossover technique similar to
solution. After creating the initial population, new solutions Hartmann (1998) which preserved the order thus not violating
are generated by mating two existing solutions (Crossover) or the precedence constraint. The main difference between the
by altering an existing one (Mutation). The fittest of these two was in the way the initial population is generated. A serial
survive and move on to the next generation by the means of a SGS and Activity list representation were used.
selection process while the rest are discarded. Fitness value
measures the quality of solution, depending on the objective GA developed by Toklu (2002) is applied on the schedules
function of the problem to be solved. directly and a penalty function is used to evaluate the
violation of constraints as infeasible schedules may be
Pseudo code of basic Genetic Algorithm generated.

Debels and Vanhoucke (2005) proposed a genetic algorithm

Begin Genetic Algorithm which considers two populations and hence named as bi-
population Genetic algorithm (BPGA). Both left-justified,
Generate initial population with random parents
(forward) which sort activities in the increasing order of the
Evaluate each parent in the population start time and right justified (backward), which sort activities
For (j =1 to NGEN) in the decreasing order of the finish times population are
Apply crossover with probability Pc on parents considered. The above method thus took into account features
Apply mutation with probability Pm on child of both Forward/Backward scheduling local search technique.
Add children to the current population
Debels and Vanhoucke (2007) suggested a Decomposition-
Retain best parents Based Genetic Algorithm (DBGA) for RCPSP. This method
Evaluate the new generation divides the RCPSP problem into smaller problems and
End For obtained the solution for the problem by combining the
End solution of sub problems. It was shown that the decomposition
based approach finds satisfactory near-optimal solutions and
Researchers have developed different representation schemes, gives a better solution.
genetic operators (crossover and mutation) and algorithms to
solve the RCPSP. Hartmann (1998) developed an algorithm Franco et al. (2007) developed algorithms in which initial
based on permutation of activities in an Activity List population was generated based on different features like
representation. A Serial Schedule Generation Scheme (SGS) staring date, ending date and makespan. The algorithm used
two point crossover and Serial SGS.

35
 International Journal of Computer Applications (0975 – 8887)
Volume 78 – No.9, September 2013

Kim and Ellis (2008) presented a permutation-based elitist 4. PROPOSED GENETIC ALGORITHM
genetic algorithm and observed the performance of the One of the biggest advantages of Hartmann’s approach is that
algorithm on large-sized projects. Initial population is the permutations generated from the crossover operation are
generated randomly and then Elitist selection mechanism is guaranteed to be precedence feasible. However, this also
used to discard the worse solutions thus keeping just the best restricts the exploratory power of the GA because of limited
solutions in the population. opportunities for mutation and crossover. Hence, the above
RCPSP GA algorithms may not really capture the potential of
Valls et al. (2008) suggested a hybrid genetic algorithm with the schema building blocks proposed by Holland (1975).
activity list representation and which makes use of forward– Holland’s GA performs mutation and crossover at bit level.
backward improvement of Valls et al. (2005). They developed Bit string representations may help prevent premature
a cross over scheme (Peak Crossover where instead of random convergence and give better solutions. For permutation
generation of the crossover point, properties of the schedule problems like RCPSP, obtaining feasible solutions using bit
were used when combining the parents. During first phase strings is not simple using the bit string representation.
parts of the parent schedule corresponding to peaks in the
resource usage is inherited. During the second phase evolution In case of binary strings, during the crossover stage, parent
is carried out using the neighbors, corresponding to best chromosomes are swapped as shown in Fig. 1. Child 1 is
individual generated from first phase. The neighbors are produced by taking initial part of parent 1 and tail part of
constructed with the approach used in Valls et al. (2003). parent 2. Similarly for child 2, initial part is taken from parent
2 and tail end from parent1. This allows greater freedom in
Goncalves et al. (2009) proposed a random key based genetic generating children. In case of Hartmann’s one point
algorithm which used the concept of parameterized active crossover, child 1 is produced by taking initial part from the
schedule (Goncalves et al., 2005). In the first phase priorities parent 1 and then the parent 2 list is scanned from the
are given to the activities and delay time and in the second beginning and those activities which are not part of the initial
phase these priorities are used to construct a parameterized part and considered as shown in Fig. 2. This ensures
active schedule. Instead of one point or two point crossover precedence feasibility of child.
parameterized uniform crossover (Spears and Dejong, 1991)
was employed.
Parent 1
Mehdi and Fariborz (2009) developed a Genetic algorithm
with new crossover strategy which uses combination of order 1001101100111110
crossover and partially mapped crossover. The crossover is Crossover Point
similar to Hartmann (2002) but it creates only one child. Parent 2
1000111001101010
Goncalves et al. (2011) developed a biased random key
genetic algorithm with forward-backward improvement.
Active schedule was created using serial scheduling scheme
Child 1 1001101101101010
based on the priorities of activities and then schedule was
improved using forward backward improvement procedure.
After schedule improvement gene adjustment was carried out
Parent 1 Parent 2
to reflect the changes in priorities. In this study instead of
using gene by gene mutation, some new individuals are
Child 2 1000111000111110
introduced in the next generation.

Agarwal et al. (2011) proposed an approach in which GA and

Parent 2 Parent 1
NN iterations are interleaved, feeding their best solutions to
each other alternately. The study shows that hybrid approach
Fig 1: Bit level encoding and Genetic Operations
gives better result than using GA and NN independently.

Liu and Yang (2011) developed a new serial insertion SGS

Parent 1 1 2 4 5 3
The algorithm builds an active schedule inserting an
unscheduled activity inside the partial schedule.
Crossover Point
Kim (2012) developed a hybrid genetic algorithm to solve the 4 1 2 3 5
Parent 2
construction resource constrained project to address the effect
of GA parameters like crossover probability, mutation
probability, population size and number of generation
1 2 4 3 5
Child 1
Kanchan and Karuna (2012) modified the Hartmann’s (1998)
crossover strategy and incorporated precedence with temporal Child 2 4 1 2 5 3
relationship amongst activity to produce feasible offspring i.e
for the child 1 the best schedule forms the father and Fig 2: Permutation Encoding and Genetic Operations
randomly selected schedule forms the mother and vice versa
for Child2. Then a standard two point crossover is carried out
The existing Genetic Algorithms based methods proposed by
on the selected parents. The initial population was generated
Hartmann (1998) are very powerful for large test problems,
using the priority rules Minimum Late Finish Time (LFT),
and even for some of the worst instances of J120 dataset in
Minimum Slack (MINSLK) and Greatest Rank Position
PSPLIB, the Genetic Algorithm of Hartmann (HGA) with
Weight (GRPW).
single point crossover gives solutions within 20 percent of the

36
 International Journal of Computer Applications (0975 – 8887)
Volume 78 – No.9, September 2013

upper bound value. This paper presents the variation of the 5. COMPUTATIONAL EXPERIENCE
simple Genetic Algorithm of Hartmann (HGA) with different
representation scheme and crossover logic to see if better WITH PROPOSED GA
results can be obtained for the worse case instances. Proposed For carrying out the computational analysis of Genetic
variant of genetic algorithm described below attempts to Algorithms, a select list of difficult instances from the
combine the power of bitwise representation while ensuring standard benchmark instances of PSPLIB were identified. The
precedence feasibility of schedules generated. identification is done as follows. Only the problems with large
number of activities, that is, those in J120 set are considered
for reporting because many of the smaller problem instances
4.1 Details of Proposed Genetic Algorithms in J30 and J60 were readily solved. It is seen from the PSP
The proposed variant of HGA called as DGA; uses the Library that the 600 instances of J120 data set are grouped
classical binary representation to denote activity priority. In into 60 sets of 10 instances each. The possible optimality gap,
this variant (DGA), the chromosome for each activity has 8 UB-LB, is calculated for each of the 600 instances. Then the
bits and thus can represent up to 255 values of priority. The average gap over the 10 instances in each of the 60 instance
activity priorities are used to generate the precedence feasible sets is calculated. The instance sets are then sorted in
activity list through a roulette wheel selection process. The descending order on average gap as shown in Table 1. Only
activity list is then used to generate an active schedule using a instances with larger average gaps are considered for the
serial SGS. The makespan for the schedule is computed and analysis. the Table shows only the top 10 instance sets with
stored as a fitness function for that chromosome. Once the higher average gap.
initial population of parents is generated, they are sorted in
fitness order. Using an elitist scheme, two best parents are For detailed computational analysis, the worst three instance
used to obtain new offspring by one point binary crossover of sets in Table 1, that is, J120_51, J120_56 and J120_31 are
the priority chromosomes. Occasionally a mutation operator is taken up. For comparison purpose, results were also obtained
applied to change the existing priority sequence. The by GA coded in C language based on Hartmann’s (1998)
offspring are added to the parent pool, and the best of these single point crossover rule which we call as HGA.
are retained as parents for the next generation.
Table 1: Average Gap between Upper and Lower Bounds
Pseudo Code for Generate Parents for DGA for the 10 instances in each instance set sorted in
descending order
For ( i=1 to POPSIZE)
Average Gap
Generate Binary Random sequence Sl. Instance Set
Generate Activity Priority using binary random (UB-LB)
Priority
Generate Precedence Feasible activity list through 1 J120_51 23.5
Roulette Wheel Selection
2 J120_56 20.1
Evaluate each parent in the population
End For 3 J120_31 18.8
4 J120_36 16.1
4.1.1 Initialization 5 J120_11 16.0
Priority for each activity is encoded as 8 bit binary string. The
string is generated by random assignment of 0’s and 1. This 6 J120_16 13.8
can represent upto 255 values of priority. An activity list is 7 J120_52 13.3
generated from precedence feasible candidates based on the
priority value using a roulette wheel selection. Using the 8 J120_46 12.8
activity list, an active schedule is generated using SGS and
makespan value is calculated. 9 J120_26 12.7
10 J120_57 12.0
4.1.2 Crossover and Mutation
Using an elitist strategy, pairs of adjacent parents in the
population are selected for crossover. A one point binary
crossover is carried out on the priority parents as shown in The proposed GA discussed in the earlier section 4.1 was
Fig. 1. The generated children are then converted to tested on the instance set j120_51, j120_56 and j120_31 from
precedence feasible activity list and makespan computed as instance J120 as this set consists of some instances with high
before. Then they are added to the parent population. gap between the upper and lower bound values for makespan.
Occasionally, a binary mutation is used on the child priority The instances were solved using the following parameter
sequence. values: Population size as 200, number of Generations as 500,
crossover probability as 0.95 and mutation probability as 0.95.
The proposed GA is compared with the HGA and the average
4.1.3 Selection makespan obtained from five replications (using different
The combined pool is then sorted based on minimum
random seeds) and a summary of results is reported in Table 2
makespan value (fitness value) and only the best ones are
retained for the next generation.

4.1.4 Termination
The process is repeated for several generations so that
convergence is obtained.

37
 International Journal of Computer Applications (0975 – 8887)
Volume 78 – No.9, September 2013

Table 2: Comparison of Average Makespan obtained from 7. REFERENCES

five different replications [1] Agarwal, A., Colak, S., Erenguc, S. “A neurogenetic
approach for the resource-constrained project scheduling
Instance Set Avg. Makespan Avg. Makespan problem”, Computers and Operations Research, 38,
J120 (HGA) (DGA) 2011, 44-50.
[2] Alcaraz, J., Maroto, C. “A robust genetic algorithm for
J120_51 253.86 244.26 resource allocation in project scheduling”, Annals of
Operations Research, 102, 2001, 83–109.
J120_56 287.48 279.02
J120_31 232.64 225.34 [3] Alcaraz, J., Maroto, C., Ruiz, R. “Improving the
performance of genetic algorithms for the RCPS
problem”, Proceedings of the Ninth International
As can be readily seen in Table 2, the modified algorithm, Workshop on Project Management and Scheduling,
DGA outperform HGA. The solution obtained from the above 2004, 40– 43.
mentioned GA was better than the one obtained from HGA [4] Coelho, J., Tavares, L. “Comparative analysis of meta
heuristics for the resource constrained project scheduling
In order to look at the best values obtained using different problem”, Technical report, Instituto Superior Tecnico at
methods, Table 3 shows the smallest value of makespan Portugal, 2003.
obtained among the 5 replications. The corresponding
percentage deviation of these values from the upper bound is [5] Debels, D., Vanhoucke, M. “A Bi-population Based
shown in Table 4. Genetic Algorithm for the Resource-Constrained Project
Scheduling Problem”, ICCSA, Vol. 4, 2005, 378-387.
Table 3: Best average makespan value out of the five
replications [6] Debels D, Vanhoucke M: A Decomposition-Based
Genetic Algorithm for the Resource-Constrained Project-
Best Avg. Best Avg. Scheduling Problem. Operations Research 2007; 55: 457-
Instance Set 469
Average UB Makespan Makespan
J120
(HGA) (DGA) [7] Franco, E.G., Zurita, F.T., and Delgadillo, G.M. “A
Genetic Algorithm for the Resource Constrained Project
J120_51 216.0 249.5 242.5 Scheduling Problem (RSPSP)”, Bolivia Research and
J120_56 251.7 283.8 277.2 Development, Vol.7, 2007, 41–52.
J120_31 199.0 229.9 222.9 [8] Goldberg, D.E. “Genetic algorithms in search,
optimization, and machine learning”, Addison-Wesley,
1989.
Table 4: Percent. Deviation of J120 instance set from the
[9] Goncalves, J., Mendes, J., Resende, M.G.C. “A hybrid
upper bound
genetic algorithm for the job shop scheduling problem”,
Instance Set Upper % Deviation % Deviation European Journal of Operational Research, 167, 2005,
J120 Bound, UB HGA DGA1 77-95.
[10] Goncalves, J., Mendes, J., Resende, M.G.C. “A random
J120_51 216.0 15.51 12.27 key based genetic algorithm for the resource-constrained
J120_56 251.7 12.75 10.13 project scheduling problem”, Computers and Operations
research, 36, 2009, 92-109.
J120_31 199.0 15.53 12.01
[11] Goncalves, J., Resende, M.G.C, Mendes, J. “A biased
random key genetic algorithm with forward-backward
6. SUMMARY AND CONCLUSION improvement for resource-constrained project scheduling
The objective of the research was to study the single mode problem”, Journal of Heuristics, 17, 2011, 467-486.
RCPSP with minimization of project makespan as [12] Hartmann, S. “A competitive genetic algorithm for
performance measure. As a NP hard problem, it is difficult to resource-constrained project scheduling”, Naval
solve for optimality when the number of activities in the Research Logistics, 45, 1998, 733–750.
project increases. This report presents modified GA for
resource constrained project scheduling problem (RCPCP) [13] Hartmann, S. “Self-adapting genetic algorithm for
with an objective of minimizing the makespan or total project scheduling under resource constraint”, Naval
duration of the project. Research Logistics, 49, 2002, 433-448.
[14] Hindi, K.S., Yang, H., Fleszar, K. “An evolutionary
It can be seen that by using binary encoding of the activity
algorithm for resource-constrained project scheduling”,
priority, and creating a precedence feasible activity list, one
IEEE Transactions on Evolutionary Computation, Vol.6,
can obtain better results than HGA for some difficult
2002, 512–518.
instances in J120 data set. There is thus scope for further
tuning these algorithms so as to get better results closer to the [15] Holland, J. H. “Adaptation in Natural and Artificial
upper bound using genetic algorithms. Systems”, University of Michigan, Press, Ann Arbor, MI
1975.

38
 International Journal of Computer Applications (0975 – 8887)
Volume 78 – No.9, September 2013

[16] Jin lee Kim. “Hybrid genetic algorithm parameter effects [22] Spears, W.M., and Dejong, K. A. “On the virtues of
for optimization of construction resource allocation parameterized uniform crossover”, Proceedings of the
problem”, Construction research Congress ASCE, 2012. Fourth International Conference on Genetic Algorithms,
1991, 230-236.
[17] Kanchan, J., Karuna, J. “A modified genetic algorithm
for resource constrained project scheduling problem”, [23] Toklu, Y.C. “Application of genetic algorithms to
International journal of Computer Applications, 57, construction scheduling with or without resource
2012, 41-45 constraints”, Canadian Journal of Civil Engineering, 29,
2002, 421–429.
[18] Kohlmorgen, U., Schmeck, H., Haase, K. “Experiences
with fine-grained parallel genetic algorithms”, Annals of [24] Tormos, P., Lova, A. “A competitive heuristic solution
Operations Research, 90, 1999, 203–219. technique for resource-constrained project scheduling”,
Annals of Operations Research, 102, 2001, 65–81.
[19] Kolisch, R., Hartmann, S. “Experimental investigation of
heuristics for resource-constrained project scheduling: [25] Valls, V., Ballestin, F., Quintanilla, M.S. “A Resource
An update”, European Journal of Operational Research, Constrained Project Scheduling: A critical activity
174, 2006, 23-37 reordering heuristic”, European Journal of Operational
Research, 149, 2003, 282–301.
[20] Lee, J.K., Kim, Y.D. “Search heuristics for resource
constrained project scheduling”, Journal of the [26] Valls, V., Ballestin, F., Quintanilla, M.S. “Justification
Operational Research Society, 47, 1996, 678–689. and RCPSP: A technique that pays”, European Journal of
Operational Research, 165, 2005, 375–86.
[21] Mehdi, D. and Fariborz, J. “A new efficient Genetic
algorithm for project scheduling under resource [27] Valls, V., Ballestin, F., Quintanilla, M.S. “A hybrid
constraints”, World Applied Sciences Journal, 7, 2009, genetic algorithm for the RCPSP”, European Journal of
987-997. Oerational Research, 185, 2008, 496-508.

IJCATM : www.ijcaonline.org 39

View publication stats