Optimizing Time Cost Trade Off Scheduling by Genetic Algorithm

Download as pdf or txt
Download as pdf or txt
You are on page 1of 9

International Journal of Application or Innovation in Engineering & Management (IJAIEM)

Web Site: www.ijaiem.org Email: [email protected], [email protected] Volume 2, Issue 9, September 2013 ISSN 2319 - 4847

Optimizing Time Cost Trade off Scheduling by Genetic Algorithm


Dr. Mamta Madan1, Rajneesh Madan2
1

Vivekananda Institute of professional studies- GGSIPU, Delhi 2 NIIT technologies ltd, Gurgaon, India

Abstract
The Time Cost Trade off Scheduling Problem represents an important research area. Many heuristic methods have been proposed to solve TICOTS (Time cost tradeoff scheduling problem) .It is an NP hard problem. The solution is implemented on Genetic Algorithm and has been named as GASolver. Dependency Injection is being used to make the problem loosely coupled, so that other areas of scheduling like Resource Constrained, Payment Scheduling etc. can be merged with same solution in the future. We should be able to design a common framework for all the scheduling problems.

Keywords: Genetic Algorithm, Dependency Injection, GASolver.Core, Time Cost Trade off Scheduling, Chromosome, Cost Slope.

1. INTRODUCTION
In Software Project Management, it is often important to shorten the duration of some activities through expending extra budget in order to compress the software project total completion time. The procedure can be conducted under either fixed available budget or a desirable threshold of software project completion time. This procedure is known as Time/Cost Trade-off. There are some procedures which can expedite the activity durations. These procedures can be summarized as a unique cost function corresponding to each activity. The aim of this paper is to find activity durations and a schedule that optimises the total software project cost.

2. TERMS RELEVANT TO THE CONTEXT


Normal Duration: The time that an activity is completed using the least direct cost is known as normal duration and is denoted by Dn. Forced Duration: The shortest time that an activity takes possibly by supplying more resources is known as forced duration of an activity and is denoted by Df. Normal Cost: The total cost of an activity for its normal duration is called as normal cost and is denoted by Cn.

Forced Cost: The cost of an activity when it is quickly completed due to forced deadline. It is called forced cost and is denoted by Cf. Cost Slope: The additional direct cost that one has to pay to reduce one unit of time from the normal duration of an activity is defined as cost slope. Cost Slope = (Cf-Cn)/ (Dn-Df). Assumptions for the above problem The normal cost of an activity is more than the forced cost. There is a linear relationship between activity time and cost. We can apply resources to shorten the activity.

3. GENETIC ALGORITHM
Genetic algorithms (GAs) are search algorithms that are conceptually based on the methods that living organisms adapt to their environment. These methods, known as natural selection or evolution, combine the concept of survival of the fittest among string structures with a structured yet randomized information exchange to form a search algorithm with some of the innovative flair of human search. In each generation, a new set of string structures is created from (bits and pieces of) the fittest strings from the previous generation and occasionally a randomly altered new part. This process of exploiting historical data allows the GA to speculate on new search points that will improve performance thus producing better solutions. Genetic algorithms were initially developed by JohnH.Holland, a professor of psychology and computer

Volume 2, Issue 9, September 2013

Page 320

International Journal of Application or Innovation in Engineering & Management (IJAIEM)


Web Site: www.ijaiem.org Email: [email protected], [email protected] Volume 2, Issue 9, September 2013 ISSN 2319 - 4847
science at the University of Michigan. As an optimization tool, the Genetic Algorithm attempts to improve performance leading to an optimal solution. In this process, there are two distinct steps, (1) the process of improvement and (2) reaching the optimum itself. Of these two steps, the most important is the process of improvement. In complex systems, due to the potential high costs involved, reaching the optimum solution may not be justified as long as continuous improvement is being made and an optimal (desirable) solution can be found. Genetic algorithm (GA) [i] [ii] [iii] is a pioneering method of metaheuristic optimization which originated from the studies of cellular automata of Holland in the 1970s. It is also known as an evolutionary algorithm and a search technique that copies from biological evolution. In Genetic Algorithm, a population of candidate solutions called individuals evolves toward better solutions from generation to generation.

4. WHY GENETIC ALGORITHM


1. GA can quickly scan a vast solution set. Bad proposals do not affect the end solution negatively as they are simply discarded. 2. The inductive nature of the GA means that it doesnt have to know any rules of the problem - it works by its own internal rules. This is very useful for complex or loosely defined problems. 3. They efficiently search the model space, so they are more likely (than local optimization techniques) to converge toward a global minima. 4. There is no need of linearization of the problem. 5. There is no need to compute partial derivatives. 6. More probable models are sampled more frequently than less probable ones

5. RELATED WORK
Erenguc et al. (1993) [ iv ]were the first to consider the DTCTP( Discounted Time Cost problem) with discounted cash flows throughout the life of the software project, and where shorter activity durations can be attained by incurring greater direct costs. The objective of this problem was to determine the activity durations and a schedule of activity start times so that the NPV of all cash flows is maximized. The problem was formulated as a mixed integer nonlinear program which was amenable to solution using the generalized Benders decomposition technique. The algorithm was tested on 140 software project scheduling problems, the largest of which contains 30 nodes and 64 activities. The computational results were quite encouraging since 123 of the 140 problems require less than 1 CPU second of solution time. Vanhoucke et al. (2002) [v] described a solution procedure for the DTCTP (Discounted Time Cost Tradeoff problem) in which three special cases of time-switch constraints were involved. These constraints imposed a specified starting time on the software project activities and force them to be inactive during specified time periods. The authors propose a branchand-bound (B&B) algorithm and a heuristic procedure which both made use of a lower bound calculation for the DTCTP (without time-switch constraints). The procedures were validated on a randomly generated problem set. The authors also discussed an illustrative example based on a real-life situation. Tareghian and Taheri in 2006 [vi ] developed a solution procedure to study the trade-offs among time, cost, and quality in the management of a software project. This problem assumes the duration and quality of software project activities to be discrete, non-increasing functions of a single nonrenewable resource. The objective was to minimize the total cost of the software project while maximizing the quality of the software project and also meeting a given deadline. Three interrelated integer programming models were developed such that each model optimizes one of the given entities by assigning desired bounds on the other two. Various forms of quality aggregations and effect of activity mode reductions were also investigated. The computational performance of the models was presented using a numerical example.

6. SOLUTION FOR TIME/COST TRADE OFF SCHEDULING


We have implemented the Time/ Cost trade off problem using Genetic Algorithm. We need to address the following objectives for its implementation:A. Specifying the relationships between the tasks. B. Description of normal Duration, forced duration, normal cost, forced cost and cost slope. C. The representation of the chromosome. D. Implementation of selection, crossover and mutation function. E. Calculation of an objective function to evaluate the best schedule and optimal cost. F. Class Diagram and Implementation of Time/Cost Trade off scheduling problem. G. Test Results and Analysis. 6.1 Specifying the Relationships between the Tasks A project is best represented as a Task Precedence Graph (TPG). A TPG is an acyclic directed graph consisting of a set of tasks and a set of precedence relationships. With the help of Task Precedence Graph we set the order of execution for each task.

Volume 2, Issue 9, September 2013

Page 321

International Journal of Application or Innovation in Engineering & Management (IJAIEM)


Web Site: www.ijaiem.org Email: [email protected], [email protected] Volume 2, Issue 9, September 2013 ISSN 2319 - 4847
The Task Precedence Graph is shown below in the form of Table. Table 2Task Precedence Graph for TICOSP T4 T5 T6 T7 T8 T9 T10 T11 1 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1 1 0 0 0 0 0 0 1 1 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

T1 T2 T3 T4 T5 T6 T7 T8 T9 T10 T11 T12 T13 T14 T15

T1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

T2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

T3 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0

T12 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0

T13 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0

T14 0 0 0 0 0 0 0 0 0 0 0 1 1 0 0

T15 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0

The above task precedence graph explains that for task T3 to finish, Task T1 should be completely finished. Similarly for task T4 to complete, Task T1 and T3 should finish and so on. Thus Task Precedence Graph enables us to set the order of execution for various tasks. 6.2 Description of Duration, Cost and Cost Slope We have already defined Dn, Cn, Df, and Cf in Introduction.We have maintained a table which holds the task code, the normal duration of the cost which is the upper duration and upper cost, similarly needs to maintain the forced duration and forced cost which is known as lower cost and lower duration. We have taken a scenario in which project consists of 8 tasks varying from T1 to T8, their respective Normal duration (Dn), Forced duration, Normal cost, Forced cost is also maintained in the Table 2. Cost Slope is also calculated for each task and also stored. Table 3 : Tasks with Respective Duration and Cost Slope Tasks Normal Duration(Dn) 4 5 3 6 4 2 3 5 Forced Duration(Df) 3 4 3 5 3 1 1 4 Normal Cost(cn) 11 13 10 18 22 15 8 16 Forced Cost(cf) 14 18 10 25 32 17 10 24 Cost Slope C=(CfCn)/(Dn-Df) 3 5 0 7 10 2 1 8

T1 T2 T3 T4 T5 T6 T7 T8

6.3 Representation of Chromosome The most important issue in Genetic Algorithm is to design appropriate chromosomes for a specific problem. The first part of the chromosome consists of activities that are arranged and validated against Task Precedence Graph. The second part of the chromosome consists of duration which is varying from lower bound (the forced duration) and the upper bound (i.e. the normal duration). Let us assume we have 8 activities varying from T1 to T8. Table 4: Task Vs Duration T3 T4 T5 3 6 4

Tasks Duration(Df,Dn)

T1 3

T2 4

T6 2

T7 2

T8 4

Volume 2, Issue 9, September 2013

Page 322

International Journal of Application or Innovation in Engineering & Management (IJAIEM)


Web Site: www.ijaiem.org Email: [email protected], [email protected] Volume 2, Issue 9, September 2013 ISSN 2319 - 4847
The duration which is mentioned in the above table is the random generation of duration. The value of the random number is between lower and the upper bound of duration. Substitute this calculated duration and can use the respective change in the forced cost and hence can get the optimal cost increase due to optimal shrink in the duration after applying all the three operators. 6.4 Implementation of Selection, Crossover and Mutation Function The three critical functions of Genetic Algorithm are selection, crossover and mutation. They are explained below.
a) Selection

We initially generate a two dimensional array of the above mentioned genome of task and randomly generate duration varying from lower bound to upper bound. We have checked the validity of genome by checking: Have obeyed the Task Precedence Relationship Have fitness better than deathfitness variable Here the death fitness variable signifies the fitness of the genome. We have made the function to calculate the fitness of the Genome. If the fitness of the genome is -1, (value of death fitness) then the genome is an invalid genome. We select only those Genomes which are good reproducers. If the fitness is better, only then it will reproduce otherwise it is removed from the genome list and if it is able to reproduce it will be added to the list of genomes which will be further utilized for crossover and mutation. This way we are able to select the genomes which have the capability to reproduce further. We have functions like candie() and canreproduce() which will check the survival of the genome.
b) Crossover

The crossover operator mimics the way in which bisexual reproduction passes along each parents good genes to the next generation. Normally, two parents solution create two new offspring solutions by combining their genes using one point crossover. Below shown is an example of two genomes which are successfully randomly generated. They have been taken from the list of successful reproducers. Before crossover Randomly we have chosen two genomes from the list of selected genomes which can reproduce well. They are represented as Genome1 and Genome2. The Genome1which is taken randomly from the list of reproducers is shown below in Table4. We can observe that the chosen genome has the normal sequence of task precedence and their respective random durations are also shown in the same Table. Table 5 : Genome1

Similarly we have picked up another genome randomly, the task is not generated in the sequence but they follow the Task Precedence Relationships as they are already validated in the process of selection. Table 6 : Genome2

Volume 2, Issue 9, September 2013

Page 323

International Journal of Application or Innovation in Engineering & Management (IJAIEM)


Web Site: www.ijaiem.org Email: [email protected], [email protected] Volume 2, Issue 9, September 2013 ISSN 2319 - 4847
As per the crossover operator, two genomes mom and dad are made crossover at a certain point known as crossover point, generating two baby genomes. We have Genome1 and Genome2 as Mom and Dad respectively and the crossover point randomly taken as 4. The two Baby Genomes are created. BabyGenome1 is shown in Table 6 and BabyGenome2 is shown in Table 7. Table 7: Baby Genome1

Task Prec A B C D E F G H

Duration 3 4 3 3 4 3 6 4

Table 8: Baby Genome2 Task Precedence D E F G A B C H Duration 3 4 3 6 3 4 3 4

We will pick randomly one of the baby genomes i.e. either baby genome1 or baby genome 2 depending on the generation of the random number. This genome will be stored in the list of successful genomes.
c) Mutation

Following the crossover operator the offspring may be mutated by the mutation operator. Mutation is basically to get some variation in the result. Similar to random mutation in the biological world, this function is intended to preserve the diversity of the population, thereby expanding the search space into regions that may contain better solutions. To perform mutation for time cost trade off problem, we randomly select a genome, and perform mutation. Due to mutation we perform variation in the duration by adding 1 from duration or we may subtract 1 from the duration. Depending on the random Boolean value we may decide to give it a positive variation or a negative variation. We have genome1 in Table 8 and we are supposed to mutate it. Table 9 : Genome before Mutation Task Precedence Duration 3 A B C D E F G H 4 3 6 4 2 2 4

Assume that we randomly pick 6th position of this genome, we generate a boolean value say that comes to true i.e. we have to generate +ve variation i.e. we will add +1to the respective task i.e. duration for task F will become 3.We have

Volume 2, Issue 9, September 2013

Page 324

International Journal of Application or Innovation in Engineering & Management (IJAIEM)


Web Site: www.ijaiem.org Email: [email protected], [email protected] Volume 2, Issue 9, September 2013 ISSN 2319 - 4847
made a function to validate that this value does not exceed the upper bound for duration, if it exceeds the upper range then we have to go for ve variation. Thus the Genome after mutation is as follows: Table 10: Genome after Mutation Task Precedence Duration A B C D E F G H 3 4 3 6 4 3 2 4

We have done the +ve variation in the above Genome. We can also perform ve variation, by subtracting -1from the randomly picked value and then checking it with the lower limit of duration. 6.5 Objective Function to Evaluate the Best Schedule and Optimal Cost Depending on the purpose of the problem different categories of fitness function can be applied. For example, if the purpose is to minimize the duration of the project, only duration can be used in the fitness value. If the objective is to minimize the cost of the project, only cost of the project should be considered as the fitness value. But if the objective is to reach the optimum point of the project, both cost and the project duration should be used in the fitness function. Fitness function for time/cost tradeoff problem Fitness Function =Total duration + (direct cost- indirect cost) Where total duration is the reduced duration of the project in reference to the normal duration. Indirect cost and direct cost are the reduction in the indirect cost and addition of direct cost. Table 11 : Tasks with Cost Slopes

To visualize the calculation of objective/fitness function, lets take a Genome consisting of Tasks, randomly generated duration, their respective normal duration and their cost. This is shown above in Table 10. Dn= 32 weeks Df= 28 weeks From the above table we can see that the project normally is suppose to be finished in 32 weeks but the random forced duration generated by GA chromosome required to be finished in 28 weeks. Therefore for this example total duration to be reduced is 4 weeks.

Volume 2, Issue 9, September 2013

Page 325

International Journal of Application or Innovation in Engineering & Management (IJAIEM)


Web Site: www.ijaiem.org Email: [email protected], [email protected] Volume 2, Issue 9, September 2013 ISSN 2319 - 4847
Therefore value for Total Duration= 4 Lets say 5 units is the direct cost per unit time Direct cost = 5*4= 20 And indirect cost which got increased was 17$(3$ for Task T1, 5$ for Task T2, 1$ for Task T7 and 8$ for Task T8) Fitness Function = 4+ (20-17) =7 The fitness functions for this random generation for duration comes out to be 7, similarly it is been applied for other generations, the minimum the value of fitness function, the better the situation of random generations. We have made a function which will calculate the fitness of all the Genomes and the Genome having the least fitness value (i.e. the cost is minimized) will be considered and will be shown at the console. 6.6 Class Diagram and Implementation Details GASolver.Core GASolver.Core is the main component of the solution. It is responsible for implementing all the three operators namely selection, crossover and mutation on various generations. It also provides a contract IGenome to be implemented in different genomes who wish to use GASolver for optimizing their problem. Following is the class diagram of GASolver.Core. It has a population (generation) class which essentially is collection of similar genomes.

Figure 1 : Class diagram for GAsolver GASolver.TICOTOSchedule GASolver.TICOTOSchedule is an implementation of Time Cost Tradeoff Project Scheduling Problem. This class is representation of genome and has methods for mutation, crossover and calculating fitness of genome. TICOTODataConnection class is responsible for making the connection to database and fetching different data from TICOTO database. GASolver.Core population class is also responsible for creating collection of genomes objects, it must know about the actual implementation of IGenome e.g RCProjectSchedulingGenome. But we cannot instantiate the actual genome object since it will tightly couple the population class to that genome implementation and the population class could not be used for other Genomes. We have used dependency injection object oriented programming principle to overcome this problem.

Figure 2: Class Diagram of Time Cost Scheduling

Volume 2, Issue 9, September 2013

Page 326

International Journal of Application or Innovation in Engineering & Management (IJAIEM)


Web Site: www.ijaiem.org Email: [email protected], [email protected] Volume 2, Issue 9, September 2013 ISSN 2319 - 4847
We are using Microsoft Unity 1.1 as our dependencies container. Figure 3 depicts that population class does not have any reference (dependency) for various genomes (in this case RCProjectSchedulingGenome) and yet be able to instantiate them with the help of dependency injection container. Use of IGenome contract and DI container make the solution generic and work with different genomes meant for different problems.

Figure 3 : GASolvers Dependencies Diagram 6.7 Test Results and Analysis Test Case 1 We have tested the results in two situations. Firstly we kept the upper duration as 50. So the random generation is varied between 43 to 50 weeks and their respective delta cost is shown below in the Table11.The Figure 4 shows the analysis of the same. Upper duration =50 Table 12 : Time Cost Trade off Scheduling Generation 1 269 318 7922 13673 13969 15212 15217 18214 Duration 45 43 44 45 45 45 46 47 50 Delta Cost 22 34.16 30 24.5 24.83 25 20 15 0

Figure 4: Time Cost Trade off Scheduling Test Case 2 Upper Duration= 32 Here the upper duration is changed and maintained at 32. The random durations are generated between 28 to 32 ManDays. The results are shown below in the Table 12. The Figure 5 shows the analysis of the same.

Volume 2, Issue 9, September 2013

Page 327

International Journal of Application or Innovation in Engineering & Management (IJAIEM)


Web Site: www.ijaiem.org Email: [email protected], [email protected] Volume 2, Issue 9, September 2013 ISSN 2319 - 4847
Table 13 : Time Cost Trade off Scheduling Generatio n 1 2 4 4956 Duration 28 29 31 32 Delta Cost 20 15 5 0

Figure 5 : Time Cost Trade off Scheduling

7. CONCLUSION AND FUTURE DIRECTIONS


Time/Cost Trade off Scheduling is an important problem as studied in literature survey. We have implemented this with Genetic Algorithm using C#.net. Most of the solutions that existed earlier for Time Cost tradeoff were not extendable. We have implemented GASolver .core using which any specific problem domain genome can be constructed. The fitness function is only to be specified by the project manager for their own specific domain. The same GASolver.core can be extended to other important research areas like Resource Constrained Scheduling , Payment Scheduling problem etc. Once all these areas will be part of GASolver, it will be the complete solution to project scheduling problems.

REFRENCE
[1] Chambers LD (ed.) (1999) Practical handbook of genetic algorithms: complex coding systems. CRC Press, Boca Raton [2] Davis L (1991) Handbook of genetic algorithms. Van Nostrand Reinhold, New York [3] David E. Goldberg Genetic Algorithm, in search Optimisation and Macine Learning [4] Demeulemeester, B. Dodin and W.S. Herroelen ,A random network activity generator,Demeulemeester et al., 1993 E.L. , Operations Research (5) (1993), pp. 972980. [5] Mario Vanhoucke and Dieter Debels, The discrete time/cost trade-off problem: extensions and heuristic procedures, , Journal of Scheduling Volume 10, Numbers 4-5, 311-326, DOI: 10.1007/s10951-007-0031-y. [6] Tareghian, H.R. & Taheri, H. 2006. On the discrete time, cost and quality trade-off problem, Applied Mathematics and Computation, 181, pp 1305-1312.

AUTHOR
Mamta Madan has over 16 years of experience in teaching UG and PG courses at various institutes. Currently she is a Professor in Department of Information technology at Vivekananda Institute of Professional Studies, Guru Gobind singh Indraprastha University. She did her Doctorate in Computer Science from Banasthali University. She has published several papers in reputed research journals and read at many conferences. Her current areas of interest include Bioinspired computing algorithms, Object oriented analysis and design, design principles and patterns, cloud computing, optimization of software engineering processes, Artificial intelligence.

Volume 2, Issue 9, September 2013

Page 328

You might also like