You are on page 1of 3


International J. of Recent Trends in Engineering and Technology, Vol. 3, No. 2, May 2010

Survey on Scheduling With Learning Techniques

E. Grace Mary Kanaga 1, M. L. Valarmathi 2, Juliet A Murali 3
Karunya University/Computer Science and Engineering, Coimbatore, India
Government College of Technology, Coimbatore, India
Karunya University/Computer Science and Engineering, Coimbatore, India

Abstract—  This paper describes about scheduling of jobs that SPT was the best choice for optimizing waiting time
using learning techniques. It also describes about different and system utilization [19,23].
learning effect. The evaluation on different learning effect B. Multi Agent Systems for Scheduling
shows that the experience based learning (EBL) gives better Multi agent systems can be used to solve scheduling
solution for calculating processing time. The EBL can be problems. The participants in this system are considered
incorporated with scheduling problems such as job shop
as software entities called agents. Two types of entities
scheduling, meeting scheduling, patient scheduling etc. As
the experience increases the time required to process a job is
are included in this agent based systems, task agents and
reduced. Hence it reduces the make span of the schedule. resource agents. The agents have proactive and reactive
The EBL can be applied to solve scheduling problem of nature because of that it allows the communication
single as well as multiple machines. This paper also includes between agents and also allows taking coordinated
a framework for patient scheduling that incorporates the decisions [20].
Index Terms—Scheduling, Learning, Multi agent systems III. CATEGORIES OF LEARNING EFFECT

I. INTRODUCTION Scheduling is the process of allocating tasks in order to

achieve optimal solution. The time required to perform a
Industrial scheduling techniques are studied in the field task (job) decreases as the experience related to it
of operations research (OR). OR techniques are very increases ,even then each task takes some time for its
effective for solving well-defined centralized processing. When it have maximum experience then it
optimization problems, where the algorithm can takes only minimum processing time [1].The learning
determine the optimal schedule for all parties involved model used for scheduling are based on learning curve
[24]. Scheduling is the process of allocating tasks in order introduced by Wright [3,14].
to achieve optimal solution. The time required to perform
a task decreases as the experience related to it increases, A. Scheduling Problem with Position Based Learning
even then each task takes some time for its processing. In position based learning model the processing time of
When it has maximum experience then it takes only job is depends on the position of the job in the sequence.
minimum processing time. The actual processing time is a function of jobs position
in the schedule [2, 13]. The processing time of a job j if
II. SCHEDULING TECHNIQUES scheduled in position r is given by
A scheduling problem can be characterized by a set of Pjr=Pjra j,r =1…n (1)
jobs, each with one or more operations. The operations of Where, a is the constant learning index,a≤0, Pjr is the
a job are to be performed in a specified sequence on actual processing time and r is the position of job in the
specific machines [21]. The performance measures that schedule.
are related to scheduling include machine utilization,
cycle time, throughput rate and inventory level [19]. B. Sum of Processing Time Based Learning Effect
A. Scheduling Rules The sum of processing time based learning effect model
the actual processing time of a job as a function of total
The procedures are designed to provide good solutions normal processing time of jobs that are already processed
to complex problems in real time when scheduling and of the job’s scheduled position.[5,12] .
problems are considered. These procedures are called
scheduling rules. The scheduling rules have been C. The Time Dependent Learning Effect
classified into class 1, class 2, class 3 etc, according to The Time dependent learning effect of a job is to be a
performance criteria [22]. Class 1 contains simple priority function of the total normal processing time of the jobs
rules, which are based on information related to the jobs. scheduled in front of the jobs [6,9]. Let pjr be the
Class 2 consists of combinations of rules from class one. processing time of job j if it is scheduled in position r in a
Class 3 contains rules that are commonly referred to as sequence. Then
Weight Priority Indexes. Conway and Maxwell showed

© 2010 ACEEE
DOI: 01.IJRTET.2010.03.02.179
International J. of Recent Trends in Engineering and Technology, Vol. 3, No. 2, May 2010

Pjr=pj(1+ P[1]+ P[2]+…+ P[r-1)a (2) considered as unary. But in case of time based experience
model the experience is calculated in terms of normal
Where, pj is the normal processing time of job j, p[k] is
processing time.
the normal processing time of a job if it is scheduled in TABLE I.
the kth position in a sequence and a is the constant Input Parameters
learning index, a≤0. Tasks Arrival Time
Processing Time
D. Experience Based Learning Effect T1 0 7
According to A. Janiak and R. Rudek the learning T2 1 5
effect is modeled by the job processing time pj(E) T3 1 2
formulated as a non-increasing positive function of the T4 2 10
experience E possessed by the is equal to its normal TABLE II.
processing time aj processor. If the processor does not COMPARISON OF DIFFERENT PROCESSING TIME WITH DIFFERENT
possess experience (E =0), the processing time of a task LEARNING EFFECT
(say j), i.e., Pj(0)= aj.The processing time of job j is Proc
scheduled in the vth slot in a sequence is given as follows essin
Processing g
Pj(Eβ(v))=aj-bj(min{ Eβ(v),gj})αj (4) Time without Time
learning with
Where, aj is the normal (sequence-independent) learn
processing time of task j, αj anb bj are the exponential and ing
linear learning ratios of task j and gj is the learning Position Based Learning Model 10 6.39
threshold [10,11].
Time Based learning Model 10 4.18
For the above model, the parameters aj > 0, bj > 0, aj >
0, gj > 0 and βj=[0, 1] are assumed to be rational and since Experience Based Learning
10 3.5
Model(position based approach)
the job processing time is some positive value [1,7]. Experience Based Learning
The learning factor is represented in terms of 10 4.9
Model(time based approach)
experience and is given as,
Eβ(v)= [l] +βv e[v] (3)
The comparison given in Table II shows that the
Where, [l] is the experience already possessed by the experience based learning gives better result in reducing
processor, e[l] is the experience provided to the processor make span as compared to other techniques. Hence EBL
by a job scheduled in the lth position e[l]≥ 1 and β[v] is the is incorporated in patient scheduling and the proposed
amount of experience (percentage of e[v]) provided to the framework is shown in Fig. 1.
processor by job [v], β[v] = [0,1] .
  RS1 RSm


Resource 1 Resource m

This section presents the comparison of different

learning technique in terms of processing time. The
learning index has to be calculated fist and by using this CA
the processing time is calculated. The learning index is
the logarithm to the base two of learning rate. If the rate PS1 PSn

of improvement in the processing time is 20% means the ...

learning percentage would be 80%. Here consider

learning percentage as 80%.So the learning index is
calculated as a= αj =-0.322. Consider four tasks. Three RA: Resource Agent
tasks are already scheduled. Now the task having PA: Patient Agent
1: Send patient details CA: Common Agent
processing time 10 is trying to schedule in 4th position. 2: Send resource details PS : Patient Schedule
The input parameters are shown in Table I. The 3: Specifies the tasks for patients LA: Learning Agent
processing time is calculated for different learning effects 4: Request for resources RS : Resource
based on the equation discussed above are given in Table Schedule
II .The experience based learning model provides less Figure 1. Framework for patient scheduling using experience based
processing time. Even though time based learning model learning
is also have less processing time, but it is not consider the The framework consists of Patient Agent (PA),
threshold value for processing time. In experience based Resource Agent (RA) and Common Agent (CA).The CA
model it is considered. Here the experience calculation is a physician they collect and maintain information about
may be position based, where the experience depends on patients and resources. It also determines the tasks that
number of jobs. The experience gained by a job is have to be taken, which consists of consultation of doctor,

© 2010 ACEEE
DOI: 01.IJRTET.2010.03.02.179
International J. of Recent Trends in Engineering and Technology, Vol. 3, No. 2, May 2010

diagnostic procedures like MRI scan, CT scan lab tests based and job-position-based learning effects”,
etc. Now the PA knows what are all the tasks it has to Information Sciences 178 (2008) 2476–2487.
perform and then it request for the resources. The [6] J.-B. Wang,C.T. Ng, T.C.E. Chengb, L.L. Liu,” Single-
Resource Agent may be X-Ray, CT Scan, Lab Tests, machine scheduling with a time-dependent learning effect’,
Int. J. Production Economics 111 (2008) 802–811.
consultation with Physician etc.In this framework a [7] Adam Janiak a, Władysław Adam Janiak b, Radosław
special agent called Learning Agent(LA) is included. The Rudek a,1, Agnieszka Wielgus a,1“Solution algorithms for
LA calculates the experience of each resource and find the makespan minimization problem with the general
out the processing time according to the experience they learning mode”,Computers & Industrial Engineering 56
possessed. (2009) 1301–1308.
[8] Tamer Eren, Ertan Gu¨ner” A bicriteria flowshop scheduling
VI. PERFORMANCE ANALYSIS with a learning effect”, Applied Mathematical Modelling
32 (2008) 1719–1733.
There are several performance metrics for the [9] D. Biskup, “Single-machine scheduling with learning
scheduling problem. considerations,”Eur. J. Oper. Res., vol. 115, no. 1, pp.
173–178, May 1999.
A. Total completion time [10] T. Eren and E. Güner, “A bicriteria flowshop scheduling
This represents the total amount of time used by the with a learning effect,” Appl. Math. Model., vol. 32, no. 9,
schedule. For scheduling problems the main objective is pp. 1719–1733, Sep. 2008.
to minimize this metric. As the experience increases the [11] Tamer Eren,” A bicriteria parallel machine scheduling with
a learning effect of setup and removal times”, Applied
total time requires to process a job is reduced hence the
Mathematical Modelling 33 (2009) 1141–1150.
total completion time. [12] W.-C. Lee, C.-C. Wu, and H.-J. Sung, “A bi-criterion
Ci=Ci-1+Pi (4) single-machine scheduling problem with learning
considerations,” Acta Inform., vol. 40,no. 4, pp. 303–315,
B. Total tardiness Feb. 2004.
It means how late a job is completed. Sometimes the [13] T. C. E. Cheng, C.-C. Wu, and W.-C. Lee, “Some
scheduling problems with deteriorating jobs and learning
jobs that are not completed within specified time. This
effects,” Comput. Ind. Eng., vol. 54,no. 4, pp. 972–982,
tardiness can be reduced by the introduction of learning May 2008.
effect, since it also involves the processing time. [14] C. Koulamas and G. J. Kyparisis, “Single-machine and
Ti=max{0,Ci-di} (5) two-machine flowshop scheduling with general learning
functions,” Eur. J. Oper. Res.,vol. 178, no. 2, pp. 402–407,
Apr. 2007.
VII. CONCLUSION [15] G. Mosheiov and J. B. Sidney, “Scheduling with general
job-dependent learning curves,” Eur. J. Oper. Res., vol.
This paper describes the various categories of learning 147, no. 3, pp. 665–670,Jun. 2003.
effects and their use in scheduling problems. Comparison [16] J.-B. Wang and T. C. E. Cheng, “Scheduling problems with
of different learning technique such as position based the effects of deterioration and learning,” Asia-Pac. J.
learning, time based learning and experience based Oper. Res., vol. 24, no. 2,pp. 245–261, 2007.
learning has been done. Among this techniques [17] D. Biskup, “A state-of-the-art review on scheduling with
learning effects,”Eur. J. Oper. Res., vol. 188, no. 2, pp.
experience based learning is considered as better for
315–329, Jul. 2008.
scheduling problems. The experience based learning [18] J.-B. Wang, “Single-machine scheduling problems with the
model can be used to solve any type of the scheduling effects of learning and deterioration,” Omega, vol. 35, no.
problems like patient scheduling, job shop scheduling etc. 4, pp. 397–402,Aug. 2007.
[19] C. S. Chong and Malcolm Yoke Hean Low “A Bee
REFERENCES Colony Optimization Algorithm To Job Shop Scheduling”,
Proceedings of the 2006 Winter Simulation Conference.
[1] A. Janiak and R. Rudek, “The learning effect: Getting to the [20] A.Jones and L.C. Rabelo “Survey of Job Shop Scheduling
core of theproblem,” Inf. Process. Lett., vol. 103, no. 5, pp. Techniques “, National Institute of Standards & Tech.
183–187, Aug. 2007. [21] Jain and Meeran. S., “Deterministic job shop scheduling:
[2] G. Mosheiov, “Scheduling problems with a learning effect,” past, present and future,” European Journal of Operational
Eur. J. Oper.Res., vol. 132, no. 3, pp. 687–693, Aug. 2001. Research, Vol. 113, No. 2 (1999), pp.390-434.
[3] A Janiak and R. Rudek , “Experience-Based Approach to [22] Wu, D. (1987), “An Expert Systems Approach for the
Scheduling Problems With the Learning Effect”, IEEE Control and Scheduling of Flexible Manufacturing
Transactions on Systems, man, and Cybernetics—part a: Systems,” Ph.D. Dissertation, Pennsylvania State
systems and humans, vol. 39, no. 2, March 2009. University.
[4] W. C. Lee, Chin-Chia Wu” A note on single-machine group [23] Conway, R. and W. Maxwell (1967), Theory of Scheduling,
scheduling problems with position-based learning effect”, Reading, Massachusetts: Addison-Wesley.
Applied Mathematical Modelling 33 (2009) 2159–2163. [24] I Vermeulen · Sander Bohte · Koye Somefun Han La
[5] T.C. Edwin Cheng, Chin-Chia Wub, Wen-Chiung Lee Poutré”Multi-agent Pareto appointment exchanging in
“Some scheduling problems with sum-of-processing-times- hospital patient scheduling”, SOCA (2007) 1:185–196 DOI

© 2010 ACEEE
DOI: 01.IJRTET.2010.03.02.179