Upload
jj-yuan
View
215
Download
2
Embed Size (px)
Citation preview
European Journal of Operational Research 158 (2004) 525–528
www.elsevier.com/locate/dsw
Short Communication
A note on the single machine serial batchingscheduling problem to minimize maximum lateness
with identical processing times
J.J. Yuan a,b, A.F. Yang a,b, T.C.E. Cheng b,*
a Department of Mathematics, Zhengzhou University, Zhengzhou, Henan 450052, People�s Republic of Chinab Department of Management, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong, People�s Republic of China
Received 29 August 2002; accepted 5 May 2003
Available online 2 August 2003
Abstract
We consider the single machine, serial batching scheduling problem 1jprec; pj ¼ p; s-batch; rjjLmax. The complexity ofthis problem is reported as open in the literature. By reducing this problem to the version without precedence con-
straints, we show that the problem is polynomially solvable.
� 2003 Elsevier B.V. All rights reserved.
Keywords: Scheduling; Precedence constraints; Batches
1. Introduction and problem formulation
Let n jobs J1; J2; . . . ; Jn and a single machinethat can handle only one job at a time be given.
There are precedence relations � between the jobs.Each job Jj has a release date rj, a processing timepj and a weight wj. The jobs are processed inbatches. The completion time of all jobs in a batch
is defined as the finishing time of the last job in the
batch, i.e., the processing time of a batch is equal
to the sum of the processing times of all jobs in the
batch. If a new batch is formed, a constant setup
time s will be incurred. If Ji and Jj are two jobs
* Corresponding author. Tel.: +852-27665215; fax: +852-
23645245.
E-mail address: [email protected] (T.C.E. Cheng).
0377-2217/$ - see front matter � 2003 Elsevier B.V. All rights reserv
doi:10.1016/S0377-2217(03)00361-8
such that Ji � Jj, we only require that Ji is pro-cessed before Jj, and so it is allowed that Ji and Jjare processed in the same batch. Jobs cannot start
before their release dates. We assume that the
starting time of the setup of each batch cannot be
before the maximum release date of the jobs in the
batch. Hence, the starting time of a batch B is atleast maxJj2B rj þ s. Following [3,6], this model iscalled the serial batching scheduling problem and
is denoted by
1jprec; s-batch; rjjf ;
where f is the objective function to be minimized.Coffman et al. [5] provide an algorithm to solve
the problem 1js-batchjP
Cj in Oðn log nÞ time.Albers and Brucker [1] provide an Oðn2Þ timealgorithm to solve the problem 1jprec; pj ¼ p;
ed.
526 J.J. Yuan et al. / European Journal of Operational Research 158 (2004) 525–528
s-batchjP
Cj, and prove that the problems
1js-batchjP
wjCj and 1jchains; pj ¼ 1; s-batchj �PwjCj are unary NP-hard. Webster and Baker [8]
give an algorithm to solve the problem 1js batchjLmax in Oðn2Þ time. Ng et al. [7] reduce the problem1jprec; s-batchjLmax to the problem 1js-batchjLmaxand show that the problem 1jprec; s-batchjLmax canbe solved in Oðn2Þ time by using a revised Web-ster–Baker algorithm. Recently, Baptiste [2] gives
an algorithm to solve the problem 1jpj ¼ p;s-batch; rjjLmax in Oðn14 log nÞ time. Brucker andKnust [4] report that the complexity of the prob-
lem 1jprec; pj ¼ p; s-batch; rjjLmax is still open.We show in this paper that, by appropriately
modifying the release dates and due dates of the
jobs, the problem 1jprec; pj ¼ p; s-batch; rjjLmaxcan be polynomially reduced to the problem
1jpj ¼ p; s-batch; rjjLmax in Oðn2Þ times. Conse-quently, the problem 1jprec; pj ¼ p; s-batch; rjjLmaxis polynomially solvable.
2. Reduction
For the problem 1jprec; pj ¼ p; s-batch; rjjLmax,a feasible schedule (solution) is a pair ðp;BSÞ suchthat p ¼ ðpð1Þ; pð2Þ; . . . ; pðnÞÞ is a sequence of jobindices satisfying the precedence constraints, i.e.,
for any pair of jobs Ji and Jj, if Ji � Jj, thenp1ðiÞ < p1ðjÞ, and BS ¼ fB1;B2; . . . ;Bkg is abatch sequence that partitions the set of all jobs
such that, for each index t with 16 t6 k,fj : JpðjÞ 2 Btg consists of consecutive integers, i.e.,there are two integers x and y with y ¼ xþ jBtj 1such that Bt ¼ fJpðxÞ; Jpðxþ1Þ; . . . ; JpðyÞg. The startingtime of a batch B under a schedule ðp;BSÞ is de-noted by SBðp;BSÞ. The completion time of a jobJj under a schedule ðp;BSÞ is denoted by Cjðp;BSÞ(or simply Cj when there is no ambiguity).
If Ji and Jj are two jobs such that Ji � Jj, thenthe starting time of the batch that contains Jj mustbe at least ri under any feasible schedule. Modi-fying the release date of each job Jj by setting
r0j :¼ maxfrj; rig;
we see that the maximum lateness will not change
under any feasible schedule. Hence, we can re-
cursively modify the release dates of the jobs such
that, for each pair of jobs Ji and Jj, if Ji � Jj, thenr0i 6 r0j. This procedure can be done in Oðn2Þ timeby the standard ‘‘Algorithm Modify rj’’ in thebook by Brucker [3]. The modified release dates
obtained above can be denoted in the followingform:
r0j :¼ maxfri : Ji � Jj or i ¼ jg for 16 j6 n:
Again, if Ji and Jj are two jobs such that Ji � Jj,then the completion time of Ji must be less than orequal to that of Jj under any feasible schedule.Modifying the due date of job Jj by setting
d 0i :¼ minfdj; dig;we see that the maximum lateness will not change
under any feasible schedule. In fact, if ðp;BSÞ is afeasible schedule, thenCiðp;BSÞ6Cjðp;BSÞ, and somaxfCiðp;BSÞ di;Cjðp;BSÞ djg
¼ maxfCiðp;BSÞ d 0i ;Cjðp;BSÞ djg;
i.e., the maximum lateness of Ji and Jj is notchanged after we modify the due date di of Ji to d 0
i .Hence, we can recursively modify the due dates of
the jobs such that, for each pair of jobs Ji and Jj, ifJi � Jj, then d 0
i 6 d 0j. This procedure can be done in
Oðn2Þ time by the standard ‘‘Algorithm Modify
dj’’ in the book by Brucker [3]. The modified duedates obtained above can be denoted in the fol-
lowing form:
d 0i :¼ minfdj : Ji � Jj or i ¼ jg for 16 i6 n:
The above discussion leads to Lemma 2.1:
Lemma 2.1. Any optimal schedule of the problem
1jprec; pj ¼ p; s batch; r0j; d0jjLmax under the modi-
fied release dates and due dates is also an optimal
schedule of the problem 1jprec; pj ¼ p; s batch;rjjLmax under the original release dates and due
dates.
For every job subset J fJ1; J2; . . . ; Jng, a jobJi 2 J is called a minimal job of J if Ji has nopredecessor in J.The general idea of the procedure constructing
a feasible schedule for the problem 1jprec; pj ¼p; s-batch; rjjLmax from an optimal solution for theproblem 1jpj ¼ p; s-batch; r0j; d
0jjLmax can be de-
scribed as follows.
J.J. Yuan et al. / European Journal of Operational Research 158 (2004) 525–528 527
Let ðp;BSÞ be an optimal schedule for the
problem 1jpj ¼ p; s-batch; r0j; d0jjLmax under the
modified release dates and due dates. If p satisfiesthe precedence relations � between the jobs, then
the schedule ðp;BSÞ is also optimal for the versionwith precedence relations, and thus, we havenothing to do. Otherwise, let JpðiÞ be the first job
under p such that JpðjÞ � JpðiÞ for some j > i. LetJp�ðiÞ be the minimal job of the job subset
fJpðjÞ : j > i and JpðjÞ � JpðiÞg:
Since Jp�ðiÞ � JpðiÞ, we must have r0p�ðiÞ 6 r0pðiÞ andd 0
p�ðiÞ 6 d 0pðiÞ. Hence, by exchanging the positions of
jobs JpðiÞ and Jp�ðiÞ in the schedule ðp;BSÞ, thecompletion times of JpðiÞ and Jp�ðiÞ under the new
schedule are not greater than the completion times
of Jp�ðiÞ and JpðiÞ under ðp;BSÞ , respectively, and sothe maximum lateness of the jobs JpðiÞ and Jp�ðiÞ is
not increased under the new schedule. By noting
that the completion time of any job apart from JpðiÞand Jp�ðiÞ is not increased under the new schedule,
the resulting new schedule is still optimal for
1jpj ¼ p; s-batch; r0j; d0jjLmax. This procedure is re-
peated until a schedule ðp�;BS�Þ satisfying theprecedence relations � between the jobs is ob-
tained.
Formally, given any optimal schedule ðp;BSÞfor the problem 1jpj ¼ p; s-batch; r0j; d
0jjLmax, we can
recursively obtain a feasible schedule ðp�;BS�Þ forthe original problem in the following way:
(i) If Jpð1Þ is a minimal job of fJ1; J2; . . . ; Jng, thenset Jp�ð1Þ ¼ Jpð1Þ, and the schedule is not chan-
ged. Otherwise, let Jp�ð1Þ be a minimal job of
J1 ¼ fJj : Jj � Jpð1Þg. Then we exchange thepositions of jobs Jp�ð1Þ and Jpð1Þ in the job se-quence p and batch sequence BS. The presentschedule is denoted by ðp1;BS1Þ.
(ii) Suppose that ðp1;BS1Þ; ðp2;BS2Þ; . . . ; ðpi;BSiÞand also Jp�ð1Þ; Jp�ð2Þ; . . . ; Jp�ðiÞ (16 i6 n 1)have been obtained. If Jpiðiþ1Þ is a minimal
job of fJpiðiþ1Þ; Jpiðiþ2Þ; . . . ; JpiðnÞg, then set
Jp�ðiþ1Þ ¼ Jpiðiþ1Þ, and the schedule is not chan-
ged. Otherwise, let Jp�ðiþ1Þ be a minimal job ofJiþ1 ¼ fJpiðjÞ : j > iþ 1 and JpiðjÞ � Jpiðiþ1Þg.Then we exchange the positions of jobs
Jp�ðiþ1Þ and Jpiðiþ1Þ in the job sequence pi and
batch sequence BSi. The present schedule is
denoted by ðpiþ1;BSiþ1Þ.(iii) Set ðp�;BS�Þ ¼ ðpn;BSnÞ:
Clearly, the creation of the schedule ðp�;BS�Þtakes at most Oðn2Þ time. By the definition ofðp�;BS�Þ and the above discussion, we can easilysee the correctness of the following lemma.
Lemma 2.2. ðp�;BS�Þ is an optimal schedule for the
modified problem
1jpj ¼ p; s batch; r0j; d0jjLmax
under the modified release dates and due dates.
Furthermore, if Jp�ðiÞ and Jp�ðjÞ are two jobs such that
Jp�ðiÞ � Jp�ðjÞ, then i < j.
Combining Lemmas 2.1 and 2.2, we deduce that
ðp�;BS�Þ is an optimal schedule for the originalproblem 1jprec; pj ¼ p; s-batch; rjjLmax. Hence,
the problem 1jprec; pj ¼ p; s-batch; rjjLmax can bepolynomially reduced to the problem 1jpj ¼ p;s-batch; rjjLmax in Oðn2Þ time.Recall that Baptiste [2] gives an Oðn14 log nÞ
algorithm for the problem
1jpj ¼ p; s-batch; rjjLmax:This implies that the problem 1jprec; pj ¼ p;s-batch; rjjLmax can be solved in Oðn14 log nÞ time aswell.
3. Conclusions
In this paper, we have studied the single
machine, serial batching scheduling problem
1jprec; pj ¼ p; s-batch; rjjLmax. The complexity ofthis problem is reported as open in the literature.
By reducing this problem to the version without
precedence constraints, we show that the problem
is solvable in Oðn14 log nÞ time. Since the highorder of the polynomial is not practical, any
improvement in the complexity is interesting.
Acknowledgements
This research was supported in part by The
Hong Kong Polytechnic University under the
528 J.J. Yuan et al. / European Journal of Operational Research 158 (2004) 525–528
ASD in China Business Services. The first author
was also supported in part by the National Nat-
ural Science Foundation of China under the grant
number 10371112.
References
[1] S. Albers, P. Brucker, The complexity of one-machine
batching problems, Discrete Applied Mathematics 47 (1993)
87–107.
[2] P. Baptiste, Batching identical jobs, Mathematical Methods
of Operations Research 52 (2000) 355–367.
[3] P. Brucker, Scheduling Algorithms, Springer Verlag, Berlin,
2001.
[4] P. Brucker, S. Knust, Complexity results for scheduling
problems, http://www.mathematik.uni-osnabrueck.de/re-
search/OR/class/, 2002.
[5] E.G. Coffman, M. Yannakakis, M.J. Magazine, C. Santos,
Batch sizing and sequencing on a single machine, Annals of
Operations Research 26 (1990) 135–147.
[6] J.K. Lenstra, A.H.G. Rinnooy Kan, P. Brucker, Complexity
of machine scheduling problems, Annals of Discrete Math-
ematics 1 (1977) 343–362.
[7] C.T. Ng, T.C.E. Cheng, J.J. Yuan, A note on the single
machine serial batching scheduling problem to minimize
maximum lateness with precedence constraints, Operations
Research Letters 30 (2002) 66–68.
[8] S.T. Webster, K.R. Baker, Scheduling groups of jobs on
a single machine, Operations Research 43 (1995) 692–
703.