co @® Palmer's algorithm (1965)
— (Idea) Observations from Johnson's algorithm.
— (1) The jobs that appear in the front in optimum schedule are the jobs that have
longer processing times at later machines.
— (2) The jobs that appear in the back in optimum schedule are the jobs that have
shorter processing times at later machines.
— Assign indexes for each job.
% The index values are designed such that the jobs may have higher values if it has longer processing
times at later machines = Assign jobs with higher values to the front in a schedule.
— Slope index $,
— Let
IM = VE jy FON = BE yy Hoe
=(m=3)t;.—On= ty,
— Then the permutation schedule is 5,,, > Sis, =.= Syay +
% According tothe experiments condueted on 1580 problems( n<6, m<10 ), we found optimum for
about 30% of the cases(Based on. Dannenbring, 1977, Mgt. Science. V.23, pp1174-1182)Heuristics for Minimization of Makespan for (Fs||Cmax) Problem
Palemer’s Heurisitic:
This heuristic comprises two steps as follows.
Step 1: For n job and m machine static flow shop problem, compute slope Aj for j"
job as follows;
m
-¥ {m ~(2i-1)}py
ist
Step 2: Order the jobs in the sequence based on descending (decreasing) order of Aj
values.
AExample 4.3
Solve F;|| Cimx problem for the data shown in Table using Palmer's heuristic.
Machines! ji be ds je
M 6 8 3 4
M 5 1 3 4
Ms 4 4 4 2Solution
a
> tm -@-D} py =-Y B-Ci-Dhpy.
‘a »
Table 4.1 Calculation for the job's slope.
2p nse
Z 3-(2x2-1)=0 3-(2x3-1) = -2
Job j Pa Px Aj
1 6 3 4 4
2 8 1 4 8
3 3 5 4 2
4 4 4 2 4+
Arranging slope values in descending order: there are two sequences;
Sequence 1 = { js. jrjs-i2}
Sequence 2= { js.js.ir siz}Directed graph for sequence | is;
Process time js
Figure 3.3 Directed graph for seq. { j3 . j1. is. i2 }Directed graph for sequence 2 is;
is is k jt
Proce:
MCG
Completion Time
GD
Ms Ga +) Gd)
Me Gi) + G)—Gay-
Figure 3.4 Directed graph for sequence { js, js. jn «iz }
»
Conclusion: Note the Cyyx=26 for both sequencesa ® Capmbell, Dudek & Smith(1970)'s:
Sometime called the " Heuristic nN
— Recall that applying Johnson's Algorithm blindly to 3 m/c problem as if it were a
2mic problem with 4 worked as a good heuristic.
— CDS Algorithm we convert n job m machine
problem into (m-1) sub-problems and then select the best one among these (m-1)
schedules.
= Sayset Qty, B= $y. 0 m1 +then apply Johnson's 2 m/e
Algorithm 7
— According to Dannenbring's experiment(1977), the algorithm found optimal
solutions for 55% of cases for 1580 problem sets. Furthermore, the possibility of
finding an optimal solution is large by a simple neighborhood search
method(say, from 1, 2, 3, ...m, we get 2,1, 3, um OF 1, 3,2, a)