NN help please
enas
please I need help very ergent.
I'm dealing with a scheduling jobshop problems using neural network .
Jobshop scheduling problem is a problem in shopfloor in factories and
its description is:
consists of
- a set of machines.
- a collection of jobs to be scheduled, where each job must pass
through each machine once only.
each job has its own processing order and this may bear no relation to
the processing order of any other job.
each job should be processed through the machines in a particular
order(precedence constraints)
and in neural using multi layered, backpropagation. , your input is
the description of the problem, and your desired is the sequences of
jobs on each machine after scheduling using some algorithms.
in literature there is different type of how coding the problem to
before entering the neural algorithm like:
example:
job processing order(processing time)
1 1(3) 2(3) 3(2)
2 2(3) 3(1) 1(4)
after scheduling using SPT heuristic:
Machine jobs sequence on machine
1 1,2
2 2,1
3 2,1
Data encoding:
1- Input:(X1, X2,....,X2n) where X2i-1 is the processing order of the
ith job, X2i is the processing time of the ith job, n is the number of
jobs
Output (Y1, Y2,...,Ym) shows the processing order of a waiting job on
a machine j , m the number of machines.
according to this is I enter the data as follows:
J1Order J1 Pt j2 order J2Pt m/c1 m/c2 m/c3
123 332 231 314 12 21 21
..
..
..
2-Input processing time of job on a various machine, second input is
the precedence order of for this job based on the branch tree concept
as 1 represents the precedence order of machines 123, 2 represnts 132
sequence
Output corresponds to the job sequence on every machine applying
branching tree 1 represents sequence of job s 12, 2 corresponds to
sequence of jobs 21
I enter to neural
332 1 314 4 1 2 2
..
..
3-putting a description of operation attributes for example
inputfeatures : operation(operation s clusstered into 3 classes:{1}
first, {2} Middle, {3} last),
process time(Short, Medium, Long),
remaining time(denoates thesum of processing times for the remaining
operations of that job, remaining time ranges from 0 to 8 , 3 classes
{0-2.67} Short, {2.67-5.33}Middle, {5.33-8.67} Long ), machine
load( Low and high)
Output the description of the position of the job on the
machine(Zero , one)
so I enter the description of each operation in the problem in one row
operation pt rt ml output
first Medium Long high zero
..
As above I etryed these different types of coding but the result is
sooooooooooooooo bad the error in testing is 6 digits, the net doesn't
learn atall in every method the error in testing is high and the
learning is trapped in local minima.
i tried problems of 6 jobs and 6 machines
2 jobs and 2 machines
3 jobs, 3 machines
I even created from 100-1000 different problems but with the same
range of processing time say (1-9) and even other ranges.
I tried to fix the number of jobs that I create of the problem
and of course different lr and momentum, one layer, 2 layer.
and other methods of describing and coding the problems
I've been 2.5 years in this and my master has to come to an end, and
my supervisiors will exclude me.
please help what to do please? or anyone can give me thoughts or try
with me.
Thanks in advance.
Enas Esh