Vous êtes sur la page 1sur 2

This is a summary of the paper Fast Simulated Annealing hybridized with Quenching for solving job

shop scheduling problem by Kashim Akraf, Khurram Kamal, and Alam Zeb that published in Applied Soft
Computing Journal 49 (2016) 510-523.

In this paper, Akraf begins by explaining Job Shop scheduling problem (JSSP) which is a well-known NPhard combinatorial optimization problem. Because this problem is categorized as NP-hard, considering
efficiency of time needed to get global optimal solution, it is reasonable for the author to choose
approximation method rather than exact method to optimize the solution.

Scheduling objective in this paper is minimization of makespan in Job Shop type scheduling (in this case
is static/offline JSSP). Makespan is the total amount of time needed to complete a group of jobs. In this
paper, permutation with repetition proposed by Bierthwirth is used for solution representation.
Candidate solution is represented by 1 dimensional array of number that shown a sequence of
operation (for all jobs) performed during makespan. By using simple calculation, we can calculate
makespan for every combination of sequence operations. But because sequence of operations for a
makespan is not really happen in 1 row (some operations can be happen at the same time in different
machine), we cant calculate the makespan by just sum directly processing time of all operations in a
sequence. The paper explains an easy step to calculate the makespan from the sequences in Fig 3c.

The authors proposed a novel hybrid approach that called HFSAQ (Hybridize Fast Simulated Annealing
with Quenching) to solve JSSP. Fast Simulated Annealing (FSA) is a variant of Simulated Annealing which
use different cooling approach (Inversely linear temperature instead of geometric cooling schedule) and
probability density function (Cauchy PDF instead of Boltzmann PDF). FSA use Inversely linear
temperature cooling schedule because this method is much faster than geometric cooling rate. Cauchy
PDF is used because it more sensitive to changes in fitness of current neighboring solution than
Boltzmann PDF, therefore FSA is less likely stuck in local optima. Quenching is used to perform local
search in near vicinity of current solution. Quenching cycle will be invoked if there is no improvement in
best solution made by SA over a predefined number of iterations. The authors also add Tabu List
mechanism in their implementation, this approach will block the candidate solution if it founded in Tabu
List (exceptional for candidate solution with makespan better than current best solution, aspiration
criterion rule). Candidate solution (Neighborhood) is generated using generation methodology proposed
by Nowicki and Smutnicki. Based on their research, makespan of existing schedule can be reduced by

swapping operations which on critical path. The authors of this paper use algorithm proposed by CruzChavez to estimated path during makespan that can be categorized as critical path.

Based on the authors result of experiments, HFSAQ is taken slightly more time compare than FSA, but
found more global optimum solutions. About time computation that longer than FSA is not something
surprising, due to Quenching mechanism used in this paper (maximum iteration at one temperature is
5000). The authors run the experiments for problem LA01 to LA30, FT06, FT10, FT20, ORB01, to ORB10
and ABZ5 to ABZ9 20 times each and the result is HFSAQ is able to solve 44 out of 48 problems optimally,
with 44 zero ARD values (in the paper it said only 25 zero ARD values, but when we count from the table
2a actually it should be 44, maybe the author miss detail about this), 0 ARD values indicate global
optimal solution was found in every run. Even the worst solution of HFSAQ is better than the best
solution of FSA for the entire solved problem. This result shown HFSAQ is outperformed FSA in terms of
effectiveness to get the optimal solution still in reasonable time.

Vous aimerez peut-être aussi