2013 Fourth International Conference on Digital Manufacturing & Automation

Based on Tabu Search and Particle Swarm Optimization algorithms solving Job
shop Scheduling optimization problems

Liang Xu1, Li Yanpeng1, Jiao Xuan2

1
Software Technology Institute, Dalian Jiaotong University, Dalian, Liaoning, 116028, China
2
School of Management, Dalian University of Technology , Dalian, Liaoning,116023, China
[email protected]

Abstract—Solving the Job shop Scheduling problem, the design search algorithm, reduce the dependence on initial solution
is based on Particle Swarm Optimization and Taboo Search to the tabu search, because of the tabu search algorithm of
which is a fast algorithm, And in this algorithm, bring in strong local search ability, it is very good to solve the
particle swarm strategy and taboo search strategy, A hybrid difficult to convergence ability of particle swarm
intelligence algorithm based on Particle Swarm algorithm and optimization later, so as to speed up the convergence speed,
the taboo Search algorithm(TS-PSO) is designed. It overcomes and improves the accuracy of convergence and improves the
particle swarm optimization algorithm in solving optimization performance of the algorithm.
combinatorial optimization problem, and better to avoid the
tabu search algorithm falling into local optimum, and III. THE REALIZATION OF TS-PSO
convergence speed has also been increased. Through particle
swarm and taboo search algorithm combined, the results show TS-PSO is presented according to job shop scheduling
that this algorithm has very good accuracy of convergence, problem, At the beginning of the hybrid algorithm, the
and is feasible, and compared with the traditional scheduling particle swarm algorithm is global search, at the end of the
algorithm, Embodies the obvious superiority . algorithm, the convergence of tabu search algorithm is
adopted to improve the local search.
Keywords-Particle Swarm Optimization; Taboo Search; Job
Shop Scheduling Problem; Hybrid scheduling A. The standard of updating the position and speed
According to the references[3][4] in the definition of the
I. INTRODUCTION speed and position, The speed of the particle is said that the
With the development of intelligent technology, more particle probability value, the variation of the speed and
and more bionic intelligent optimization algorithm has been position has the close relation, So the definition of speed
applied to the Job shop scheduling problem, and has got must limit in [0, 1] (Rand () function is in the programming
many good results. Because the single intelligent algorithm language). Take the vmax = 4 and c1 = c2 = 2, w as the inertia
has its own insurmountable defects, the hybrid intelligent weight, according to the literature [4], the inertia weight is
algorithm can compensate for each other's deficiencies, that affect the global and local search ability of PSO, The greater
composites and uses the Particle Swarm Optimization of w for global search, the smaller is advantageous to the
algorithm and tabu search algorithm, the advantages of the local search. TS - PSO algorithm in this paper, PSO
mixed tabu search of particle swarm optimization algorithm algorithm is used in the global search, so w set to a larger
is proposed. At the same time in order to further solve the value, namely the w value set to 1 in this article.
algorithm of the optimal solution, the two algorithms are B. The steps of TS-PSO
analyzed, designed the comprehensive algorithm based on
these two algorithms. So it overcomes the drawback of two Step 1 Initialize the parameters in the first place.
algorithms, realizes the complementary advantages, makes Speed and position of the particles in the initialization
the algorithm more vitality. parameter, or the w, c1, c2, vmax initialization, the
generation of random initial population, and give the
II. THE SYMPHYSIC THOUGHT OF TS-PSO initial value of each particle's position and speed.
The tabu search algorithm and particle swarm Parameter setting tabu search algorithm[5], set tabu table is
Optimization algorithm[1] fusion is the based on the idea of empty, and the largest number of iterations is given.
particle swarm algorithm and use the global optimal solution Step 2 Solving the fitness of each particle, the local
is introduced in to the tabu search algorithm. Because of the optimal solution of each particle set the P_best and
dependence of the tabu search algorithm for the initial value G_best’s global optimal solutions[6].
is bigger, at the beginning, the algorithm is lack of Step 3 Comparing the particles to adapt to the value of
information, resulting in a loss of the speed of convergence f (xi) and local optimal solution P_best, if f (xi) better than
and solving speed, and particle swarm algorithm[2] is simple P_best solution, makes P_best = f (xi), also need to
and easy to implement operation, good global search ability compare the current fitness G_best f (xi) and the global
is strong, but in the end its characteristics such as slow optimal solution, if the value of f (xi) than G_best better,
convergence speed, therefore particle swarm algorithm of makes G_best = f (xi).
the global optimal solution as the initial solution of tabu Step 4 Update the particle's velocity and position, and
calculate the adaptive value of updated particles.

978-0-7695-5016-9/13 $26.00 © 2013 IEEE 322

DOI 10.1109/ICDMA.2013.78
 Step 5 According to the above steps to determine F = min(max(Fr )), r = 1, 2, 3, …m (1)
whether achieve maximum number of iterations or local
search criteria, if accord with local search conditions to s.t. Bpq - Bp ( q1) + E pq ” 0, p = 1, …n; q = 1, …U i-1 (2)
achieve maximum number of iterations, you will get the
global optimal solution of G_best as initial solution of TS
Bp1 • 0ˈp = 1, 2, 3, …n (3)
Initialization, and then perform the first, otherwise go to step Wpqr = Wijr = 1ϨBij - Bpq + E ij ” 0 i  p (4)
3.
Step 6 From Initialization to produce a number of The objective function F is the latest completion machine
neighborhood solutions, tabu search, and adapt to the finished as soon as possible, and get the shortest processing
optimal values of some candidate solutions. time; Equation (2) is the only one operation can complete a
Step 7 Judgment for each candidate solution, whether it process before the next process is processed to do; Equation
fits the amnesty criterion, if fit the amnesty criterion, will (3) represents the beginning of the processing time’s first
meet the current rule of amnesty candidate solution instead process and any job is nonnegative; Equation (4) represents
of the current solution, and the replace of candidate any machine can only process one job.
solutions to join tabu table, finally put the candidate solution Table 1 gives the results of TS and PSO algorithm, The
instead of Initialization. table shows the TS can obtain the optimal a big part of the
Step 8 Judgment tabu search[7] process termination solution, But it takes a lot of running time, the operation
condition is satisfied, if meet the termination conditions the effect and the initial TS of solution have great relevance.
termination search TS, and enter the Initialization, the Table 1 gives the comparison of TS-PSO algorithm with the
optimal solution algorithm, if do not meet the termination standard FT and LA, indicating that this algorithm can find
conditions,then in step 6. the optimal solution in the most.
From the experimental results of TS-PSO algorithm
IV. TS-PSO SOLVES JSSP AND DOES EXPERIMENT comparing to TS and PSO algorithm which has been
ANALYSIS improved a lot.
According to TS - PSO algorithm , it is proposed in this
paper has chosen the different dimension of n * m standard
numerical example to show the effectiveness of the TS -
PSO algorithm. We use c + + programming in this paper, it
adopts the standard JSP FT classes and class LA test
problem.
In the JSSP[8],it needs satisfy: (1) Every task in the
machine processing has the same order, and we have to
know that. (2) Each machine has and only one task at one
time. (3) One task cannot be processed on different
machines at the same time. (4)It allows tasks to wait
between processes, and machines can be idle. (5) Each
working procedure is not allowed to interrupt. Completion
time of the minimum value of the workshop scheduling
problem of the optimization goal is to complete this task n
and find one of the optimal machining process, and making
the completion time is the minimum.
The order of the machine MM×MM(i, j) shows Processing
i part j the operation of the machine, MM(i, *) shows all
operations on the i’s workpiece in priority order processing
of all machine number arrangement.
Processing time matrix T×T(i, j) for the processing time
in the j on the i of machine workpiece.
The array AJ×AJ(i, j) for the i machine, j machining, AJ(i,
*) said the workpiece in the machine processing
arrangement.
So we solve to minimize the maximum completion time
for the task: making (p, q) represent job p and its the process
of q, Bpqand Epq Respectively show ( p, q ) processing the
starting time and the process of the completion time. Wpqr
shows that whether (p, q ) processes in the r machine, if (pˈ
q) processes in the r machine, Wpqr=1; otherwise, Wpqr=0. Fr Table 1. The results of TS-PSO and other algorithms
shows the time in the r machine, F is the objective function.

323
 Using the TS-PSO ,we get the convergence curves of [4] Jing Zhao, Cong-zhao Han-Bin Wei, Binary particle swarm
FT10(Fig 1)ˈFrom the Fig 1,we can get the message, when optimization with multiple evolutionary strategies. Science
China(Information Sciences), 2012. (in Chinese)
the number of iterations is 25,we get the optimal solution,
[5] Baykasoglu A, Ozbakir L, Sinmez A. Using Multiple Objeetive Tabu
and keep the same solution . so we can see that the TS-PSO Seareh and Grammars to Model and Solve Multi-objeetive Flexible
get the solution easily and the convergence speed is fast, this Job Shop Seheduling Problems. Journal Intelligent Man-
algorithm is viable. ufacturing,2004,15:777-785.
[6] Rao R S, Narasimham S, Ramalingaraju M. Optimization of
Distribution NetworkConfiguration for Loss Reduction Using
Artificial Bee Colony Algorithm s. International Journal of Electrical
Power and Energy Systems Engineering (IJEPESE),2008,1:709-715.
[7] F.Pezzella, G.Morganti, G.Ciaschetti. A genetic algorithm for the
Flexible Job-shop Scheduling Problem. Computers&Operations
Research,2008,35:3202-3204.
[8] J H Chen, S Y Ho. A Novel Approach to Production Planning of
Flexible Manufacturing Systems using an Efficient Muti-objective
Genetic Algorithm. International Journal of Machine Tools and
Manufacture,2005,45: 949-957.
[9] Si-fang Tong, Zeng-chuan Tong,Jun-jian Ma, Improved PSO
algorithm based on chaos theory and its application to design flood
hydrograph. Water Science and Engineering, 2010. (in Chinese)

Figure 1. The convergence curves of FT10

V. CONCLUSIONS
This paper uses a hybrid algorithm combining TS
algorithm and PSO algorithm, provides an effective way for
solving the JSSP problem. In this paper, we use the particle
swarm optimization algorithm in the initial stage of solving
JSSP, Because the accuracy of particle swarm optimization
algorithm is not very good, easy to fall into local premature
phenomenon , so at the end of the algorithm , we introduced
TS algorithm, the TS algorithm can avoid falling into local
optimal solution itself, and it used a tabu list, the tabu list
avoid to repeat searching. The use of the advantages of
PSO[9] and TS are two integration, Thus construct the TS-
PSO algorithm, This algorithm shows some advantages in
solving the JSSP problem. So this paper presents the TS-
PSO algorithm can solve the job shop scheduling problem, It
is a kind of scheduling for job shop good optimization
method.
VI. ACKNOWLEDGMENT
I would like to express my sincere gratitude to National
863 Plan Project (2012AA041402-4), Liaoning Province
Science and Technology Plan Project (20102017), Dalian
Science and Technology Plan Project(2010J21DW009),
which provide financial aid and help for this paper.
REFERENCES
[1] Rong-Hwa Huang, Chang-Lin Yang, Ant colony system for job shop
scheduling with time windows, The International Journal of
Advanced Manufacturing Technology,2007,9:2010-2018.(in Chinese)
[2] Jun-Zhan, Zhen-Wu, Chun-tian Cheng, Improved particle swarm
optimization algorithm for multi-reservoir system operation, Water
Science and Engineering, 2011. (in Chinese)
[3] Ellips Masehian, Davoud Sedighizadeh. Multi-objective robot motion
planning using a particle swarm optimization model. Journal of
Zhejiang University-Science C(Computers & Electronics), 2010.

324