Professional Documents
Culture Documents
Two-Stage Method and Application for Molten Iron Scheduling Problem between
Iron-Making Plants and Steel-Making Plants
Hui Huang*, Tianyou Chai*,**, Xiaochuan Luo*, Binglin Zheng*, Hong Wang*,***
* State Key Laboratory of Integrated Automation for Process Industry (Northeastern University),
Shenyang, China, 110819 (Tel: +86-24-83678332; e-mail: huanghui1978@sohu.com).
** National Metallurgical Automation Engineering Technology Research Center, Northeastern University,
Shenyang, China, 110819 (e-mail: tychai@mail.neu.edu.cn)
*** Control Systems Centre, School of Electrical and Electronic Engineering, the University of Manchester,
P O Box 88,Manchester M60 1QD, UK
Abstract: In this paper, a two-stage method is proposed to solve the molten iron scheduling problem
(MISP). Firstly, a 0-1 mixed-integer nonlinear programming model is established for special type of
molten iron scheduling problem (STMISP). A method to solve mathematical model of STMISP is
presented, which include assigning machine strategy of STMISP and determining mathematical model of
eliminating machine conflicts. Secondly, on the basis of results of STMISP, a heuristic algorithm based
on the rule of first come first serve (FCFS) principle is established to solve normal type of molten iron
scheduling problem (NTMISP). The molten iron scheduling system with the two-stage method has been
successfully applied to Shanghai Baosteel Company to realize quickly optimal molten iron scheduling.
Application results show that the two-stage method is both effective and feasible.
Keywords: Molten iron scheduling problem, Two-stage method, Hybrid flow shop, Mixed-integer
nonlinear programming model, Heuristic algorithm.
problem. However, the focus was not on the molten iron practical potentials. It plays an important role in increasing
scheduling problem. Due to the complexity of MISP, production efficiency and reducing production cost.
research into the MISP has not only theoretic value, but also
B la s t F u r n a c e s m p ro cesses
R e la d lin g (R L )
(B F ) M o lte n ir o n p r e tr e a tm e n t s ta tio n s
s ta tio n s
1 # P R s ta tio n 1 # S P s ta tio n 1 # P O s ta tio n
1# B F 1#PR 1#SP 1# PO 1# R L Mj
m a c h in e s
...
...
...
...
1
1 1 1 M #RL
M # PR
q M #SP
p M h # PO d
………...
n p o ts
...
...
...
...
S # P R s ta tio n S # S P s ta tio n S # P O s ta tio n
1# PR 1#SP 1# PO 1# R L
...
...
...
...
M 1# BF S M S
#RL
M q # PR M Sp # S P M S
h # PO d
9477
18th IFAC World Congress (IFAC'11)
Milano (Italy) August 28 - September 2, 2011
determines in what sequence, at what time and on which determined by the state and residual capacity of pre-
device molten iron should be arranged at various logistics processing and post-processing machines. This means that
stages from receiving iron to reladling. the solution to the MISP is very difficult. In addition, on the
molten iron production site, the requirements of real-time
As it can be seen from Fig.1, molten iron logistics process scheduling is very high, the average response time of
has both the features of parallel flow shop and hybrid flow obtaining scheduling plan is 10 seconds. This paper presents
shop. At the molten iron pretreatment and reladling stages, a two-stage method to solve MISP. Firstly, a 0-1 mixed-
there are two parallel production lines. In addition, there are integer nonlinear programming model is established for
many parallel devices in various processes of each STMISP, and a method including assigning machine
production line. Therefore, the MISP is neither parallel flow strategy and eliminating machine conflicts is presented to
shop scheduling problem, nor is hybrid flow shop solve the model of STMISP. Secondly, on the basis of
scheduling problem. It is called parallel hybrid flow shop results of STMISP, a heuristic algorithm based on the rule of
scheduling problem (PHFSSP) in this paper. The solution to FCFS is established to solve NTMISP.
PHFSSP is not only to determine jobs processed in each
production line, but also to determine on which device jobs
should be arranged in various processes of each production 3.1 Mathematical model and solution method of STMISP
line, and then to determine processing sequence and the start
time of jobs in various units. Since the parallel flow shop 3.1.1 Mixed integer nonlinear programming model of
and hybrid flow shop scheduling problem have been proved STMISP
to be NP-hard problem, it is clear that the PHFSSP is also a The following symbols are used to define the problem
NP-hard problem.
parameters and decision variables.
With reference to the solution steps of parallel flow shop
Parameters
scheduling problem, the procedure for solving PHFSSP can
be divided into two steps. In the first step, the jobs are Mj the total number of parallel machines of unit j ;
allocated to each production line, where the production
sequence of jobs is determined. In the second step, the m i
s
the initialization unit number of charge i , mis ≤ mi ;
devices of jobs are determined in various processes of each ptijk the processing time of pot i on machine k of unit
production line, and the production time of jobs is
j;
determined. Therefore, the procedure for solving MISP is
accordingly divided into two steps. Firstly, the processing t i
j , j +1
the transportation time of pot i from the machine
route of molten iron is determined in accordance with the of unit j to the machine of unit j + 1 ;
indices of molten iron, such as time, weight and composition
etc. Secondly, the production devices and processing time of etijk the end time of pot i on the machine k of unit j ;
pots of molten iron are determined, thus the scheduling plan
of pots of molten iron is generated. This paper will focus on SI(i, j, k) the immediate successor pot of pot i processed on
the second step. MISP can be simplified as hybrid flow shop the machine k of unit j ;
scheduling problem after determining molten iron Ti RL the target starting time of pot i on the machine of
processing route. In addition, as the capacity of pre- and
reladling unit;
post-processing machines is finite, the process paths of pots
Ti BF the starting time of pot i on the unit of receiving
of molten iron are subjected to uncertainties. Therefore, the
difficulty of MISP is further increased. iron;
C1ijk the coefficient of penalty cost for the waiting time
3. TWO-STAGE METHOD FOR MISP of pot i after finishing process on the machine
MISP can be solved in two stages according to the different k of unit j ;
compositions and types of molten iron. Molten iron can be C2i the coefficient of penalty cost for the processing of
divided into two categories according to their component pot i being late with respect to the targeted
types. One of categories is the special composition type of starting time of pot i on the machine of
molten iron, such as silicon steel type of molten iron and reladling unit;
demanganization type of molten iron. Another category is l the number of the intervals of equipment available
normal composition type of molten iron. Special
composition type of molten iron has high production quality time, 1 ≤ l ≤ q , where qk is the total interval
k
requirement and high production priority. It must be number of equipment available time;
processed by all the processes, so it needs to be scheduled
firstly. As processing capacity of pre-processing and post- US kl the starting time of the l th interval of available
processing machines is finite, the normal composition type time of machine k ;
of molten iron may be not processed by either pre- UE kl the end time of the l th interval of available time
processing or post-processing machines. Whether normal
composition type of molten iron is processed by pre- of machine k ;
processing machines or post-processing machines is U a sufficiently large positive integer.
9478
18th IFAC World Congress (IFAC'11)
Milano (Italy) August 28 - September 2, 2011
Decision variables processed at the receiving iron stage, assign n pots to m units
and find the optimal production path. In other words, assign
xijk the variable of model denotes the starting time of each pot to all the machines, and determine the processing
pot i on the machine k of unit j , sequence of each pot in the same machine. The aim is to
ensure the conflicting time on each machine minimum so as
yijk 1 if pot i is arranged on the machine k of unit j , to obtain a better initial schedule. The core of assigning
0 otherwise. machine strategy is how the machine can be chosen for the
Using the above notations, the MISP can be formulated as pot when there are other pots assigned at the corresponding
the following 0-1 mixed-integer nonlinear programming time periods in each unit.
model.
n m−1 M j (a) Fore conflict (b) Hind conflict
Min ∑∑∑C1ijk (xi( j+1)k′ − xijk − ptijk − t ij, j+1 )
i =1 j =1 k =1
(c) Inside inclusive conflict (d) Outside inclusive conflict
n M5
⎫
+ ∑∑∑ C 2 i ( xijk − Ti DG
)⎬ (1)
i =1 j = 5 k =1 ⎭ (e) Complete conflict (f) Mixed conflict
S .T . Processing time which has been occupied in the machine
Processing time of pots of molten iron
M
i = 1,L, n; j = 1,L, m
j
∑
k =1
y ijk = 1 (2) Fig. 3. The types of time conflict on machines
et ijk y ijk = x ijk y ijk + pt ijk y ijk The types of possible time confliction on machines are
shown in Fig.3. To describe the assigning machine strategy,
i = 1,L, n; j = 1,L, m; k = 1,L, M j (3) we first define the confliction function as follows:
xSI(i, j,k) jk ySI(i, j,k) jk −etijk yijk +U(2 − ySI(i, j,k) jk − yijk ) ≥ 0 ⎧ min( et ijk , et rjk ) − max( x ijk , x rjk ),
⎪ (10)
i, SI(i, j, k) = 1,L, n; j = 1,L, m; k = 1,L, M j (4) f j (i, r , k ) = ⎨ [ x ijk , et ijk ] I [ x rjk , et rjk ] ≠ Φ
⎪
xi ( j+1)k′ yi ( j+1)k′ − etijk yijk + U (2 − yi ( j+1)k′ − yijk ) ≥ t ij , j+1 ⎩ 0 , other
3.1.2 Method to solve mathematical model of STMISP The 0-1 variable yijk of the model of STMISP was solved
For solving the mixed integer nonlinear programming model by assigning machine strategy. The transportation time
of STMISP, the method includes two parts. t ij , j +1 and processing sequence of pots are determined. Then,
(1) Assigning machine strategy of STMISP constraint (3), (4) and (5) can be expressed as follows:
et ijk = x ijk + pt ijk i = 1, L, n; j = 1, L, m; k = 1, L, M j (11)
The problem of machine assignment of STMISP can be
described as this: based on the starting time of pots xSI(i, j,k) jk −etijk ≥ 0 i, SI(i, j, k) = 1,L, n; j = 1,L, m;k = 1,L, Mj (12)
9479
18th IFAC World Congress (IFAC'11)
Milano (Italy) August 28 - September 2, 2011
xi( j +1)k′ − etijk ≥ t ij , j +1 i =1,L, n; j =1,L, m −1; k, k′ =1,L, M j (13) allowable waiting time, the operation time of the
conflict pot is adjusted backward.
Under the assumption of no changing processing sequence Step6: Judge the type of the conflict, if it is a hind conflict,
of pots, the mixed integer nonlinear programming model of and the conflict time is longer than the allowable
STMISP can be simplified to give: waiting time, then the current pot does not select the
Min (1) current machine. If the conflict time is less than the
S.T. (6)-(9),(11),(12),(13) allowable waiting time, then the operation time of
It can be seen that the above eliminating machine conflicts the current pot is adjusted backward. If it is a fore
can be solved using the well-known linear programming (LP) conflict, and the conflict time is longer than the
method. allowable waiting time, the current pot does not
select the current machine. If the conflict time is less
3.2 Heuristic algorithm of NTMISP than the allowable waiting time, then the operation
time of the conflict pot is adjusted backward.
Since the capacity of pre- and post-processing machines is Step7: After above adjustment on the operation time of the
finite, not all of pots of normal type of molten iron can be current pot, if the current pot does not give up
handled by the pre- and post-processing machines in the selecting machine of the current process, then we
planning cycle. Thus only part of pots of normal type of select the machine in which the start time of current
molten iron can be processed. Which pot of molten iron can pot is the earliest in the current process. Meanwhile
be processed is determined by whether there is idle machine the operation time of the current pot in this machine
when it arrives at the pre- or post-processing process. is determined. If the current pot gives up selecting all
Therefore, pots of normal type of molten iron are required to machines of current process, then the next pot is
be determined for their treatment paths in the scheduling scheduled according to Steps 2-7.
process. Meanwhile, the processing machine and the start Step8: Check whether all pots have been arranged in the
and end times of the machine in each process are determined. current process. If they are, go to Step 9. Otherwise,
By summarizing the practice experience of molten iron go to Steps 2-7 and schedule other pots.
scheduler, this paper presents a heuristic algorithm based on Step9: Schedule the next process according to Steps 2-8.
the rule of first come first serve (FCFS) bases to solve the Stop until all pots of normal type of molten iron have
scheduling problem of normal type of molten iron. The been arranged in all processes.
algorithm is as follow.
4. APPLICATION
Step1: In the planning cycle, the pots of normal type of
molten iron are arranged in accordance with the 4.1 Background
order of receiving iron, pre-processing,
desulphurization, post-processing and reladling. The Shanghai Baosteel Company of China is a large iron
Step2: Compute the start/end time of pots in all machines of and steel enterprises. The treatment process of molten iron in
the current process, and sort the scheduling this company is comprehensive and includes all kinds of
according to the start time of pots in each machine. typical treatment processes of molten iron. Baosteel
Company has large molten iron treatment equipments and
Step3: Select the pot with the earliest start time in all
processes of molten iron.
machines of current process, compute conflict time
between operation time of the pot and maintenance
time of the machine, and between operation time of 4.2 System development and application results
the pot and available time of the machine. If there is
a conflict, the start time of the pot on the current Based on the optimal two-stage method presented in the
machine is adjusted according to the conflict time. paper, the molten iron scheduling system software is
Step4: Calculate the conflict time between the operation developed using the integrated development environment of
time of the current pot and those of other pots that Microsoft Visual Studio.NET 2008. The mathematical
have been arranged on the current machine. If there model and algorithm are developed using C + +, and the
is a conflict, the current pot is scheduled in system application interface is developed using C # language.
accordance with the rule of FCFS, and judge the type ORACLE 10g is used for the database software of system.
of the current process. If the current process is either
An example from practical production data on 04/16/2009 is
desulphurization or reladling process, go to step5;
used to show the effectiveness of the proposed two-stage
otherwise, go to step6.
method. There are 36 pots that need to be arranged,
Step5: Judge the type of the conflict, if it is a hind conflict, including 6 pots of special type of molten iron, 30 pots of
the operation time of current pot is adjusted normal type of molten iron. After assigning machine, the
backward. If it is a fore conflict, and the conflict time initial scheduling results of pots of special type of molten
is longer than the allowable waiting time, the iron are shown in Table 1, where data with underline are
operation time of current pot is adjusted backward. If processing machine and the start time of pots of special type
it is a fore conflict, and the conflict time is less than of molten iron, which have conflicts on the machine. After
eliminating machine conflicts, the scheduling results of pots
9480
18th IFAC World Congress (IFAC'11)
Milano (Italy) August 28 - September 2, 2011
of special type of molten iron are shown in Table 2, where within 10 seconds. This may satisfy the real time
data with underline are the processing machine and the start requirement, effectively reduces the work intensity of
time of pots of special type of molten iron, which have been molten iron schedulers and improves the work efficiency. In
modified after eliminating conflicts. On the basis of addition, the processing rate of pots processed by pre- and
scheduling results of pots of special type of molten iron, pots post-processing has increased, which means that more pots
of normal type of molten iron are arranged. The scheduling can be processed by the pre- and post-processing machines.
results of all pots in the planning cycle are thus obtained. Therefore, the production efficiency of pre- and post-
processing machines, and the composition index and quality
After a long period of system operation, the application of molten iron has been improved, which leads to significant
results show that the computing time of molten iron economic benefits.
scheduling that used the proposed two-stage method is
Table 1 The initial scheduling results of pots of special type of molten iron
Desulphurization
Receiving iron Pre-processing Pre-processing Reladling
(dephosphorization)
Pot
Start Start Start Start Start
machine machine machine machine machine
time time time time time
7 09:00 3#BF 10:30 50 11:10 51 12:00 57 12:55 53
8 09:10 4#BF 10:20 50 11:00 52 11:50 B5 12:45 54
19 11:00 3#BF 12:30 50 13:10 T5 14:00 57 14:55 55
20 11:10 4#BF 12:20 50 13:00 51 13:50 B5 14:45 56
31 13:00 3#BF 14:30 50 15:10 52 16:00 57 16:55 53
32 13:10 4#BF 14:20 50 15:00 T5 15:50 B5 16:45 54
9481