SchedulingSeminar LixinTang

Data Analytics and Optimization
in Steel Industry
Lixin Tang
Key Laboratory of Data Analytics and Optimization

for Smart Industry (Northeastern University),
Ministry of Education, China
April 28 2021
http://schedulingseminar.com
Outline
Research Background
System Modeling and Optimization Method
Production Scheduling
Logistics Scheduling
Energy Optimization
Data Analytics
2
1. Research Background —— Steel is a Key Driver of the World’s Economy
Construction Automotive
Home
Shipbuilding Appliances
Machinery Logistics
Data Analytics and Optimization in Steel Industry

3
1. Research Background —— China is the Largest Steel Producer
❖ China has been the largest steel producer in the world for the last twenty
consecutive years
❖ In 2020, China's steel output has reached 1.05 billion tons, accounting for
56.5 percent of the world's steel output
❖ Steel industry has been one of the pillar industries in China’s national
economy
World Steel Production China's industrial output ratio
Steel 11.7%
900
800
700
China Japan USA
600
500 China
400
300
200
100
0
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015

4
1. Research Background —— Steel Production Process
Features: continuous and discrete production, huge devices, high-temperature

operations, massive consumption of energy and resource.
Hot Rolling
Iron-making Steelmaking Continuous Casting Slab Yard
Shipping Coil Yard Cold Rolling Coil Yard

5
1. Research Background —— Challenges Faced by Steel Industry
High
Resource High
Consumption CO2 Emission
Steel
Production
High
Energy High
Consumption Inventory
Steelmaking Logistics Hot rolling Cold rolling

6
1. Research Background —— Analytics and Optimization in Steel Industry
Production Logistics
Steel
Manufacturing
Engineering
Energy Data

7
Outline
Research Background
Energy Optimization
Data Analytics
8
2. System Modeling and Optimization Method
➢ New Characteristics
⚫ Complex physical and chemical processes
⚫ Large variety and low volume products
⚫ Complicated logistics structure
➢Complicated Large Variety

Modelling and Algorithmic and
ChallengesHuge Chemical Complicated
Production Process Low Volume Equipment Logistics Structure
⚫ Conflicting Iron
multiple objectives
Modelling
⚫ Complex technological and managerial constraints
Steel
⚫ Large scale Slab

integer variables and strongly NP-hard
Algorithmic
⚫ Cannot
Rolling Process directly
Coil apply or generalize existing algorithms

9
2. System Modeling and Optimization Method — System Modeling
Various Demands
Product Customer
Physical Chemical Mechanical
Quality performance performance performance Demand
Engineering Object
Production Logistics Energy Equipment
Management
Management
Space
Time
Steelmaking Hot rolling Cold rolling Logistics
Various Products
Time Space
Scheduling semi-finished finished Scheduling
raw materials
products products
Data Analytics and Optimization in Steel Industry 10

10
Set-Packing modeling
❖ The problem is transformed into the job
1
job job job job job job job
optimization combination of multiple

2 3 4 5 6 7 8
batch schemes of jobs, and the Set-

job job job job job job job
9 10 11 12 13 14 15
Width
Packing model is formulated; Cast 1 Cast 2 Cast 3
❖ A batch scheme of jobs is defined as

an element that includes the
combination of jobs;
❖ The sub-problems are to describes the Batch 1 Batch 2 Batch 3
generation rules of batch schemes of

C1 C2 C3 C |W |-1 C |W |

jobs; 
Batch
Batch ... 
❖ Effectively reduce the number of
scheme
scheme of
...
...
...
...
...
of one
variables and constraints and improve  multiple
cast casts

...
the solving efficiency of the model. 

1 1 1 1 1 = m
L. Tang, G. Wang, Z. Chen. Integrated charge batching and casting width selection at Baosteel.
Operations Research, 2014, 62(4): 772-787. 11
11
Space-time network flow modeling

Engineering perspective
❖ The space-time is discretized into Crane Removal
coils
grid and depicted based on 2 4 7 9 Obstacle
coils
network graph. Each node 1 3 5 6 8 10

Other coils
Null
represents a location, each edge

position
Modeling perspective
indicates a crane's move between Space
two locations in a stage; 1

2
❖ The spatial location includes all 3
the locations in the storage area 4
and the entry, exit and initial 5
location of the crane; 6

7
❖ The scheduling of task sequence 8
is transformed into the allocation 9
of crane movement in stages, and 10

Route of coils
an event-based space-time removal
crane
Route of crane
network model is established.

0 1 2 3 4 5 6 7 8 Time
Y. Yuan and L. Tang. Novel time-space network flow formulation and approximate dynamic programming approach for
the crane scheduling in a coil warehouse. European Journal of Operational Research, 2017, 262(2): 424-437.
Continuous-time based modeling

❖ Continuous-time modeling allows Engineering perspective
tasks take place at any point in the

continuous domain of time, and
Molten Al
thus improve the accuracy and
efficiency of modeling;
❖ Unit-specific event-based approach
is used for network-represented Machine
Modeling perspective
process, which allows batches to

Machine 3 8Event1 2 9
Event2
merge/split.
Machine 2 4 Event16 7
Event2
❖ This model needs fewer event

Machine 1
points describing beginning and
5
Event1 3 Event2 1
end of events and has better 1 2 3 4 5

Time
Order
computational performance. 6 7 8 9
Q. Guo, L. Tang, J. Liu, S. Zhao. Continuous-time formulation and differential evolution algorithm for an integrated
batching and scheduling problem in aluminium industry. International Journal of Production Research, 2020.
系统优化
（OR）
数据解析
（AI）
Data Analytics and Optimization (DAO)
L. Tang, Y. Meng. Data analytics and optimization for smart industry. Frontiers of Engineering
Management, 2021, 8(2): 157-171. 14
❖ Mathematical modeling is used to formulate the identifiable and
quantifiable parts of the production, logistics and energy scheduling
problems. Meanwhile, data analytics supplements to the mathematical
model for constructing the parts that are hardly to model and forming
the parameters of the model.
Mathematical Modeling Data Analytics
Industrial Data
系统优化
（OR）
数据解析
（AI）
Complicated Model Technological

constraints Parameter procedure
Management, 2021, 8(2): 157-171. 15
15
Integer optimization Evolutionary learning
Data analytics method

optimization method
Convex and sparse

OR Statistical physics
optimization (System based learning
Optimization)
System
AI
Intelligent Information theory
optimization
(Data based learning
Analytics)
Reinforcement
Dynamic optimization
learning
Management, 2021, 8(2): 157-171. 16
Optimization Features
Optimal Near-Optimal
Integer
Exact Programming Computational
Algorithms Intelligent
Algorithms
Large scale Multi-objectives Dynamics
Production and logistics

system
Assigning +
Sequencing
Iron- Steel- Hot Cold
Production making making rolling rolling Logistics
Scheduling Scheduling
Coil
Slab yard
Warehouse

17
Integer Optimization Methods and Improvement
LR/Benders （OA）Outer
Algorithm Branch & Price Branch & Cut
Decomposition Approximation
Casting problem with Integrated problem of Crane scheduling

width decision in steel-
Problem
making and continuous ship scheduling and MINLP problem in slab
casting production storage space allocation warehouse
Multilayer Low
Multiple Valid Variable Multiple Quadratic Model Valid Variable
Theory branching dimension
cuts inequality reduction linear cuts cuts tightening inequality reduction
strategy DP
Superior to traditional
Superior to commercial Superior to commercial
Superior to traditional OA and commercial
Effect optimization software optimization software
Benders in performance optimization software
CPLEX in performance CPLEX in performance
in performance
❖ The proposed algorithms are superior to international best commercial solving

software in terms of solving time, precision and stability.
18
Integer Optimization — Brach & Price

❖ A Branch & Price approach based on
set packing model; Branch & Price
❖ Discover the trapezoidal feature of Reduce Mixed Integer Programming CPLEX
Optimal solution
（MIP） Solver
the cost structure, and construct a
sol space
Molding tech
UR
property
Dantzig-Wolfe
( x, y ) → λ
Decomposition
new low-dimensional dynamic Reduce Set Patitioning 削减
Valid
inequality
（SPP） VI)
(
programming algorithm, which
sol space 松弛空间
UR
overcomes the high-dimensional Cast subset W  W Singe cast   W
Colum Generation
Performance Analysis
Update cast
Restricted subset
Pricing Sub-
feature of the conventional dynamic Master Problem

（RMP） W =W {1* , , u*}
problem
（SP）
programming algorithm;
State space  (k ; x1 , x2 ,..., xg )
Dual
solution ( π, σ, τ, θ) Dynamic
Programming
❖ Propose a multi-layer branching Relaxation
Adding
solution  ,   W
Branch & Price

Branching tree Update Pkj
strategy with sub-problem structure;
22)
( 23)
(
24)
(
RMP Branching Branching

ijs
variabel
 sj ,i ,  lkjs
variabel
Branching
Branching W *
❖ For the first time, optimal solving of Optimal cast

plan
the same kind of problem is realized.
Integer Optimization — Lagrangian Relaxation

❖ The coupling/complex constraint is relaxed into the objective function by Lagrangian
multiplier, thus decouple and decompose the full problem into several independent
sub-problems
➢ Decomposition：batch decoupling strategy; stage-based decomposition
➢ Dual problem solution: hybrid backward and forward dynamic programming;
Lagrangian Relaxation Algorithm
Multiplier relaxation
Lower bound decomposition

solve subproblems
optimally
1, if d j  0
x*j = 
0 , otherwise
L. Tang, H. Xuan, J. Liu. A new Lagrangian relaxation algorithm for hybrid flowshop scheduling to minimize total
weighted completion time. Computers & Operations Research, 2006, 33(11): 3344-3359. 20
Benders Decomposition Algorithm Structure
Various Valid Inequalities Polytope

Polytope of
Convex hull of solutions
Improving relaxation
relaxation space cut 3
cut 2
lower bound
cut 1
Combinatorial Benders Cuts

Accelerating
convergence
Feasible Benders cut
Mixed Integer Program

Integer Continuous
Variable Variable
Variable Reduction MP Values of variables SP

Reducing LB UB Infeasible
search space Optimal Benders cut

Feasible
Values of dual variables
L. Tang, D. Sun and J. Liu. Integrated storage space allocation and ship scheduling problem in bulk cargo
terminals. IISE Transactions, 2016, 48(5): 428-439. (Featured Article) 21
Outer Approximation(OA) Algorithm Structure
Multi-generation Cuts
Accelerating NLP NLP … NLP
convergence
MILP + MC
Partial Surrogate Cuts Tightening

lower bound
Hybrid Strategy of OA and

GBD
Improving
efficiency
Scaled Quadratic Cuts with
Multi-generation Cuts
L. Su, L. Tang and I.E. Grossmann. Computational strategies for improved MINLP algorithms. Computers &
Chemical Engineering, 2015, 75: 40-48. 22
Integer Optimization — Branch & Cut

❖ Branch & Cut developed； Column Index
A
4(1) 8(3) 12(4)
❖ The model tightening technique

Exit_Col
6(2) 16(5)
7(3)
is proposed based on the 10(4) 3(1)

5(2)
11(4) Time
reformulation with compact lower Mid_Col

14(5)
9(4)
2(1) 15(5)
bound； 13(5)
❖ A serial of valid inequalities (e.g.

LR, Crane 1
Init_Col 1(1)
LR, Crane 2
subtour elimination) to accelerate CPLEX B&C

the convergence of the algorithm; Instance time Gap time number of
sol sol
(s) (%) (s) cuts
❖ Variable reduction; 1 47 5.273 0 47 2.902 4
2 82 123.225 0 82 73.586 10
❖ The algorithm can solve the real 3 92 232.270 0 92 55.427 8
scale problems to optimal, and is 18 432 85.099 0 432 73.554 4

19 460 248.010 0 460 81.979 26
superior to CPLEX in performance. 20 73 3.978 0 73 3.728 12
Avg 142.119 0 70.180 30
X. Cheng, L. Tang and P.M. Pardalos. A Branch-and-Cut algorithm for factory crane scheduling problem.
Journal of Global Optimization, 2015, 63(4): 729-755. 23
Computational Intelligent Optimization
Population based Algorithm Neighborhood based Algorithm
❖ Independent of prior knowledge ❖ Dependent of prior knowledge
❖ Crossover, mutation operators ❖ Neighborhood construction
❖ Generate new individuals by ❖ Generate new solution by

crossover & mutation operators neighborhood
Differential evolution, particle swarm Tabu search, local search, annealing

24
Differential Evolution with An Individual–

Performance
Dependent Mechanism
Dimension of Benchmark Functions

Individual–Dependent Algorithms
Parameters Setting Self-adaptive -
10-D
= + -
30-D
= + -
50-D
= +
i i allocation IDE1 13 15 0 20 8 0 19 6 3
Fi = randn( ,0.1) CRi = randn( , 0.1) IDE2 9 15 4 20 7 1 19 8 1
NP NP IDE3 6 22 0 16 9 3 17 8 3
IDE4 3 22 3 6 22 IDE0 4 DE1
22 2
1.55E+01 DE2 DE3
IDE5 23 5 0 23 5 DE4 0 17DE5 10 1
Fitness Error Values

1.45E+01
CoDE 23 5 0 11 9 DE6 8 14DE76 8
Individual–Dependent ESPDE
1.35E+01
JADE
17 7
14 8
4 18 6 DE8
6 12 9 DE10
4 19DE95
7 14 8
4
6
Mutation Operator Self-adaptive jDE
1.25E+01
19 7 2 16 7 5 19 5 4
SaDE 18 8 2 19 6 3 18 4 6
1 N
1 N
2
1 N
 1 N

2 selection DE1
1.15E+01 15 9 4 14 7 7 14 6 8
DI =  xi − xj DF =   f (xi ) −
i =1 
 f (x j )  DE2 19 4 5 19 4 5 16 6 6
N i =1 N j =1 N N j =1  DE3
1.05E+01 17 4 7 21 4 3 20 4 4
DE4 18 6 4 23 3 2 21 4 3
9.50E+00
DE5 17 8 3 16 8 4 20 2 6
DE6 0.E+00
19 7 1.E+05
2 17 52.E+05 6 16 3.E+05
6 6
Perturbations DE7 15 Function
7 6 Evaluations
18 6 4(FES)
15 6 7
with Small Probability Global
DE8 16 4 8 17 5 6 16 5 7
The experiment
DE9 9 11 8 12 6 10 11 7 10
d = L + rand ( 0,1) * (U − L ) search DE10 20 7 1 23 5 0 24 4 10

demonstrates the algorithm’s
CLPSO
CMA-ES
16 7
20 3
5 23 5
5 19 4
0 24 3
5 18 4
1
6
outstanding performanceGL-25 22 6 0 24 4 0 23 4 1
L. Tang, Y. Dong and J.Y. Liu. Differential evolution with an individual–dependent mechanism. IEEE
Transactions on Evolutionary Computation, 2015, 19(4): 560-574. ( ESI Highly Cited Paper) 25
Improved Differential Evolution Algorithm

for Dynamic Scheduling
Performance
290000
Incremental Mechanism Improving 275000

PIDE
PIDERIM
Fitness Value
for Initial Population efficiency
Generation
260000
245000
Real-coded Matrix
230000
1 6 11 16 21 26 31 36
Representation
Iteration
Avoiding
invalid 500,000
The box-plot
solutions
400,000
Fitness Value
300,000
Randomly Mutation Operator Expanding 200,000
vi , g = xi , g + F (xr1, g − xr 2, g ) search 100,000
space
DE/rand/1 DE/current-to-best/1 DE/best/1 JADE PIDE
+ F (xbest
M
, g − xi , g ) + F ( x r 3, g − x r 4, g ) + F ( x r 5, g − x r 6, g )
M M
Algorithm has a fast convergence speed
L. Tang, Y. Zhao and J.Y. Liu. An improved differential evolution algorithm for practical dynamic scheduling in steelmaking-continuous
casting production. IEEE Transactions on Evolutionary Computation, 2014, 18(2): 209-225. ( ESI Highly Cited Paper)
26
Hybrid Multi-objective Evolutionary

Algorithm Performance
Incorporating the Avoiding

Concepts of Personal Best local
and Global Best optimum
Multiple Crossover
Increasing
Operators to Update the
robustness
Population
Improving
Propagating Mechanism
diversity
L. Tang and X. Wang. A hybrid multiobjective evolutionary algorithm for multiobjective optimization problems.
IEEE Transactions on Evolutionary Computation, 2013, 17(1): 20-45. 27
Adaptive Multi-objective Differential

Evolution with Reference Axis
Performance
Reference Axis Vicinity Avoiding

Mechanism to Guide local
Evolution Process optimum
Restoring Good Distribution

Improving
of the Population Before
diversity
Evolution Starting
Hybrid Control Strategy for

Accelerating
Both Parameters and
convergence
Mutation Operators
L. Tang, X. Wang, and Z. Dong. Adaptive multiobjective differential evolution with reference axis vicinity
mechanism. IEEE Transactions on Cybernetics, 2019, 49(9): 3571-3585. 28
Outline
Research Background
Energy Optimization
Data Analytics
29
3. Production Scheduling —— Steel Production
rolling
ironmaking steelmaking continuous casting slab yard hot rolling mill coil yard
continuous annealing coil yard electro-galvanization
picklig-rolling coil yard
thermo-galvanization coil yard
Unit Warehouse
Production: Iron-making/Steelmaking/Hot Rolling/Cold Rolling

30
3. Production Scheduling —— Steelmaking Stage
Charge Batching Cast Batching
ladle
Cast
tundish
Order 1
Charge
Order 2 Charge
Steel making Continuous Casting
Steelmaking Scheduling
Convertor CF-1
Convertor CF-2
Refining RF-1
Refining RF-2
Caster CC-1 Waiting time
Caster CC-2 t
Cast 1 Cast 2 Cast 3

31
3. Production Scheduling —— Charge Batching of Steelmaking
Group all the slabs of

different customer
Open-order Part orders into batches
Open-order Slabs
Customer-order Part p-median clustering

High variety with capacity and additional
Low volume Charge Customer-order Slabs technical constraints
Charge1 Charge4 Charge7
Charge8 Charge9
⚫ Minimize assignment cost
Charge2 Charge3
CF-1 Charge5 Charge6 ⚫ Minimize open-order slabs
CF-2
⚫ Minimize unfulfilled cost of order
RF-1
RF-2
CC-1 Waiting Waiting Waiting ⚫ Lagrangian relaxation

CC-2 time time time
⚫ Column generation
Cast 1 Cast 2 Cast 3 t
L. Tang, G. Wang, J. Liu, J. Liu. A combination of Lagrangian relaxation and column generation for order batching
in steelmaking and continuous-casting production. Naval Research Logistics, 2011, 58(4): 370-388. 32
3. Production Scheduling —— Cast Batching of Steelmaking
ladle Decisions
• Batch and sequence charges to
Cast tundish form casts for the given tundishes
• Select a casting width for each
Charge charge in a cast
Steel making Continuous Casting Objectives
• Maximize tundish utilization

C C C C C C C C
• Minimize total grade switch and
width switch cost
1 2 3 4 5 6 7 8
C C C C C C C
9 10 11 12 13 14 15 C= Charge
Width CAST 1 CAST 2 CAST 3

Constraints
• Grade switch constraint

• Width switch constraint
• Lifespan of tundish
serial-batch 1 serial-batch 2 serial-batch 3 time
3. Production Scheduling —— Steelmaking Scheduling
Just-in-time idea
ladle
Solve machine conflicts in (SCC)
Cast tundish production scheduling based on
JIT idea
Charge
Four-level scheduling
Steel making Continuous Casting (CC)
• Level 1: cast sequences on the casters
Charge 1 time
• Level 2: sub-scheduling
Charge 2 Charge 3
• Level 3: rough scheduling
CF-1
CF-2 • Level 4: elimination of machine conflicts
RF-1 Beneficial effects

RF-2
Waiting time • Improve productivity of large devices
CC-1
CC-2 • Shorten waiting-time between operations
cast • Cut down production costs
L. Tang, J. Liu, A. Rong, Z. Yang. A mathematical programming model for scheduling steelmaking-continuous
casting production. European Journal of Operational Research, 2000, 120(2): 423-435. 34
3. Production Scheduling —— Semi-continuous Batch Scheduling
Characteristics of Semi-Continuous ❖ A new kind of batch scheduling

Batching Scheduling
❖ We analyze the semi-continuous
Classical Batching
The new Semi batch scheduling problem, and
-Continuous Batching
Machine Scheduling
Machine Scheduling present optimal algorithm
Preheating Heating Soaking

zone zone zone
Begin and finish Enter and leave the
processing together Handle machine one by one
several jobs
simultaneously
The same batch
begin processing Respective start
at the same time time
Input Output
The same Respective

completion time completion time Measure Nozzle
The heating process of Tube-billets in

Traditional batching machines are mainly divided into three
heating furnace
types: (1) burn-in (2) fixed batch (3) serial batching
L. Tang, Y. Zhao. Scheduling a single semi-continuous batching machine. Omega, 2008, 36(6):992-1004.
35
3. Production Scheduling —— Hot Rolling Scheduling
Decision Objective
warm up
slab Slab width material section
Sequence of adjacent Minimize the total
jobs to be processed changeover costs
N +M N +M
Minimize  C
i =1 j =1
ij X ij
A turn
staple material
section
N +M
Subject to X
i =1
ij = 1, j{1, 2, ..., N+M }
N +M
X
j =1
ij = 1, i{1, 2, ..., N+M }

iS jS \{i }
X ij | S | −1, S  {1, ..., N+M }, 2  |S|  N+M -2 Structure and components of a turn
The The
width 1 2 first i last M
slab slab
The first turn Turns within a shift
L. Tang, J. Liu, A. Rong, Z. Yang. A multiple traveling salesman problem model for hot rolling scheduling in Shanghai Baoshan Iron &
Steel Complex. European Journal of Operational Research, 2000, 124(2): 267-282.
36
3. Production Scheduling —— Slab Allocation at Hot Rolling Stage
Unfulfilled
Orders
Open-order Part
Open-order Slabs
Customer
Customer-order Part
Orders
High variety Customer-order Slabs
Low volume Charge
Allocate the
Open-order Slabs to
Unfulfilled Orders
Open-order Slabs Order 1 Order 2 Problem 1
This work was awarded INFORMS Franz Edelman Award Finalist, 2013
37
3. Production Scheduling —— Parallel Batch Scheduling at Cold Rolling
Minimize
Maximize Equipment Matching
Mismatching
Reward Constraints Constraints
cost
Form batches for Select a median coil

each empty furnace for each batch
L. Tang, Y. Meng, Z. Chen, J. Liu. Coil batching to improve productivity and energy utilization in steel 38
production. Manufacturing & Service Operations Management, 2016, 18(2): 262-279. 38
Outline
Research Background
Energy Optimization
Data Analytics
39
4. Logistics Scheduling —— Logistics in Steel Plant
rolling
ironmaking steelmaking continuous casting slab yard hot rolling mill coil yard
◼ 研究背景
continuous annealing coil yard electro-galvanization
picklig-rolling coil yard
thermo-galvanization coil yard
Unit Warehouse
Logistics: (Un)Loading/Transportation/Shuffling/Storage/Stowage

40
4. Logistics Scheduling —— Crane Scheduling in Loading Operation
Track
Crane scheduling problem
Determines the transportation sequence for all
demanded coils and shuffled position for each
m8 m7 m6 m5 blocking coil.
m1 m2 m3
m0 m4
Parts
Decisions Objectives
Loading / Tank 1 Tank 2 Tank 3 Tank 4 Tank 5 Tank 6 Tank 7 Tank 8
Unloading
Retrieval sequence of Minimize the time by
the target coils and which the retrieval of
Bridge
shuffled positions for all target coils is
blocking coils completed
Track
Track
Row 1 Row 2 Row 3 Row 4 For general case For special cases
Heuristic algorithm & Polynomial algorithms

worst-case analysis (optimal solutions)
L. Tang, X. Xie, J. Liu. Crane scheduling in a warehouse storing steel coils. IISE Transactions, 2014, 46(3): 267-282.
4. Logistics Scheduling —— Coordinated Transportation Scheduling
Offline problems involving a

Integrated Production & Two-Stage
Distribution Scheduling single production line
⚫ Optimal dynamic programming
⚫ Obtain a joint schedule of job processing at
the plant and two-stage shipping algorithms
⚫ Optimize a performance measure that takes
into account both delivery timeliness and total Offline problems involving
transportation costs multiple production lines
⚫ Fast heuristics
⚫ Worst-case & asymptotic performance
Online problems
……
⚫ Online algorithms
⚫ Competitive ratios analytics
L. Tang, F. Li, Z. Chen. Integrated scheduling of production and two-stage delivery of make-to-order products:
offline and online algorithms. INFORMS Journal on Computing, 2019, 31(3):493-514. 42
4. Logistics Scheduling —— Shuffling
Slabs to
Shuffling Problems in Steel Plants
be
shuffled Assign a storage slot for each shuffled item during
Stack retrieving all target items in the given sequence
height Target
slab
Bottom of
the stack
(slab 1) Decisions Objectives
The structure of a slab stack
Suitable storage
Minimize shuffling and
positions for shuffled
crane traveling
items
Upper
level
Lower 2 1
level
For general case For special cases
Shuffling coil of coil 1 Demanded
Polynomial algorithms
Shuffling coil of coil 2 Non-demanded Greedy heuristic
(optimal solutions)
The structure of a coil stack
L. Tang, R. Zhao, J. Liu. Models and algorithms for shuffling problems in steel plants. Naval Research
Logistics, 2012, 59(7): 502-524. 43
4. Logistics Scheduling —— Reshuffling and Stacking
❖ For statistic and dynamic The layout of a block

reshuffling problem, an improved a bay
a column
mathematical formulation and a a lane
simulation model are established,

blocking
objects
Retrieving
object
respectively； height
width length
a position
a tier
❖ Five polynomial time heuristics
and their extended versions are Arrival
container
proposed and analyzed the-

oretically； Blocking
container
❖ The proposed heuristic outper- Blocking

container
forms existing methods. Retrieving

container
L. Tang, W. Jiang, J. Liu, Y. Dong. Research into container reshuffling and stacking problems in container terminal
yards. IISE Transactions, 2015, 47(7): 751-766. (IISE Transactions Best Applications Paper Award) 44
4. Logistics Scheduling —— Ship Stowage Planning
Minimize the
Minimize the Minimize the
dispersion of coils for
moment imbalance shuffling
the same destination
right
Upper
level
2
Lower 2 1
fore 3 stern level
4 1 row
Shuffling coil of coil 1 Demanded
column Shuffling coil of coil 2 Non-demanded
left
Structural Weight restriction Operational

constraints constraints constraints
L. Tang, J. Liu, et al. Modeling and solution for the ship stowage planning Modeling and solution for the ship
stowage planning problem of coils in the steel industry. Naval Research Logistics, 2015, 62(7): 564-581.45
Outline
Research Background
Energy Optimization
Data Analytics
46
5. Energy Optimization —— Energy Analytics
Description Diagnosis Prediction
Solution
Energy saving
Benchmark
Energy consumption and regeneration data
Analysis Analyze cause
Filter out Complement Bottleneck

exceptional data Identification Identify bottleneck
the missing data
Energy
Obtain actual process data Analysis Analyzeconsumption
Energy regeneration
Iron Steel Heating Hot Acid Continuous
smelting making furnace rolling rolling annealing
Liquid Hot Cold
iron Slab Slab coil coil
High High High High Normal
temp temp temp temp temp
Coal Gas Pulverized

Oxygen Natural Gas Electricity
Coal
5. Energy Optimization —— Dynamic Energy Allocation
Objectives Constraints ADP Algorithm

⚫ Minimizing emission ⚫ Production ⚫ Balance Accomplish dynamic
⚫ Minimizing cost
⚫ Demand ⚫ Price energy allocation
⚫ Maximizing income
Process-dimension
Objectives emission cost income
minimization minimization maximization
Production Energy demand Balance Price/cost
plan in-out ratio storage balance emission penalty

Constraints
priority mix requirement supply-demand purchase price
capacity conversion pressure balance sale price

❖ The proposed energy allocation
method shows obvious superiority
Restrictions pressure
emission
in terms of effectiveness and
unit demand holding capacity
calorific value
limitation stability than static method.

48
5. Energy Optimization —— Gas Allocation
Comprehensive allocation of gas system

Constraint
⚫ Determine: allocation plan of BFG, COG, LDG
Definition
⚫ Multi-objective: minimizing consumption cost,
purchase cost, emission cost, energy holding cost

Soft
Constraint
⚫ Solution method: soft constraint handling NSGA-II Definition
Y. Zhang, G. G. Yen, and L. Tang. Soft constraint handling for a real-world multiobjective energy distribution
problem. International Journal of Production Research, 2020, 58(19): 6061-6077. 49
5. Energy Optimization —— Steam Scheduling
Steam scheduling by coordinating Objectives

demand and electricity generation ⚫ Maximize electricity generation upon demand
z = max  ( ui + vi xti , j =1 + wi Rti )
t i
Make full use of Supply capacity constraints

User Electricity
excess steam
4
ai0   xtij  ai1，bij0  xtij  bij1，ri0  Rti  min ( xti1 , ri1 )，qi0  Qti  min ( xti1 , qi1 )
demand generation
resources j =1
   
xtij = min ai1 , max  ai0 , StD −  (x
ti ,3 + Rti + Qti )  
  iI1 I 2 I3  
Results comparison Fluctuation, safe flow constraints

   
Ft D = max  0,   ( xtij + Rti + Qti ) − e D  Ft Z = max  0,  xtij − e Z 
 i jJ3   i 
 ( x
i j
tij + Rti + Qti ) −   ( xt −1,ij + Rt −1,i + Qt −1,i )   D
i jJ 3
Steam demand constraints

 Z  xtij  StZ  D  ( xtij + Rti + Qti )  StD
i i j

50
5. Energy Optimization —— Oxygen Scheduling
Task Minimize operating cost of oxygen system
Dynamically balance and optimize  1 

Z =   ci  Fti + ciA  Ati + ciY  Yti +  ti  ci  0.7 Bi 
t iE  2 
the oxygen system
Oxygen generators capacity, operating
requirements
Supply Modes Oti − Ot −1,i  ti Gti = Gt −1,i + Yti − Dti , Gi0  Gti  Gi1 ,
⚫Supplied by oxygen generator  ti = max 0, ( ti − t −1,i ) dt =  Dti , dt   Gt −1,i

iE iE
⚫Supplied by liquid oxygen system Pipeline pressure, fluctuation limitations
Oxygen generation system Holding system consumption

( H t − H t −1 ) +  Stj   Ati H 0  Ht  H 1
j =1 iE
H t − H t −1
Oxygen hold and pipe Steelmaking  Ati   ti ai Ati  Oti
Compressure H t −1
Ironmaking
Oxygen
generator Oxygen demand constraints
Emission
Other users
Liquid
oxygen
Evaporator
 S + Y + ( H
j
tj
iE
ti t − H t −1 ) + Ft =  Oti
iE
External users

51
Outline
Research Background
Energy Optimization
Data Analytics
52
6. Data Analytics and Process Optimization for Quality
Case 1. Iron-making —— Iron Quality Prediction

Multi-objective Evolutionary Ensemble Learning
Fusion of thermodynamic model Sub-learner based on fusion of

(meso) and process data (macro) meso and macro data
Multi-objective evolutionary Evolving the structure and

algorithm parameters of ensemble model
Macroeconomic
Process data and image
Iron quality
Multi-objective
evolutionary learning
Mesoscopic
Thermodynamic model
X. Wang, T. Hu, and L. Tang. A multiobjective evolutionary nonlinear ensemble learning with evolutionary feature selection for
silicon prediction in blast furnace. IEEE Transactions on Neural Networks and Learning Systems, 2021. 53
Case 2. Steelmaking —— Dynamic Prediction

Oxygen
Waste gas
Challenges
Fume hood
Water-cooled
oxygen lance
Blow oxygen
Molten iron and steel scrap Auxiliary materials
Measurement
⚫ Continuous prediction requirement
Pour out
molten steel
Refractory
⚫ Unstable performance of single model

from spout
lining First stage Third stage Reblow
Impurities are Second stage
oxidized on Tapping
the surface Molten steel
Whole blowing process
Blowing at bottom
⚫ Dynamic adjustment requirement
Production process of BOF steelmaking
Dynamic analytics method

Stage 1 Stage L
⚫ Multi-stage modeling strategy
⚫ Dynamic model with feedback
Stage 1 Stage 2 Stage L
⚫ Hybrid kernel function
⚫ Differential evolution algorithm
Principle of multi-stage modeling in BOF steelmaking process
C. Liu, L. Tang, J. Liu, Z. Tang. A dynamic analytics method based on multistage modeling for a BOF steelmaking process.
IEEE Transactions on Automation Science and Engineering, 2019, 16(3): 1097-1109. 54
Case 3. Hot Rolling —— Temperature Prediction of Reheat Furnace
Continuous casting Fuel burner

Features of Heating Process
⚫Dynamic ⚫ Non-linear
Slab
Slabs Hot slabs
Mechanism Model
Hot rolling
Slab yard
Support beam
⚫Difficult to obtain
⚫Obvious prediction error
Reheating furnace
Deviation Mechanism Model

Compensation
LS-SVM Mechanism ⚫LS-SVM is used to compensate for
Model Model the prediction deviation of the slab
temperature
⚫Significantly improve the model
Mixed Model prediction accuracy

55
Case 4. Cold Rolling —— Strip Quality Analytics

Multi-objective Ensemble Learning
Least square support vector Sub-learner in the ensemble

machine (LSSVM) learning
Multi-objective evolutionary Evolving the ensemble learning

algorithm model

56
Conclusion and On-going Research
Optimization Analytics
Plant-wide Production and Inventory Planning
Decision-
making Production/Logistics Batching and Scheduling
Execution Process Optimization + Optimal Control
Knowledge 诊
+ Diagnosis
断 + 预测 + Prediction
Discovery
Production process Product quality
Understanding + Description
Perception
Industry image Speech IOT Sensing Visualization
Production data Equipment data Energy data Logistics data
Steel/Nonferrous Petroleum-Chemical Energy Mining/Logistics
Management, 2021, 8(2): 157-171. 57
Conclusion and On-going Research
Dedicated to Science
Without Physical Background
Optimization
Data Analytics Data Analytics

Rooted in Industry
Refine
With Physical Background
Logistics, Resources Petrochemistry, Energy
Steel Industry

SchedulingSeminar LixinTang

Uploaded by

Copyright:

Available Formats

SchedulingSeminar LixinTang

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

SchedulingSeminar LixinTang

Uploaded by

Copyright:

Available Formats

Data Analytics and Optimization

Key Laboratory of Data Analytics and Optimization

System Modeling and Optimization Method

Data Analytics and Optimization in Steel Industry

World Steel Production China's industrial output ratio

Data Analytics and Optimization in Steel Industry

Features: continuous and discrete production, huge devices, high-temperature

Iron-making Steelmaking Continuous Casting Slab Yard

Shipping Coil Yard Cold Rolling Coil Yard

Data Analytics and Optimization in Steel Industry

Steelmaking Logistics Hot rolling Cold rolling

Data Analytics and Optimization in Steel Industry

Data Analytics and Optimization in Steel Industry

System Modeling and Optimization Method

⚫ Large variety and low volume products

⚫ Complicated logistics structure

➢Complicated Large Variety

⚫ Large scale Slab

Data Analytics and Optimization in Steel Industry

Steelmaking Hot rolling Cold rolling Logistics

Data Analytics and Optimization in Steel Industry 10

optimization combination of multiple

batch schemes of jobs, and the Set-

❖ A batch scheme of jobs is defined as

generation rules of batch schemes of

Space-time network flow modeling

network graph. Each node 1 3 5 6 8 10

represents a location, each edge

two locations in a stage; 1

and the entry, exit and initial 5

location of the crane; 6

of crane movement in stages, and 10

an event-based space-time removal

network model is established.

Continuous-time based modeling

tasks take place at any point in the

process, which allows batches to

❖ This model needs fewer event

end of events and has better 1 2 3 4 5

Data Analytics and Optimization (DAO)

Complicated Model Technological

Integer optimization Evolutionary learning

Data analytics method

Convex and sparse

Large scale Multi-objectives Dynamics

Production and logistics

Data Analytics and Optimization in Steel Industry

Integer Optimization Methods and Improvement

Casting problem with Integrated problem of Crane scheduling

❖ The proposed algorithms are superior to international best commercial solving

Integer Optimization — Brach & Price

overcomes the high-dimensional Cast subset W  W Singe cast   W

feature of the conventional dynamic Master Problem

Branch & Price

RMP Branching Branching

❖ For the first time, optimal solving of Optimal cast

the same kind of problem is realized.

Integer Optimization — Lagrangian Relaxation

Lagrangian Relaxation Algorithm