Data & Intelligence Global One Team ‘Optimization’ › english › optimization.pdf ·...
Transcript of Data & Intelligence Global One Team ‘Optimization’ › english › optimization.pdf ·...
© 2019 NTT DATA Mathematical Systems Inc.
15 Feb. 2019NTT DATA Mathematical Systems, Inc.Takahito Tanabe ([email protected])
Data & Intelligence Global One Team
NTT DATA Mathematical Systems, Inc. ‘Optimization’
2© 2019 NTT DATA Mathematical Systems Inc.
1.What is Optimization • Control & Constraint
2.Examples• Production Planning ( maxworkload .vs. inventry)
• Production Planning ( costly equipment )
• Cutting Stock
• Stock Allocation
3.Optimization project in Reality• Our Standpoint
• Manager & Practitioner
• Driver Required!
• Let’s PoC together
‘Optimization’ : NTT Data Mathematical Systems, Inc.
3© 2019 NTT DATA Mathematical Systems Inc.
What is Optimization ?
4© 2019 NTT DATA Mathematical Systems Inc.
Maximize compatibility (objective) under matching constraint
Constrained Recommendation
Job ApplicantCompany
Typical Recommendation
Job ApplicantCompany
ConstrainedRecommendation
#applicants ≧ N (∝ pay)for each Company
5© 2019 NTT DATA Mathematical Systems Inc.
What is Mathematical Optimization ?
Find the optimal CONTROL or DECISIONthat satisfies CONSTRAINTS
6© 2019 NTT DATA Mathematical Systems Inc.
Examples
7© 2019 NTT DATA Mathematical Systems Inc.
Production management (Orders with various due date)
8© 2019 NTT DATA Mathematical Systems Inc.
Production management (Orders with various due date)
When to produce ?
9© 2019 NTT DATA Mathematical Systems Inc.
Production planning : Balance workload ?
Control the production date to balance the machine load
workload (no control)
days
10© 2019 NTT DATA Mathematical Systems Inc.
Production planning : Constrain the workload
workload(optimized)
days
max workloadconstraint
Production was pushed forward according to optimization result.
11© 2019 NTT DATA Mathematical Systems Inc.
Production planning : Inventory minimized
Inventory cost
days
minimized
12© 2019 NTT DATA Mathematical Systems Inc.
Greedy strategy tested with Our Simulation Tool S4
Push forward all the production to the workload limit !With various inventory maximum (parameter)
13© 2019 NTT DATA Mathematical Systems Inc.
Simulation and Optimization
Simulation results
optimizationresults
max workload
inventorycost
max workload / cost
Inventory cost and max workload are offset
14© 2019 NTT DATA Mathematical Systems Inc.
Production Resource Planning at a Japanese Manufacturer (2013-)
You can control the product flow
Each product requires certain machine resources we should allocate
Each machine needs costly special equipment to handle each product’s process
Reduce equipment keeping the production output within the available machine resource
product A
product flow of B
product flow of A
machineproduct B
Achievement: Reduced special equipments by 30~50%# of machines : 60
# of products : 48
15© 2019 NTT DATA Mathematical Systems Inc.
A Cutting Stock Problem for Film Manufacturer (2017-)
Cutting Patterns Example
width of ‘1’ width of ‘2’
fixed size
pattern ‘1’ pattern ‘2’
…
Determine Cutting Pattern (2D region tiled by rectangular products) Product requirement constrained Reduce (# of Cutting Pattern) ⇔ (amount of loss)
loss
16© 2019 NTT DATA Mathematical Systems Inc.
Data Maintenance Schedule for IT company (2018-)
Assign Data maintenance/counting task (#500) to either of 2 machines Due date / precedence / capacity constraint should be fulfilled.
precedence
capacity
17© 2019 NTT DATA Mathematical Systems Inc.
Stock allocation
Control stock allocation for orders at {A,B,C,D} Allocate from neighbouring storage is cost-minimum Quality balance offsets shipping cost
1 1i
i
x =orderAmount
Dj D
j
x Stock
,
ij ij
i j
c x
storageA
storageB
storageC
order#1
order#2
storageDorder#6
order#3
order#7
……
…
Balancequality
mixed quality
Control Allocation
18© 2019 NTT DATA Mathematical Systems Inc.
Stock allocation demo
19© 2019 NTT DATA Mathematical Systems Inc.
Optimization in Reality
20© 2019 NTT DATA Mathematical Systems Inc.
Solve real world problem for business practitioners
⇒ Our packaged softwares are ‘stock in trade’ for this purpose (Numerical Optimizer is one of them)
Our Standpoint
AcademiaBusiness
Practitioners MSI
Sponser
mathematicalmodels
Businessrequirements
algorithms
theories
software
21© 2019 NTT DATA Mathematical Systems Inc.
‘Optimization’ everywhere ! (supply chain)
demand
feedstock
factory
warehouse/storage
customer
blend raw materials roster staffs select line for production allocate production resource manage storage for feedstock create cutting pattern
plan shipment
arrange (half-finished) goods
stock allocation loading
routing configuration assign route to vehicle
22© 2019 NTT DATA Mathematical Systems Inc.
Managers and Practitioners
IT System Integrator
Practitioner#1
Practitioner#2
Practitioner#3
commitment!
our own ..
Fulfill ..
managerial personnel
Save Cost!
driven by
Constraintdriven by
Objective
GAP
cost,revenue,investment..
due-dateresource
regulation..
23© 2019 NTT DATA Mathematical Systems Inc.
Traditionally in Japan ..
Practitioner#1
Practitioner#2
Practitioner#3
commitment!
our own ..
Fulfill ..
driven by
Constraint
due-dateresource
regulation..
untouchable
24© 2019 NTT DATA Mathematical Systems Inc.
But ..
Practitioner#1
Practitioner#2
Practitioner#3
commitment!
our own ..
Fulfill ..
driven by
Constraint
due-dateresource
regulation..
untouchable
MathematicalOptimization
appeals!
25© 2019 NTT DATA Mathematical Systems Inc.
Mathematical Optimizaion promotes communication
IT System Integrator
Practitioner#1
Practitioner#2
Practitioner#3
commitment!
our own ..
Fulfill ..
managerial personnel
GO Digital
driven by
Constraintdriven by
Objective
Numerical Optimizer
cost,revenue,investment..
due-dateresource
regulation..
26© 2019 NTT DATA Mathematical Systems Inc.
Collect Data ;Build Model ;Solve ;
Process of Mathematical Optimization
27© 2019 NTT DATA Mathematical Systems Inc.
while (!satisfy(output)) {Collect Data ;Build Model ;Solve ;Examine Result;if ( budget? ) exit;
}
Process of Mathematical Optimization in Reality
Something is wrong with
this solution ..
I’ll fix it bymodifying the
model !
Trial & Error
28© 2019 NTT DATA Mathematical Systems Inc.
Responsible Leader Required !
Who knows value of the final outcome
≠ data scientist
• Go to that goal !• What happened ? Now
choose plan B !
29© 2019 NTT DATA Mathematical Systems Inc.
Aiming for Deploy …
CHASMline balancing
resource planning
routing
stockallocation
!
!
Our optimization engine is working in Domestic sites:
JR-stationChain StoreCall CenterProduction center Air conditioning …
EVERY DAY
optimization integrated system
30© 2019 NTT DATA Mathematical Systems Inc.
Aiming for Deploy …
line balancing
resource planning
routing
stockallocation
!
!
Why don’t we start OptimizationProof of Concept
by applying it to your actual situations?
CHASM
optimization integrated system
31© 2019 NTT DATA Mathematical Systems Inc.
We can also provide materials to introduce optimization
© 2019 NTT DATA Mathematical Systems Inc.