Supply chain optimization Jose Pinto - CEPAC
Transcript of Supply chain optimization Jose Pinto - CEPAC
1
Part 3
Process and Supply Chain Operations
Supply chain optimization
Jose Pinto
2
I. Strategic Framework to the Analysis of Supply Chain Networks
II. Design of Supply Chain NetworksIII. Planning of Supply ChainsIV. Planning, Scheduling and Supply Chain
Management for Operations in Oil Refineries
AppendicesI. Demand ForecastII. Transportation IssuesIII. The Role of Inventory
Outline
3
Supply Chain
Scope: a supply chain covers the flow of materials, information and cash across the entire enterpriseSupply chain management: process of integrating, planning, sourcing, making and delivering product, from raw material to end customer, and measuring the results globallyTo satisfy customers and make a profitWhy a ‘supply chain’?
4
Traditional View: Logistics in the Economy
1990 1996Freight transportation $352 $455 billionInventory expense $221 $311 billionAdministrative expense $ 27 $ 31 billion
Logistics related activity 11% 10.5% GNP
Source: Cass Logistics
5
Traditional View: Logistics in the Manufacturing Firm
Profit: 4%
Logistics cost : 21%
Marketing cost: 27%
Manufacturing cost : 48%
ProfitLogistics
Cost
Marketing Cost
Manufacturing Cost
6
Supply Chain Management: The Magnitude in the Traditional View
The grocery industry could save $30 billion (10% of operating cost by using effective logistics and supply chain strategies
A typical box of cereal spends 104 days from factory to sale
A typical car spends 15 days from factory to dealership
Compaq estimates it lost $0.5 billion to $1 billion in sales in 1995 because laptops were not available when and where needed
P&G estimates it saved retail customers $65 million by collaboration resulting in a better match of supply and demand
Laura Ashley turns its inventory 10 times a year, five times faster than 3 years ago
7
Objectives of a Supply ChainMaximize overall value generated
Satisfying customer needs at a profitValue strongly correlated to profitabilitySource of revenue – customerCost generated within supply chain by flows of information, product and cashFlows occur across all stages – customer, retailer, wholesaler, distributor, manufacturer and supplierManagement of flows key to supply chain success
8
Supply Chain Stages
Supplier Manufacturer Distributor Retailer Customer
Supplier Manufacturer Distributor Retailer Customer
Supplier Manufacturer Distributor Retailer Customer
9
Decision phases in a supply chainSupply chain strategy or design
Location and capacity of production and warehouse facilities?Products to be manufactured, purchased or stored by location?Modes of transportation?Information systems to be used?Configuration must support overall strategy
Supply chain planningOperating policies – markets served, inventory held, subcontracting, promotions, …?
Supply chain operationDecisions and execution of orders?
10
Cycle View of Supply Chains
Customer Order Cycle
Replenishment Cycle
Manufacturing Cycle
Procurement Cycle
Customer
Retailer
Distributor
Manufacturer
Supplier
11
Process view of a supply chain
Customer order cycleTrigger: maximize conversion of customer arrivals to customer ordersEntry: ensure order quickly and accurately communicated to all supply chain processesFulfillment: get correct and complete orders to customers by promised due dates at lowest costReceiving: customer gets order
12
Process view of a supply chain
Replenishment cycleReplenish inventories at retailer at minimum cost while providing necessary product availability to customerRetail order:
Trigger – replenishment point – balance service and inventoryEntry – accurate and quick to all supply chainFulfillment – by distributor or mfg. – On timeReceiving – by retailer, update records
13
Process view of a supply chain
Manufacturing cycleIncludes all processes involved in replenishing distributor (retailer) inventory, on time @ optimum costOrder arrival Production schedulingManufacturing and shippingReceiving
14
Process view of a supply chain
Procurement cycleSeveral tiers of suppliersIncludes all processes involved in ensuring material available when required
15
Supply chain macro processes
CRM (Customer Relationship Management) – all processes focusing on interface between firm and customers
ISCM (Internal Supply Chain Management) – processes internal to firm
SRM (Supplier Relationship Management) – all processes focusing on interface between firm and suppliers
16
Push/Pull View of Supply Chains
Procurement,Manufacturing andReplenishment cycles
Customer OrderCycleCustomer
Order arrives
PUSH PROCESSES PULL PROCESSES
Pull – processes in response to a customer orderPush – processes in anticipation of a customer order
17
Supply chain performance –Strategic fit and scope
NewProduct
Development
Marketingand
Sales
OperationsSupply and
ManufactureDistribution Service
Finance, Accounting, Information Technology, Human Resources
Business Strategy
New ProductStrategy
MarketingStrategy Supply Chain Strategy
18
Achieving strategic fit
Quantity - lot sizeResponse timeProduct varietyService levelPriceInnovation
ImpliedDemand
UncertaintyRegular Demand
Uncertainty due to customers demand and
Implied Demand Uncertainty due to
uncertainty in Supply Chain
Understanding the Customer and Demand Uncertainty
19
Levels of Implied Demand Uncertainty
Low High
Price ResponsivenessCustomer Need
DetergentLong lead time steel
High FashionEmergency steel
Implied Demand Uncertainty Low HighProduct Margin Low High Average Forecast Error 10% 40-100% Average Stockout rate 1-2% 10-40%Average markdown 0% 10-25%
Fischer (1997) Harvard Bus. Rev, March-April, 83
20
Supply source uncertaintySupply uncertainty increases with...
Frequent breakdownsUnpredictable and/or low yieldsPoor qualityLimited supplier capacityInflexible supply capacityEvolving production processes
Life cycle position of productNew products high uncertaintysalt vs existing automobile model vs new communication device
21
Understanding the Supply Chain
High Low
Low
High
Cost (efficient)
Responsiveness to Quantity, Time, Variety, Innovation, Service level
DELL
AUTOMOTIVES
INTEGRATED STEEL MILLS
22
Achieving Strategic Fit
Implied uncertainty spectrum
Responsive supply chain
Efficient supply chain
Certain demand
Uncertain demand
Responsiveness spectrum Zone o
f
Strateg
ic Fit
Low Cost
High Cost
23
Responsive and Efficient Supply ChainsEfficient Responsive
Primary goal demand at lowest cost respond quickly
Product designstrategy
maximize performanceat minimum cost
create modularity
Pricing strategy Lower margins higher margins
Manufacturingstrategy
lower costs (highutilization)
maintain flexibility
Inventory strategy minimize to lower cost maintain buffer inventory
Lead time strategy reduce but not atexpense of costs
aggressively reduce
Supplier strategy select based on costand strategy
select based on speed,flexibility, reliability andquality
24
Product life cycle
Implied uncertainty spectrum
Responsive supply chain
Efficient supply chain
Product Maturity
Product Introduction
Responsiveness spectrum Zone o
f
Strateg
ic Fit
25
Strategic Scope
Suppliers Manufacturer Distributor Retailer Customer
Competitive Strategy
Product Dev. Strategy
Supply Chain Strategy
Marketing Strategy
26
Drivers of Supply Chain Performance
Efficiency Responsiveness
Supply chain structure
Competitive Strategy
Supply Chain Strategy
Inventory Transportation Facilities Information
DriversTRADE OFF FOR EACH DRIVER
27
Inventory‘What’ of supply chainMismatch between supply and demandMajor source of costHuge impact on responsivenessMaterial flow time
i = d t (i – inventory, d – throughput, t – flow time)
Role in competitive strategyComponents
Cycle inventory – average inventory between replenishmentsSafety inventory - to cover demand and supply uncertaintySeasonal inventory – counters predictable variation
Overall trade off: responsiveness vs. efficiency
28
Transportation
‘How’ of supply chainLarge impact on responsiveness and efficiencyRole in competitive strategyComponents
Mode – air, truck, rail, ship, pipeline, electronicRoute selectionIn house or outsource
Overall trade off: responsiveness vs efficiency
29
Facilities‘Where’ of supply chainTransformed (factory) or stored (warehouse)Role in competitive strategyComponents
Location - central or decentralCapacity – flexibility vs. efficiencyManufacturing methodology – product or process focusWarehousing methodology – storage – sku, job lot, crossdocking
Overall trade off: responsiveness vs. efficiency
30
Affects every part of supply chainConnects all stagesEssential to operation of all stages
Role in competitive strategySubstitute for inventory
ComponentsPush vs. pullCoordination and information sharingForecasting and aggregate planningEnabling technologies
EDI, Internet, ERP, SCMOverall trade off: responsiveness vs. efficiency
Information
31
Considerations for Supply Chain Drivers
Driver Efficiency Responsiveness
Inventory Cost of holding Availability
Transportation Consolidation Speed
Facilities Consolidation /Dedicated
Proximity /Flexibility
Information What information is best suited foreach objective
32
Obstacles to achieving strategic fit
Increasing variety of products
Decreasing product life cycles
Increasingly demanding customers
Fragmentation of supply chain ownership
Globalization
Difficulty executing new strategies
All increase uncertainty
33
Major obstacles to achieving fitMultiple global owners / incentives in a supply chain
Information Coordination & Contractual Coordination
Increasing product variety / shrinking life cycles / demanding customers/customer fragmentation
Increasing demand and supply uncertainty
Local optimization and lack of global fit
34
Dealing with Product Variety: Mass Customization
MassCustomization
Low
HighHigh
Low
Long
Short
Lea
d T
ime
Cost
Customization
35
Fragmentation of Markets and Product Variety
Are the requirements of all market segments served identical?
Are the characteristics of all products identical?
Can a single supply chain structure be used for all products / customers?
No! A single supply chain will fail different customers on efficiency or responsiveness or both.
36
II.Designing the supply chain network
37
FACILITY DECISIONS: Network Design Decisions
Facility roleWhat processes are performed
Facility locationWhere should facilities be located
Capacity allocationHow much capacity should be allocated to each facility
Market & supply allocationWhat markets should each facility serveWhat supply sources should feed each facility
38
Factors Influencing Network Design DecisionsStrategic
Cost or Responsiveness focus
TechnologicalFixed costs and flexibility determine consolidation
MacroeconomicTariffs and Tax incentives. Stability of currency
Political stability - clear commerce & legal rulesInfrastructure
sites, labor, transportation, highways, congestion, utilities
CompetitionLogistics and facility costs
39
The Cost-Response Time Frontier
Local FG
Mix
Regional FG
Local WIP
Central FG
Central WIP
Central Raw Material and Custom production
Custom production with raw material at suppliers
Cost
Response Time HighLow
Low
High
40
Logistics and facilities costs
Inventory costsTransportation costs
Inbound and outboundFacility (setup and operating) costsTotal logistics costs
41
Service and Number of Facilities
Number of Facilities
ResponseTime
Facilities Costs
42
Costs and Number of Facilities
Costs
Number of facilities
Inventory
Facility costs
TransportationFrequent inbound
43
Cost Build-up as a function of facilities
Percent Service Percent Service Level Within Level Within
Promised TimePromised Time
TransportationTransportationCos
t of O
pera
tions
Cos
t of O
pera
tions
Number of FacilitiesNumber of Facilities
InventoryInventoryFacilitiesFacilities
Total CostsTotal Costs
LaborLabor
44
Framework for network design decisions
Define a supply chain strategyCOMPETITIVE strategy
Define a regional facility strategyLocation, roles and capacity
Select desirable sitesHard infrastructure – transport, utilities, suppliers, warehousesSoft infrastructure – skilled workforce, community
Choose locationPrice location and capacity allocation
45
A Framework for Global Site Location
PHASE ISupply Chain
Strategy
PHASE IIRegional Facility
Configuration
PHASE IIIDesirable Sites
PHASE IVLocation Choices
Competitive STRATEGY
INTERNAL CONSTRAINTSCapital, growth strategy,existing network
PRODUCTION TECHNOLOGIESCost, Scale/Scope impact, supportrequired, flexibility
COMPETITIVEENVIRONMENT
PRODUCTION METHODSSkill needs, response time
FACTOR COSTSLabor, materials, site specific
GLOBAL COMPETITION
TARIFFS AND TAXINCENTIVES
REGIONAL DEMANDSize, growth, homogeneity,local specifications
POLITICAL, EXCHANGERATE AND DEMAND RISK
AVAILABLEINFRASTRUCTURE
LOGISTICS COSTSTransport, inventory, coordination
46
Manufacturer Storage with Direct Shipping
Manufacturer
Retailer
Customers
47
Factories
Retailer
Customers
In-transit merge by carrier
In-Transit Merge Network
48
Factories
Customers
Warehouse storage by distributor/retailer
Distributor Storage with Carrier Delivery
49
Factories
Customers
Distributor/retailer warehouse
Distributor Storage with “Last Mile” Delivery
50
Factories
Customers
Cross Dock DC
Pickup sites
Retailer
Manufacturer or Distributor Warehouse with Consumer Pickup
51
Tailored Network: Multi - Echelon Finished Goods Network
RegionalRegionalFinishedFinished
Goods DCGoods DC
RegionalRegionalFinishedFinished
Goods DCGoods DC
Customer 1Customer 1DCDC
Store 1Store 1
NationalNationalFinishedFinished
Goods DCGoods DC
Local DCLocal DCCrossCross--DockDock
Local DC Local DC CrossCross--DockDock
Local DCLocal DCCrossCross--DockDock
Customer 2Customer 2DCDC
Store 1Store 1
Store 2Store 2
Store 2Store 2
Store 3Store 3
Store 3Store 3
52
Network Optimization ModelsAllocating demand to production facilitiesLocating facilities and allocating capacity
Speculative StrategySingle sourcing
Hedging StrategyMatch revenue and cost exposure
Flexible StrategyExcess total capacity in multiple plantsFlexible technologies
Which plants to establish? How to configure the network?
Key Costs:Fixed facility costTransportation costProduction costInventory costCoordination cost
53
Capacitated Plant Location Models
Max , ,1 1 1
n n m
i i i j i ji i j
Profit f y c x= = =
= ⋅ + ⋅∑ ∑∑
,1
1,...n
i j ji
x D j m=
= =∑
,1
1,...m
i j i ij
x K y i n=
≤ ⋅ =∑{ }0,1 1,...iy i n∈ =
n number of potential plant locationsm number of marketsfi annualized fixed cost of keeping plant i opencij cost of producing and shipping from i to j
Dj annual demand from market j
Ki potential capacity of plant i
yi 1 if plant i is open; 0 otherwise
xij quantity shipped from plant i to market jDecisions
54
Gravity Location Models
x, y Warehouse Coordinatesxn, yn Coordinates of delivery location ndn Distance to delivery location nfn Cost per ton mile to delivery location nDn Quantity to be shipped
−− +=n yyxxd nn
22 )()(
Min Total Cost1
k
n n nn
TC D d f=
= ⋅ ⋅∑
ASSUMPTION: TRANSPORT COSTS GROW LINEARLY WITH SHIPMENTS
55
Demand Allocation ModelWhich market is served by which plant?Which supply sources are used by a plant?
s.t.
All mkt demand satisfied
No factory capacity exceed
xij quantity shipped from plant site i to customer j
Cij cost to produce & ship one unit from factory i to market j
n no. of factory locationsm no. of marketsDj annual demand from
market j Ki annual capacity of factory i
,1
1,...n
i j ji
x D j m=
= =∑
,1
1,...m
i j ij
x K i n=
≤ =∑, 0 1,... ; 1,...i jx i n j m≥ = =
, ,1 1
n m
i j i ji j
MinC c x= =
= ∑∑
56
Warehouse and Plant Location ModelPlant and warehouse locations?Quantities shipped between various points?
, , , , , ,1 1 1 1 1 1 1 1
n t l n n t t m
i i e e h i h i i e i e e j e ji e h i i e e j
MinC f y f y c x c x c x= = = = = = = =
= + + + +∑ ∑ ∑∑ ∑∑ ∑∑
Fixed costsplants
warehouse
Shipping costsSupply source to plant
Plant to warehouseWarehouse to market
57
Warehouse and Plant Location Model,
11,...
n
h i hi
x S h l=
≤ =∑
, ,1 1
0 1,...l t
h i i eh e
x x i n= =
− ≥ =∑ ∑
,1
1,...t
i e i ie
x K y i n=
≤ =∑
, ,1 1
0 1,...n m
i e e ji j
x x e t= =
− ≥ =∑ ∑
,1
1,...t
e j je
x D j m=
= =∑
,1
1,...m
e j e ej
x W y e t=
≤ =∑
Supplier capacity
Balance supply-plant
Supplier capacity
Balance plant-warehouse
Warehouse capacity
Demand satisfaction
58
Network design decisions in practiceDo not underestimate the life span of plants
Long life hence long term consequencesAnticipate effect future demands, costs and technology changeStorage facilities easier to chance than production facilities
Do not gloss over cultural implicationsLocation – urban, rural, proximity to others
Do not ignore quality of life issuesWorkforce availability and morale
Focus on tariffs & tax incentives when locating facilities
Particularly in international locations
III.Supply chain management of flexible process networks - Lagrangean-based
decomposition techniques
Peter ChenJose M. Pinto
Content
MotivationDecomposition techniques considered
Lagrangean and Lagrangean/surrogate relaxationSubgradient and Modified subgradient optimization
Problem StatementCPFN Model
StructureObjective and constraints
ObservationDecomposition Applied
Lagrangean relaxationLagrangean/surrogate relaxation
Strategies ProposedResultsConclusionFuture Work
Motivation
Difficulties faced by most chemical companiesIncreasing number of competitorsIncreasing product variety from customer demandLarger and more complicated process networkMore efficient management is needed to survive and stay competitive
Why decomposition techniques are neededThe optimization of process networks are very difficult to solve using standard (full-scale) method.
Large computational effortTechnological barriers
Comparison of techniques are necessaryVarious decomposition techniques existThe effectiveness of the techniques is not standardized
Problem StatementA process network interconnects in a finite number of ways.
Processes I1~I4I1 is dedicatedI2, I3, and I4 are flexible
Chemicals J1~J6J1 and J2 are purchasedJ3 is consumed or sold as productJ4 and J6 are purchased or producedJ5 is sold as product
Sites C1 & C2C1: consists of all the processes and production schemes, contains byproduct J6.C2: doesn’t have I1, I3 contains only 3 schemes, and J6 is not produced.
Markets L1~L4L1 and L2 sells raw materialsL3 and L4 buys products
Bok et al. (2000) Ind. Eng. Chem. Res. 39, 1279-1290.
L1
L2 L4
L3
CFPN StructureI1.K1
I2.K1
I2.K2
I3.K1
I3.K2
I3.K3
I3.K4
I2.K1
I2.K2
I3.K1
I3.K2
I3.K3
I4.K1
I4.K2
I4.K1
I4.K2
Site 1
Site 2
J1 J2 J3 J4 J5 J6
CFPN Model – Objective FunctionObjective – Maximize the operating profit of the network
∑∑∑
∑∑∑∑
∑∑∑
∑∑∑∑∑
∑∑∑
∑ ∑ ∑∑∑
∑∑∑∑
∑∑∑∑
∈ ∈ ∈
∈ ∈ ∈ ∈
∈ ∈ ∈
∈ ∈ ∈ ∈ ∈
∈ ∈ ∈
∈ ∈ ∈ ∈ ∈
∈ ∈ ∈ ∈
∈ ∈ ∈ ∈
−
−
−
−
−
−
−
=
Jj Ll Ttjctjct
Dd Ll Cc Ttdlctdlct
Jj Cc Ttjltjlt
Ii Kk Kk Cc Ttctikkcikk
Jj Cc Ttjctjc
Ii JMj Kk Cc Ttijkctikct
Jj Ll Cc Ttjlctjlt
Jj Ll Cc TtjlctjltCPFN
F
YP
SF
Z
V
W
P
SZMax
i i
ik i
φ
ω
θ
ζ
ξ
δ
ϕ
γ
'''
Amount chemical sold
Amount chemical bought
Process operation cost
Inventory cost
Changeover cost
Shortfall penalty cost
Transportation cost
Raw material delivery cost
List of Assumptions
AssumptionsMass balance of raw materials and byproducts are proportional to the main product of the process and respective production scheme.The operating cost of a process is proportional to the amount of main product produced.Changeover only implies in cost and the overall time spent is negligible.Only one delivery of chemicals from one market over τc time interval is allowed. Demand is given by a range of values, having an upper and a lower bounds
CFPN Model – Constraints
Ratio of input chemicals to the main product
Ratio of output chemicals to the main product
Limits production under available capacity
Indicates when changeover occurred ( Zikk’ct = 1)
Allows only one production scheme per time period
TtCcKkJMjJIjIiWW iikikcjkctijijkcijkct ∈∈∈∈∈∈= ,,,',,'µ
TtCcKkJMjJOjIiWW iikikcjkctijijkcijkct ∈∈∈∈∈∈= ,,,',,'µ
TtCcKkJMjIiQW iikjicijckijkct ∈∈∈∈∈≤ ,,,,ρ
TtCcKkKkIiZYY iijctikkctikikct ∈∈∈∈∈≤−+ + ,,',,1 '1'
TtCcKkIiY ijKk
ikcti
∈∈∈∈=∑∈
,,,1
CFPN Model – Constraints (Cont’d)
Mass balance of chemicals in the network
Delivery of raw materials
Limits one delivery of chemicals over a τc time interval.
Prevents the purchase of raw material from exceeding the available amount
TtCcLlJjPYPPDd
Ujlctdlctjlct ∈∈∈∈≤ ∑
∈
,,,
TtCcLlDdYPYPYP dlcttdlctdlc ∈∈∈∈≤++ −− ,,,11,2,
TtLlJjaP Ujlt
Ccjlct ∈∈∈≤∑
∈
,,
{ }TtCcJjFWSVFPWV
cCctjc
Ii Kkijkct
Lljlctjctjct
Lljlct
Oi Kkijkcttjc
j ij i
∈∈∈+++=+++ ∑∑∑∑∑∑∑−∈∈ ∈∈∈∈ ∈
− ,,'
'1,
CFPN Model – Constraints (Cont’d)
Limits the product sales below the maximum allowable demand
Shortfall penalty if the minimum demand is not met
Bounds
TtLlJjSdSFSFCc
jlctLjlttjljlt ∈∈∈−+≥ ∑
∈− ,,1,
{ }1,0,0,,,,,,
1
'
'
∈
≥≤
≤
dlctikct
ctikkijkctjctjltjlctjlctjct
ctikk
Ujctjct
YPYZWVSFSPF
ZVV
TtLlJjdS Ujlt
Ccjlct ∈∈∈≤∑
∈
,,
Decomposition techniques considered
RelaxationLagrangean relaxation
Easier to solveRelaxing the right constraintObtaining a good multiplier
Lagrangean/surrogate relaxationReduction in the oscillating behavior
Updating multipliersSubgradient optimization
Simple algorithm structure
Modified subgradient optimizationAccelerating convergence
A more suitable step sizeImproved search direction
Solving relaxed problems
Updating the multipliers
Terminating criteria
Returning Solutions
Yes
No
Lagrangean and Lagrangean/surrogate relaxation
Lagrangean relaxation
Lagrangean/surrogate relaxation
n
m
yx
eDyCxbByAxdycxZ
}1,0{0
max
∈
≥
≤+≤+
+=
( )( )
n
m
yx
eDyCxByAxbucxuZ
}1,0{0
max)(
∈
≥
≤++−+=
( )( )
{ } .1,00
max)(
n
m
u
y
xeDyCx
ByAxbutcxtZ
∈
≥
≤++−⋅+=
General MIP
Narciso and Lorena, EJOR 1999, 114, 165
Subgradient and modified subgradient optimization
Subgradient optimization
Modified subgradient optimization
kk
kk gtuu +=+1( )( )2k
kk
kg
uLwlt −= ( )0,2 2121 >−≤≤ εεεε kl
kk
kk dtuu +=+1
;1−+= kk
kk dgd ξ
;)1(0 crr LLL αα −+=
( ) ≥
= − otherwiseerrif
rrr 26.316933.0
20εα
≥−
=−
−
−
Otherwise
gdIfd
gd kk
k
kk
k
0
,0121
1
γξ
−
= 2
)(1k
k
kk
d
uLLtβ
Fumero, Comp. & Oper. Res. 2001, 28, 33-52.
Observation
Constraints that link variables at different time period
The model can be decomposed into |T| separate problems if the variables at different time periods in these constraints are treated as different variables.
TtCcKkKkIiZYY iijctikkctikikct ∈∈∈∈∈≤−+ + ,,',,1 '1'
{ }TtCcJjFWSVFPWV
cCctjc
Ii Kkijkct
Lljlctjctjct
Lljlct
Oi Kkijkcttjc
j ij i
∈∈∈+++=+++ ∑∑∑∑∑∑∑−∈∈ ∈∈∈∈ ∈
− ,,'
'1,
TtCcLlDdYPYPYP dlcttdlctdlc ∈∈∈∈≤++ −− ,,,11,2,
TtLlJjSdSFSFCc
jlctLjlttjljlt ∈∈∈−+≥ ∑
∈− ,,1,
Decomposition applied
Following equations are declared and converted to the equivalentinequality form
Following variable replacement are done
The model is decomposed into |T| sub-problems through relaxation
Ctdlctdlc
Btdlctdlc
Adlctdlct
Bikctikct
Atikctikc
Ctjltjl
Bjltjlt
Ctjctjc
Bjctjct
YPYPYPYPYPYP
YYYY
SFSFSFSF
VVVV
2,2,1,1,
1,1,
1,1,
1,1,
,,
,
,
,
−−−−
++
−−
−−
→→→
→→
→→
→→
Cdlct
Bdlct
Cdlct
Bdlct
Cdlct
Bdlct
Bdlct
Adlct
Bdlct
Adlct
Bdlct
Adlct
Bikct
Aikct
Bikct
Aikct
Bikct
Aikct
Cjlt
Bjlt
Cjlt
Bjlt
Cjlt
Bjlt
Cjct
Bjct
Cjct
Bjct
Cjct
Bjct
YPYPandYPYPYPYP
YPYPandYPYPYPYP
YYandYYYY
SFSFandSFSFSFSF
VVandVVVV
≤≥→=
≤≥→=
≤≥→=
≤≥→=
≤≥→=
Lagrangean relaxation
Relaxing the following inequalities into objective
Adding the remaining inequalities as constraints
Resulting objective function
Cdlct
Bdlct
Bdlct
Adlct
Bikct
Aikct
Cjlt
Bjlt
Cjct
Bjct YPYpYPYPYYSFSFVV ≥≤≥≤≤ ,,,,
Cdlct
Bdlct
Bdlct
Adlct
Bikct
Aikct
Cjlt
Bjlt
Cjct
Bjct YPYpYPYPYYSFSFVV ≤≥≤≥≥ ,,,,
∑ ∑∑∑∑ ∑∑∑∑ ∑∑
∑ ∑∑∑∑ ∑∑∑ ∑∑
∑ ∑∑∑∑ ∑∑∑ ∑∑∑∑∑ ∑∑
∑ ∑ ∑ ∑∑∑ ∑∑∑∑ ∑∑∑
∈ ∈ ∈ ∈∈ ∈ ∈ ∈∈ ∈ ∈
∈ ∈ ∈ ∈∈ ∈ ∈∈ ∈ ∈
∈ ∈ ∈ ∈∈ ∈ ∈∈ ∈ ∈ ∈ ∈∈ ∈ ∈
∈ ∈ ∈ ∈ ∈∈ ∈ ∈ ∈∈ ∈ ∈ ∈−
−+
−+
−+
−+
−+
−
−
−
−
−
−
−
=
Tt Dd Ll Cc
Cdlct
Bdlct
YPdlct
Tt Dd Ll Cc
Adlct
Bdlct
YPdlct
Tt Jj Ll
Bjlt
Cjlt
SFjlt
Tt Ii Kk Cc
Bikct
Aikct
Yikct
Tt Jj Cc
Bjct
Cjct
Vjct
Tt Jj Lljctjct
Tt Dd Ll Cc
Adlctdlct
Tt Jj Cc
Bjltjlt
Tt Ii Kk Kk Ccctikkcikk
Tt Jj Cc
Bjctjc
Tt Ii JMj Kk Ccijkctikct
Tt Jj Ll Ccjlctjlt
Tt Jj Ll CcjlctjltLRCFPN
YPYPuYPYPuSFSFu
YYuVVuF
YPSFZV
WPSZMax
i
i i
ik i
)()()(
)()(
21
'''
φ
ωθζξ
δϕγ
Lagrangean relaxation (Cont’d)
Resulting objective function at time period t
The total profit is equal to the summation over the time periods
∑∑∑∑∑∑
∑∑∑∑∑∑∑
∑∑∑∑∑∑∑∑∑∑∑
∑∑∑ ∑ ∑∑∑∑∑∑∑∑
∈ ∈ ∈∈ ∈ ∈
∈ ∈∈ ∈ ∈∈ ∈
∈ ∈∈ ∈ ∈∈ ∈∈ ∈ ∈ ∈
∈ ∈∈ ∈ ∈ ∈∈ ∈ ∈∈ ∈ ∈−
−+−+
−+−+−+
−−−−
−−−=
Dd Ll Cc
Cdlct
Bdlct
YPdlct
Dd Ll Cc
Adlct
Bdlct
YPdlct
Jj Ll
Bjlt
Cjlt
SFjlt
Ii Kk Cc
Bikct
Aikct
Yikct
Jj Cc
Bjct
Cjct
Vjct
Jj Lljctjct
Dd Ll Cc
Adlctdlct
Jj Cc
Bjltjlt
Ii Kk Kk Ccctikkcikk
Jj Cc
Bjctjc
Ii JMj Kk Ccijkctikct
Jj Ll Ccjlctjlt
Jj Ll Ccjlctjlt
tLRCFPN
YPYPuYPYPu
SFSFuYYuVVu
FYPSFZ
VWPSZMax
i
i i
ik i
)()(
)()()(
21
''' φωθζ
ξδϕγ
∑∈
−− =Tt
tLRCFPNLRCFPN ZZ
Lagrangean relaxation - Constraints
CcKkJMjJIjIiWW iikikcjkctijijkcijkct ∈∈∈∈∈= ,,',,'µ
CcKkJMjJOjIiWW iikikcjkctijijkcijkct ∈∈∈∈∈= ,,',,'µ
CcKkJMjIiQW iikjicijckijkct ∈∈∈∈≤ ,,,ρ
CcKkKkIiZYY iijctikkCtcik
Bikct ∈∈∈∈≤−+ + ,',,1 '1,'
CcKkIiY ijKk
Bikct
i
∈∈∈=∑∈
,,1
{ }CcJjFWSVFPWV
cCctjc
Ii Kkijkct
Lljlct
Bjctjct
Lljlct
Oi Kkijkct
Ctjc
j ij i
∈∈+++=+++ ∑∑∑∑∑∑∑−∈∈ ∈∈∈∈ ∈
− ,'
'1,
CcLlJjPYPPDd
Ujlct
Cdlctjlct ∈∈∈≤ ∑
∈
,,
CcLlDdYPYPYP Adlct
Btdlc
Ctdlc ∈∈∈≤++ −− ,,11,2,
List of constraints
Lagrangean relaxation – Constraints (Cont’d)
LlJjaP Ujlt
Ccjlct ∈∈≤∑
∈
,
LlJjdS Ujlt
Ccjlct ∈∈≤∑
∈
,
LlJjSdSFSFCc
jlctLjlt
Ctjl
Bjlt ∈∈−+≥ ∑
∈− ,1,
1,, ' ≤≤≤ ctikkUjct
Cjct
Ujct
Bjct ZVVVV
Cdlct
Bdlct
Bdlct
Adlct
Bikct
Aikct
Cjlt
Bjlt
Cjct
Bjct YPYpYPYPYYSFSFVV ≤≥≤≥≥ ,,,,
0,,,,,,,, ' ≥ctikkijkctCjct
Bjct
Cjlt
Bjltjlctjlctjct ZWVVSFSFSPF
{ }1,0,,,, ∈Cdlct
Bdlct
Adlct
Bikct
Aikct YPYPYPYY
Lagrangean/surrogate relaxation
Lagrangean/surrogate relaxation is done in a similar fashion like the Lagrangean relaxationResulting objective function at each time period t
Subject to the same constraints as the Lagrangean relaxation Total Profit:
−+−+
−+−+−×+
−−−−
−−−=
∑∑∑∑∑∑
∑∑∑∑∑∑∑
∑∑∑∑∑∑∑∑∑∑∑
∑∑∑ ∑ ∑∑∑∑∑∑∑∑
∈ ∈ ∈∈ ∈ ∈
∈ ∈∈ ∈ ∈∈ ∈
∈ ∈∈ ∈ ∈∈ ∈∈ ∈ ∈ ∈
∈ ∈∈ ∈ ∈ ∈∈ ∈ ∈∈ ∈ ∈−
Dd Ll Cc
Cdlct
Bdlct
YPdlct
Dd Ll Cc
Adlct
Bdlct
YPdlct
Jj Ll
Bjlt
Cjlt
SFjlt
Ii Kk Cc
Bikct
Aikct
Yikct
Jj Cc
Bjct
Cjct
Vjct
Jj Lljctjct
Dd Ll Cc
Adlctdlct
Jj Cc
Bjltjlt
Ii Kk Kk Ccctikkcikk
Jj Cc
Bjctjc
Ii JMj Kk Ccijkctikct
Jj Ll Ccjlctjlt
Jj Ll Ccjlctjlt
tLSCFPN
YPYPuYPYPu
SFSFuYYuVVut
FYPSFZ
VWPSZMax
i
i i
ik i
)()(
)()()(
21
''' φωθζ
ξδϕγ
∑∈
−− =Tt
tLSCFPNLSCFPN ZZ
General Algorithm Setting bounds and
determining parameters 1
Fix: Yikc,1Vjc,1SFjl,1YPdlc,1
2
Fix: Yikc,2Vjc,2SFjl,2YPdlc,2
3
Fix: Yikc,3Vjc,3SFjl,3YPdlc,3
T…Fix: Yikc,t-1Vjc,t-1SFjl,t-1YPdlc,t-1
|T| sub-problems
Fix Yikct and YPdlct
Terminating Criteria
Starting next iteration
Updating the multipliers:
Subgradient optimization
Modified subgradient optimization
No
Return lower bound and corresponding solution
Yes
t = 2t = 3
t = T
Linking Constraints
Full Scale
t = 1
Proposed strategy
Strategy ALagrangean relaxation with subgradient optimization
Strategy BLagrangean relaxation with modified subgradient optimization
Strategy CLagrangean/surrogate relaxation with subgradient optimization
Strategy DLagrangean/surrogate relaxation with modified subgradient optimization
Results – Calculation time
Calculation time vs time period
0
200400
600800
1000
0 20 40 60 80
time period
Cal
cula
tion
time
(sec
)
Full-scaleStrategy AStrategy C
Results – Solution value
Solution Value vs time period
05000
1000015000200002500030000
0 20 40 60 80
time period
Solu
tion
valu
e ($
)
Full-scaleStrategy AStrategy C
Results – Percent difference from the optimum
1.3131.31742
0.8240.83335
0.8260.82928
1.7231.19521
0.5360.55414
0.0000.0007
Strategy CStrategy Atime period
Conclusion
Calculation timeTime spent is much less than the full scale method
Solution valueThe percentage deviation from the optimum is below 2%
Lagrangean vs. Lagrangean/surrogate relaxationLagrangean/surrogate always used equal or more iterationsLagrangean/surrogate spent slightly more timeFor same number of iterations, Lagrangean relaxation gave equal or better solution value
Alternative approaches
Decentralized approachModel predictive control strategies
Multiproduct, multiechelon distribution networks with multiproduct batch plants (Perea, Ydstie and Grossmann, 2003)
Comparison with integrated approachPoorer coordination of the supply chain decisionsSmaller computational time
Future work
Implementing modified subgradient optimizationTesting strategies B and D and compare them with A and CSearch for new strategies
Other decomposition methodsApplying search_t* algorithm for each set of multiplier values
JOSÉ M. PINTO
IV.Planning, Scheduling and Supply
Chain Management for Operations in Oil Refineries
OUTLINE
• Introduction • Planning Models
– refinery diesel production
• Scheduling models– crude oil scheduling– fuel oil / asphalt area
• Logistics– oil supply model
• Petroleum Supply Chain• Conclusions
MOTIVATION
Refinery TargetsProfitability Maximization (Pelham and Pharris, 1996; Ramage, 1998)
Minimization of Operational Costs(Bodington and Shobrys, 1996)
Beginning of Computational Applications for Planning/Scheduling:
Petrochemical Industry: 1950s (Linear Programming) (Symonds, 1955; Bodington, 1992) (Dantzig, 1963)
CPI in general: 1970s (Reklaitis, 1991; Kudva and Pekny; 1993)
ADVANCESAvailability of more powerful and less expensive computers;Mathematical Developments:
Time representation; (Moro and Pinto, 1998)
Combinatorics in MIP;(Raman and Grossmann, 1994)
Non-convexities in MINLP;(Viswanathan and Grossmann, 1990)
Consequences for the Petroleum Industry: (Ramage, 1998)
Unit Level Optimization (FCC, Crude Unit, etc..)
Large Portions of the Plant or Plant-wide Optimization
1980’s 1990’s
OPTIMIZATION IN REFINERY OPERATIONS
LPs in crude blending and product pooling (50´s) (Symonds, 1955)
Advanced control : MPC (Cutler, DMCC,1983)
Planning models (Coxhead, Moro et al, 1998.)
crude selection, crude allocation for multi refinery
partnership models for raw material supply
OVM Refinery, Austria (LP) (Steinschorn and Hofferl, 1997)
In-house simulation models for scheduling (Magalhaes et al., 1998)
Scheduling optimization models
gasoline blending (Bodington, 1993)
gasoline production, TEXACO(NLP) (Rigby et al., 1995)
crude oil unloading (Lee et al., Shah, 1996)
OUTLINE
• Introduction • Planning Models
– refinery diesel production• Scheduling models
– crude oil scheduling– fuel oil / asphalt area
• Logistics– oil supply model
• Conclusions
Objectives
• To develop a general representation for refinery units
– streams with multiple inputs and destinations
– nonlinear mixing and process equations
– bounds on unit variables
• To apply to the production planning of a real world refinery
– diesel production
– to satisfy multiple specifications
PLANNING MODEL FOR REFINERIES
TYPICAL PROCESS UNIT
UNIT EQUATIONS
- Feed flowrate:
- Feed Properties:
Pu,F,j = fj ( Qu’,s,u , Pu’,s,j ) u’ Uu, s Su’,u, j Js
- Total flowrate of each product stream:
Qu,s = f ( Qu,F , Pu,F,j , Vu ) j JF, s SU
- Unit product stream properties:
Pu,s,j = fj ( Pu,F,j , Vu ) j Js, s SU
- Product streams flowrates (splitter):
s SU
Qu F Qu s us Su uu Uu
, ' , ,,
=∈ ′′∈∑∑
Q Qu s u s uu Us,u
, , ,= ′′∈∑
∈∈
∈
∈∈
∈∈∈
HYDROTREATING UNIT (HT)
HT MODEL
Feed flowrate:
QHT,F=QCD1,HD,HT+QCD2,HD,HT+QFCC,LCO,HT+QCK,CGO,HD
Feed properties:
PHT,F,j = fj (Qu’,s,HT , Pu’,s,j ) u’ UHT , s Su’,HT, j JHT
Example - Flash Point (FP)
Iu,HD,FP = exp[10006.1/(1.8 Pu,HD,FP+415) - 14.0922] u UHT
( )PHT F FP, , ..
ln .=
+−
0 5510006 1
14 0922415
αα =
∈
∈
∑
∑
Q I
Q
u HD HTu HT
u HD FP
u HD HTu HT
U
U
, , , ,
, ,
∈
∈∈∈
REAL-WORLD APPLICATION
Planning of diesel productionPetrobras RPBC refinery in Cubatão (SP, Brazil).
Three types of diesel oil:
Metropolitan Diesel. Low sulfur levelsmetropolitan areas
Regular Diesel. Higher sulfur levelsother regions of the country
Maritime Diesel. High flashing point.
DIESEL SPECIFICATIONS
PropertyREGULAR
DIESELMETROPOLITAN MARITIME
DENSITYmin / max
0.82/0.88
0.82/0.88
0.82/0.88
FLASH POINTmin (°C)
- - 60.0
ASTM 50%min / max (°C)
245.0/310.0
245.0/310.0
245.0/310.0
ASTM 85%max (°C)
370.0 360.0 370.0
CETANE NUMBER min 40.0 42.0 40.0SULFUR CONTENT max
(% WEIGHT)0.5 0.2 1.0
MAIN RESULTS
Potential Improvement US$ 23,000 / day or US$ 8,000,000 / yrImplemented with on-line data acquisition
OUTLINE
• Introduction • Planning Models
– refinery diesel production• Scheduling models
– crude oil scheduling– fuel oil / asphalt area
• Logistics– oil supply model
• Conclusions
PROPOSED APPROACH FOR PLANNING AND SCHEDULING
CrudeScheduling
FRACTIONATION
LPG Scheduling
Fuel Oil / AsphaltScheduling
REFINERY PLANNING
SHORT TERM CRUDE OIL SCHEDULING Crude Oil System
OBJECTIVES
maximize total operating profitrevenue provided by oil processing cost of operating the tanks
generate a schedule for crude oil operations receiving oil from pipelinewaiting for brine settlingfeeding the distillation units
TIME SLOT REPRESENTATION
MILP OPTIMIZATION MODEL
Max total operating profitsubject to:
Timing constraintsPipeline material balance equationsPipeline operating rules
Pipeline always connected to a tank
Material balance equations for the tanksVolumetric equationsComponent volumetric balance
Tank operating rulesMinimum settling time
Rules for feeding the distillation unit
DECISION VARIABLES
slot k
Ypj,kYdf,j,k
fraction f
REAL-WORLD EXAMPLE
Oil parcel Volume
(m3)
Start time
(h)
End Time
(h)
Composition
1 60,000 8 20 100% Bonito
2 50,000 48 58 100% Marlin
3 1,000 58 58.2 100% Marlin
4 60,000 100 112 100% RGN
Tank initial conditions
Distillation target flowrate = 1500 m3/h
RESULTS
MODEL SOLUTION
• GAMS / OSL
• CPU time2.80 hrs (Pentium II 266 MHz 128 MB RAM)
• Variable size time slot model912 discrete variables
3237 continuous variables
5599 equations
• Fixed size time slot model
21504 discrete variables !
OUTLINE
• Introduction • Planning Models
– refinery diesel production• Scheduling models
– crude oil scheduling– fuel oil / asphalt area
• Logistics– oil supply model
• Conclusions
FUEL OIL/ASPHALT PRODUCTION SCHEDULING PROBLEM
•The plant produces ≅ 80% of all Brazilian fuel oil;
•The plant has relevant storage limitations;
•Complexity of distribution operations;
•End of the monopoly in the Brazilian oil sector.
Product Base Diluent usedFO1FO2FO3FO4UVO1UVO2CAP07CAP20
OCC+LCO or OCC or LCOOCC+LCO or OCC or LCOOCC+LCO or OCC or LCOOCC+LCO or OCC or LCOpure LCOpure LCOpure HGpure HG
RASF RASF RASF RASF RASF RASF RASF RASF
major specification:viscosity
MATHEMATICAL MODELS
non-convex MINLP (5 bilinear products in the viscosity constraints);
Uniform Discretization of Scheduling Horizon;
Objective Function: Minimize the Operational Cost.
First Approach:
MILP;
Second Approach:
Linear Transformation
COST = Raw-Material Costs + Inventory Costs + + Pumping Costs + Transition Costs
MINLP MODEL
) CHANGETRAN()]FIDCB(
)VQCINVD()VQCINVQ()VICINVI(
FRASFMCRFRLCOCDFOCCRCD)FDCCD([COST
p,n
O
1o
P
1p
N
1nn,p,ot,o,i
I
1i
O
1oi
t,d
D
1ddt,q
Q
1qqt,i
I
1ii
t3t2t,s
T
1t
S
1ss
⋅+⋅+
+⋅+⋅+⋅+
+⋅+⋅+⋅+⋅=
∑∑∑∑∑
∑∑∑
∑ ∑
= = == =
===
= =
Minimize:
Material Balance Constraints:
Subject to:
volume in i at t’ = initial volume in i + [ inputs in i up to t’ - ( outputs from i up to t’ ) ]
the volume capacities of all tanks are also subject to bounds
Operating Rules for the Plant:
at each t, the plant production must be stored in one single tank
,...,T1 t1)(XQC)(XICQ
1qq,t
I
1ii,t ==+∑∑
==
FO area UVO/asphalt area
Demand Supply of Plant Products
∑=
==≤+O
1oi,o,ti,t ,...,T1,...,I; t1 i1)(XIDXIC
simultaneous tank loading and unloading is not allowed (exception: HG storage tank)
FO area
Operating Rules for the Plant (continuation):
UVO / Asphalt may be sent to truck terminals only between 6:00 a.m. and 6:00 p.m.
,...,Q1 q;b)DT/12(t1)1b()DT/12()];DT/12/(T[,...,1b
HTXQD bq,t
=⋅≤≤+−⋅=
≤
while a asphalt is produced, the RASF diluent must be HG
,...,T 1 t0XDRASF)(XQC8
5q,t1q,t ==−∑
=
while asphalt is produced, the OCC stream from UFCC must be directed to storage in TK-42208
Material Flow Constraints:
,...,T1 t1)XZ1(XDRASF t,t1 =≤−+
flowrates to oil-pipelines must obey pump limitations
,...,T1,...,O; t1,...,I; o1 iFIDXIDFID0 maxi,o,ti,o,t ===⋅≤≤
flowrates to truck terminals must obey pump limitations
,...,T1,...,Q; t1 qFQDXQDFQD0 maxq,tq,t ==⋅≤≤
Viscosity Constraints:
at each t, the viscosity adjustment must be done regarding the kind of product
,...,T1) tXIC(MIFO)XQC(MIUVVISC i,t
I
1i Pppq,t
Q
1q Vvvt
iq
=⋅+⋅= ∑ ∑∑ ∑= ∈= ∈
also, the availability of diluents should be considered
,...,T1t]}FRLCOFOCCRFRASFU)(FDRASF([VISC
]MIDFRLCOMIDFOCCRMIRASFFRASFU)MID(FDRASF{[
,...,T 1t
VISC]}FRLCOFOCCRFRASFU)(FDRASF([
]MIDFRLCOMIDFOCCRMIRASFFRASFU)MID(FDRASF{[
ttt
D
1dd,tt
3t2tt
D
1d
2s
Sssd,t
t
ttt
D
1dd,t
3t2tt
D
1d
2s
Sssd,t
d
d
=+++⋅=
=⋅+⋅+⋅+⋅
=
=+++
⋅+⋅+⋅+⋅
∑
∑ ∑
∑
∑ ∑
=
=
≤
∈
=
=
≤
∈
or
5 bilinear products non-convex MINLP
EXACT MILP MODEL
Similar MINLP model structure; More continuous variables than MINLP model;More constraints than MINLP model;Combinatorial feature of the MINLP model preserved.
Characteristics:
Structure: non-convex
MILP Model = MINLP Model +
Nonlinear Viscosity Constraints
+ Linearized Constraints
CONSTRAINTS FOR LINEAR TRANSFORMATION
,...,T1',...,I; t1)] i(FIDK[FIRASFKMIVIZVIKO
1oi,o,t
't
1ti,tii'i,t ==−+⋅= ∑∑
==
,...,T1',...,Q; t1) qFQDK(FQRASFKMQVQZVQK q,t
't
1tq,tqq'q,t ==−+⋅= ∑
=
,...,T1 t)FQRASFK()FIRASFK(MIDFRLCO
MIDFOCCR)MIDFDRASF( MIRASFFRASFU
I
1i
Q
1qt,qt,i3t
2t
D
1ds
2s
Sst,dt
d
=+=⋅+
+⋅+⋅+⋅
∑ ∑
∑ ∑
= =
=
≤
∈
,...,T1,...,I; t1t itan consU U XICFIRASFK i,ti,t ===⋅≤
,...,T1,...,Q; t1t qtanconsU UXQCFQRASFK q,tq,t ===⋅≤
,...,T1,...,I; t1 i VIKMIVI t,iit,i ===⋅
,...,T,...,Q; t qVQKMQVQ q,tqq,t 11 ===⋅
,...,T1t ;O,...,1,...,I; o1 i FIDKMIFID t,o,iit,o,i ====⋅
,...,T1,...,Q; t1 qFQDKMQFQD q,tqq,t ===⋅
Mat
eria
l Bal
ance
Flow
Vis
cosi
ty
(48)
(49)
(50)
(52)
(51)
(54)
(53)
(47)
(46)
REAL-WORLD EXAMPLE
instanceevaluated
Scheduling horizon: 3 days
Time span: 2 hours
Nominal production: 200,000 m3/month
PRODUCTION SCHEDULE AND STORAGE INFORMATION
0
1
TK-4
4113
TK-4
3307
TK-4
3301
TK-4
3301
TK-4
3307
TK-4
3301
TK-4
3302
TK-4
4108
TK-4
3302
TK-4
4108
TK-4
3301
TK-4
4108
TK-4
3301
TK-4
4108
TK-4
3302
TK-4
3307
TK-4
3301
TK-4
4113
TK-4
3301
TK-4
3301
TK-4
3307
TK-4
3307
TK-4
3301
TK-4
4108
TK-4
4108
TK-4
3307
TK-4
3301
TK-4
3301
TK-4
3301
TK-4
3303
TK-4
3302
TK-4
3307
TK-4
3301
TK-4
3301
TK-4
3301
TK-4
3301
START END
UVO1 UVO2 CAP07 CAP20 FO1 FO2 FO3 FO4
020406080
100120140160180
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35
Each
inte
rval
= 20
m3 /h
diluent from TK-42221 (HG) diluent from TK-42208 (OCC+LCO)pure OCC from UFCC pure LCO from UFCC
FO1 (p=1)
02468
10
1 4 7 10 13 16 19 22 25 28 31 34 37
TK-43301 (i=1)
TK-43302 (i=2)
FO2 (p=2)
02468
1012
1 36
TK-43303 (i=3)TK-43304 (i=4)
FO3 (p=3)
02468
10
1 36
TK-43305 (i=5)TK-43306 (i=6)
FO4 (p=4)TK-43307(i=7)
02468
10
1 36
CAP-20 (v=4)
0
2
4
1 36
TK-44110 (q=6)TK-44115 (q=7)TK-44116 (q=8)
CAP-07 (v=3)TK-44108 (q=5)
0
2
4
1 36
UVO2 (v=2)
0
2
4
1 36
TK-44111 (q=3)TK-44112 (q=4)
UVO1 (v=1)
0
2
4
1 36
TK-44113 (q=1)TK-44114 (q=2)
Schedule of diluents in the mixer
Volume (x 10-3 m3) in product storage tanks
OIL-PIPELINE TO SÃO PAULO - CASE A(RESTRICTED PERIOD: 6:00 a.m./12:00 p.m. - 1st day)
0
100
200
300
400
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35
(m3/h) FO1FO2FO3FO4
OIL-PIPELINE TO SÃO PAULO - CASE B(RESTRICTED PERIOD: 0:00 a.m./6:00 p.m. - 2nd day)
0
100
200
300
400
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35
(m3/h) FO1FO2FO3FO4
OIL-PIPELINE TO SÃO PAULO - CASE C(RESTRICTED PERIOD: 6:00 p.m./12:00 a.m. - 2nd/3rd days)
0
100
200
300
400
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35
(m3/h) FO1FO2FO3FO4
OIL-PIPELINE TO SÃO PAULO - CASE D(RESTRICTED PERIOD: 12:00 a.m./6:00 a.m. - 3rd/4th days)
0
100
200
300
400
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35
(m3/h) FO1FO2FO3FO4
UVO1 TRUCK TERMINAL
020406080
100
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35
(m3/h)
from TK-44113 (q=1)
from TK-44114 (q=2)
UVO2 TRUCK TERMINAL
020406080
100
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35
(m3/h)
from TK-44111 (q=3)
from TK-44112 (q=4)
CAP-07 TRUCK TERMINAL
from TK-44108 (q=5)
020406080
100
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35
(m3/h)
CAP-20 TRUCK TERMINAL
020406080
100
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35
(m3/h)
from TK-44110 (q=6)from TK-44115 (q=7)from TK-44116 (q=8)
Dispatch schedule for ultra-viscous oilTransfer schedule for fuel oils
TRANSITION PROCESS IN OIL-PIPELINES
to
t1 > to
t2 > t1
t3 > t2
TRANSITION COST
Final Product A
Final Product B
Final Product C
FLOW
UNDESIRABLE MIX
Refinery Market
4514
6890
15121512
4465
2629
MINLP model MILP model
number of 0-1 variablesnumber of constraintsnumber of continuous variables
68907985
1512 19684465
6733
WITHOUT TRANS.CONSTRAINTS
WITH TRANS.CONSTRAINTS
number of 0-1 variablesnumber of constraintsnumber of continuous variables
case MIP model nodes iterations CPU time (s) objectiveMILP 937 15674 570.46 969.61A MINLP - 13815 335.36 966.99MILP 1296 16626 711.01 965.72B MINLP - 15508 391.45 961.14MILP 764 13086 490.86 954.99C MINLP - 23792 531.98 956.99MILP 1197 23080 851.78 950.65D MINLP - 12845 299.30 959.49
OIL-PIPELINE TO SÃO PAULO(WITH TRANSITION MODELING)
0
100
200
300
400
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35
(m3/h) FO1FO2FO3FO4
LOCAL OIL-PIPELINE(WITH TRANSITION MODELING)
0
100
200
300
400
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35
(m3/h) FO1FO2FO3FO4
MILP models
OUTLINE
• Introduction • Planning Models
– refinery diesel production• Scheduling models
– crude oil scheduling– fuel oil / asphalt area
• Logistics– oil supply model
• Conclusions
CRUDE OIL SUPPLY PROBLEM
• Solution of oil supply problems among crude oil terminals and refineries
MOTIVATION
• Increasing utilization of the system– Larger crude oil demand for crude oil in refineries– Outsource of transportation
• Potential economic impact– No systematic scheduling– Operations involve high costs and aggregated values
• Petrobras distribution complex– 4 refineries in the State of Sao Paulo
PETROBRAS DISTRIBUTION COMPLEX
Types of crude oil
PROBLEM SPECIFICATION
iDetermined by the petroleum origin
iApproximately 42 types of crude oil may be processed
Types ofcrude oil
Classes of crude oil
iSets of crude oil types with similar properties
iNecessary due to limited amount of tanks
i7 classes
PROBLEM SPECIFICATION
Types of crude oil
Classes of crude oil
Tankers
iTransport types of crude oil
iOverstay incurs in additional costs- US$ 10 k to US$ 20 k per day
PROBLEM SPECIFICATION
Types of crude oil
Classes of crude oil
Tankers
Piers
iDifferent capacities
PROBLEM SPECIFICATION
TanksPiersiStore classes of crude
oil
iMinimum storage levels
iSettling time between loading and unloading operations
Types of crude oil
Classes of crude oil
Tankers
PROBLEM SPECIFICATION
Tanks
Pipe-lines
Classes of crude oil
Tankers
Piers
iFlow rate at each pipeline limited by the density of the heaviest crude oil class
iPossible to connect to at most one tank at everytimeTypes of
crude oil
PROBLEM SPECIFICATION
Tanks
Pipe-lines
Sub-stations
Types of crude oil
Classes of crude oil
Tankers
Piers
iBuffer operations between terminal and refineries
iStore difference in flow rate between inlet and outlet pipelines
PROBLEM SPECIFICATION
Mathem
atical formulation
Tanks
Tankers
Substations
Refineries
Pipelines
Crude types
Piers
• Input parameters– Operating
constraints– Initial
inventory– Costs– Possible
allocations
System operation
iSchedules- Allocation of
crude oil types to classes
- Assignment of tankers to piers
- Loading- Unloading- Settling
OBJECTIVES
Terminal
Pipelines
Substations
RefineriesImpossible to solve complete problem
PROPOSED STRATEGY
iDecomposition of the problem in three formulations
- Port Model
- Substation Model
- Algorithm to adjust timing of pipelines
PROPOSED STRATEGY
• Port Model
• Results– Allocation of tankers to
piers– Loading and unloading
profiles of tanks– Loading of pipelines– Timing of interfaces in
pipelines
PROPOSED STRATEGY
Port
- Algorithm to adjust timing of pipelines
Port Results for this refinery
PROPOSED STRATEGY
Pipelines
• Substation Model
Port
Pipelines
PROPOSED STRATEGY
Substations
Terminal
Pipelines
Substations
- Algorithm to adjust timing of pipelines
Refineries
… results for other refineries are determined
PROPOSED STRATEGY
MATHEMATICAL FORMULATION
• MILP model formulation• Time representation
– Continuous– Based on events
time
Vi Vi+1 Vi+2 Vi+3
Qi Qi+1 Qi+2 Qi+3
Xi Xi+1 Xi+2 Xi+3
Inventory level (cont. variable)
Amount generated (cont. variable)
Decision to produce (disc. variable)
Ti Ti+1 Ti+2 Ti+3Time events (cont. variable)
PROPOSED MODEL - VARIABLES
• Binary variables – Decisions– Assignment of ship n to pier p:– Unloading of ship n to tank t:– Unloading of tank t to oil pipeline o:
• Continuous variables– Timing– Inventory– Flowrates– Operating profit
pnA ,
etnLT ,,
eotUT ,,
MODEL ASSUMPTIONS AND OPERATING RULES
– Ships with earlier arrival date unload first in the same pier
– Each ship unloads to only one tank at any time
– Each pipeline receives crude oil from at most one tank at any time
– Each refinery is connected to the docking stations from one and
only one oil pipeline
– The same crude oil class has constant flowrates
PROPOSED MODEL - TIMING
• Ships, tanks and pipelines– Timing variables in each time event
• Initial• Final
• Matching of the timing variables– Unloading from ship n to tank t
– Unloading from tank t to pipeline o
fet
fen TTTN ,, =
seo
set TDTT ,, = f
eofet TDTT ,, =
set
sen TTTN ,, =
PORT MODEL - CONSTRAINTS
• Decisions
– Assignment of tanker n:
– Operation of tank t:
– Operation of tanker n:
– Operation of oil pipeline o:
1, =∑∈ nPp
pnA
1,,,, ≤+ ∑∑∈∈ tt Oo
eotNn
etn UTLT
1,, ≤∑∈ oTt
eotUT
1,, ≤∑∈ nTt
etnLT
PROPOSED MODEL – CONSTRAINTS
• Material balances– Tanks, Refineries
• Operational constraints– Ships and tanks: flowrate bounds
• Timing– Ships
– Tanks
( )∑∈
+≥nPp
entpnpn
startpn
sen ATN ,,,, .θτ
∑∈
≤nPp
endpn
fenTN ,, τ
fen
sen TNTN 1,, −≥
HTT fet ≤,
fet
set TTTT 1,, −≥
)1.( ,,,,,, −+∆+≥ ∑∑∈
′∈
′tt Nn
etnOo
eotdect
fet
set LTUTTTTT
PROPOSED MODEL – CONSTRAINTS
• Matching of timing variables– Ships ↔ Tanks
– Tanks ↔ Pipelines
)1.()1.( ,,,,,,, etns
ensetetn
sen LTHTNTTLTHTN −+≤≤−−
)1.()1.( ,,,,,,, etnfen
fetetn
fen LTHTNTTLTHTN −+≤≤−−
)1.()1.( ,,,,,,, eotset
seoeot
set UTHTTTDUTHTT −+≤≤−−
)1.()1.( ,,,,,,, eotfet
feoeot
fet UTHTTTDUTHTT −+≤≤−−
PROPOSED MODEL – OBJECTIVE FUNCTION
( )
( )
( )
cost) (interface .
tankers)oil theofcost (overstay .
cost)n utilizatio(pier .
tanks)in thecost (oil .
port) in the revenue oil initial - (final .
)refineries the torevenue (oil .max
1
1,,,,
,,
,
0,
1
1'',,,
∑ ∑ ∑ ∑
∑
∑ ∑
∑ ∑
∑ ∑
∑ ∑ ∑ ∑ ∑
∈≠′
∈′
−
=′′
∈
∈
∈
∈ ∈ ∩∈
−
=
−
−
−−
−
−+
=
o CLOclcllc
CLOlc
E
eeolccl
facelccl
sen
n
sen
p Nn
startpn
endpn
pierp
c Nncn
crudec
cl Ttt
TEt
classcl
r CLRcl Oo TTt
E
eeot
classrcl
o o
p
c
cl
r r ocl
INTCOST
TCOST
COST
CCOST
VVREVP
QutREVR profit
ττ
SUBSTATION MODEL - MAIN ASSUMPTIONS
iTanks cannot be loaded and unloaded simultaneously
iOutlet pipelines cannot be loaded by inlet pipelines and tanks simultaneously
iSubstation must receive crude oil at the flow ratesgenerated by the Pot Model- Lots of crude oil
SUBSTATION MODEL – SUMMARY
iMinimize- Cost = cost of tank loading/unloading +
cost of pipeline alignment +cost of interface
- Subject to:- Assumptions of the substation model- Operating constraints 8 Tank loading/unloading 8 Pipeline operation
- Timing constraints8 Inlet pipelines8 Tanks8 Outlet pipelines
REAL-WORLD PROBLEMREPLAN
RECAP
SEBATRPBC
GEBAST
SEGUA
REVAP
OSBAT III
OSBAT II
OSVAT I
OSVAT II
OSvAT III
OSVAT IV
OSBAT IV
P4
P2
P3
P1
São Sebastião
São José dosCampos
Paulínia
Guararema
Capuava
Cubatão
• Problem 1– Port Model
• Problem 2– Substation
Model
Problem 3– Substation
Model
• Problem 4– Substation
Model
COMPUTATIONAL RESULTS
• Smaller optimality gaps for the Port Model
• Large variation on computational times
Problem 1 Problem 2 Problem 3 Problem 4
Number of continuous variables 1996 4954 712 703Number of binary variables 1039 759 66 123Number of constraints 7203 10337 1158 1682Relaxed LP solution 21,768.32 23.00 11.00 11.39Best Integer Objective 20,073.96 42.00 21.00 15.00Optimality gap 7.78% 82.61% 90.91% 31.74%Nodes 1118 3784 3921 422Iterations 62313 74410 19321 5244CPU time (Pentium III 450MHz) 1,457.51 s 3,602.07 s 134.69 s 28.28 s
Port Model Substation Models
PROBLEM 1 – TANKERS AND TANKSPEDREIRAS
08
27
01020304050
30 35 40 45TBN01
B
0
20
40
60
60 62 64 66 68 70
FRONT BREA
05
0
30
60
90
120
10 15 20 25 30
REBOUÇAS
38
0
10
20
30
40
40 45 50
MURIAÉ
01a
01b
0
20
40
60
80
45 55 65 75 85
120
TBN02
C
0
5
10
15
20
85 86 87 88 89 90
RAVEN
D
0
20
40
60
90 95 100 105
TBN03
B
0
10
20
30
139 141 143 145
VERGINA II
38
0
30
60
90
85 90 95 100
NORTH STAR
26
0
50
100
150
95 100 105 110 115TBN04
E
0
20
40
60
154 161 168
PRESIDENTE
29a29b
0
20
40
60
80
90 100 110 120
CANTAGALO
A
0
5
10
15
20
0 1 2 3 4 5
TQ3210
Vmin
Vmax
07
14212835
0 24 48 72 96 120 144 168
TQ3208
Vmin
Vmax
07
14212835
0 24 48 72 96 120 144 168
TQ3214
Vmin
Vmax
020406080
0 24 48 72 96 120 144 168
TQ3219
Vmin
Vmax
020406080
0 24 48 72 96 120 144 168
TQ3237
Vmin
Vmax
02040
6080
0 24 48 72 96 120 144 168
TQ3241
Vmin
Vmax
02040
6080
0 24 48 72 96 120 144 168
TQ3215
Vmin
Vmax
02040
6080
0 24 48 72 96 120 144 168
TQ3233
Vmin
Vmax
020406080
0 24 48 72 96 120 144 168
TQ3238
Vmin
Vmax
07
14212835
0 24 48 72 96 120 144 168
TQ3242
Vmin
Vmax
020
4060
80
0 24 48 72 96 120 144 168
TQ3217
Vmin
Vmax
020406080
0 24 48 72 96 120 144 168
TQ3234
Vmin
Vmax
020406080
0 24 48 72 96 120 144 168
TQ3239
Vmin
Vmax
020
4060
80
0 24 48 72 96 120 144 168
TQ3243
Vmin
Vmax
02040
6080
0 24 48 72 96 120 144 168
TQ3218
Vmin
Vmax
020406080
0 24 48 72 96 120 144 168
TQ3235
Vmin
Vmax
020406080
0 24 48 72 96 120 144 168
TQ3240
Vmin
Vmax
020
4060
80
0 24 48 72 96 120 144 168
TQ3244
Vmin
Vmax
020
4060
80
0 24 48 72 96 120 144 168
17.4
58.2
58.2
17.4
35
16.9
38.0 (P4)
27.8
3.7 (P4)
11.9 (P4)
0.6 (P4)
25.4 4.7 (P4)
15 (P4)
5.3 (P3)
24.7 (P3)
37.8 (P2)
19.6
CANTAGALO
PRESIDENTE
FRONT_BREA NORTH_STAR
48.8
2.2
28.06.1 (P4)
28.0
25.4
53.3
24 Hrs
12.2 (P2)
42.4 (P2)
63.6 (P2)
26.0 (P2)
2.2
19.6
29.4 (P4)
1.4 (P2)
51.1 (P3)
30L (P-4)
2.8
19.6
55.5 (P4)
60.4
PEDREIRAS
MURIAE
6.2 (P2)
24.3 Hrs
26.0 Hrs 66.0 (P2)
50.3Hrs
27.8
34.4
24 Hrs
60.0 (P4)
11.5 (P4)
42.4
63.6
61.8
2.8
28.9 (P3)
24 Hrs
43.2 Hrs
REBOUCAS
VERGINA II
67.7
55.7 (P1)
62.4 (P2)
48.8
34.8
11.0
58.7
11.5
24 Hrs 60.0 (P4)
42.4
19.6
TBN01
16.9 (P1)
62.4
24 Hrs
63.6
11.5
28.0
2.8
4.5 (P4)
24 Hrs
62.4 (P1)
60.0
58.7
2.2
TBN02
16.9
46.4 Hrs
67.7
60.4
27.8
33.9
14.1
62.4
RAVEN
48.8
16.9
41
TBN03
60
19.6
61.8
65
TBN04
62.4
70.2
65.1 2.6
26.5
62.4 16.9 60 59.1
0 24 48 72 96 120 144 168
P-1
P-2
P-3
P-4
TQ3208
TQ3210
TQ3214
TQ3215
TQ3217
TQ3218
TQ3219
TQ3233
TQ3234
TQ3235
TQ3237
TQ3238
TQ3239
TQ3240
TQ3241
TQ3242
TQ3243
TQ3244
OSVAT
OSBAT
SEGUA/E
OSBATII/S
Tempo (horas)
PROBLEM 1 – GANTT CHART
PROBLEM 1 – REFINERIES AND PIPELINESREPLAN+REVAP
Min
Max
0
400
800
1200
1600
0 24 48 72 96 120 144 168
REPLAN+REVAP (detailed)
850
900
950
1000
0 24 48 72 96 120 144 168
RECAP+RPBC
Min
Max
0
200
400
600
0 24 48 72 96 120 144 168
RECAP+RPBC (detailed)
310
320
330
340
350
0 24 48 72 96 120 144 168
Profile Pipeline OSBAT_I e OSBAT_II
0 24 48 72 96 120 144 168Tempo (hours)
Flux
de
Oil
Stationde
Guaratuba
RPBC
CL-7
CL-6
CL-7
CL-6
CL-7 CL-6
GEBASTProfile Pipeline OSVAT_I e OSVAT_II
0 24 48 72 96 120 144 168Tempo (hours)
Flux
de
Oil Station
deRio
Pardo
SEGUA CL-3
CL-5
CL-6
CL-5
CL-3
CL-4
CL-5 CL-4
CL-5
CL-5
CL-6
CL-5
GEBAST
CL-4
Flow rate OSVAT_I e OSVAT_II (1000 m^3/h)
012345
0 24 48 72 96 120 144 168
Flow rate OSBAT_I e OSBAT_II (1000 m^3/h)
0
1
2
0 24 48 72 96 120 144 168
PROBLEM 4 – GANTT CHART
1
1
1
2.6
2.6
8.5
10.8
7.5
12.6
10.3
1.4
8.5
1
12.6
7.5
11.2
2.3
8.3
15.2
4.3
1.4
8.4
11.2
11.2
4.6
8.4
8.4
8.3
2.3
4.6
4.3
4.6
15.2 4.60 24 48 72 96 120 144 168
OSBATIII/S
TQ40
TQ41
TQ42
OSBATIV/E
RECAP/E
Time (hours)
PROBLEM 4 – TANKERS, REFINERIES AND PIPELINES
TQ40
Vmin
Vmax
05
10152025
0 24 48 72 96 120 144 168
TQ41
Vmin
Vmax
05
10152025
0 24 48 72 96 120 144 168
TQ42
Vmin
Vmax
05
10152025
0 24 48 72 96 120 144 168
RECAP
Min
Max
0
40
80
120
0 24 48 72 96 120 144 168
RECAP (detailed)
60
70
80
0 24 48 72 96 120 144 168Flow rate OSBAT_IV (1000m^3/h)
0
0,34
0 24 48 72 96 120 144 168
Profile Pipeline OSBAT_IV
0 24 48 72 96 120 144 168
Flux
de
Oil
CL-7 CL-6
CL-7
SEBAT
CL-7 CL-6
RECAP
OUTLINE
• Introduction • Planning Models
– refinery diesel production• Scheduling models
– crude oil scheduling– fuel oil / asphalt area
• Logistics– oil supply model
• Conclusions
CONCLUSIONS
The LP based Branch and Bound Method (solver OSL): is satisfactory to generate “good” feasible solutions;no guarantee of global optimum solutions for all instances;
Issues:time representationblending/pooling transitions
Problems can be modeled as large scale MILPs / non-convex MINLP;
The OA/ER/AP Method (solver DICOPT++): is efficient to circumvent the non-convexity problem;is satisfactory to generate feasible solutions;has computational performance similar to MILP model.
CONCLUSIONS - CHALLENGES
The understanding of the problem itself can constitute the major difficulty;
Large Scale Systems - Main theoretical difficulties:
The cooperation between the modeler level and the plant floor level is essential and remains as the main challenge for the Operational Research
Continuous work necessary due to the dynamic nature of scheduling problems.
Complex problems with high combinatorial features;
NP-Complete Problems
Large Scale Systems - Main practical difficulties
Infeasible computational times
ACKNOWLEDGEMENTS
Researchers
M. Joly R. MásL. MoroR. RejowskiP. Smania
Financial Support
Petrobras
CAPESCNPqFAPESP
Engineers
C.A. GrattiM. F. LehnerM.V. MagalhãesA.C. Zanin
General Petroleum Supply Chain
• Iyer et al. (1998).• Van den Heever and
Grossmann (2000).• Van den Heever et al.(2000).
Ierapetritou et al. (1999). • Kosmidis (2002).• Barnes et al. (2002).
Oil field Infrastructure
InternationalPetroleumCommerce
• Operations
Research.
• Lee et al. (1996).
• Pinto et al. (2000).
• Más and Pinto (2002).
Crude Oil
Supply
• Ponnambalam (1992).
• Bok et al. (1998).
• Pinto and Moro (2000).
Refinery -Planning/Scheduling
• Ross (2000).
• Iakovou (2001).
• Magatão et al. (2002).
• Stebel et al. (2002).
• Rejowski and Pinto (2003).
Distribution
Development of an optimization model
that is able to represent
a petroleum supply chain to support the
decision making planning process of
supply, production and distribution
Objective
Refinery - Processing Unit Model
( ) u
u ,t u ',s ,u ,tu ',s
QF Q∈
= ∑US
( )u
u ,s ,t u ,t u ,s u ,p ,t u ,s ,v u ,v ,tu
QS QF . f PF QGain .V∈
= + ∑VO
u ,s
u ,s ,t u ,s ,u ',tu '
QS Q∈
= ∑UO
( )
( )
u
u
u',s ,u ,t u ',s , p ,tu ',s
u ,p ,tu ',s ,u ,t
u ',s
Q .PSPF
Q∈
∈
=∑
∑US
US
u ,s ,p ,t u ,s ,p u ,p ,t u u ,t u ,v ,t uPS f ( PF | p ,QF ,V | v )= ∈ ∈PI VO
L Uu u ,t uQF QF QF≤ ≤ L U
u ,v u ,v ,t u ,vV V V≤ ≤
Feed mixing
Unit
model
Bounds
Product splitting stream
Supply, Distribution – Storage Model
Supply, Distribution - Pipeline Model
( ) u
u ,t u ',s ,u ,tu ',s
QF Q∈
= ∑US u ,s ,t u ',s ,u ,tQS Q=
u ,s ,t u ,s ,u ',tQS Q=
Uu ,t uQF QF≤
( )
( )
u dem SB
u f
p pipe
u ,t u,t u ,t u ,t u,tt u t
u u,v u ,v ,t u,t u u ,tu t v u t
u u ,t u u ,tu t u t
Max Z Cp . QF Vol .Cpet .Lot
[ Cr Cv .V ] .QF Cinv .Vol
Cinv .Vol Ct .QF
∈ ∈ ∈ ∈
∈ ∈ ∈ ∈ ∈
∈ ∈ ∈ ∈
= − −
− + −
− −
∑ ∑ ∑ ∑
∑ ∑ ∑ ∑ ∑
∑ ∑ ∑ ∑
U T U T
U T VO U T
U T U T
PSC
Supply Chain Model
Large Scale MINLP
Supply Chain Model – cont. from previous slide
subject to the models of:
• processing units
• tank
• pipeline
{ }QF ,QS ,Q,Vol ,Lot PF ,PS ,V y 0,1+∈ℜ ∈ℜ ∈
units that compose refinery topologyrefineries that compose the supply chain
••
petroleum and product tanks that compose refineries petroleum and product tanks that compose terminals refineries and terminals that compose the supply chain
•••
pipeline network for petroleum supply pipeline network for product distribution••
Supply Chain Components
Petroleum Distribution Overview
Product Distribution Overview
RPBC flowsheet
Intermediate connections
Modeling Example
Tanks and CD1 Model
tan k1, petroleum,t tan k1, petroleum,CD1,tQS Q , t= ∀
tan k1, petroleum,CD1,t500 Q 20000, t≤ ≤ ∀
{ }
{ }
{ }
u ', petro leum ,C D 1 , p ,t u ', petro leum , pu ' tan k 1 ,2 ,3
C D 1 , p ,tu ', petro leum ,C D 1 ,t
u ' tan k 1 ,2 ,3
Q . P r opP F , t
Q
p YC 3C 4 ,YL N ,YH N ,YK ,YL D ,YH D ,YA T R
∈
∈
= ∀
∈
∑
∑
{ }C D 1 ,t u ', petro leum ,C D 1 ,t
u ' tan k 1 ,2 ,3Q F Q , t
∈= ∀∑
1 1CD1,s ,t CD1,t CD1, p ,t CD1,s ,V CD1,V ,tQS QF .PF QGain .V , t
s pC3C4 YC3C4
LN YLNHN YHNK YK
LD YLDHD YHDATR YATR
= + ∀
→ →
→ → →
→ →
CD1,s ,p ,t CD1,s ,p CD1,s,p CD1,V1,t CD1 CD1,sPS Prop +PGain .V ,s , p ,t= ∀ ∈ ∈SO PO
CD1,s
CD1,s ,t CD1,s ,u ',tu '
QS Q∈
= ∑UO
Refinery Multiperiod Planning – REVAP resultsDICOPT(NLP CONOPT++)(MIP OSL, CPLEX)
Demand profile - GLN
Steep
Smooth
Steep
Smooth
Refinery Planning – Model with Uncertainty
( )u p f u
c ,t u ,t ,c u,t,c u ,t ,c c ,t u ,t ,c u,s,t,cc t c t u s
Max Z prob Cp QF Vol prob Cf QS∈ ∈ ∈ ∈ ∈ ∈ ∈
= − −
∑ ∑ ∑ ∑ ∑ ∑ ∑
C T U C T U SO
{ }f uf p
u u ,t ,c u u,v u ,v ,t u,tc t u c t vu ,
Cb y [ Cr ( Cv .V )] QF∈ ∈ ∈ ∈ ∈ ∈∈
− − + ∑ ∑ ∑ ∑ ∑ ∑ ∑
C T U C T VOU \ U U
p
u ,t ,c u ,t ,ct u
Cinv Vol∈ ∈
−
∑ ∑
T U s.t. refinery constraints
c1 c2 cN
Discrete Scenarios
1
P1% P2% PN%
c,tprob =∑Cc∈
Planning under Uncertainty - REVAP results
Proposed Strategies and Results
Primal subproblem Dual subproblems Multipliers update
Strategy 1 Fixed assigment Lagrangean Subgradient
Strategy 2 Fixed inventory Lagrangean Subgradient
Strategy 3 Fixed inventory Surrogate Subgradient
Strategy 4 Fixed inventory Lagrangean Modified Subgradient
0
100
200
300
400
500
600
700
0 10 20 30 40 50 60
Number of time periods.
CPU
seco
nds
Problem RMP Strategy 1 Strategy 2
Strategy3 Strategy4
Cases:1: Complete model
2: Pre-selection of some suppliers
3: Interruption of pipeline segment SG-RV
General constraints:Planning horizon: one / two time periods
Supply of 20 oil types
Generation of 32 products (6 transported with pipelines)
Supply Chain Example
Supply Chain Example – Petroleum Selection
LarabMarlinRgnCabiunAlbacoBicudoCondosoBonitCabiunaLarabeCoso
Case 1 Case 2
Case 3
Supply Chain Example – REVAP10905114047551
169451648911400
176419331001
00
871
600600660
0060
00
6551
Supply Chain Example – RPBC
138914081389
9824100759924544951775449
363636
151515
Supply Chain Example – Oil Supply
542005420054200
360003600025992
902009020080192
850085008500 35500
3550035500
Supply Chain Example – Product Supply
1764218311362
33452889
10800
2454025208
0
4400440011660
177001770017700
640364036403
65456089
11000200182001920018
Supply Chain Example – Intermediate Streams
60224666
48058
PFO1APFO1
A PFO4A
0530
Case Case 1 Case 2 Case 3
Number of time periods 1 2 1 2 1 2
Constraints 2304 4607 2306 4611 2304 4607
Variables 2544 5087 2544 5087 2544 5087
Discrete variables 195 390 195 390 195 390
Solution time (CPU s) 116.8 656.2 152 915.6 157.8 2301
Objective Value ($ x106) 20.4 41.3 20.3 41.1 18.0 36.3
Supply Chain Example – Computational Results
Mathematical programming
-General refinery topology
-General petroleum supply chain representation
-Representation of nonlinear properties and multiple periods
-Non-convex Large-Scale MINLP
Real-world applications
-General planning trends along multiple periods
-Analysis of scenarios (discrete probabilities)
-Intensive computational effort
Conclusions
Modeling
Upstream-Downstream Integration
Multi country supply chains (royalties, tariffs)
Modeling of uncertainties
Efficient solution methods
Decomposition (spatial, temporal, functional)
Approaches (Lagrangean Relaxation, Cross Decomposition)
Research needs