global supply chain planning under demand and freight rate
TRANSCRIPT
Page 1
Global Supply Chain Planning under Demand and Freight Rate UncertaintyFengqi YouIgnacio E. Grossmann
Nov. 13, 2007
Sponsored by The Dow Chemical Company (John Wassick)
Page 2
• Objective:Developing Models and Algorithms for Global Multiproduct Chemical Supply Chains Plannning under Uncertainty
Motivation
• Global Chemical Supply ChainCosts several billion dollars annuallyPlanning under uncertain environment
Introduction
Changes of customer orders
Fluctuations of gas prices
Page 3
Supply Chain Tactical Planning
Plants Distribution Centers Customers
Introduction
Page 4
Problem Statement
• GivenMinimum and initial inventoryInventory holding cost and throughput costTransport times of all the transport linksUncertain customer demands and transport cost
• DetermineTransport amount, inventory and production levels
• Objective: Minimize Cost
Problem Statement
Page 5
Outline
• Stochastic Programming ModelTwo-stage stochastic programming modelSimulation-optimization framework
• Model Extensions for Robustness and RiskRobust optimizationRisk management
• AlgorithmsMulti-cut L-shaped method
Page 6
Uncertain Parameters
• Demand and Freight Rate UncertaintyNormal distributionForecast value as mean, variances come from historical data
» Demand uncertainty has 3 level variances» Freight rate uncertainty has 2 level variances
Stochastic Programming Model
Page 7
Scenario Planning• Discretize the probability distribution function
Approximate all the uncertainties to discrete distribution (scenarios)Each scenario represents a possible outcomeGenerate scenarios by Monte Carlo sampling
» Assign each scenario the same probability (i.e. For N sampling, Ps=1/N)» Combine statistical methods for “good” approximation
Stochastic Programming Model
Page 8
Decision Stages under Uncertainty• Here-and-now
Decisions (x) are taken before uncertainty ω resolute• Wait-and-see
Decisions (yω) are taken after uncertainty ω resolute as “corrective action” -recourse
xyω
ω= 1ω= 2ω= 3ω= 4ω= 5ω= Ω
Stochastic Programming Model
Uncertainty reveal
Page 9
Two-stage Stochastic Programming• First stage decisions
Here-and-now: decisions for the first month (production, inventory, shipping)• Second stage decisions
Wait-and-see: decisions for the remaining 11 months
Minimize E [cost]
Stochastic Programming Model
Page 10
Multi-period Planning Model
• Objective Function:Min: Total Expected Cost
• Constraints:Mass balance for plantsMass balance for DCsMass balance for customersMinimum inventory level constraintCapacity constraints for plants
Stochastic Programming Model
Page 11
Second stage cost
Objective Function
Inventory Costs
Throughput Costs
Freight Costs
Demand Unsatisfied
First stage cost
Stochastic Programming Model
Probability of each scenario
Page 12
revealed
Rolling Horizon Strategy
uncertain
Inter-facility shipment from the previous time periods considered as pipeline inventoryFacility-customer shipment considers as part of demand realizationInventory level in the previous time period considers as the initial inventory for t =1Consider uncertainty reduction as time period moving forward
Simulation-optimization Framework
Page 13
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
-2.5 -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5
Robust Optimization
Cost
Probability ExpectedCost
Desirable Penalty Undesirable Penalty
Objective: To find out the optimal solution that yields similar results under the uncertain environment – Robust solution!
Robust Optimization
Cost of robustness
Page 14
Robust Optimization using Variance
• Goal Programming FormulationNew objective function: Minimize E[Cost] + ρ ·V[Cost]Different ρ can lead to different solution (multi-objective optimization)
Expected Cost Expected Variance of each scenario
Weighted coefficient
Robust Optimization
Page 15
Robust Optimization via Variability Index
• First Order Variability indexConvert NLP to LP by replacing two norm to one norm
Robust Optimization
Page 16
Variability Index (cont)
• Linearize the absolute value termIntroducing a first order non-negative variability index Δ
Robust Optimization
Page 17
Risk Management (cont)
• Risk: The probability of exceeding certain target cost level Ω
Histogram of Frequencies Cumulative Risk Curve
Ω
Cost
Cumulative Probability = Risk (x, Ω)
Risk Management
Page 18
Risk Management (cont)
• Risk Management for scenario planningCalculation by
Binary variables Prob
abili
ty
Risk Management
Page 19
Risk Management Model Formulation
Prob
abili
ty
Risk Management Constraints
Risk Objective
Economic Objective
Multi-objective
Risk Management
Page 20
Downside Risk• Definition: Positive Profit Deviation
Binary variables are not required, pure LP (MILP -> LP)
Risk Management
Page 21
Standard L-shaped Method
Scenario sub-problems
Masterproblem
x
1y
Sy
2y
Master problem
Scenario sub-problems
Algorithm: Multi-cut L-shaped Method
3y
Page 22
Expected Recourse Function
The expected recourse function Q(x) is convex and piecewise linearEach optimality cut supports Q(x) from below
Algorithm: Multi-cut L-shaped Method
Page 23
Multi-cut L-shaped Method
YesNo
Solve master problem to get a lower bound (LB)
Solve the sub problem to get an upper bound (UB)
UB – LB < Tol ? STOP
Add cut
Algorithm: Multi-cut L-shaped Method
Page 24
Conclusion• Current Work
Develop a two-stage stochastic programming approach for global supply chain planning under uncertainty. Simulation studies show that 5.70% cost saving in average can be achievedPresent two robust optimization models and two risk management model. Develop an efficient solution algorithm to solve the large scale stochastic programming problem
• Future WorkMulti-site capacity planning under uncertainty
Page 25
Questions?