Download presentation
Presentation is loading. Please wait.
Published bySamuel Stevens Modified over 9 years ago
1
Forward and Backward Deviation Measures and Robust Optimization Peng Sun (Duke) with Xin Chen (UIUC) and Melvyn Sim (NUS)
2
Agenda Overview of Robust Optimization Forward and Backward Deviation Measures Uncertainty Sets and Probability Bounds Conclusions
3
Uncertain linear constraints Uncertain linear constraint Chance constraint Hard to solve when is small
4
Linear constraints with uncertainty Affine Uncertainty : zero mean, independent but not necessarily identically distributed
5
Overview of Robust Optimization Worst case Relies only on the distribution support Easy to solve (Soyster 1973) Extremely conservative
6
Overview of Robust Optimization Goal of Robust Optimization: Easy to obtain feasible solutions that satisfy chance constraint Not as conservative as worst case
7
Overview of Robust Optimization Design of Uncertainty Set Subset of worst case set, W : Budget of uncertainty Control radius of uncertainty set, S max : Worse case budget For reasonable probability bound, << max
8
Overview of Robust Optimization Robust Counterpart Semi-infinite constraint Tractability Polyhedral Uncertainty Set Linear Programming (LP) Conic Quadratic Uncertainty Set Second Order Cone Programming (SOCP)
9
Overview of Robust Optimization Ellipsoidal Uncertainty Set Ben-Tal and Nemirovski (1997), El-Ghaoui et al. (1996) Robust Counterpart is a SOCP Probability bound for symmetrically bounded uncertainties:
10
Overview of Robust Optimization Polyhedral Uncertainty Set Bertsimas and Sim (2000) Robust Counterpart is a LP Probability bound for symmetrically bounded uncertainties: More conservative compared to Ellipsoidal Uncertainty Set
11
Overview of Robust Optimization Norm Uncertainty Set Robust Counterpart depends on the choice of norm Probability bound for symmetrically bounded uncertainties:
12
Overview of Robust Optimization Modeling limitations of Classical Uncertainty Sets Uncertainty set is symmetrical, but distributions are generally asymmetrical Can we use more information on distributions to obtain less conservative solution in achieveing the same probability bound? --- Forward and Backward Deviation Measures
13
Forward and Backward Deviation Measures Set of forward deviation measures Set of backward deviation measures
14
Alternatively, Forward deviation Backward deviation
15
Forward and Backward Deviation Measures Key idea
16
Important property Subadditivity
17
Least conservative forward and backward deviation measures Suppose distribution is known p * = q * if distribution is symmetrical p *, q * ¸ (standard deviation) p * = q * = if distribution is Normal
18
Least conservative forward and backward deviation measures p *, q * can be obtained numerically E.g: p * = q * = 0.58 for uniform distribution in [-1,1] E.g:
19
Forward and backward deviation measures – two point distribution example
20
Distribution not known Suppose distribution is bounded in [-1,1] (but not necessarily symmetrical)
21
Only know mean and support
22
Forward and Backward Deviation Measures: Function g( )
23
General Deviation Measures
24
General Deviation Measure
25
Agenda Overview of Robust Optimization Forward and Backward Deviation Measures Uncertainty Sets and Probability Bounds Conclusions
26
Uncertainty Sets and Probability Bounds Model of Data Uncertainty Recall: Affine Uncertainty : zero mean, independent but not necessarily identically distributed
27
Uncertainty Sets and Probability Bounds Model of Data Uncertainty
28
Uncertainty Sets and Probability Bounds New Uncertainty Set (Asymmetrical)
29
Uncertainty Sets and Probability Bounds Recall: Norm Uncertainty Set (Symmetrical)
30
Uncertainty Sets and Probability Bounds Generalizes Symmetrical Uncertainty Set
31
Uncertainty Sets and Probability Bounds
33
Robust Counterpart
34
Uncertainty Sets and Probability Bounds Probability Bound More information achieve less conservative solution while preserving the bound. E.g p j = q j = 1 for symmetric distribution in [-1,1] p j = q j = 0.58 for uniform distribution in [-1,1]
35
Conclusions RO framework Affine uncertainty constraints Independent r.v.’s with asymmetric distributions Forward and backward Deviation measures Defined from moment generating functions – implying probability bound Sub-additivity – linear combinations Easy to calculate and approximate from support information Advantage -- less conservative solution Next Stochastic programming with chance constraints
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.