  


Risk Optimization in Probabilistic Programs with Single or Multiple Chance Constraints
Siqian Shen (siqianumich.edu) Abstract: Problems of probabilistic programming, also referred to as chanceconstrained programming (CCP), evaluate valueatrisks of uncertain events, and permit certain tolerances for such risks. In this paper, we optimize risk tolerances as decision variables in CCP models, which by assumption only contain random discretely distributed righthand sides. The paper first considers a bilevel interdiction problem, in which an inner problem contains one singlerow chance constraint, while an outer problem decides the risk tolerance associated with the constraint, to maximize the minimum objective computed by the inner problem. For optimizing the risk tolerance, we show that only finite possible values matter to the inner CCP objective, and interpret the risk variable as Special Ordered Set of Type 1 (SOS1) binary variables. We then transform the bilevel program into a deterministic integer programming formulation, whose size depends on the set cardinality of all possible realizations of righthand sides. We also extend our investigation to CCP models with multiple singlerow chance constraints, each of which has an associated risk tolerance variable. The goal is to seek an optimal combination of risks to minimize the overall objective value in the resulting CCP. This paper reformulates the problem as an integer program in multiple SOS1 binary variables, and employs a decomposition strategy for solving the reformulation. We demonstrate the two model variants by respectively solving a stochastic minimumcost flow problem and a multicommodity flow capacity expansion problem on a set of randomly generated graph instances. Keywords: Chanceconstrained programming, Special Ordered Set of Type 1 (SOS1), mixedinteger programming, Benders decomposition, risk optimization Category 1: Stochastic Programming Category 2: Integer Programming ((Mixed) Integer Nonlinear Programming ) Category 3: Integer Programming (Cutting Plane Approaches ) Citation: Department of Industrial and Operations Engineering, University of Michigan Download: [PDF] Entry Submitted: 06/10/2012 Modify/Update this entry  
Visitors  Authors  More about us  Links  
Subscribe, Unsubscribe Digest Archive Search, Browse the Repository

Submit Update Policies 
Coordinator's Board Classification Scheme Credits Give us feedback 
Optimization Journals, Sites, Societies  