Techniques for incorporating expected value constraints into stochastic programs

In this dissertation, techniques for solving a class of stochastic programs that are characterized by constraints on the expected value of some uncertain quantity are explored. There expected value constraints cause the structure of the problem to deviate from the dual angular structure required for using Benders decomposition, so other approaches must be considered.; One approach is to reformulate the problem to return it to a dual angular structure. Second-stage variables that are represented in the expected value constraints are converted to first-stage variables, and the expected value constraint is moved into the master problem. The result is a dual angular problem with many extra first-stage variables. Now classical Benders decomposition can be applied.; In many cases, this is still not sufficient to make the problem practical to solve. Modifications are made to the Benders decomposition algorithm that allow the original first-stage variables to be solved independently of the variables from the expected value constraints and to simultaneously solve for the variables in the expected value constraints and the other second-stage variables. Subject to some conditions, the computational effort needed to apply the modified algorithm is comparable to the effort needed to solve similar problems without expected value constraints. Implementation of the algorithm is described in the context of DECIS, an existing software implementation of Benders decomposition.; Computational results from samples and real world applications are presented. The thesis concludes with a discussion of other applications for these techniques, and with suggestions for additional research.