On the Regularized Decomposition Method for Two Stage Stochastic Linear Problems.

A new approach to the regularized decomposition (RD) algorithm for two stage stochastic problems is presented. The RD method combines the ideas of the Dantzig-Wolfe decomposition principle and modern nonsmooth optimization methods. A new subproblem solution method using the primal simplex algorithm for linear programming is proposed and then tested on a number of large scale problems. The new approach makes it possible to use a more general problem formulation and thus allows considerably more freedom when creating the model. The computational results are highly encouraging.