 |
Center for Nonlinear Studies |
We present a new decomposition framework for continuous nonlinear nonconvex two-stage optimization problems. A barrier-based smoothing technique is used to render the second-stage recourse function differentiable, so that efficient standard nonlinear optimization methods and software can be utilized for the solution of both the master and subproblems. We explore the convergence properties of the framework, which is particularly challenging because of the nonconvexity of the subproblems that could lead to local minima in addition to global optima. We demonstrate the practical performance of the framework by solving large-scale optimal power flow problems where the network is decomposed into a transmission and several distribution systems.
Host: Russell Bent