Lawrence Berkeley National Laboratory

We consider the electric distribution system expansion planning problem of selecting the least-cost set of upgrades to alleviate voltage magnitude violations. To this end, we use the full nonlinear power flow model to account for losses and represent meshed networks. The underlying problem is highly nonconvex due to the nonlinear power flow equations and the presence of discrete variables. Leveraging the efficacy of local search methods for handling continuous subproblems, a bilevel heuristic algorithm is proposed based on outer-approximation mixed-integer programming techniques. The proposed algorithm is shown to achieve the same optimal value as off-the-shelf mixed-integer nonlinear solvers with much less computation time on five benchmark IEEE distribution feeders.