Distributed Deep Neural Networks for the OPF Problem

The Optimal Power Flow (OPF) problem is a crucial optimization task in electrical grids, guiding operational decisions across many applications. Centralized OPF raises concerns regarding privacy, scalability, and single points of failure. Distributed OPF addresses these issues but can be slow due to iterative local optimizations and coordination. We propose a distributed Deep Neural Network (DNN) surrogate: each subsystem deploys a local DNN trained on trajectories from a conventional distributed solver, and only boundary variables are exchanged until consensus. Focusing on DCOPF, we evaluate the approach on the IEEE 5-bus and IEEE 30-bus systems. In both cases, the DNN-based procedure converges to operating points close to the ADMM baseline and substantially reduces computational time, indicating the applicability of the approach.