Distributed generation, electric vehicles, and variable demand complicate distribution-network reconfiguration. We propose a hybrid architecture combining graph convolutional inference, symbolic ...