As part of the European project SenergyNets, we aim to optimally operate the heating and cooling network of the Gaîté–Montparnasse district. The resulting optimization problem is nonlinear and non-convex, reflecting the real behaviour of the system and its equipments. While such a formulation is required for accuracy, its scale makes efficient solving challenging. We evaluate several techniques to improve the performance of the nonlinear solver Knitro on this large-scale problem.