This paper offers a novel approach for computing globally optimal solutions to the pump scheduling problem in drinking water distribution networks. A tight integer linear relaxation of the original non-convex formulation is devised and solved by branch and bound where integer nodes are investigated through non-linear programming to check the satisfaction of the non-convex constraints and compute the actual cost. This generic method can tackle a large variety of networks, e.g. with variable-speed pumps. We also propose to specialize it for a common subclass of networks with several improving techniques, including a new primal heuristic to repair near-feasible integer relaxed solutions. Our approach is numerically assessed on various case studies of the literature and compared with recently reported results.
unpublished: technical report 2019-06-18 Mines ParisTech, Center for Applied Mathematics, June/2019