Optimization of Large‐Scale Water Transfer Networks: Conic Integer Programming Model and Distributed Parallel Algorithms

We address in this paper the optimization of a multi-echelon water transfer network and the associate transportation and inventory systems with demand uncertainty. Optimal network structure, facility locations, operation capacities, as well as the inventory and transportation decisions can be simultaneously determined by the MINLP model which includes bilinear, square root and nonlinear fractional terms. By exploiting the properties of this model, we reformulate the MINLP problem as a conic integer optimization model. To overcome the memory and computing bandwidth limitations caused by the huge number of active nodes in the branch-and-bound search tree, novel distributed parallel optimization algorithms based on Lagrangean relaxation and message passing interface as well as their serial versions are proposed to solve the resulting conic integer programming model. A regional water transfer network in China is studied to demonstrate the applicability of the proposed model and the performance of the algorithms. This article is protected by copyright. All rights reserved.