A parallel adaptive memory algorithm for the capacitated modular hub location problem

Abstract

The capacitated modular hub location problem is a realistic variant of the popular hub location problem arising from the design of telecommunications networks. Given a set of demand nodes, the problem consists in selecting a subset of nodes to represent hubs and assigning the rest of the nodes to the hub nodes, such that the transportation cost is minimized while satisfying the capacity constraints. The adaptive memory algorithm is a hybrid evolutionary heuristic that uses a central memory to store blocks of solutions. At each iteration, the recombination operator uses these solution blocks to create an offspring solution. In this paper, we present a parallel adaptive memory algorithm for the capacitated modular hub location problem that stores both solution blocks and complete solutions in a shared memory for the creation of an offspring. Other distinguishing features of our proposed algorithm include specially designed recombination and mutation operators for search diversification, an effective tabu search procedure for search intensification, and parallel computing for global optimization. Extensive computational results on three well-known data sets of 170 benchmark instances show that the proposed algorithm competes very favorably with the state-of-the-art heuristics from the literature. In particular, it finds 115 improved best-known solutions (for more than 67% of the cases). Furthermore, we analyze the key algorithmic components and shed lights on their impact on the proposed algorithm.