A GRASP algorithm with RNN based local search for designing a WAN access network
One of the problems in the design of a Wide Area Network consists in laying out the access network, that is, choosing the concentrator sites and the communication lines which will connect the terminal sites to the switches in the backbone network. We model the problem of finding a minimum cost access network as a variant of the Steiner Problem in Graphs, which can be approximately solved by heuristics such as the Greedy Randomized Adaptive Search Procedure (GRASP), which comprises a construction phase and a local search phase. We consider two different alternatives for the construction phase. One method works iteratively selecting from a candidate list of the terminal nodes nearest to the current partial solution; the other chooses from among a list of k shortest paths from a single terminal node. Both methods are experimentally compared over 206 problem instances of different topological characteristics, generated using the problem classes in the SteinLib repository, and with known lower bounds for their optimal values. Although both methods obtained good results, the method based on finding nearest nodes is less computationally expensive and gives lower cost (average 2% better) solutions over all problem classes; a significative amount in network design problems.