Aggregation method experimentation for large-scale network location problems

We present demand point aggregation procedures for the p-median and p-center network location models. A coarse aggregation structure is initially obtained by partitioning the demand points according to a grid imposed over the demand region. A “row-column’’ aggregation algorithm is used to determine the spacing of rows and columns of the grid to exploit the problem structure. A second step involves locating aggregate demand points on the subnetworks induced by the cells of the grid partitioning. The aggregate demand point set so obtained then defines an approximating location model; alternatively, it may initialize an iterative network location–allocation procedure to find the aggregate demand points. We have tested our procedures on data sets based on maps from the TIGER/Line database of the United States Census Bureau, and report on our computational experience.