Spatial Interaction Model Optimisation on Parallel Computers
Felicity George, Nicholas J. Radcliffe, Mark A. Smith, Mark H. Birkin, Martin Clarke · 1994
This paper discusses both the parallelisation of the original optimisation scheme and the use of genetic algorithms in place of this method. 1 INTRODUCTION Geographical Modellingand Planning (GMAP) Ltd have developed a computer model that simulates the pattern and volume of business that can be expected from a geographical network of sales or service outlets. When coupled with some optimisation scheme, this software can identify business resource planning solutions that provide maximum sales, profit or service results. Ford uses GMAP software --- an optimisation scheme called the Idealised Representation Plan (IRP)[10] --- to help to decide where to locate its car dealerships. The underlying spatial interaction model simulates car distribution and sales on the basis of information about the location of dealers and supply points, together with other geographic, demographic and statistical data. This allows prediction of the sales and flow patterns that result from a given configuration of dealerships. Running this software on a Sun SPARCstation 1 it would take around three months of computer time to find a network of 1000 Ford dealers covering the entire UK; thus the serial code has only ever been used to model one region of Britain at a time. In a collaborative project between GMAP Ltd and Edinburgh Parallel Computing Centre, the GMAP IRP code has been parallelised for the CM-200. In addition, GA techniques have been studied as a replacement for the existing heuristic optimisation scheme. The major objectives of this work were: ffl To design and implement a parallel version of a spatial interaction model used by Ford UK to plan its sales delivery network. ffl To achieve a reduction in run-time by at least two orders of magnitude, via this parallel solution, in order t...