Spatial Autocorrelation informed approaches to solving location-allocation problems: spatial autocorrelation-informed sampling for initial heuristic solutions

Keywords: Spatial statistics, Spatial optimization, Heuristic algorithm, Spatial autocorrelation
Abstract Type: Paper Abstract

Abstract

This study intends to address this issue by exploring the possibility of using spatial statistics for spatial optimization.
This study seeks to help delineate an initial solution space that can lead to optimal or near-optimal Location-allocation(L-A) solutions exploiting spatial autocorrelation (SA) in the geographical weights. L-A problems have intrinsic limitations known as NP-hard complications. It also often takes sizable time to find an optimal solution because many L-A problems entail a massive exploration of possible feasible solutions. However, location problems are associated to SA. Local SA quantifies the pattern of high and low weights in its objective function. This feature suggests that the property of SA may be able to contribute to solving L-A problems.
To date, heuristics have furnished non-SA-based methods to surmount the aforementioned limitations of L-A problems. Although the Teitz and Bart heuristic and ALTERN are widely used heuristics to find (near) optimal solutions, neither is guaranteed to produce a global optimum, and their performances are sensitive to randomly selected starting solutions. An initial solution close to an optimal can be a key to avoiding the local optima dilemma, as well as to reducing the solution time for sizable L-A problems. Accordingly, this study examines initial solutions that consider latent SA to see what role it can play when solving L-A problems.