Ant Colony Algorithms for Optimal Site Selection in Large Regions

何晋强,黎夏,刘小平,陶嘉

DOI: https://doi.org/10.3321/j.issn:1007-4619.2009.02.011

2009-01-01

Abstract:Optimal site search for sitting facilities is crucial for the effective use and management of resources and it is also a common task for urban planning. The brute-force method has difficulty in solving complex site search problems especially in large scale areas. In this study, a location model is proposed based on ant colony algorithms (ACO). It combines ant colony intelligence and GIS to solve the problems of complicated spatial optimal allocation. ACO has strong search ability for a huge volume of spatial data. At first, the algorithm is modified about the strategy of pheromone update and Tabu table adjusting to fit the sites location problem. Spatial allocation problems usually have a large set of spatial data and only a few targets. The pheromone evaporates very fast because the selected cells only amount to a small percentage of all the cells. The positive feedback is too weak to play a role in the optimization. A modification is to incorporate the strategy of neighborhood pheromone diffusion in defining pheromone updating. At the same time, an optimal result for sites selection usually does not include two near candidate cells together, so a restricted Tabu table updating strategy is adopted which resembles the strategy of neighborhood pheromone diffusion. Then another important modification is to adopt a multi-scale approach to alleviate the computational demand in conducting large-scale spatial search. This includes two phases of optimization. First, a coarser resolution is used for the identification of rough locations of targets using ACO. Then the next round of optimization is implemented by just searching the neighborhood around these initial locations in the original resolution. This two-phase procedure of optimization can thus significantly reduce the computation time. The study area is located in the city of Guangzhou. This optimization problem considers two spatial variables, population distribution and transportation conditions, which are obtained from GIS. The raster layers have a resolution of 100 × 100 m with a size of 250 × 250 pixels. A comparison experiment is conducted among the multi-scale ACO, ACO, genetic algorithms and simple search algorithms. Experiments indicate that this multi-scale ACO method can produce similar results but use much lesser computation time, compared with the single ACO method. This method has better performance than the single search method and the genetic algorithm method for solving site search problems. ACO has obtained the utility improvement of 2.7%—5.5%, compared with the single search method. ACO has slight improvement of the total utility value over the GA method, but is able to reduce computation time significantly. Its computation time is only 12.5%—29.5% and 8.8%—20.3% of those of the single search method and the GA method respectively. In conclusion, integrating ACO with GIS is important for solving real-world site selection problems. The integration allows these two techniques to be mutually benefited from each other. ACO provides an efficient distributed computation algorithm while GIS provides useful spatial information. The comparison experiment indicates that the proposed model has better performance than GA and SSA especially in the computing time for solving site search problems.