Multi-objective optimisation algorithms for GIS-based multi-criteria decision analysis : an application for evacuation planning

Abstract: Geographic Information Systems (GIS) have acquired greater relevance as tools to support decision-making processes, and during the last decades they have been used in conjunction with Multi-Criteria Decision Analysis techniques (GIS-MCDA) to solve real-world spatial problems. GIS-MCDA can be generally divided in Multi-Attribute and Multi-Objective techniques. Until now most of the applications of GIS-MCDA have been focused only on using the Multi-Attribute approach, and less than 10% of the research has been related to a specific type of Multi-Objective technique: the use of heuristic algorithms. The present study explores how different heuristic methods solve a spatial multi-objective optimisation problem. To achieve this, four algorithms representing different types of heuristics were implemented, and applied to solve the same problem related with an evacuation planning situation. The implemented algorithms were Standard Particle Swarm Optimisation (SPSO), Non-dominated Sorting Genetic Algorithm II (NSGA-II), Archived Multi-Objective Simulated Annealing (AMOSA) and Multi-Objective Grey Wolf Optimiser (MOGWO). The results show that the four algorithms were effective on solving the given problem, although in general AMOSA and MOGWO had a higher performance in terms of number of solutions, effectiveness of the optimisation, diversity, execution time and repeatability. However, the differences in the results were not clear enough to state that one type of heuristic is superior than others. Since AMOSA and MOGWO are the most recent algorithms among the implemented ones, they include several improvements achieved by the latest research, and their superior performance could be linked to these improvements more than to the specific type of algorithms they belong to. Further research is suggested to explore the suitability of these methods for many-objectives spatial problems, to consider the variability and dynamism of real-world situations, to create a standard set of algorithms to be used for benchmarking, and to integrate them with the currently available GIS-MCDA tools. Despite this, from the performed research it is possible to conclude that heuristics methods are reliable techniques for solving spatial problems with multiple and conflictive objectives, and future research and practical implementations in this field can strengthen the capacities of GIS as a multi-criteria decision-making support tool.