The increase in the frequency of natural disasters such as earthquakes, landslides, and floods is becoming a critical problem globally due to their effects on humans and the environment [1
]. The total deaths from natural disasters recorded worldwide between 2006 and 2016 was 1.2 million, twice that from the 1990s [2
]. Thus, there is an obvious need to efficiently plan evacuation as a strategy, among others, to handle emergency situations and reduce disaster risks.
Evacuation plans are developed to ensure the safety of affected people by efficiently and quickly moving them away from dangerous places to safe places in order to reduce the loss of life and damage. However, evacuation planning is a complex process, involving many stakeholders and management aspects. In disaster management, evacuation planning is tackled as a Multi-Objective Optimization Problem (MOOP) with consideration of the spatial component [3
]. Many studies presented multi-objective evacuation models [7
], in which the main goal of the models was to find the best alternatives for allocating facilities and the optimal distribution of the affected population to appropriate safe places (also called “shelters”). These models often involve multiple conflicting criteria and objectives, such as distance and distribution to shelters; by optimizing one of the criteria or objectives, the other criteria or objectives are assigned inappropriate values.
Recently, Multi-objective Optimization Methods (MODM) integrated with Geographical Information Systems (GIS) were introduced to provide better solutions for spatial optimization problems [10
]. The most used approaches can be classified into two categories: exact methods and metaheuristics methods [12
]. The exact methods, including goal programming, linear programming, weighting, and constraint methods, are the oldest and traditional methods. These methods transform an MOOP into a scalar problem and solve it as a single optimization problem [15
]. For example, Coutinho-Rodrigues et al. [16
] presented a multi-objective approach for an urban location/routing problem using a mixed-integer linear programming model. Furthermore, Horner et al. [17
] and Kocatepe et al. [18
] designed a GIS-based network optimization approach for siting medical special needs hurricane relief shelters and used the GIS-based spatial p-median optimization technique to solve the evacuation and sheltering of pets and the human population with special needs in the state of Florida, United States of America (USA). Although exact methods guarantee finding an optimal solution, the results are highly dependent on the knowledge of experts who assign weights to the criteria. Having limited knowledge about the criteria and their effects or being biased toward some preferences both make the final result unreliable.
Unlike exact methods, metaheuristic algorithms are known to be efficient for solving more complex problems by providing a set of optimal solutions in a reasonable amount of time [13
], without being influenced by the preferences of experts. Although they do not guarantee finding global optimal solutions, metaheuristic algorithms iteratively improve the feasible solutions through several heuristic techniques. A heuristic is an approach based on the rules, strategies, or ad hoc procedures to solve an optimization problem [19
]. Many metaheuristics are inspired by natural processes. For example, Evolutionary Algorithms (EAs) and Swarm Intelligence algorithms (SIs) solve problems by mimicking the behavior of natural species or the rules of natural phenomena [20
]. In the spatial optimization domain, many studies applied these techniques due to their potential to optimize multiple and conflicting objectives and provide non-dominated solutions as outputs [13
]. These methods include Multi-objective linear programming [22
], Genetic Algorithm (GA) [23
], Particle Swarm Optimization (PSO) [27
], Ant Colony Optimization (ACO) [28
], Tabu Search [29
], and the Artificial Bee Colony (ABC) [31
]. Of these, Multi-Objective ABC is less used and tested in comparison to other techniques, mainly due to its relatively recent introduction.
Many studies on evacuation planning applied metaheuristics and mostly EAs [4
]. For instance, Garrett et al. [34
] used GAs to address the problem of evacuation planning to find optimal door locations for a building. Saadatseresht et al. [4
] proposed multi-objective optimization for evacuation planning using Non-Dominated Sorting Genetic Algorithm II (NSGA-II). Georgiadou et al. [35
] used an Evolutionary Algorithm to optimize the response to an emergency situation such as a major accident. For SIs, Hu et al. [36
] and Xu et al. [37
] applied modified PSO algorithms to find the optimal allocation of earthquake emergency shelters. A massive pedestrian evacuation problem was solved using a Multi-Objective ACO [38
]. The proposed approach efficiently minimized three objective functions including total evacuation time, total routes risk degree, and total crowding degree. Saeidian et al. [39
] proposed an approach to solve location allocation of earthquake relief centers using PSO, ACO, and Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) models, and GIS.
ABC, proposed by Karaboga [40
], is another metaheuristic algorithm that mimics the foraging behavior of honey bees in nature. The ABC is reported to be an efficient algorithm for solving optimization problems with a single objective function, and it was applied in many fields [41
]. The standard ABC was modified and improved for better performance in solving different problems [31
], which includes developing and modifying the MOABC to solve multi-objective problems [43
]. With this in mind, in this paper, we used and modified MOABC and propose a model for evacuation planning. One of the main issues with MOABC that influences evacuation planning is trapping into local optimum solutions. We addressed the issue by modifying the neighborhood strategies for local search, as well as testing and adopting selection strategies to choose from among the variety of existing strategies [12
Our aim for evacuation planning was to find the optimal distribution of evacuees to safe places. To solve this problem, two conflicting objective functions to be minimized were defined. The model was tested and evaluated in Kigali, the capital of Rwanda.
This paper is organized as follows: Section 2
briefly introduces the concept of the Multi-Objective Optimization Problem, the mathematical model of evacuation problem, and the ABC algorithm. Section 3
describes the proposed MOABC approach in detail. Section 4
presents a case study and data preparation. Experimental results are provided and discussed in Section 5
. The conclusions and recommendations are presented in Section 6
In this paper, we presented a modified ABC algorithm to address evacuation planning by minimizing the total capacity constraints violations in shelters and the total evacuation distance to the assigned shelter.
We proposed a Multi-Objective Artificial Bee Colony (MOABC) based on the modified original ABC. The four important strategies, including worker bees for improving the neighborhood search of ABC and simultaneously optimizing two conflicting objectives, were: (1) the discrete random generation of the initial population; (2) combining random swap (RS) and random insertion (RI) neighborhood strategies; (3) a crossover operator for exchanging information between worker bees; and (4) the Pareto-based approach (non-dominant sorting method) for evaluation and fitness assignment of the MOABC. The proposed algorithm was compared to NSGA-II and standard MOABC. The experimental results showed that the proposed MOABC algorithm can perform better and is suitable for multi-objective evacuation problems in urban areas. In the future, instead of using a Pareto-based ABC approach, the ε-dominance multi-objective-based ABC can be studied to seek better optimal solutions for evacuation optimization problem applications. Future work could also focus on the improvement of encoding and representation of solutions for spatial optimization problems.
MOABC for an evacuation model was applied to the selected study area of Kigali, which was chosen due to the high frequency occurrences of flood and landslide hazards. The area is large, highly populated, and has some improperly located settlements. The characteristics of urban areas can increase the complexity of the model. Some of the candidate shelters used in this study cannot accommodate all evacuees from the nearest building blocks, so some may walk a long distance to be evacuated and reach safe areas. For future research, there is a need to conceptualize the model based on the characteristics of the area, including topography and behavior of evacuees, and to include other factors such as traffic, risks along evacuation paths, and socioeconomic conditions.