DIMMA: A Design and Implementation Methodology for Metaheuristic Algorithms - A Perspective from Software Development

Metaheuristic algorithms will gain more and more popularity in the future as optimization problems are increasing in size and complexity. In order to record experiences and allow project to be replicated, a standard process as a methodology for designing and implementing metaheuristic algorithms is necessary. To the best of the authors’ knowledge, no methodology has been proposed in literature for this purpose. This paper presents a Design and Implementation Methodology for Metaheuristic Algorithms, named DIMMA. The proposed methodology consists of three main phases and each phase has several steps in which activities that must be carried out are clearly defined in this paper. In addition, design and implementation of tabu search metaheuristic for travelling salesman problem is done as a case study to illustrate applicability of DIMMA. taheuristic methods are genetic algorithm (Goldberg, 1989) and ant colony optimization (Dorigo & Stützle, 2004) in which collective intelligence play the important role (Wang, 2010). Tabu search (Glover & Laguna, 1997) and simulated annealing (Kirkpatrick, Gelatt, & Vecchi, 1983) are the two popular singlesolution based metaheuristics that improve a single solution in an iterative algorithm. With growing scale and complexity of optimization problems, metaheuristics will gain more and more popular. According to significant growth in using metaheuristics as optimization tools, there must be a standard methodology for design and implementing DOI: 10.4018/jamc.2010040104 International Journal of Applied Metaheuristic Computing, 1(4), 58-75, October-December 2010 59 Copyright © 2010, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited. them. Such a methodology is used for recording experience and allows projects to be replicated. Moreover, this standard methodology can be a comfort factor for new adopters with little metaheuristic experience, and can show the guidelines to everyone who want to design and implement metaheuristics. To the best of our knowledge, no methodology has been proposed in literature for design and implementation metaheuristic algorithms. There are many software frameworks in the literature for metaheuristics (Voss & Woodruff, 2002; Fink et al., 1999), in which framework means reusable programming codes and components for metaheuristics (Talbi, 2009). Hence, the meaning of frameworks in these references is different from our proposed methodology. Although there are several tutorials as lectures on how to design meheuristics (Thierens, 2008), they are sometimes for special metaheuristic and do not consider this process as a whole. The proposed methodology in this paper, a Design and Implementation Methodology for Metaheuristic Algorithms (DIMMA), shows guidelines to everyone who wants to design and implement a metaheuristic algorithm. Webster’s collegiate dictionary defines methodology as “a body of methods, rules, and postulate employed by a discipline” or “the analysis of the principles or procedures of inquiry in a particular field” (Merriam-Webster, 1997). DIMMA includes several phases, steps, disciplines, and principles to design and implement a specific metaheuristic for a given optimization problem. In other words, DIMMA refers to the methodology that is used to standardize process of design and implementing a metaheuristic algorithm. In Sections 2-5 weWe explain the architecture of DIMMA and its phases and steps, In Section 6 wefollowed by a description of each step of DIMMA using design and implementation of Tabu Search (TS) metaheuristic for Travelling Salesman Problem (TSP) as a case study. 2. ArchItecture oF dImmA The architecture of DIMMA has been inspired from Rational Unified Process (RUP) which is a methodology for software engineering (Kroll & Krutchten, 2003). DIMMA has two dimensions including dynamic and discipline dimension (Figure 1). Dynamic dimension is the horizontal dimension, which includes phases of the methodology: initiation, blueprint, and construction. Discipline dimension is the vertical dimension that shows the disciplines, which logically group the steps, activities, and artifacts. DIMMA has three sequential phases that each of them has several steps (Figure 2). In each step, we define several activities, which must be done to complete the steps. These phases are as follows: initiation, in which the problem in hand must be understood precisely, and the goal of designing metaheuristic must be clearly defined. The next phase is blueprint, the most important goals of this phase are selecting metaheuristic solution method, defining performance measures, and designing algorithm for our solution strategy. The last phase is construction in which implementing the designed algorithm, parameters tuning (parameter setting), analyzing its performance, and finally documentation of results must be done. In some steps, it is necessary to review pervious steps to justify and improve decisions and algorithm. For example, it is common for the algorithm to be modified after the performance evaluations. These backward movements are illustrated in Figure 2.