Conference Paper

Achieving memetic adaptability by means of fuzzy decision trees

Dept. of Math. & Comput. Sci., Univ. of Salerno, Salerno, Italy
DOI: 10.1109/FUZZY.2010.5584336 Conference: Fuzzy Systems (FUZZ), 2010 IEEE International Conference on
Source: IEEE Xplore


Evolutionary Algorithms are a collection of optimization techniques that take their inspiration from natural selection and survival of the fittest in the biological world and they have been exploited to try to resolve some of the more complex NP-complete problems. Nevertheless, in spite of their capability of exploring and exploiting promising regions of the search space, they present some drawbacks and, in detail, they can take a relatively long time to locate the exact optimum in a region of convergence and may sometimes not find the solutions with sufficient precision. Memetic Algorithms are innovative meta-heuristic search methods that try to alleviate evolutionary approaches' weaknesses by efficiently converging to high quality solutions. However, as shown in literature, memetic approaches are affected by several design issues related to the different choices that can be made to implement them. This paper introduces a multi-agent based memetic algorithm which executes in a parallel way different cooperating optimization strategies in order to solve a given problem's instance in an efficient way. The algorithm adaptation is performed by jointly exploiting a knowledge extraction process, based on fuzzy decision trees, together with a decision making framework based on fuzzy methodologies. The effectiveness of our approach is tested in several experiments in which our results are compared with those obtained by some non-adaptive memetic algorithms.

1 Follower
3 Reads
  • [Show abstract] [Hide abstract]
    ABSTRACT: This paper introduces a genetic fuzzy system for parameter control of metaheuristics. Two basic metaheuristics have been considered as examples, genetic algorithm and tabu search. The controlled parameters of the tabu search are the short and long term memories. Parameters of the genetic algorithm under control are the mutation and reproduction rates. Fuzzy rule-based models offer a natural mechanism to describe global behavior as a combination of control rules. They also inherit a means to gradually shift between control rules which jointly defines a control strategy. They are a natural candidate to construct parameter control strategies because they provide a way to develop decision mechanisms based on the specific nature of search regions and transitions between their boundaries. An application example using the classic vehicle routing problem with time windows is included to evaluate the genetic fuzzy system performance. Experimental results show that GFS-controlled metaheuristics improve search behavior and solution quality when compared against standard, constant parameters genetic and tabu search approaches. It also provides reasonably good suboptimal solutions faster than specially tailored exact methods reported in the literature.
    Evolutionary Intelligence 09/2011; 4(3):183-202. DOI:10.1007/s12065-011-0059-y