Self-adaptive mechanisms for the identification\nof the most suitable variation operator in evolutionary algorithms\nrely almost exclusively on the measurement of the\nfitness of the offspring, which may not be sufficient to assess\nthe optimality of an operator (e.g., in a landscape with an high\ndegree of neutrality). This paper proposes a novel adaptive\noperator selection mechanism which uses a set of four fitness\nlandscape analysis techniques and an online learning algorithm,\ndynamic weighted majority, to provide more detailed\ninformation about the search space to better determine the\nmost suitable crossover operator. Experimental analysis on\nthe capacitated arc routing problem has demonstrated that\ndifferent crossover operators behave differently during the\nsearch process, and selecting the proper one adaptively can\nlead to more promising results.
Loading....