Research Of Multi-Objective Artificial Bee Colony Algorithm Based On Feedback

In the human productions and livings, most of the optimization problems are both multi-objective and NP-hard. Utilizing the multi-objective optimization method, the conflicting goals in these problems can be better balanced and thus given a satisfactory optimization results. Recently, the multi-objective optimization approaches mainly focus on heuristic algorithms, among which the artificial bee colony algorithm demonstrates a strong ability in solving those problems.One important quality that makes artificial bee colony algorithm different is that it is essentially a probabilistic search process. It can solve a variety of complex issues without providing any gradient information related to the problem. Compared to other swarm intelligence algorithm, the artificial bee colony algorithm also has advantages in less control parameters, fast convergence rate and easy implementation. Although it has been gaining more and more attention, the artificial bee colony algorithm is still in its infancy when applied to multi-objective optimization problems. Existing multi-objective artificial bee colony algorithm is a basic and single method that cannot handle all kinds of problems based on the problems’own characteristics and cannot obtain instant feedback to control the algorithm. This kind of blindness to some degree wastes the resources of the algorithm itself.Feedback mechanism refers to a process that returns the output to the input and changes the input in some way, thereby affecting the overall circumstances. For the deficiencies of current multi-objective artificial bee colony algorithm, this paper first proposes the multi-objective artificial bee colony algorithm framework based on feedback mechanism and detailedly describes the various strategies under the feedback-based framework, including initialization, employed bees foraging, food source evaluation, food source formation mechanism, foraging observation, scout bees foraging and so on. The paper investigates three algorithms based on search cost, search strategy, search factor respectively.The search cost feedback-based algorithm utilizes the update frequency of limit value fed back by employed bees to dynamically adjust and balance the important global parameter of search cost, which improves the overall efficiency of the algorithm. For the search strategy feedback-based algorithm, the proportions of different strategies are adapted to the feedback of searched information for specific problems. This algorithm is more flexible and achieves a better non-dominated solution set. According to the food source probability feedback of the scout bees, the search factor feedback-based algorithm outperforms others by the dynamical adjustment factor in balancing the relationship of global and local search. Finally, the three algorithms are applied to the function optimization problems and the QoS (Quality of Service) based service selection problem. In the experiments, this paper verifies the algorithms’ performance from different aspects using several defined indicators and studies the convergence, parameter adjustment and experiment proof. Moreover, experimental results fully illustrate the effectiveness of the three proposed multi-objective artificial bee colony algorithm in comparison to other algorithms.