A Dynamic Guided Optimal Strategy Based On Preference Information

Multi-objective evolutiona ryalgorithm is accompanied by a variety ofoptimizatio nproblem’s arising. In recent years, MOEA made a considerablecontribution in the field of optimization, and it has been the focus of research in thisfield. In our real life, we often encounter optimization problem, and usually requiresmultiple targets simultaneously optimized. For example, the rapid development of ashop, not only to consider the production costs, product quality, the seller serviceattitude, but also to pay attention to customer’s satisfaction, shop’s profits, and otheraspects. As we all know, service attitude and customer’s satisfaction are mutuallyreinforcing relationship, while production costs and product quality is of twoconflicting objectives. However, in most cases, getting the optimal overall need tocompromise a plurality of sub-goals which tend to conflict with each other.For the traditional multi-objective evolutionary algorithm is concerned, its goalis to reach the global optimum solution set and do the best to achieve each sub-goaloptimal. The performance of the geometry is the widely distributed evenly, which iscurrently the contents of the study by the most scholars. However, in real life, each ofour individual requirements and goal is often personal, which is so-called individualpreferences. In MOEA, taking into account the differences of individual preferencesto obtain the optimal solution set which is satisfied decision-maker, such algorithms iscalled preference-based MOEA. However, this is an important part of my paper.Including preferences with a multi-objective evolutionary algorithm is one of the newresearch directions in the field of evolutionary computing. So it has a definitelyresearch significance.This paper proposed a dynamic heuristic multi-objective optimal strategy basedon the preference information. The strategy is to adjust a parameter Epsilon reflectingdynamic of the guided regions in searching process and set another parameter Thetacontrolling the size of preference range of DM, which employs the distance betweensolution set and the guiding regions as a factor of selection strategy. Finally, we canobtain a compromise solution set within the desired region effectively. Using thedynamic optimal strategy in classical algorithm verifies its performance. And comparewith some classic preference algorithms, it shows the algorithm with this strategyprocessing a good performance especially on the convergence and effectiveness. Overall, relatively than other algorithms, the work of this paper has a certaininnovation, performance is as follows:1) We proposed a new preference relation that it is called D-dominance, usingthe idea of dividing objective space and using the reference point to expresspreference information. Then adopting the mapping point of the reference point todivide objective space again and redefine the dominated relationship.2) We proposed a dynamic guided optimal strategy to improve the global searchability of the algorithm and to make up the limitations of the space division ofdominated relationship. The strategy is to adjust a parameter Theta controlling the sizeof preference range and increasing the diversity of preference solutions for decisionmakers. Then it sets another parameter Epsilon controlling the size of guided regionand reflecting dynamic of the guided regions.3) The proposed method in this paper is very flexible and can be applied in avariety of MOEA. NSGA-II algorithm as an example to experiment and analysis themethods, compared with the g-dominance approach, r-dominance methods to proveD-dominance algorithm has better performance.