A novel predictive method based on key points for dynamic multi-objective optimization. (15th March 2022)

• Record Type: Journal Article
• Authors: Wang, Chunfeng; Yen, Gary G; Zou, Fei
• Is Part Of: Expert systems with applications. Volume 190(2022)
• Publication Date: 2022-03-15
• Subjects: Dynamic multi-objective optimization -- TOPSIS -- Clustering strategy -- Predictive strategy; Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2021.116127
• Languages: English
• ISSNs: 0957-4174

Abstract

Highlights: TOPSIS is incorporated into the key points guided evolutionary algorithm framework. The score computed by TOPSIS is used in mating and environment selection. A prediction method is applied for tracking the locations of these key points. Abstract: Dynamic multi-objective problem is very difficult to be solved because of the variability of the objective function with time. To overcome the difficult caused by such variability, a predictive method utilizing some key points (including polar points and centroid points) is designed, which contains four critical steps. First, the whole population is automatically divided into multiple clusters, which will be used to preserve a good diversity in the process of population evolution. Second, the technique for order of preference by similarity to ideal solution (TOPSIS), a well-regarded multi-attribute decision making strategy, is exploited to improve its convergence speed further. Third, the polar point and centroid point in each cluster are utilized to obtain the initial population by using sequence predictive method when environmental changes are detected. Fourth, to accelerate the convergence speed, the quantitative value for each individual determined in the prediction process is also used in mating selection and environmental selection. The numerical results imply that the new method can deal with the change of environment effectively and track the Pareto optimal front (POF) quickly. Meanwhile, the comparison results with several selected state-of-the-art methods also show that the overall performance of the proposed method is the best on most benchmark problems.