An Improved Dynamic Multiobjective Optimization Algorithm Based On NSGA-? And It's Application

In the areas of Engineering applications and scientific research,there exists a large number of dynamic multi-objective optimization problems(DMOPs).Compared to statistic multi-objective optimization problems(SMOPs),dealing with DMOPs is a challenging and difficult task.Evolutionary algorithms(EAs)must take the convergence speed and population diversity into account when it comes to DMOPs.The EAs need imporve their overall performance by apply existing or improved strategies as long as the algorithms are used to treat DMOPs.In recent years,it has attracted more and more attentions for EAs to fully utilize the history information obtained during the evolutionary process to guide the searching of Pareto optimal solutions.At present,the reuse and selection of history information are mostly based on the direct geometrical distance between individuals for the past Pareto set.Meanwhile,in order to further improve the dynamic adaptive ability of EAs in dealing with DMOPs,researchers have designed many strategies to improve the diversity of populations.Based on the problems above,the research on the optimal decision-making of operational indices for beneficiation processes under dynamic environments has been carried out.The detail work has been summarized as follows:1.The detailed analysis and description of dynamic multi-objective optimization problems are presented.The mathematical description of DMOPs is introduced,as well as the classification method of DMOPs.By comparing the differences between dealing with SMOPs and DMOPs,the difficulties and existing problems in solving DMOPs are analyzed.2.An improved dynamic multi-objective optimization evolutionary algorithm based on NSGA-?,called RDMOEA,is proposed in this thesis,in which three kind of strategies are designed and introduced in the structure of NSGA-?.In order to improve the adaptive ability to changes,a Pareto frontier predictive strategy is proposed in this thesis.During the evolutionary process,the algorithm detect changes by calculating the gap between the values of change detection function and the predefined threshold.As long as any change is detected,the Pareto frontier predictive strategy will generate new individuals by the time series selected from history Pareto set.At the same time,an adaptive inhomogeneous mutation strategy is proposed,which generate certain number of mutation individuals in the regions where few individuals are found.Two type of benchmark functions are applied to test the overall performance of the proposed algorithm.Meanwhile,the DNSGA-II-A is utilized to compare and analyze performance in depth.The IGD indices obtained by the two algorithms for the problems utilized show that,RDMOEA owns a faster convergence speed,while the HVR indices demonstrate that the solutions obtained by the proposed algorithm spread in the whole objective space uniformly,showing great diversity.The errors indicate that the convergence speed of the algorithm is increased dramatically by applying Pareto prediction.3.The dynamic multi-objective optimal decision making of operational indices of beneficiation(DMOIP)process is described.The objectives and constraints of beneficiation process are analyzed,as well as their relationships with all the decision variables.Based on the analysis of dynamic characteristics,the mathematical model for dynamic optimal decision making of operational indices is proposed.In order to conduct the application research,considering all kinds of dynamic characteristics,six items of experimental operating conditions are selected based on real production data.4.The capacity of RDMOEA in solving DMOIP is investigated in this thesis by comparing the performances of three algorithms,including KnEA,GrEA.The results indicate that under the six environmental working conditions,RDMOEA can get the Pareto set quickly which realize the perfect balance of maximizing overall concentrate output,overall concentrate grade and its average,as well as minimizing the variance of overall concentrate grade.Meanwhile,REMOEA shows better performance than the other two algorithms,which futher demonstrates that RDMOEA is effective and excellent in dealing with DMOIP.