The Research On Prediction Model-based Dynamic Multi-objective Optimization Algorithm

Dynamic multi-objective optimization problem(DMOP) is a kind of widespreadbasic optimization problem with extensive application prospects in scientific researchand engineering practice, so it is of great scientific and engineering significance tosolve the core problems of DMOP. Currently, based on static multi-objectiveoptimization evolutionary algorithms, most of dynamic multi-objective evolutionaryalgorithms(DMOEA) take improved auxiliary strategies to deal with the problemchanges, but most of these improved strategies have just increased the randomness ofthe algorithm and reduce the convergence rate to deal with the problem changes, andthem have not improved algorithms according to the characteristics of differentDMOP, finally, it is difficult for them to achieve a satisfactory optimizationperformance.This paper presents a novel prediction model-based dynamic multi-objectiveoptimization algorithm(N-PDMOEA) for DMOP, the new prediction model(ADLM)of N-PDMOEA is designed by taking full advantages of heuristic knowledge ofDMOP to solve dynamic multi-objective optimization problem with translationalPareto-optimal set(DMOP-TPS), and then it is applied to solve the optimizationproblem in the cloud computing task scheduling. This paper mainly focuses onPDMOEA and its application on the cloud computing task scheduling, and the maincontents of this paper include:First, the related concepts of DMOP and the basic principles of several commonDMOEA are reviewed. In addition, the basic concepts of PDMOEA are build, and thereason why different prediction models needs be designed for different problems isexplained.Second, through the studies on common DMOP and prediciton models, this paperdefines a kind of dynamic multi-objective optimization problem with translationalPareto-optimal set(DMOP-TPS) and proposes the corresponding algorithmN-PDMOEA for DMOP-TPS. Based on NSGA2, which is a classical algorithm inmulti-objective field, the new prediction model ADLM of N-PDMOEA is designedaccording to the characteristics of DMOP-TPS. Comparative experiment results offour models show that the new prediction model(ADLM) is better than other threeprediciton modles in terms of convergence. Last, this paper proposes a new prediction model(ADLM)-based dynamicmulti-objective task scheduling algorithm(N-PDMOTSA) in cloud computingenvironment. According to the task scheduling in the cloud computing environments,the definition of a dynamic task scheduling problem with Pareto-optimal set (DCCTS)is given, then N-PDMOTSA is designed based on the characteristics of DCCTS. Thesimulation results show that N-PDMOTSA can solve DCCTS well.