Research On Multi - Objective Optimization Task Scheduling Based On Improved Genetic Algorithm In Heterogeneous Computing System

Task-scheduling is the key part for the heterogeneous computing system. An efficient task-scheduling can improve utilization of the resource and reduce the execution time of task. Therefore, the study of task-scheduling problem is valuable.In the paper, the research of status for multi-objective optimization is described. Especially the genetic algorithm is applied to the multi-objective optimization. The structure and principle of system and the mode of computing tasks for heterogeneous computing systems is described, laying the groundwork for the application environment of this study. Problem and classification of the task-scheduling are analyzed, clearing the task scheduling type of this paper. The multi-objective optimization problem is introduced. The multi-objective type has three classes, including Minimize goals, maximizing objectives, maximize and minimize part of the target portion of the target. At present, the multi-objective optimization algorithms based on genetic algorithm include the multi-objective genetic algorithms based on the weight, the niche genetic algorithm based on the shared function, the multi-objective genetic algorithms based on the vector evaluation.In the paper, the heterogeneous computing environments and task characteristics are set for application of the algorithm, explaining the achievement of objects, including load balancing and make-span, reliability. Three goals by formulating are unified into a goal of minimizing. This paper proposes an improved genetic algorithm based on weight values for the premature convergence and poor local search ability and local optimization problems of the genetic algorithms by research of the process and advantages and disadvantages of genetic algorithms. The algorithm is mainly done the following areas:Firstly, adaptive crossover and mutation rate are achieved by the population entropy of Optimum individuals, adjusting the search range and avoiding the premature convergence because of diversity.Secondly, the idea of single parent genetic algorithm is taken, reducing the amount of computation to a certain extent.Thirdly, the recycling pool and backtracking mechanisms and optimal preservation and optimal evolutionary strategy is proposed to prevent premature convergence and local optimal solutions effectively.Finally, this paper does experiments, evaluating the diversity of initial population. The simulation results also show the superiority of the improved algorithm. Comparing with traditional genetic algorithm and adaptive genetic algorithm and the improved algorithm, the convergence of the improved algorithm is good, and be able to get the optimum solution.