Reinforcement Learning For Fuzzy Multi-objective Cloud Resource Scheduling Problem

For solving the scheduling of cloud computing resource,scholars proposed a variety of cloud resource scheduling models.However,there are uncertainties in the process of practical application,thus we use the fuzzy theory to fuzzify the task execution time to make the model more realistic,and proposed an uncertain multi-objective cloud resource scheduling model.To solve the model,several improved Q learning algorithms in reinforcement learning were studied.In this paper,the classical triangular fuzzy number was used to fuzzify the task execution time,and an uncertain multi-objective cloud resource scheduling model was established to optimize the optimize task completion time and running cost simultaneously.To solve the model,a heuristics accelerate Q-Learning based on automatic updating weighting factor algorithm(WHAQL)was presented.The experimental results show the difference between the uncertain cloud resource scheduling model and the determined cloud resource scheduling model,the influence of uncertainty on the cloud resource scheduling is verified.The optimization ability and convergence speed of WHAQL algorithm are verified in algorithm comparison experiments.In the follow-up study,in order to make the uncertainty of execution time more realistic,Z-number fuzzy number was used to fuzzify the execution time.An uncertain multi-objective cloud resource scheduling model based on Z-number was established.On the basis of Q Learning,eligibility trace,heuristic function,weighting factor and Boltzmann mechanism were introduced.The simulation results show that compared with QL,QL(?),HAQL have better performance in the optimization ability and convergence speed.The experimental results verified that the larger the eligibility trace is,the faster the algorithm converges.HAQL(?)was verified to make the system load more balanced.In order to more convenient to apply different algorithms to different cloud resource scheduling model,on the basis of Cloudsim,using open source of the Swing framework designed a cloud resource scheduling algorithm simulation platform.The platform provides a simple interface,and has good extensibility.The difference between the algorithm and the model can be compared by using simple operation.It provides reliable experimental data for comparing the performance of model and algorithm.