REAL-TIME SCHEDULING OF PARALLEL TASKS BASED ON REINFORCEMENT LEARNING
Existing parallel tasks scheduling algorithms do not consider the environmental instability,adaptability and real-time performance simultaneously.In view of this,we propose a parallel tasks scheduling algorithm based on reinforcement learning.It took the scheduling process as a Markov decision process.Through the interactions between agents and the environment,policies were optimized by the proximal policy optimization method.A simulation method was used to construct the reward function.By adding empirical terms to the advantage estimators by denoising autoencoders,the agents could learn efficient and reliable arrangement policies.The results of simulation experiments in two scenarios show that the proposed method can schedule in milliseconds,and improve the time utilization by more than 17％and the output by more than 16％compared with existing algorithms.
国家科技期刊平台
NSTL
万方数据
