Deep Reinforcement Learning-based Co-Optimization Method for Energy Consumption and Arithmetic Power in Cloud Computing Data Centers

As the scale of data centers continues to expand, they face the challenge of rapidly increasing data volume.To solve the problem, this study constructs a system model based on cloud computing data center scenario, designs a task scheduling model using the improved Double DQN algorithm, and proposes a co-optimization method for energy consumption and arithmetic power in cloud computing data centers.Through simulation experiments on Cloud Sim cloud simulation platform, it is found that this paper's method has smaller energy consumption compared with other algorithms, and the energy consumption values are reduced by 23.92% and 17.62% compared with the Q-learning algorithm and the Q-learning() algorithm with different numbers of virtual machines, and it has a faster convergence speed.Meanwhile, this paper's method performs better in reward value, average latency and load balancing, and the average latency is reduced by 30.30%~53.33% and 67.76%~90.59%than the comparison method in regular traffic and high traffic environment.The results show that the optimized Double DQN algorithm in this paper can effectively reduce the energy consumption and latency of cloud computing data centers, and has some practical value in the co-optimization of energy consumption and arithmetic power.