Prediction Method of Energy Consumption Based on Multiple Energy-Related Features in Data Center

As the scale of cloud data center continues to expand, issues such as energy consumption and resource utilization are becoming more and more prominent. To improve the energy efficiency of the data center, energy consumption prediction methods are critical to data center energy conservation efforts. However, due to complexity and heterogeneity in cloud computing scenarios, it is difficult to precisely estimate the energy consumption using conventional approaches. To this end, this work presents an energy consumption method based on multiple energy-related features to cope with low energy efficiency. Unlike other methods that focus only on a few performance features, the proposed method screens out 12 key features related to energy and uses the deep learning model for adequate training. In particular, this approach is composed of three main phases including (i) energy-related features acquisition, (ii) essential feature selection, and (iii) energy consumption model establishment. The experimental evaluation shows that the proposed energy consumption prediction method is superior to other energy prediction methods in most cases.