A Predictive Model for Power Consumption Estimation Using Machine Learning

The reduction of the power consumption of IT devices (computers, sensors, etc.) represents a challenging task for the research in Green IT. A number of research works on energy management have been proposed to ensure a reduction of the energy consumed by a software or an IT device. Nevertheless, to optimize the energy consumption of IT devices, it is essential to have an accurate model for power consumption estimation. Therefore, we propose in this study a new solution for estimating the power consumption of IT devices with a high accuracy. Our proposed solution is based on a statistical model called ARIMA to enable the evaluation of the power consumption. In particular, the proposed model uses historical data related to resources usage (CPU, memory, etc.) collected on a Linux server to perform the prediction. This prediction can be used by a management system in order to take accurate decision of reconfiguration of IT devices parameters. We provide an evaluation of the performance of our proposed solution against three types of machine learning algorithms, namely linear regression, support vector regression and multi-layer perceptron regressor. The results of performance evaluation confirm that the proposed model outperforms other models in terms of accuracy.