Energy Consumption Monitoring and Prediction System for IT Equipment.

This paper focuses on the monitoring and prediction of the energy consumption of IT equipment to make informed decisions in terms of energy efficiency. The challenge with current monitoring systems lies in their specialization, scalability and integration complexities. To overcome these challenges, we propose a monitoring and prediction system for energy consumption of IT equipment. The proposed solution combines an adaptable, cost-effiective and energy-Efficient embedded device with open source software and a service-oriented architecture (SOA), which offers flexibility and integration capabilities, facilitating the easy inclusion of several workstation working from different environments. Several traditional Linear Regression (LR) models were evaluated prediction models using a temporal window of hour taking into account several features. As a result of the LR models evaluation, it is established that the Bayesian Ridge model was the best model since it presented the lowest error and the highest coefficient of determination. Finally, two approaches were evaluated to predict energy consumption: a Kernel Density Estimation (KDE)-based mechanism proposed to generate predictor variables in order to predict future energy using the best LR model, and a KDE-based energy model. Numerical results show that the energy prediction using KDE for the energy measurements provides lower time response than the LR based mechanism for the available dataset. [ABSTRACT FROM AUTHOR]