Enhancing Power Efficiency Prediction in Small-Scale Data Centers using Attention-Based Deep Neural Networks

The growth of Internet-based applications, the rise of enormous data storage, and the evolution of Internet of Things (IoT) technologies have significantly increased the number of data centers. They are crucial in various industries for storing data either on-site, off-campus, or both. Temperature control is crucial in these buildings, as they house a large number of running machines. Several factors affect their performance, including heat generated by IT equipment. Data centers consume a sizable amount of electricity due to the complexity of big data workloads that require extensive computing. The IT infrastructure side of data centers accounts for 60% of power consumption, while systems for heating, ventilating, and air conditioning (HVAC) consume the remaining 40%. To improve energy efficiency, Power Usage Efficiency (PUE) is used to evaluate and enhance the energy performance of data centers. In this study, a prediction model was developed using data from EnergyPlus and IoT sensors, focusing on neural net-based machine learning models such as artificial neural network (ANN), deep neural network (DNN), long short-term memory (LSTM), and attention-based LSTM (Att-LSTM) to predict power usage effectiveness (PUE) values. The accuracy of these algorithms was evaluated using real data from IoT sensors in a data center, and by applying a comparative strategy, the Att-LSTM model outperformed other models. By utilizing Att-LSTM models, data center operators can optimize energy consumption and enhance overall performance, providing valuable insights into improving energy efficiency in data centers.