A Bidirectional Gated Recurrent Unit Model for PUE Prediction in Data Centers

A Bidirectional Gated Recurrent Unit Model for PUE Prediction in Data Centers 1st Dhivya Dharshini Kannan Department of Electrical and Computer Engineering National University of Singapore Singapore e1322630@u.nus.edu 2nd Anupam Trivedi Department of Electrical and Computer Engineering National University of Singapore Singapore eleatr@nus.edu.sg 3rd Dipti Srinivasan Department of Electrical and Computer Engineering National University of Singapore Singapore dipti@nus.edu.sg Abstract—Data centers account for significant global energy consumption and a carbon footprint. The recent increasing demand for edge computing and AI advancements drives the growth of data center storage capacity. Energy efficiency is a cost- effective way to combat climate change, cut energy costs, improve business competitiveness, and promote IT and environmental sustainability. Thus, optimizing data center energy management is the most important factor in the sustainability of the world. Power Usage Effectiveness (PUE) is used to represent the opera- tional efficiency of the data center. Predicting PUE using Neural Networks provides an understanding of the effect of each feature on energy consumption, thus enabling targeted modifications of those key features to improve energy efficiency. In this paper, we have developed Bidirectional Gated Recurrent Unit (BiGRU) based PUE prediction model and compared the model perfor- mance with GRU. The data set comprises 52,560 samples with 117 features using EnergyPlus, simulating a DC in Singapore. Sets of the most relevant features are selected using the Recursive Feature Elimination with Cross-Validation (RFECV) algorithm for different parameter settings. These feature sets are used to find the optimal hyperparameter configuration and train the BiGRU model. The performance of the optimized BiGRU-based PUE prediction model is then compared with that of GRU using mean squared error (MSE), mean absolute error (MAE), and R-squared metrics. Index Terms—data centers, power usage effectiveness, energy- plus, neural networks, gated recurrent unit, bidirectional gated recurrent unit, recursive feature elimination with cross-validation I. INTRODUCTION The capacity of data centers (DC) in Southeast Asia is expected to increase by 19% annually between 2021 and 2026 due to the present focus on edge computing [1]. By 2030, the global DC market will grow to reach US$554.4 billion, with Indonesia, Malaysia, and Singapore serving as the main development areas [2]. Singapore intends to increase its current 1.4 GW capacity of more than 70 DCs by 300 MW. The goal established by the Singaporean government is for all © 2025 IEEE. This is the author’s accepted version of the work published in the Proceedings of the 2025 International Joint Conference on Neural Networks (IJCNN). D. D. Kannan, A. Trivedi, and D. Srini- vasan, “A Bidirectional Gated Recurrent Unit Model for PUE Predic- tion in Data Centers,” IJCNN 2025, Rome, Italy, pp. 1–8, 2025. DOI: 10.1109/IJCNN64981.2025.11227238. The final published version is available at: https://ieeexplore.ieee.org/document/11227238 DCs in the country to have a PUE of 1.3 or below at 100% IT load [3]. By 2027, the demand for DC storage capacity is expected to double as a result of advances in AI [4]. According to the International Energy Agency, in 2022, DCs consumed about 460 TWh of electricity, which is 2% of the global electricity consumption [5] and equivalent to the total electrical usage of Japan [6]. By 2026, DC energy consumption is expected to be about 1000 TWh. By 2027, 75% of the companies will have established sustainable data center initiatives, up from only 5% in 2022, according to Gartner [7]. Power Usage Effectiveness (PUE) is the direct representa- tion of energy consumed by DCs and is the ratio of total DC energy usage to IT equipment energy usage [8]. Modern DCs are complex systems with huge amounts of interdependent data from various sensors such as IT load, cooling system, environmental parameters, etc, which makes it difficult to estimate PUE using traditional mathematical methods. Prede- fined feature interactions are not necessary for neural networks (NN). It creates the best-fit model by automatically looking for patterns and interactions between features [9]. Thus, capturing the complex inter-dependencies between features and their impact on PUE. Various PUE prediction models based on Neural Networks [9] [10], Deep Neural Network [11] [12], Gated Recurrent Unit [13], Multilayer Perceptron, Resilient Backpropagation- based Deep Neural Network, and Attention-based Long-Short Term Memory [14] were proposed in previous studies. RNN (Recurrent Neural Network) is a type of NN applied for time-series-based data for sequence prediction. It maintains a hidden state that is updated at each time step to keep the information from the previous steps in memory [15]. Since the data set for the PUE prediction is time-series based, RNN is the best choice for developi

…(Full text truncated)…

This content is AI-processed based on ArXiv data.