Fernández Cerero, DamiánFernández Montes González, AlejandroOrtega Rodríguez, Francisco JavierJakóbik, Agnieszka2025-01-132025-01-132024-03Fernández Cerero, D., Fernández Montes, P., Ortega Rodríguez, F.J. y Jakóbik, A. (2024). Sequence-to-sequence Architectures for Estimating Long-Term Usage in Data Centre Digital Twins. Engineering Applications of Artificial Intelligence, 24.952-1976https://hdl.handle.net/11441/166442Data centres constitute the core infrastructure of global Internet services, as they are responsible for processing and storing huge amounts of data around the world. The performance and energy efficiency of such infrastructures are crucial to enabling services that require computational and storage resources to be available, scalable, and ready to serve incoming workloads. This paper proposes three Sequence-to-Sequence architectures to predict long-term usage in data centre digital twins, and performs a thorough comparative of their behaviour. We focus on three prominent neural network models: Recurrent Neural Network (RNN) Encoder-Decoder, Temporal Convolutional Network (TCN), and Transformer. The efficacy of these architectures in simulating and predicting data centre operations over extended periods is examined and studied on the well-known Alibaba 2018 machine usage dataset. The analysis emphasises the accuracy, computational efficiency, and scalability of each model to handle large-scale data centre operational data. The findings highlight the distinct advantages and limitations of each architecture, providing information on its suitability for integration into digital twins of data centres. The results aim to guide the development of more efficient predictive models for data centre management, ultimately contributing to improved operational efficiency, reduced energy consumption, and improved decision making in real-time data centre operations.application/pdf38 p.engArtificial Intelligence-Driven Data Centre ManagementSequence-to-Sequence ArchitecturesArtificial Intelligence-Based Long-Term Use PredictionRecurrent Neural NetworkTemporal Convolutional NetworkTransformer ModelsScalable Artificial Intelligence Models for Infrastructure ManagementReal-Time Operationalinfo:eu-repo/semantics/articleinfo:eu-repo/semantics/openAccesshttps://doi.org/10.2139/ssrn.4767239