Optimized Resource Forecasting for Carbon-Intelligent Data Centers With TempoSight: A Hybrid Deep Learning Approach

Precise resource utilization forecasting is paramount for enabling carbon-intelligent operation in Industrial Internet of Things (IIoT) environments; Cloud data centers (DCs) are no exception. Optimizing their resource allocation not only reduces operational costs but also minimizes energy consumption and associated carbon emissions, contributing to sustainable computing. This paper introduces <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">TempoSight</i>, a novel hybrid Deep Learning (DL) architecture designed for multivariate (MV) time series forecasting, specifically targeting carbon-intelligent resource provisioning in Cloud DCs. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">TempoSight</i> synergistically combines the strengths of Patch Time Series Transformers (PatchTST), excelling at capturing global context and long-range dependencies, and Long Short-Term Memory (LSTM) networks, mastering sequential dynamics. To address the critical need for efficient hyperparameter tuning in complex DL models, we propose a Cuckoo Search Algorithm (CSA) based optimization approach. This enables efficient training and optimization of <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">TempoSight</i> for MV time series forecasting, leading to improved resource utilization predictions. Rigorous evaluation on the Alibaba and Bitbrains datasets, representing diverse real-world IIoT workload patterns, demonstrates <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">TempoSight</i>’s superior accuracy and robustness compared to state-of-the-art DL models. Notably, under high-load conditions, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">TempoSight</i> achieves a remarkable 10-15% reduction in Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE). This research highlights the critical role of hybrid DL models and intelligent optimization in achieving accurate and efficient resource provisioning, paving the way for carbon-aware IIoT applications and contributing to the broader goals of sustainable and green computing. Simulation studies also suggest that <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">TempoSight</i>’s high-fidelity CPU forecasts significantly improve the Green Energy Utilization Factor (GEUF) when guiding Predictive Carbon-Aware (PCA) scheduling, quantitatively connecting forecasting precision with concrete sustainability advantages in DC operations.