SpectraNet: a lightweight hybrid time–frequency deep learning framework for sustainable cloud workload forecasting

The rapid proliferation of cloud computing has created the demand for sustainable resource management strategies to address the increasing energy consumption and carbon footprint of data centers. Accurate CPU workload forecasting is a cornerstone of this effort, enabling proactive resource allocation that minimizes energy waste from over-provisioning and prevents performance degradation from under-provisioning. This study introduces SpectraNet, a lightweight hybrid model designed to advance sustainability in cloud computing by delivering accurate CPU usage predictions with minimal computational overhead. SpectraNet integrates time-domain and frequency-domain analysis to capture both transient and periodic workload dynamics, enabling robust and fine-grained forecasting. SpectraNet processes time-series data through parallel time- and frequency-domain branches, which are then fused and refined using a CNN with attention to capture local patterns. Finally, an LSTM models long-term dependencies to generate the forecast. Extensive experiments on the vmCloud Dataset and Alibaba Cluster Trace datasets demonstrate that SpectraNet achieves a Mean Absolute Error (MAE) of 0.0549 on long-range forecasts, delivering comparable or superior accuracy to larger baselines while using roughly twenty times fewer parameters and offering noticeably faster inference. The model’s efficiency and small footprint make it a practical and scalable solution for real-world, resource-constrained cloud environments. By enabling more precise and energy-efficient resource management, SpectraNet provides a valuable tool for building more sustainable and cost-effective cloud infrastructures. The implementation code for this study has been made publicly available.