Accurate Cloud Data Center Workload Forecasting Using CEEMDAN-VMD and BiLSTM-BiGRU Networks

Abstract

Managing cloud data center resources— high-dimensional, chaotic operational data—requires forecasting workload. In order to enhance prediction without temporal repeating features in the CVCBM model, this paper presents a lightweight Bidirectional GRU (BiGRU) and Bidirectional LSTM (BiLSTM). Workload signals are concurrently denoised and decomposed in two phases (CEEMDAN and VMD). While K-Means clustering favors high-workload data, SE identifies key components. Conv1DBiLSTM-BiGRU is a hybrid approach that acquires both short-term and long-term temporal patterns. Using input datasets, the Flask-built trained model forecasts workloads in real time. According to experimental study, the improved model offers reliable, scalable, and real-time forecasts for cloud data center resource allocation while lowering computing costs and improving forecast accuracy.