Cloud Workload Prediction Based on Bayesian-Optimized Autoformer

Abstract

Accurate workload forecasting plays a pivotal role in the management of cloud computing resources, enabling significant enhancement in the performance of the cloud platform and effective prevention of resource wastage. However, the complexity, variability, and strong time dependencies of cloud workloads make prediction difficult. To address the challenge of enhancing accuracy in contemporary cloud workload prediction, this paper employs empirical and quantitative research methods, introducing a cloud workload prediction method based on Bayesian-optimized Autoformer, termed BO-Autoformer. Initially, the cloud workload data were divided according to the time-sliding window to construct a continuous feature sequence, which was used as the input of the model to construct the Autoformer prediction model. Subsequently, to further enhance the model’s performance, the Bayesian optimization method was employed to identify the optimal combination of hyperparameters, resulting in the development of the Bayesian optimization-based Autoformer cloud workload prediction model. Finally, experiments were conducted on a real Google dataset to evaluate the model’s effectiveness. The findings reveal that, compared to alternative models, the proposed prediction model demonstrates superior performance on the cloud workload dataset, and can effectively improve the prediction accuracy of the cloud workload.