Deep Learning-Based Container Lifecycle Prediction in Cloud Computing Environments

Abstract

In cloud computing environments, containers serve as the core units for resource scheduling and service deployment. Accurate prediction of their lifecycles is critical for improving system resource utilization and service stability. This paper addresses the challenge of dynamic factors affecting container lifecycles and the limited modeling capacity of traditional methods. It proposes a lifecycle prediction model based on deep regression networks. The proposed model takes multidimensional runtime monitoring metrics and system state information as inputs. It employs a multi-layer nonlinear structure to extract temporal features. An attention mechanism and embedding fusion module are integrated to enhance the modeling of key behavioral patterns. During training, regularization strategies are applied to improve model generalization. Mean squared error is used as the optimization objective to ensure stability and accuracy in continuous time prediction. To comprehensively evaluate the effectiveness of the proposed method, a container-level lifecycle prediction dataset was constructed based on the Google Cluster Trace. Experimental analysis was conducted from multiple perspectives, including hyperparameter sensitivity, system structure complexity, and sampling strategies. The results show that the proposed model outperforms several mainstream baseline methods in terms of MAE, RMSE, and R². It accurately reflects lifecycle trends and demonstrates strong adaptability and modeling stability.