Container load prediction method based on a hybrid model of ARIMA and Bi-LSTM

With the rapid development of cloud computing and container technology, containerization has become a core component of modern IT infrastructure. Container load prediction is crucial in improving resource scheduling efficiency and optimizing system performance. However, traditional prediction methods often fail to achieve the desired accuracy in large-scale cloud environments due to the dynamic and complex nature of container loads. To address this problem, this paper proposes a container load prediction method based on a hybrid model of ARIMA and Bi-LSTM. First, a periodic classification method based on container resource consumption characteristics is designed to classify container load into periodic and non-periodic. Then, the ARIMA and Bi-LSTM models are used to predict the periodic and non-periodic loads, respectively. To further enhance the prediction accuracy, the results from both models are fused using a dynamic weighted average method. The proposed method is validated using the Alibaba Cluster Trace 2018 dataset. Experimental results demonstrate that, compared to ARIMA, LSTM, and Bi-LSTM models, the hybrid model improves CPU load prediction accuracy by 25.19%, 45.90%, and 37.60%, respectively, and memory load prediction accuracy by 29.05%, 39.36%, and 24.67%. These results highlight the significant predictive advantage of the hybrid model.