Research on LSTM-Based Load Prediction for Edge Data Centers

In order to meet the demands of the application scenarios for the fifth-generation (5G), such as high transmission rate and low delay, the concept of Mobile Edge Computing (MEC) was proposed in industry. By deploying some servers with computing and storage capabilities in mobile access networks, MEC could meet the above demands. However, there is a problem of uneven resource utilization in edge data centers (EDC) currently. Considering the mobility-related factors of EDC to predict the workloads of EDC, the prediction results are conducive to resource allocation, so that resources can be reasonably allocated to improve resource utilization. In order to perform accurate load prediction, this paper jointly considers user mobility and geographic location information of EDC, and proposes a method of load prediction based on Long-Short-Term Memory Networks (LSTM). Experiments show that the prediction accuracy of the proposed method is averagely improved by 4.21%, compared with the state-of-the-art Autoregressive Integrated Moving Average Model (ARIMA).