Deep multi-task learning model for time series prediction in wireless communication

Abstract

Making phone calls, sending messages and surfing the Internet all depend on wireless communication. Too many users connect to a same base station at the same time, which would slow network speed down. To address this issue, telecom operators can tune the network capacity in advance according to predicted Maximum Connections. Therefore, predicting Maximum Connections is necessary. Traditional time series model and machine learning can be utilized to address time series prediction task. However, these methods do not take multi-task learning into consideration, and related tasks can promote each other actually. In this paper, we propose a deep learning model based on LSTM for time series prediction in wireless communication, employing multi-task learning to improve prediction accuracy. We conducted several critical features and utilized training signal of related task as inductive bias to promote the generalization performance of main task. Through experiments on several real datasets, we found that the proposed model is effective, and it outperforms other prediction methods.