华东师范大学学报(自然科学版) ›› 2025, Vol. 2025 ›› Issue (4): 1-14.doi: 10.3969/j.issn.1000-5641.2025.04.001

• •    

基于流量感知的移动网络授权信息在线预测方法

傅宇飞1, 徐泽松1, 冯安昊2, 魏同权1,*()   

  1. 1. 华东师范大学 计算机科学与技术学院, 上海 200062
    2. 河南大学 软件学院, 河南 开封 475004
  • 收稿日期:2024-01-17 出版日期:2025-07-25 发布日期:2025-07-19
  • 通讯作者: 魏同权 E-mail:tqwei@cs.ecnu.edu.cn
  • 基金资助:
    国家自然科学基金 (62272169); 上海市科技重大专项 (2021SHZDZX)

Online grant prediction for mobile networks based on traffic perception

Yufei FU1, Zesong XU1, Anhao FENG2, Tongquan WEI1,*()   

  1. 1. School of Computer Science and Technology, East China Normal University, Shanghai 200062, China
    2. School of Software, Henan University, Kaifeng, Henan 475004, China
  • Received:2024-01-17 Online:2025-07-25 Published:2025-07-19
  • Contact: Tongquan WEI E-mail:tqwei@cs.ecnu.edu.cn

摘要:

在移动网络中, 移动设备需要持续监听控制信道来保证接收到基站发送的调度授权信息, 这在基站没有发送授权信息的情况下会导致大量的能量浪费, 预测授权信息的发送规律可有效减少这种浪费. 现有的授权信息预测方法在特定的流量场景下可实现较高的准确率, 但仍难以应对动态变化的通信环境. 为解决上述问题, 提出了一种基于流量场景感知的在线调度授权信息预测方法: 首先, 对历史授权信息序列进行特征提取并对所提取的特征值进行聚类, 以划分不同的流量场景; 然后, 对每一类场景单独构建并离线训练预测模型. 在线预测阶段, 系统根据流量场景的感知结果选择合适的模型进行预测. 实验结果表明, 与基准方法相比, 所提方法在预测错误率方面最多可降低81.52%, 且在仿真和真实流量轨迹上均可有效节省能耗. 同时, 所提流量场景感知方法能够大幅降低计算复杂度, 从而更适合对流量场景的实时感知.

关键词: 机器学习, 时间序列预测, 移动网络, 聚类

Abstract:

In mobile networks, mobile devices need to continuously monitor the control channels to ensure that they receive all the grants for scheduling sent by the base station, which results in enormous energy waste when the base station does not send any grants. Predicting the pattern of grant sending can effectively reduce such waste. Existing approaches to grant prediction can achieve high accuracy in specific traffic scenarios, but it’s still difficult for them to cope with dynamically changing communication environments. To solve this problem, an online grant prediction method based on traffic scenario perception is proposed. Firstly, the method extracts features from historical grant information sequences and perform a clustering of these features to classify various traffic scenarios. Then, a single prediction model is designed and trained offline for each class of scenarios. During the online prediction stage, the system selects the most appropriate model for prediction based on the results of traffic scenario perception. Experimental results show that, the proposed method is able to reduce prediction error rate by up to 81.52% compared with the benchmarking methods and can achieve effective energy savings on both simulated and real traffic traces. In addition, the proposed approach to traffic scenario perception is able to significantly reduce the computational complexity compared with the benchmarking approach, which makes it more suitable for real-time traffic scenario perception.

Key words: machine learning, time series forecasting, mobile network, clustering

中图分类号: