基于改进长短时记忆神经网络的5G通信网络流量预测  

Flow prediction of 5G communication networks based on improved long and short term memory neural network

作　　者：赵巍[1] ZHAO Wei(Institute of Technology,East China Jiaotong University,Nanchang 330100,China)

机构地区：[1]华东交通大学理工学院,南昌330100

出　　处：《沈阳工业大学学报》2022年第6期672-676,共5页Journal of Shenyang University of Technology

基　　金：江西省教育厅科学技术项目(GJJ209301);江西省高校人文社会科学研究项目(GL20147);南昌市5G无线网络优化重点实验室项目(2020-NCZDSY-015);南昌市移动通信重点实验室项目(2018-NCZDSY-008).

摘　　要：针对5G通信网络流量预测研究中常存在的忽略流量序列内所包含变化趋势与数据间的关联特性而导致预测结果不准确、与用户需求不符等问题,通过采用基于压缩感知的改进长短时神经网络算法,对通信数据流的采样建立具有一致收敛的约束条件稀疏矩阵,将实际流量序列与待预测的流量序列作为LSTM模型输入同时进行深度训练,实现对未来流量数据的精度预测,从而有效提高了5G通信网络流量预测精度.经TensorFlow仿真实验显示,所提算法可以将通信网络数据流的预测误差均值控制在5%~10%,绝对误差均值为3%~8%,相较于常规LSTM通信流量算法的预测结果,平均精度提高了近5%.Aiming at the problems during 5G communication network flow prediction,such as the inaccurate prediction which is inconsistent with the users’needs due to ignoring the correlation between the change trend and the data contained in the flow sequence,by using an improved long and short neural network algorithm based on compressed sensing,the constraint sparse matrix with uniform convergence was established for the sampling of communication data stream to analyze actual data.The actual flow series and the flow series to be predicted were used as the input of LSTM(long and short term memory)model,and the intensive training was carried out at the same time to realize the accurate prediction of future flow data and improve the flow prediction accuracy of 5G communication networks.The TensorFlow simulation results show that the as-proposed algorithm can control the average prediction error of communication network data flow within the range from 5%to 10%and the average absolute error ranges within 3%to 8%.Compared with the communication flow predicted by conventional LSTM algorithm,the average prediction accuracy is improved by nearly 5%.

关 键 词：压缩感知 LSTM模型 5G通信网络 流量预测 神经网络 稀疏矩阵 基站 TensorFlow仿真 

分 类 号：TN915[电子电信—通信与信息系统]