5G EN-DC场景下LTE基站下行速率预测方法研究  

Research on downlink throughput prediction method of base station in 5G EN⁃DC

作　　者：陶倩昀 袁三男 张艳秋 TAO Qianyun;YUAN Sannan;ZHANG Yanqiu(College of Electronics and Information Engineering,Shanghai University of Electric Power,Shanghai 200090,China)

机构地区：[1]上海电力大学电子与信息工程学院,上海200090

出　　处：《南京邮电大学学报（自然科学版）》2022年第2期72-78,共7页Journal of Nanjing University of Posts and Telecommunications：Natural Science Edition

基　　金：国家自然科学基金(61401269)资助项目。

摘　　要：在EN-DC Option 3x双连接中,5G gNB能否在数据分流时准确地获取LTE eNB下行速率,影响着5G E-UTRA和NR双连接(E-UTRA-NR Dual Connectivity,EN-DC)实际性能的高低。文中提出了一种结合贝叶斯超参数优化的双层堆叠长短时记忆时序预测模型(BO_SLSTM)对LTE eNB下行速率进行实时高精度预测。研究了不同自适应学习率优化算法和时间步长对模型预测精度及速度的影响,实现算法的进一步优化。实验结果显示,经过优化后的模型预测准确性达到了99.8%,在LTE eNB下行速率预测中具有良好的预测性能和较好的适用性。In EN⁃DC Option 3x,the ability of 5G gNB to obtain the LTE eNB downlink throughput accurately during data offloading affects the practical performance of 5G EN⁃DC.A time series prediction model of double stacked long⁃short term memory based on Bayesian super⁃parameter optimization(BO_SLSTM)was proposed to predict the downlink throughput of LTE eNB in real time and with high precision.The effects of different adaptive learning rate optimization algorithms and time steps on the prediction accuracy and speed of the model were studied to further optimize the algorithm.The experimental results showed that the prediction accuracy of the optimized model,which has good prediction performance and good applicability in LTE eNB downlink throughput prediction,was up to 99.8%.

关 键 词：双连接 Option3x 下行速率预测 BO_SLSTM 贝叶斯超参数优化 时序预测 

分 类 号：TN929.5[电子电信—通信与信息系统]