基于多头自注意力机制的LSTM-TCN基站流量预测算法  

LSTM⁃TCN base station traffic prediction algorithm based on multi⁃head self⁃attention mechanism

作　　者：李维烨 贾海蓉[1] 申陈宁 吴永强 LI Weiye;JIA Hairong;SHEN Chenning;WU Yongqiang(College of Electronic Information and Optical Engineering,Taiyuan University of Technology,Jinzhong 030600,China;Shanxi Communication Tongda Microwave Technology Co.,Ltd.,Taiyuan 030000,China)

机构地区：[1]太原理工大学电子信息与光学工程学院,山西晋中030600 [2]山西通信通达微波技术有限公司,山西太原030000

出　　处：《现代电子技术》2024年第23期125-130,共6页Modern Electronics Technique

摘　　要：基站流量预测对于蜂窝网络的规划、资源分配和用户体验优化至关重要。为提高基站流量预测精度,文中设计一种结合多头自注意机制(MHSA)的LSTM-TCN基站流量预测算法。其中:MHSA能够从多个角度强化基站流量数据的内在关联,增强了模型对流量数据重要特征的表达能力;LSTM-TCN模型中长短期记忆(LSTM)网络捕捉流量数据中的长短时依赖性;时间卷积网络(TCN)进一步捕捉流量数据中的全局特征,使得模型能够提取基站流量数据在不同时间尺度上的变化模式和时间依赖关系,提高基站流量预测模型的拟合能力和预测精度。实验结果表明,该流量预测算法与其他算法相比,在运营商基站流量数据的预测中有效降低了均方根误差和平均绝对误差,提高了决定系数,验证了该流量预测算法的有效性,从而为基站休眠节能提供决策支持。Base station traffic prediction is crucial for the planning,resource allocation and user experience optimization of cellular networks.An LSTM⁃TCN base station traffic prediction algorithm that incorporates a multi⁃head self⁃attention(MHSA)mechanism is designed in order to improve the accuracy of base station traffic prediction.The MHSA can strengthen the intrinsic correlation of base station traffic data in multiple perspectives,which enhances the model's ability to express important features of traffic data.The long short⁃term memory(LSTM)network in LSTM⁃TCN model captures the long and short⁃term dependencies in the traffic data,while the temporal convolutional network(TCN)captures the global features of the traffic data,which allows the model to extract the change pattern and time dependence of base station traffic data on different time scales,so as to improve the model's fitting ability and prediction accuracy.Experimental results show that the proposed traffic prediction algorithm reduces both the root mean square error(RMSE)and the mean absolute error(MAE)effectively in the prediction of operator base station traffic data and improves the coefficient of determination(R2)in comparison with the other algorithms,which verifies the validity of the traffic prediction algorithm.Therefore,the proposed algorithm can provide decision support for the dormant and energy saving of the base station.

关 键 词：5G流量 基站 流量预测 混合神经网络 多头自注意 LSTM-TCN 

分 类 号：TN929.5-34[电子电信—通信与信息系统]