基于轻量化卷积神经网络的蜂窝流量低复杂度预测方法
Low-complexity Cellular Traffic Prediction Algorithm Based on Lightweight Convolutional Neural Network
郑淞之 1张兴 1张妍 2王兴瑜 3袁国翔1
作者信息
- 1. 北京邮电大学信息与通信工程学院,北京 100876
- 2. 中国电信股份有限公司北京分公司,北京 100032
- 3. 中国人民解放军93216 部队,北京 100085
- 折叠
摘要
随着蜂窝网络数据流量需求的高速增长,对于未来时刻蜂窝流量情况的精准预测,可以帮助改善网络资源分配、实现流量负载均衡,并部署基站节能与休眠策略.基于轻量化线性瓶颈结构,提出了一个具有多个并列分支结构的空时预测模型,分别提取近期历史数据和周期性历史数据中的空时特征.对于网格化空时数据中的空间依赖性,额外通过K-Means算法对网格高维特征进行聚类,并提取网格基站密度信息作为跨域特征输入到模型中,实现了使用低复杂度、低算力需求模型对研究范围全域流量的精准预测.
Abstract
With the rapid growth of data traffic demand in cellular networks,accurate prediction of cellular traffic conditions at fu-ture moments can improve network resource allocation,achieve traffic load balancing,and deploy base station energy saving and sleeping strategies.Based on the lightweight linear bottleneck module,a spatiotemporal prediction model with multiple parallel branching modules is proposed to extract spatiotemporal features from recent historical data and periodic historical data,respectively.Meanwhile,for the spatial dependency in the gridded spatiotemporal data,high-dimensional grid features are additionally clustered by K-Means algo-rithm,and the grid base station density information is extracted and fed into the model as a cross-domain feature,thus realising accurate prediction of the cellular traffic in the whole area of the study range deploying a low-complexity and low-computing-power-demand model.
关键词
空时流量预测/轻量化模型/卷积神经网络/深度学习/蜂窝网络Key words
spatiotemporal traffic prediction/lightweight model/convolutional neural network/deep learning/cellular network引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据