面向6G通感算融合的网络智能感知

Network Intelligent Perception Oriented to 6G Integration of Communication,Sensing and Computing

面向6G通信-感知-计算(通感算)融合的发展需求,亟需突破网络智能感知方法,特别是基于深度学习的业务识别与流量预测。因此,首先提出了基于卷积神经网络的业务类型估计算法,以避免人工提取数据特征的复杂过程与估计误差,并减少训练模型参数数量;然后,将基于注意力机制的序列到序列(Sequence to Sequence,Seq2Seq)算法用于预测业务流量,以解决信息损失问题,并根据预测场景的差异性,使用不同的预测步长,在保证预测准确性的前提下减少预测时间和计算消耗;最后搭建基于微服务的智能内生融合实验平台,并在此平台上实现了流量预测与业务类型估计,该平台以微服务的形式将各种能力拆分为多个网络功能,并使用人工智能(AI)技术将各种能力进行融合,实现功能模块共享,减少网络功能冗余,赋予网络智能扩展能力。实验结果表明流量预测模型预测误差较小、业务类型估计模型准确率较高。

For the development of 6G communication-perception-computing(synaptic computing)fusion,it is urgent to break through the network intelligent perception methods,especially business identification and traffic prediction based on deep learning.Therefore,an estimation algorithm for business type based on convolution neural network is proposed first to avoid the complex process of extracting data characteristics manually and to reduce the number of training model parameters.Then,Sequence to Sequence(Seq2Seq)algorithm,which is based on attention mechanism,is used to predict business traffic to solve the problem of information loss.Based on the difference of prediction scenarios,different prediction steps are used to reduce prediction time and calculation consumption while ensuring prediction accuracy.Finally,an intelligent endogenous Integration Experimental Platform Based on micro-services is built,and traffic forecasting and business type estimation are implemented on this platform.The platform separates various capabilities into multiple network functions in the form of micro-services,and uses Artificial Intelligence(AI)technology to fuse various capabilities,achieve module sharing,reduce network function redundancy,and give network intelligence expansion capabilities.Experimental results show that the prediction error of traffic prediction model is small and the accuracy of business type estimation model is high.