基于深度神经网络的AI辅助5G自优化研究与实践  

AI-assisted 5G Self-Optimization Based on Deep Neural Networks:Research and Implementation

作　　者：赵川斌 孙红 尹浩 胡贵宾 罗宁[1] ZHAO Chuanbin;SUN Hong;YIN Hao;HU Guibin;LUO Ning(China Telecom Co.,Ltd.,Sichuan Branch,Chengdu 610000,China;China Communications,Beijing 100029,China)

机构地区：[1]中国电信股份有限公司四川分公司,四川成都610000 [2]《中国通信》杂志社,北京100029

出　　处：《移动通信》2025年第3期125-130,共6页Mobile Communications

摘　　要：当前的无线网络多代多频段多模共存,结构和参数日趋复杂,利用人工智能辅助无线通信是当前重要研究方向,在现有网络架构和接口下实现人工智能辅助5G网络运营优化是工程落地难点。将基于机器学习的移动网络参数自优化方法应用于5G网络优化工程实践,通过聚类算法将基站的参数和海量的手机终端测量数据多维关联并分区,运用深度神经网络深度学习建模并通过遗传算法调优,自动输出优化调整方案并实现网络问题的自查找和自优化。经过现网工程试点,解决了应用中数据波动、数据贯通等难题,效果明显,将原有依赖人工为主变成以人工智能自动给出优化方案的人工智能辅助智能自优化模式。该方案是人工智能应用5G网络侧的有效尝试,不仅提升了运营效率,更为人工智能辅助5G网络优化及6G网络AI能力协议演进提供了有益的参考方案。With the coexistence of multiple generations,frequency bands and modes in current wireless networks,their structure and parameters have become increasingly complex.Leveraging artificial intelligence(AI)to assist wireless communication is a key research direction.However,implementing AI-assisted 5G network optimization under existing network architectures and interfaces remains a significant engineering challenge.This paper applies machine learning-based mobile network parameter self-optimization methods to the practical optimization of 5G networks.Using the DBSCAN clustering algorithm,the parameters of base stations and massive mobile terminal measurement data are multidimensionally correlated and partitioned.A Back Propagation deep neural network model is trained,and optimization is fine-tuned using a genetic algorithm to automatically generate optimization adjustment plans,enabling self-diagnosis and self-optimization of network issues.A pilot implementation in an operational network resolved challenges related to dataset construction,data fluctuations,and data integration,leading to a significant improvement.The system transforms the traditional manual process into an AI-assisted intelligent self-optimization mode.This solution represents a promising application of AI in 5G network optimization,enhancing operational efficiency and providing valuable insights for AI-assisted 5G and future 6G network AI capabilities and protocol evolution.

关 键 词：人工智能 深度神经网络 栅格化 多维数据 

分 类 号：TN929.5[电子电信—通信与信息系统]