Intent Translation And Self-optimization For Intent-driven Radio Access Networks

Facing the access of massive devices and rapidly changing business needs in the future,the traditional device-centric manual operation and maintenance mode in radio access networks lacks an efficient and agile operating system.To improve the intelligence of radio access networks,decrease the operation and maintenance costs and labor expense,the intent-driven radio access network(ID-RAN)is proposed.Through integrating the artificial intelligence,network function virtualization and other technologies,the ID-RAN can directly convert user intents into network configuration strategies and automatically perform system performance estimation and configuration optimization.In order to effectively support diversified business requirements and continuously guarantee network performance in ID-RANs,the intent translation and self-optimization technologies for ID-RANs are studied in the thesis.The main work and innovations of the thesis include:1.Aiming at the difficulty of translation rigidity caused by the intent with network language description in ID-RANs,the translation method based on natural language expression is proposed in the thesis.Firstly,the wireless intent language model covering the various business requirements and network configuration operations is established,which implements the establishment of the named entity tags and network configuration declaration format required for wireless intent translation;Subsequently,the bidirectional recurrent neural network and sequence to sequence algorithm are utilized to implement the keyword recognition extraction of natural language-based intents and network configuration declaration conversion;Finally,the performance of the proposed translation method and human-computer interaction effects are formulated based on the simulation platform and manually labeled corpus.The simulation results show that the accuracy of the intent recognition and configuration declaration sequence conversion of the proposed method can reach 92%.2.Aiming at the difficulty of high complexity of the performance optimization algorithm caused by the "dimensional explosion" of various network parameters in ID-RANs,the self-optimization method based on machine learning is constructed for ID-RANs in this thesis.Firstly,a method for analyzing network data features based on deep learning is proposed to accurately evaluate the performance status of ID-RANs;Then,the adaptive performance optimization scheme for ID-RANs is proposed based on the deep reinforcement learning,which enables the complexity decrease under the premise of ID-RAN's performance requirements;Finally,based on the proposed optimization method,the use case study of the edge cache placement optimization in ID-RANs is conducted.The simulation results show that the accuracy of the proposed scheme for the cache content popularity can reach more than 90%,and the more traffic offload benefits can be obtained compared with other traditional cache placement schemes.The innovations in the above two aspects provide a theoretical foundation for the efficient work of ID-RANs,and also provide the technical support for ID-RAN's standardization and industrial applications.