Intelligent Intent Policy Refinement Of Space-Terrestrial Network

The space-terrestrial network(STN)merges satellite communication with terrestrial mobile communication,and a three-dimensional networking can be achieved through architectural integration,business integration,management integration,etc.Thereby,it expands network coverage and enhances the utilization of network frequency resources.Compared to the fifth-generation mobile network(5G),the sixth-generation mobile network(6G)STN is designed to cater to the future needs of a variety of industrial and personalized services.It offers a range of business applications,including holographic communication,sensory interconnection,and intelligent interaction,etc.This raises requirements for network operation and maintenance management to be more intelligent,efficient,and precise.The 6G STN will provide diversified business scenarios with differentiated services,which has obvious characteristics such as significant differences of network structure,high dynamism of network topology,and heterogeneity of communication protocols.Meanwhile,the STN facing challenges such as complex forms of intent expression,numerous types of network resources,and dynamically changing network status.It needs to address problems such as low efficiency and error-prone network operation and maintenance,difficulties in coordinating the multiple dimensions of multi-domain network resources,and complex and untimely fault recovery processes.Therefore,it is necessary to not only ensure intelligent and precise translation of user/network intents,but also achieve efficient utilization of multidimensional network resources according to intent demands,while actively ensuring timely fault localization and recovery when network equipment fails.To solve the problem of efficiently matching personalized service demands with multiple network resources in the 6G STN,and to precisely map heterogeneous intents to network logic policies,it is urgent to study intelligent translation methods in the STN.We investigate the intent translation method of user intent in the form of natural language and network intrinsic intents in the STN,in order to translate intents into network logic policies.In this way,it will realize timely unmanned on-orbit automatic translation and policy execution,thereby effectively improving network operation and maintenance management efficiency.The specific research is organized as follows:1.Aiming to address the intent translation problem for diverse operational requirements of multiple user classes in STN,we propose an intent translation method based on nat ural language processing and deterministic finite-state machine.It precisely converts task intent to network intent and further to network policies.Specifically,we construct a specific intent corpus for network operation and maintenance scenarios,and perform sequence labeling on intent entities according to the intent entity recognition model.Then,we utilize annotated data to accomplish intent entity recognition based on ALBERT-BiLSTM-CRF(A Lite BERT-Bidirectional Long Short-Term MemoryConditional Random Field)and extract relevant entities from user intents while evaluating the utility of intent entity recognition.To streamline the interaction between upper-layer applications and lower-layer controllers,we further design an intent representation method.It is based on the mapping relationship between intent entities and network communication metrics,facilitating the flow of intents in the network operation and management process.In addition,it generates network policies based on a deterministic finite-state machine.Finally,simulation experiments demonstrate the intelligent translation performance towards extrinsic intents and validate the necessity of realizing heterogeneous intent translation using natural language processing.2.To tackle the problem of heterogeneous intent representation across various network morphologies and device vendors in STN,we propose an ontology-based intent translation method for network intrinsic intents.It bridges the semantic gap between heterogeneous operational intents.Specifically,we design an ontology-based intent translation process in this paper,which considers both the construction of a unified intent ontology in offline mode and the real-time translation of intrinsic intents in online mode.In this way,it will enhance the compatibility for multi-domain networks.Furthermore,we propose a generic ontology-based intent model,thus network policies will be generated by querying and extending intent ontologies.Additionally,we introduce an ontology-based intent entity alignment model,which resolves conflicts in the intent ontologies caused by identical semantics during the construction of the universal intent translation model.Finally,simulation results show that intent ontology construction and intent entity alignment effectively improve the accuracy of intent translation,achieving transferable and reusable goals for heterogeneous network operational intent translation.The necessity of resolving intent ontology conflicts is also confirmed.3.To address the bottleneck problem of diverse user requirements and on-demand management of multiple resources in the 6G STN,we propose an intent-driven intelligent translation method.It realizes efficient resource management according to extrinsic intent requirements and proactive intent assurance based on network internal intents.Specifically,we propose an intent-driven resource management method for the 6G STN,laying the foundation for cross-domain network autonomy.We further introduce a CoX resource management mechanism.The intent coordination planning module decomposes intents into sub-intents of each network domain,the intent collaboration management module maps intents into the demand for resources of each domain,and the resource cooperation management module selects network nodes to form the optimal path to complete the intents.In the process of realizing user intents,we address the need for network intrinsic intent assurance for virtual network function failures.Based on deep reinforcement learning,the proactive intent assurance algorithm backups virtual network functions on-demand,and enables rapid self-healing of network functions.Finally,simulation results validate the effectiveness of resource management and the efficiency of fault recovery in the STN.The necessity of CoX resource management mechanism and proactive intent assurance is also confirmed.