更快的传输网络故障排除-中兴通讯

China Mobile is harnessing AI and digital twins to rapidly resolve faults inits transport infrastructure Executive Summary •In China, rising usage of 5G means mobile operators’ transport networks in cities now generatemillions of alarms each week.•Traditional manual fault diagnosis and resolution processes result in an end-to-end fault repair timeof two-to-three hours.•To shorten the repair cycle, ZTE and China Mobile Zhejiang have integrated large AI modeltechnology, small-model AI algorithms and digital twins into an automated troubleshooting solution.•Deployed by China Mobile Zhejiang in the summer of 2024, the solution has cut the diagnosis timefor typical faults from 30 minutes to under 5 minutes, according to ZTE.•ZTE also reports that the end-to-end resolution cycles for transport network faults has fallento under 100 minutes, with an over 30% efficiency improvement—equivalent to adding over 20experienced “digital employees” to the operations and maintenance team.•The solution’s “algorithm-platform-process” integrated design is designed to be a reusableframework for telecom network intelligence. fault troubleshooting accounts for more than 30%of frontline staff’s workload across all transportnetwork O&M scenarios. As the resolution ofcomplex faults are highly dependent on theexperience of O&M personnel, it can be difficult toensure the timeliness of a fault repair. Challenge–transportnetworksgenerate millionsofalarms As 5G scales commercially, the number of alarmsin transport networks is growing rapidly. In China,mobile operators’ transport networks in cities nowgenerate millions of alarms each week, according toZTE. For faults that require an on-site repair, fieldpersonnel typically communicate with networkmanagement maintenance personnel throughphone calls to discuss issues, such as fault location,repair methods, and confirmation, making the repairefficiency dependent on communication efficiency. Given that fault resolution efficiency directly impactsthe customer experience, operators enforce stricttime constraints for troubleshooting—particularlyfor dedicated line services, where SLA (service levelagreement) tiers mandate resolution within one, two,three or four hours. Enhancing fault troubleshootingefficiency, therefore, remains a key priority intelecom network operations and maintenance(O&M). The solution – integratingAI with a digital twin ofthe network To reduce work orders and shorten MTTR (mean timeto repair), ZTE and China Mobile Zhejiang Branchhave integrated large language model technology,small language model AI algorithms and digital twintechnologies into an automated troubleshootingsolution. The system first processes original alarmsusing high-frequency and engineering filtering rules,then automatically aggregates massive alarms basedon spatio-temporal correlation clustering algorithms. Traditional order dispatch systems rely on fixedrules to correlate alarms and cannot conductcomprehensive analysis, leading to frequent “onefault, multiple orders” issues. “Traditional manual andrule-engine O&M can no longer meet demands for‘high reliability, low cost, and fast response’, creatingan urgent need for AI-powered O&M upgrades,” ZTEnotes. As each manual diagnosis takes over 30 minuteson average, resulting in an end-to-end fault repairtime of two-to-three hours, ZTE estimates that In the next step, the system utilises small-model AIalgorithms to analyse root alarms, achieving a 100%alarm recall rate and a root alarm identificationaccuracy of over 90%, according to ZTE. Theoperation and maintenance centre (OMC) reportsroot alarms and derivative alarms to the faultcentre, providing decision support for the accurategeneration of work orders. Traditional manual and rule-engineO&M can no longer meet demands for‘high reliability, low cost, andfast response’, creating an urgentneed for AI-powered O&M upgrades. During this process, the solution categorises issuesinto three types—configuration-related, device-related, and cross-domain—and optimises each stageof the fault management workflow (identification,diagnosis, and repair) in the following ways: CASE STUDYNOVEMBER 2025 ZTE Solution Integrated SLM/LLMand mobile APP toachieve fault diagnosisand reconstruct theO&M mode —For configuration-related faults (repairablevia the OMC), ZTE says the system enablesfully automated identification, diagnosis andresolution, forming a closed-loop process thatrequires minimal human intervention. The repairagent automatically converts repair suggestionsinto corresponding network configurations,which are then simulated in the digital twinsystem. This simulation evaluates whether thefault will be repaired and whether there willother ripple effects. Based on the evaluationresults, the repair agent then decides whetherto issue the configuration. In the future, for such faults,there may even be no need togenerate work orders, and the OMCcan directly report the detailedfault disposal report for storage.