自主网络的代理人工智能协调-中国移动；中兴

China Mobile’s and ZTE’s new “Digital Employee” has significantly reducedthe time it takes to resolve network faults. Executive Summary •To further the development of autonomous networks, China Mobile and ZTE have built the“Co-Innovation+” autonomous network open laboratory and are co-building a “Digital Employee”that can oversee a group of agents to manage network faults and optimise network performance •Drawing on the combination of a knowledge graph, a graph neural network and a large languagemodel, the Digital Employee uses AI deep-reasoning capabilities to pinpoint the most likely rootcause of a fault and hence the solution. •After being assigned the work scope, target KPIs and service level agreement (SLA), the DigitalEmployee prioritises the work order (WO) and identifies a temporary solution based on serviceimpact, and then tracks the resolution. •As well as recycling the knowledge generated by each event, the Digital Employee generates a dailyknowledge summary, encompassing SLA achievements, overtime WO analysis and otherperformance data. •In a pilot in two districts of Beijing, covering 38,000 network elements, the Digital Employeeidentified the root cause of faults with more than 90% accuracy and reduced the numberof on-site visits required for corrective maintenance by 5%, according to China Mobile. •In a trial in Shandong Province, the solution managed a team of six agents, and identified multipleroot causes of faults with more than 90% accuracy, reducing the mean-time-to-diagnose (MTTD)faults from 15 minutes to 3 minutes, while lowering the overall mean-time-to-recovery (MTTR)across multiple vendors and domains by 11%, according to China Mobile. •China Mobile and ZTE are further enhancing the system, which they expect will be deployed at largescale in 31 provinces in China by 2027. Challenge–autonomousnetworksneed to mature Solution - a digitalemployee that can manageagents There is strong interest in the telecoms industry inautonomous networks, which could help operatorsto efficiently meet the growing demand for digitaltransformation across various industries. Whereastraditional network operations can be held back byhigh costs and low efficiency, autonomous networkscould leverage artificial intelligence (AI) to automateand optimise operational processes, significantlyreducing costs, while enhancing networkperformance. For China Mobile, which operates the world’s largestmobile network, greater automation could yieldmajor benefits. As it has to contend with massivenumbers of daily alarms and work orders, theoperator is looking to reduce the time it takes toresolve faults and the associated maintenance costs. To that end, China Mobile is working with ZTE toimprove fault monitoring. The two companies havelaunched an incubation project in Shandong Provincein China to test the concept of a “digital employee”.They are applying AI-based analysis methods tonetwork data, including alarm, logs, complaintsand performance, across equipment from multiplevendors, as well as operational support systems. Thegoal is to build a system that can manage a group ofagents as they analyse vast amounts of network faultdata, thereby reducing the mean-time-to-recovery(MTTR) and workload required to resolve cross-domain faults, especially in complex scenarios. In particular, autonomous networks are set toharness “agentic AI” in which specialist AI agentsautonomously perform specific tasks, such as faultdiagnosis or root cause analysis. But this technologyneeds to evolve further. Today, conventional agentstend to work in isolation with rigid linear workflows,making it difficult for the system as a whole todynamically adapt to changing conditions. Furthermore, existing agents tend to focus on localtechnical metrics, without accountability for overallbusiness outcomes or the user experience. Thiscreates a gap between technical performance andreal-world value delivery. “While individual agentscan learn, the agent team lacks mechanisms toevolve as a whole,” explains China Mobile. “There’sno ‘central brain’ to plan long-term strategies ordevelop new systemic capabilities proactively.” The resulting solution – called the Digital Employee- was developed by taking data about how networkcomponents are connected (network resource —A resource graph that shows how everything inthe network physically links together. —A problem-propagation graph that shows howproblems (alarms) in one part of the networkcan potentially trigger issues elsewhere. Thishelps to understand the connections betweenresources and how alarms propagate. While individual agents can learn,the agent team lacks mechanismstoevolve as a whole, there’s no‘central brain’ to plan long-termstrategies or develop new systemiccapabilitiesproactively China Mobile and ZTE have also developed a newway to search these graphs, combining traditionaltechniques with graph neural networks. When livealarm data is inputted into the mix, the system isdesigned to instan