Wireless O&M Agent: Shaping the Future of Network Assurance

The communications industry is accelerating its intelligence transformation to meet the challenges of growing network scales and complicated service scenarios. Traditional network O&M, which relies on human experience, suffers from low efficiency, slow responses, and complex operations, making it inadequate for assuring high-value areas such as large shopping malls and major activity venues. By integrating AI with large models, ZTE’s Nebula Telecom Large Model and its derived O&M agent technology have created a new paradigm for network assurance, driving the industry’s intelligent upgrade. For high-value area assurance, ZTE and its partners, in multiple commercial pilots in China, reduced manpower investment by 83% and increased efficiency by five times through its O&M intelligent agent solution, marking a major step toward intelligent network assurance.

Limitations of Traditional Assurance Models

Traditional network assurance models face four key challenges:

• Low O&M efficiency: Slow alignment between network resource configurations with service requirements, hindering quick response to bursty traffic or complicated scenarios.
• High expertise requirements: Assurance strategies rely on expert experience, requiring front-line personnel to master multiple systems and driving up training costs.
• Policy and objective disconnect: Network configurations focus more on troubleshooting, with weak connection with business goals such as user experience optimization.
• Insufficient dynamic response: Frequent policy adjustments are needed in important scenarios, but low automation makes real-time optimization difficult.

ZTE’s Wireless O&M Agent Solution

ZTE’s Unified Management Expert (UME) wireless O&M agent (Fig. 1) is an agent application developed by ZTE for wireless network assurance scenarios based on the large model technology. It supports diverse assurance scenarios, including concerts, sports events, emergency assurance, and tidal traffic during daily O&M operations. Built on the Nebula Telecom Large Model, it adopts a modular design with four core modules:

• Perception module: Collects real-time network, user, and external event data, providing decision inputs through feature extraction and data cleaning.  
• Large model module: Acts as an intelligent brain, analyzing intents and generating assurance policies with natural language processing (NLP) and knowledge graph technologies. Assurance can be triggered via calendar or email, enabling automatic scenario identification.
• Planning & execution module: Invokes network atomic capabilities (such as resource scheduling and parameter adjustment) to perform closed-loop operations based on reinforcement learning optimization policies.
• Feedback learning mechanism: Improves policy accuracy and adaptive capability through continuous data training.

The UME O&M agent integrates generative AI, multi-agent collaboration, and retrieval-augmented generation (RAG) to mitigate hallucinations in large models and ensure reliable decision-making. It offers the following core functions:

• Full scenario coverage: Supports major events (e.g. concerts, sports), provides emergency assurance, and manages daily tidal traffic.
• Intelligent generative interaction: Triggers workflows through natural language instructions (e.g., "provide network assurance for the XX shopping mall during the evening rush hour"), automatically generates and executes policies, boosting operational efficiency fivefold.
• Dynamic optimization and self-adaptation: Monitors network indicators (e.g., user count, PRB usage, and interference levels) in real time and dynamically adjusts policies to deal with traffic fluctuations. In the Hangzhou Olympic Center concert project, network traffic prediction accuracy improved by 20%, greatly enhancing assurance efficiency.

ZTE's O&M agent solution is not only a technological breakthrough but also a reconstruction of O&M models.

• Technology integration and innovation: Deeply integrating large models with telecom knowledge to address traditional AI’s limitations in structured data processing, such as improving fault diagnosis accuracy with the RAG signaling knowledge base.
• Ecosystem synergy: ZTE works with operators and industry partners to build agent technology standard and scenario libraries, accelerating large-scale deployment.
• Business value extension: Beyond cost reduction and efficiency gains, it enables experience-driven operations and differentiated services (e.g., dedicated assurance for live streaming and cloud games) for operators, opening up new revenue opportunities.

Application Pilot and Results

ZTE selected Wushang Dream Times Square, the world's largest shopping mall in Wuhan, as the verification site, covering 31 physical cells and 60 logical cells. The mall averaged over 1,800 daily users, with holiday traffic reaching three times the usual volume. The traditional solution required six man-days, while the agent reduced resource query time from two minutes to 15 seconds and fault location time from 15 minutes to one minute.

• Stable network indicators: User-perceived rate rose 15%, interference dropped 30%, and PRB usage stayed healthy.
• Efficiency breakthrough: Tasks are delivered through natural language instructions, with policies automatically generated and executed, reducing manpower costs by 83%.
• Economic benefits: Saved O&M costs can be reinvested into the network, helping operators explore new business models such as experience-driven operation.

With the evolution of 6G and general computing integration, the O&M agent will be further upgraded to “all-domain Autonomous Networks”, empowering everything from network assurance to business innovation and setting a benchmark for the industry’s intelligent transformation.