AI+ Intelligent O&M: Reshaping the Paradigm

With the accelerating digital transformation in communications and various industries, networks are becoming more complex, service scenarios more diverse, and O&M more challenging. Meanwhile. reducing network O&M costs has become essential, and traditional methods are no longer adequate. Automated and intelligent O&M is widely recognized as the way forward.

The emergence of intent-driven networks, large language models (LLMs), and digital twins opens up new opportunities for intelligent O&M. These technologies offer many advantages, including intuitive user interfaces, and capabilities for large-scale structured data processing, accurate analysis and prediction, and high-fidelity simulation testing. The integration of AI models, intent-driven networks, and digital twins can revolutionize network O&M models and drive operators from the L3 O&M phase to the higher-level L4+ autonomous network phase.

“Four-Layer, One-Entity” Architecture for High-Level Autonomous Core Network O&M

To cope with core network O&M challenges and meet the industry’s growing demand for automated and intelligent O&M, ZTE has built a new intelligent O&M architecture— “four-layer, one-entity”—based on large and small model agents and digital twin technologies (Fig. 1). This architecture aims for full openness and decoupling across networks, data, models, and applications, driving the evolution towards higher-level autonomous networks.

Through the coordinated decoupling and smooth evolution of large, small, and heterogeneous models, this architecture builds a flexible, scalable digital technology foundation to enable efficient O&M. It also uses digital twins that deeply integrate data, models, and applications to provide powerful support for network planning, construction, operation, and optimization, ensuring highly stable networks.

Furthermore, by integrating the O&M interfaces for four centers—monitoring, troubleshooting, emergency response, and network change—this architecture enables high-value closed-loop O&M scenarios such as network change management, monitoring and troubleshooting, and complaint handling, driving a shift in network O&M from traditional passive emergency response to a proactive and efficient model with reduced OPEX.

Bidirectional Integration of AI Agents and Digital Twins Empowering Closed-Loop, Efficient O&M and Highly Stable Networks

Relying on LLMs tailored for the communications field and agent services, ZTE's intelligent core network O&M system, based on the "four-layer, one-entity" architecture, has advanced and developed agentic retrieval-augmented generation (Agentic RAG) technology, large- and small-model collaboration technology (which combines generalized intent understanding with precise O&M), and reinforcement learning-based intelligent document generation technology. These technologies enable high-value O&M scenarios featuring intelligent interaction, analysis, and generation, facilitating the network evolution to L4+ high-level autonomy.

In addition, leveraging digital twin technology, the "four-layer, one-entity" intelligent core network O&M system integrates LLMs, AI agents, and multimodal LLMs (MLLMs) to construct digital twins and deliver digital twin models of core network systems, devices, and components. These models provide multidimensional support for O&M, including visualization, simulation, prediction and analysis, and policy feedback, and enable both qualitative and quantitative analysis at low costs, facilitating the shift from manual to machine-assisted and even machine-independent decision-making. This accelerates the evolution towards high-level autonomous O&M and the realization of fully closed-loop automation, effectively reducing O&M costs.

Fully closed-loop O&M management is a comprehensive approach that spans problem discovery, analysis, solution, and optimization. It not only handles faults immediately but also identifies potential risks in advance through data analysis and prediction, enabling proactive and intelligent O&M. In the core network O&M scenario, fully closed-loop O&M management supports three typical application scenarios during actual production processes: network changes, monitoring & troubleshooting, and complaint handling. Currently, manual steps are involved, such as manual reversal and review as well as manual operations such as work order dispatch and receipt. The core value of the fully closed-loop O&M management mode lies in building a self-learning, self-optimizing O&M ecosystem by integrating monitoring, automation tools, AI algorithms, and other technical means.

Continuous Evolution to Unmanned Autonomous O&M

In the TM Forum, the autonomous network level is clearly defined to guide the automation and intelligence of networks and services, evaluate the value and benefits of autonomous network services, and guide the intelligent upgrade of operators and vendors. To achieve full-process intelligence with the goal of L5, all scenarios must be independently completed by the system, ultimately realizing the ideal "unmanned" intelligent O&M model. Achieving "unmanned" intelligent O&M requires careful considerations of architecture, technologies, and capabilities. The integrated foundation, driven by AI and digital twins, serves as the intelligent engine behind "unmanned" O&M. Adaptive agent collaboration technology based on MLLMs plays a key role in improving the capabilities of "unmanned" O&M models. Multi-dimensional agent collaboration and orchestration technology provides an innovative solution for "unmanned" O&M applications. Only by combining these technologies can we achieve more efficient, secure, and compliant O&M management, building a new O&M model.