中国移动研究院_NGMN：自动化与自智系统架构白皮书_60页_7mb

Summary of the NGMN Alliance Document: Automation and Autonomous System Architecture Framework

Core Content

This document outlines a high-level framework for the development of autonomous systems in the context of network automation, emphasizing the integration of AI/ML models to enable zero-touch automation. It is aimed at guiding NGMN Partners and standards development organizations in creating interoperable specifications for the evolving 5G ecosystem.

The framework is built on the principles of autonomic computing, which includes self-configuration, self-healing, self-optimization, and self-protection (self-CHOP). These capabilities are essential for managing the increasing complexity and heterogeneity of modern networks, especially in IoT, URLLC, and eMBB service domains.

The document highlights the need for network slicing, virtualization, and cloud-native architectures to enable flexible and optimized resource allocation, supporting dynamic adaptability and cognitive automation. It also discusses feedback control loops, intelligent orchestration, and intent-based networking as key enablers for autonomous system behavior.

Main Points

1. Automation Levels

The document defines different levels of automation, ranging from manual operations to fully autonomic systems:

• Category 0: Manual O&M with human intervention
• Category 1: Assisted O&M with scripting-level automation
• Category 2: Partial automation with limited decision-making
• Category 3: Conditional automation within predefined autonomy limits
• Category 4: High-level automation combining categories 2 and 3
• Category 5: Fully autonomic system with end-to-end zero-touch automation

These levels reflect an evolution from open-loop to closed-loop automation, with increasing self-management and cognitive capabilities.

2. Autonomous System Architecture

The reference architecture of the autonomous system is based on autonomic computing principles, with the following key components:

• Knowledge Plane: A shared repository for system knowledge, enabling context awareness and self-CHOP behaviors.
  • Knowledge Management: Involves the discovery, analysis, and creation of system knowledge.
• Management and Orchestration: Enables intelligent orchestration, intent-based networking, and service-based architecture (SBA).
• AI/ML Models: Include supervised learning, unsupervised learning, reinforcement learning, federated learning, transfer learning, and automated ML.
• On-boarding and Certification: Ensures the security, trustworthiness, and interoperability of autonomous systems.

3. System Characteristics

• End-to-End (E2E) Network Slicing: A foundational building block for flexible resource allocation across core, edge, and RAN networks.
• Cross-Domain Cooperation: Enables multi-domain autonomic management and control (AMC), especially in the context of network slicing.
• Security and Privacy: Critical for autonomous systems, especially in zero-touch network and service management (ZTNM).
• Feedback Control Loop: Utilizes MAPE (Monitor, Analyze, Plan, Execute) and OODA (Observe, Orient, Decide, Act) models to enable dynamic adaptation.
• Bearer Plane Programmability: Allows for programmable network resources, facilitating real-time and near real-time control in 5G and future networks.

4. AI/ML in Autonomous Systems

AI/ML models are essential for enabling autonomous decision-making, resource optimization, and service personalization. These models operate in a dynamic environment, responding to wireless link conditions, mobility patterns, and service demands. The use of overlay network information exchange (ONIX) allows for information sharing between virtualized and physical networks, supporting flexible and scalable deployment.

5. Expected Benefits

• Service Innovation: Enables new business models and service scenarios.
• Operational Efficiency: Reduces operation and maintenance (O&M) costs and supports continuous improvements.
• User-Centric Experience: Facilitates personalization, flexibility, and adaptability.
• Commercial Impact: Benefits NSPs, SPs, and users through automated and optimized network operations.

Key Information

• The framework is designed to support heterogeneous access, virtualization, forward-looking service enablers, and emerging usage scenarios.
• Autonomous systems are expected to manage complex, dynamic, and interdependent network environments.
• Network slicing is a critical enabler for resource customization and service-specific optimization.
• AI/ML models are integral to self-CHOP behaviors, allowing autonomous adaptation to system and environmental changes.
• Cloud-native and microservices-based architectures are emphasized for scalability, flexibility, and cognitive capabilities.
• Feedback control loops and intelligent orchestration are key to achieving zero-touch automation.

Conclusion

The NGMN Alliance document provides a comprehensive framework for the development of autonomous systems in the 5G and beyond context. It outlines the evolution of automation, the role of AI/ML models, and the architectural requirements necessary to achieve self-CHOP behaviors and zero-touch network automation. The framework aims to guide industry cooperation, standardization, and innovation to support the next-generation mobile network ecosystem.