Mobile services will be immersive, intelligent, and global in the future. The present mode of operations such as manual awareness, analysis, decision and execution cannot meet the needs of fine-grained SLA guarantee and operations. Operators need new methods to simplify the deployment, operation and maintenance of mobile networks.
In recent years, artificial intelligence for IT operations (AIOps) has achieved high-level automated operations based on big data and AI, which has greatly reduced operating costs. Thousands of devices in mobile networks generate a large amount of data, which can play an important role in network planning, deployment, maintenance, optimization and operation.

Vision of Network Intelligence
New services and scenarios in the 5G-Advanced era, such as holographic interaction and integrated space-terrestrial network, impose higher requirements upon mobile communications. The continuous integration of AI and telecom technology will realize a new type of native intelligent network, providing enterprises and end users with zero wait, zero touch and zero trouble (Zero X) experience. In the future, network intelligence will play a huge role in user experience optimization, efficient operations, and security guarantee.
—User experience optimization: It involves assisted service QoS parameter adjustment, slice access control, user plane path selection and RAT/Frequency selection.
—Efficient operations: The intelligent analysis provides health check, anomaly detection and prediction, and fault root cause analysis, based on which the operations such as capacity optimization, configuration optimization, resource scale-in/out and fault location are performed to reduce repetitive work. The intent-based network is introduced to drive network planning, design and deployment through the intent, thus reducing the requirements for the O&M personnel.
—Security guarantee: The mobility and interaction behaviors of specific types of terminals are regular. User equipment may be maliciously hijacked, resulting in serious network security problems. The behaviors of user equipment can be monitored through big data analysis.

Intelligence Architecture and Key Technologies
To continuously enhance network intelligence and meet the distributed collaboration intelligence and secure and trustworthy intelligence required by IoT, it is necessary to achieve the self-optimizing, self-healing, and self-evolution capabilities with respect to the architecture design.
To achieve the goal of zero touch, zero wait and zero trouble in network operations, the following AI technologies can provide reference for intelligent development of 5G-Advanced network.
Machine Learning is the Basis of Network Intelligence
The traditional network operations mode relies on manual static rules, so it cannot adapt to dynamic scenarios. However, machine learning can make efficient and accurate decisions under complex dynamic changes. The machine learning technology can be introduced for the transition from "expert experience" to "machine learning". 
In a mobile network, data is scattered and may have privacy requirements. Federated learning can help multiple parties in data use and machine learning modeling while meeting the requirements of user privacy protection.
The applications of federated learning also face the challenges of communication efficiency, non-independent identically distribution data, security, and robustness. 
Intent Driven Management Defines User Intent and Helps to Implement Zero-Touch Operation
Intent specifies expectations, including requirements, objectives, and constraints on specific services or network management work flow. The intent network aims at the user's intent or business objective, emphasizing the intent of network operations and architecture personnel (the entire network users).
The application scenarios include network planning and design (such as intent-driven capacity planning, coverage optimization, and site planning), network and service deployment, intent-driven service deployment (such as service deployment at the network edge) and intent-driven network and service maintenance, optimization, and guarantee.
The intent network has problems such as declarative API creation, intent breakdown and translation, and compatibility between components and devices. 
Digital Twin Provides Better Simulation Verification Capability
The digital twin network constructs real-time mirroring of a physical network, which can enhance system simulation, optimization, verification and control ability that the physical network lacks.
Digital twin is applied to the network to create a virtual image of physical network facilities and build a digital twin network platform. Through the real-time interaction between physical network and twin network, low-cost trial and error, intelligent decision-making, and efficient innovation can be achieved.
The construction of digital twin network system is faced with compatibility problems, modeling difficulties, real-time challenges, and scale problems.

Application Scenarios
End-to-End Service Experience Assurance
When user experience is not satisfactory, operators need to adjust network resources in time. They need the service experience evaluation capability to measure user experience. The user experience can be affected by terminals (such as CPU/memory occupancy rate, air interface measurement), networks (such as user-plane bandwidth, user-plane latency, and user-plane congestion), and service servers (such as TCP sliding window at the transport layer, and cache at the application layer). Due to the privacy of user data and the isolated data islands, the data in various domains that affect service experience cannot be trained together. Therefore, federated learning is needed to train or reason the service experience model without losing the original data of terminals, networks and service servers. 
Security Guarantee in the Verticals
In the verticals, data analysis can be used to identify terminal or network exceptions. First of all, the behavior baselines of NEs and signaling access are established based on the predefined and big data AI learning. Secondly, according to the abnormal user behavior characteristics, signaling logs and traffic logs, abnormal signaling and NE attack behaviors are accurately identified, such as abnormal user terminals, base stations and MECs. Finally, the corresponding handling methods, such as de-attachment, de-activation, rate limit, blacklist, and N6 blocking, can be implemented together with PCF/EMS. 

Looking to the Future 
ZTE has cooperated closely with operators in network intelligence in many provinces and cities in China. They have also made positive progress in a national key project, the data-centered DICT deep convergence, promoting the pre-research of network intelligence.