Function Highlights
Network development starts with planning, including initial network planning and capacity expansion planning. For the 5G network, we can provide intelligent insights to identify the value areas from the dimensions of traffic, user/terminal, key service and scenario while making forecasts on future traffic/user development, identifying explicit weak coverage and exploring hidden problems of high backflow, resulting in accurate planning oriented to the value areas. Planning of a local network can be completed in one week, with efficiency improved by 70% and site quantity saved by 30%, and this solution has already been applied to more than 40 projects in China.
In the early stage of large-scale 5G network construction, automatic site commissioning and automatic drive test are very important for rapid 5G commercialization. The automatic site commissioning through mobile phones has been realized in many projects, and over 2,000 sites have been commissioned in five months in a project in Fujian Province, saving more than 3.2 million yuan. With the help of cloud server + terminal app, ZTE can achieve automatic single-site acceptance. This solution only needs one person, one car and one terminal, and allows automatic navigation and automatic acceptance report output, improving efficiency by 65% and reducing manpower by 68%. 
For the troubleshooting of faults that occupy a large amount of manpower on a daily basis, AI-based learning is introduced to automatically learn fault characteristics and related parameters, counters, and signaling. According to the characteristics of fault scenarios, intelligent drilling analysis is performed to quickly determine the root cause of the fault and enable interconnection with the work order system for closed-loop verification. At present, the fault locating accuracy rate is over 80%, and the time for determining the root cause is reduced from day level to minute level.
One of the outstanding advantages of AI is that it can predict the future trends based on the learning of a large amount of historical data, making predictive O&M possible. For example, the health of optical modules and optical links can be checked actively to identify the hardware aging and failure trends, and early warning will be given to guide proactive inspections and O&M. This is especially applicable to high security scenarios and VIP sites. 
Massive MIMO for 5G uses large-scale array antennas and multi-beam broadcasting. One cell has tens of thousands of possible antenna parameters. In case of multi-cell coordination, the optimization workload increases exponentially. Manual optimization of 2,000 cells requires 160 man-days. In fact, the parameter model can only be fixed according to the expert experience. ZTE's automatic antenna pattern control (AAPC) uses the optimized search algorithm to greatly compress the optimization time. Meanwhile, the antenna parameters can be optimized according to different optimization objectives and scenarios to realize the closed-loop process. The optimization work of 2,000 cells only require three man-days, which improves the efficiency by 10 times, significantly improves the coverage, and increases the user throughput by 10%. The scheme has already been applied to more than 60 projects in China. 
As the network scale and users continue to grow, the workload of network optimization will also increase, and it is necessary to focus on the optimization of the worst TOP N cells according to the Pareto principle. The system will automatically identify the cells with poor KPIs without manual intervention, achieving high efficiency and accuracy. For the identified poor quality cells, the system automatically analyzes the anomalous counter, determines the root cause determined in order and provides corrective suggestions. The analysis period is shortened from 10 hours to about 1 hour. 
While poor KPI cells analysis can be considered as macro-layer network optimization analysis, network fluctuation detection learns and predicts the KPIs of subnet cells with a granularity of 15 minutes. It compares predicted KPIs with the KPIs collected in real time, and automatically finds the counter that has the greatest impact on the KPI fluctuation to determine the root cause. 
An important factor that affects network quality is interference, especially the unpredictability of external interference. It takes a long time to find and locate the interference manually. A interference type feature library can be established based on expert experience and machine learning. After a task is set, the system automatically collects data of the entire network, identifies interference cells, analyzes interference types, and locates interference sources. Taking the interference analysis of 1,000 sites as an example. Compared with the traditional interference analysis that needs 15 hours, the interference analysis function only needs 5 hours and 3 minutes, improving the efficiency by two thirds. 

Future Prospects
There are still many intelligent application scenarios waiting to be explored, and the road to wireless network intelligence cannot be achieved overnight. ZTE has been conducting pilots, expanding the application scenario and scale, and exploring new algorithms and applications. ZTE will continuously improve its intelligent capabilities based on the existing network, and build an intelligent orchestration network integrating baseband, network and service, and in the longer term, an AI-native, intent-driven autonomous network. Technologies like intent-driven operations, edge AI, ICDT integration, baseband intelligence and digital twin simulation can be introduced to realize a high level of intelligence in 5G-Advanced networks to connect the physical and digital world. In the future, by introducing the intelligent plane and data plane, we will construct a AI-native and trustworthy intelligent network architecture to achieve a highly intelligent information network, laying a solid foundation for the digital economy.