中国移动通信集团设计院有限公司

In order to predict and reduce the occurrence of complaints of mobile network users in advance, the application of multidimensional data analysis in the prediction of potential complaints of mobile network users was deeply studied. By collecting a wide range of business domain and operation domain indicators of mobile network users as input feature data, a potential complaint user prediction model based on extreme gradient boosting (XGBoost) was successfully constructed, which had an average prediction accuracy of 96.35% on the test set. At the same time, the recursive feature augmented XGBoost (RFA-XGBoost) prediction model was proposed for the prediction of potential complaint users. By iteratively adding the predicted output of the previous round of XGBoost model to the feature set as a new feature and retraining the new round of XGBoost model, the average prediction accuracy after optimization could be improved to 98.89%. For mobile network operators, the research results mean that they can identify and intervene in potential complaints earlier, so as to effectively prevent the occurrence of complaints and further improve user satisfaction and service quality, which has important practical significance and commercial value.

In order to meet the demand for low-carbon development of wireless networks, it is necessary to summarize and prospectively plan the development direction of energy-saving in wireless networks. Firstly, the current status of energy saving in wireless networks were summarized. Secondly, the research on wireless network energy efficiency assessment standards of domestic and foreign standardization organizations were sorted out, and a multi-level, multi-dimensional, multi-scenario, and evolvable wireless networks energy efficiency index system was proposed to optimize the wireless networks energy efficiency evaluation method, and to adapt to the development characteristics of green wireless networks, such as “low energy consumption, high energy efficiency, and excellent perception”. Finally, based on the wireless network energy-saving system, the corresponding energy-saving schemes were summarized and analyzed in the stages of wireless networks planning, construction, operation&maintenance, and optimization. It provides a reference for building a green low-carbon wireless network with both energy efficiency and performance.

To address issues such as low availability of computing power in ultra-large scale computing clusters of intelligent computing centers, low maturity of domestically produced technologies, bottlenecks in large-scale networking efficiency, and complex operations and maintenance, a system based on cloud computing technology for constructing a ten-thousand card cluster in an intelligent computing center was proposed. A ten-thousand card cluster was constructed using 18 432 NPU units and an optimized RDMA network. A multi-plane network architecture was adopted, in conjunction with SDN technology to achieve RDMA network tenant isolation. The network load balancing strategy was optimized, resulting in a link load balancing error of less than 10% and an All-Reduce bandwidth of over 35 GB/s. By employing the optimized distributed storage protocol, the model’s breakpoint recovery time was reduced to half of its original duration. The validation results demonstrate that the domestic NPU ten-thousand card cluster, with the collaborative optimization of software and hardware, can not only meet the training needs of large models with hundreds of billions of parameters but also support the training tasks of large models with trillions of parameters.

Meeting the differentiated deterministic service quality experience of customers in different industries in an economically effective manner is conducive to promoting the construction of a vertical industry deterministic private network. In the face of various segmented industries and different dimensions of deterministic service quality require-ments,the classification and grading of SLA(service level agreement) indicators are very important. Firstly,various types of application scenarios of deterministic service in vertical industries were summarized,and indicator require-ments analysis was conducted for each type of business. Then,a deterministic SLA indicator system was proposed and a hierarchical design scheme for each indicator was summarized to solve the problem of insufficient reflection of deterministic attributes in current network SLA indicators,and help reducing the difficulty of understanding and align-ing indicators between industry customers and service providers,and enable the network to meet the deterministic ser-vice needs of different industries. Finally,prospects for future work were presented.

Facing the demand of 4G/5G users accessing public network services and private network services simulta-neously,the limitations of the existing technical schemes in the industry were reviewed,and a network-side non-inductive diversion scheme was proposed.It only needed to establish a public network IP service channel by the UE.When the network detected that users used private network services,a private network IP service channel for users on the network side was established by SMF/PGW-C to divert the private network services of users.There was only one IP channel in the mobile phone terminal to transmit public network services and private network services,correspond-ing to one public network IP channel and one private network IP channel on the network side,and SMF/PGW-C com-pleted the mapping and splicing of the user-side IP channel and the corresponding network-side public network IP channel and private network IP channel,as well as the replacement of the user's IP address,thus realizing the diver-sion of the user's public network services and private network services.This scheme not only meets the demand of dual-domain private network services,but also allows for the separate management and charging of users'public net-work service IP channels and private network service IP channels by the network side,and has no additional require-ments for terminals.