关键词: 联邦学习, 云边协同, 无线流量预测, JS散度, 合作博弈

Abstract: Abstract: Wireless communication network traffic prediction is of great significance to operators in network construction, base station wireless resource management and user experience improvement. However, the existing centralized algorithm model faces the problems of complexity and timeliness. Therefore, it is difficult to meet the traffic prediction of the whole city scale. A distributed wireless communication traffic prediction framework was proposed to realize traffic prediction based on single grid base station with low complexity and communication overhead. Based on the distributed architecture, a wireless traffic prediction model based on federated learning was proposed. Each grid traffic prediction model was trained synchronously. JS (Jensen-Shannon, JS) divergence was used to select grid flow models with similar flow distributions, and federated average algorithm was used to fuse the parameters of the grid flow models with similar flow distributions, so as to accurately describe the regional traffic. In addition, the flow in different areas within the city is highly differentiated. On the basis of the algorithm, a federation construction method based on a coalitional game was proposed. Combined with super-addivity, the grids were used as participants in the coalitional game, and screened. The core of the coalitional game and the Shapley value were introduced for profit distribution to ensure the stability of the alliance, thereby improving the accuracy of model prediction. Experiments show that compared with grid-independent training, the prediction error of the model decreases most significantly in the suburbs, by 28.7%, at most by 7.8% in the urban area, and at most by 4.7% in the downtown area. Compared with the grid-centralized training, the prediction error of the three regions decreases by 43.8% to 79.1%.

Key words: Federated learning, Cloud-Edge collaboration, Wireless traffic prediction, JS divergence, Coalitional game