Computer Engineering and Applications ›› 2021, Vol. 57 ›› Issue (9): 168-175.DOI: 10.3778/j.issn.1002-8331.2002-0150

Previous Articles     Next Articles

Spatio-Temporal Fully Connected Convolutional Neural Networks for Citywide Cellular Prediction

HUANG Dongyi, YANG Bing, WU Zihao, KUANG Jiayi, YAN Zeming   

  1. School of Software, Yunnan University, Kunming 650500, China
  • Online:2021-05-01 Published:2021-04-29

用于全市蜂窝流量预测的时空全连接卷积网络

黄冬宜,杨兵,吴子豪,匡佳一,颜泽明   

  1. 云南大学 软件学院,昆明 650500

Abstract:

Accurate prediction of urban cellular traffic flow is very important for the management and public safety of smart cellular network driven by big data in the future,but it is also very challenging. This paper proposes a method based on deep learning—ST-FCCNet to predict cellular traffic in urban areas. A ST-FCCNet unit structure is designed to capture the spatial dependence between any regions in a city. The ST-FCCNet network framework is deployed to model the temporal closeness and periodicity of cellular traffic to capture temporal dependencies. The influence of external factors(time, weather, vacation, etc.) is combined to obtain the final forecast result. In the experimental part, this paper verifies the effectiveness of ST-FCCNet through actual cellular data sets and compared with the existing 4 methods. The results show that the performance of ST-FCCNet is better than all other methods, and the prediction accuracy is improved by 7.50% to 7.76% compared with the start-of-art.

Key words: spatio-temporal date, deep learning, cellular prediction, convolutional neural networks

摘要:

准确地预测城市蜂窝交通流量对未来大数据驱动下的智能蜂窝网络的管理和公共安全非常重要,同时也非常具有挑战。提出了一种基于深度学习的方法——ST-FCCNet来预测城市范围内的蜂窝流量。设计了一种ST-FCCNet单元结构,来捕捉城市中任意区域间的空间依赖。通过部署ST-FCCNet网络框架来对蜂窝流量的时间邻近性和周期性进行建模,以此来捕获时间依赖。结合外部因素(时间、天气、假期等)得到最终的预测结果。实验部分,通过实际的蜂窝数据集验证ST-FCCNet的有效性和现有的4种方法进行了对比。结果表明,ST-FCCNet的性能优于其他所有方法,与最优模型相比在预测精度上提高了7.50%到7.76%。

关键词: 时空数据, 深度学习, 流量预测, 卷积网络