Time Series Processing Scheme Based on Deep Learning and Spatio-temporal Network Modeling

In order to study how massive data with periodic and tidal performance influences identification and processing in present network, this paper proposes a data-processing method combined with deep learning algorithm, XGBoost algorithm and spatio-temporal network model. By collecting a large amount of data from existing network and dealing it with pre-treatment, such as anomaly detection and periodicity detection, this proposed data processing method is capable of weeding out erroneous data and fulfill whole data set. In this way, we are capable of getting some characterized results and forming a result set. XGBoost algorithm and space-temporal network model, can then help with data analysis and providing reference for subsequent research such as feasibility analysis and scheme design. This scheme proposed in this paper is able to work in an effective way, especially when facing problems composed of large amount of data with periodicity.