基于时空相关性的卫星网络流量预测

针对当前流量预测模型未考虑时空相关性的问题，影响预测精度，提出一种基于梯度提升回归树的卫星网络流量预测方法。分析卫星网络流量的时空相关性，提取与目标卫星的时空相关流量作为原始的预测输入，并对得到的时空相关流量进行奇异矩阵分解，消除相关输入流量的信息重叠和冗余问题，最后通过改进的梯度提升回归树进行预测。仿真实验表明，该方法有效地提高了卫星网络流量预测的训练速度，预测精度略高。

Satellite Network Traffic Prediction Based on Spatiotemporal Correlation

In view of the current traffic prediction model does not consider the problem of temporal and spatial correlation, which affects the prediction accuracy, a prediction method based on satellite network traffic gradient boost regression tree is proposed. Firstly, the spatiotemporal correlation of the satellite network traffic is analyzed, and the spatiotemporal correlation traffic with the target satellite traffic is extracted as the original prediction input. Then, the singular matrix decomposition is carried out for the spatiotemporal correlation traffic to eliminate the information overlap and redundancy of the relevant input traffic. Finally, the improved gradient boost regression tree is used for prediction. Simulation results show that this method can efficiently improve the training speed of satellite network traffic prediction, and the prediction accuracy is slightly higher.