基于CIST-GCN的流行病数据分析与预测

Analysis and Prediction of Epidemic Data Based on CIST-GCN

文章提出了一个基于相关度交互图卷积网络的流行病预测方法,利用各城市的日感染人数变化模拟病毒在不同城市间的传播相似度,并对拓扑图进行加权处理,最后利用时空图卷积网络处理城市网络的空间特征,并对城市的流行病发展状况进行预测。方法在PeMS-Bay和PeMSD7数据集上实验的MAPE为2.498%和5.640%,优于传统ST-GCN的2.640%和8.822%,同时在PeMSD7上优于参考模型IT-GCN的8.603%,并且在中国33个城市的疫情预测中与真实数据契合度较高,特别是对“突增点”,对各类流行病的预测以及疫情突发状况的预警起到了一定的参考作用。

This paper proposes an epidemic prediction method based on the convolution network of correlation interactive graph.The daily number change of infected people in each city is used to simulate the transmission similarity of the virus between different cities,and the Topology is weighted.Finally,the spatial characteristics of the city network are processed by the spatiotemporal graph convolution network,and the epidemic development condition of the city is predicted.The MAPE of the method is 2.498% and 5.640% on PeMS-Bay and PeMSD7 datasets,which is better than 2.640% and 8.822% of the traditional ST-GCN.At the same time,it is better than 8.603% of the reference model IT-GCN on PeMSD7.It also has a high fit degree with the real data in the epidemic prediction of 33 cities in China,especially the “sudden increase point”.It plays a certain reference role in the prediction of various epidemics and the early warning of epidemic disease emergency situation.