基于时间序列神经网络的新冠肺炎疫情预测

Prediction of Epidemic Situation in COVID-19 Based on Time Series Neural Network

为揭示我国新冠肺炎(COVID-19)传播规律,建立时间序列神经网络预测模型,对2020年2月12日至4月15日全国、武汉市和北京市的新冠肺炎日累计确诊病例数和死亡病例数进行预测,采用预测值与实际值间的相对误差评估模型的预测性能。结果表明:与其他预测模型相比,时间序列神经网络预测模型的现存累计感染病例数预测值更接近实际值,平均绝对误差和均方根误差均最小,预测精度最高;全国、武汉市和北京市日累计确诊病例数和死亡病例数的预测值与实际值比较吻合,最大相对误差分别为2.0%和2.5%,时间序列神经网络预测模型准确性较高。

To reveal the transmission rule of COVID-19 in China,the time series neural network prediction model was established to predict the daily cumulative confirmed cases and death cases in the whole nation,Wuhan and Beijing from February 12 to April 15,2020 respectively,and the prediction performance of the model was evaluated by the relative error between the predicted value and the actual value.The results show that compared with other prediction models,the predicted value of existing cumulative infection cases of the time series neural network prediction model is closer to the actual value,the mean absolute error and the root mean square error are both the smallest,and the prediction accuracy is the highest.The predicted values,the daily cumulative confirmed cases and death cases in the whole nation,Wuhan and Beijing,are more consistent with the actual values with the maximum relative errors of 2.0%and 2.5%respectively,and the prediction model of time series neural network has high accuracy.