【摘要】 当发生重大公共卫生事件时，对传染病感染人数的预测有利于疫情发展趋势的判定和防疫措施的制定。但目前传统的感染人数预测方法在长期预测时准确率不高，为解决这一问题文章提出SEIDR-BiGRU-Informer模型。首先将传染病动力学与循环神经网络模型相结合，构建SEIDR-BiGRU模型，调整模型的训练步长，进行多种步长预测，再将同一日期下不同步长的预测结果交由Informer模型进行整合，得到该日期最终预测结果。实验表明，与传统的LSTM预测模型相比，文章提出的SEIDR-BiGRU-Informer模型能够取得更准确的预测结果，与常用的循环神经网络模型相比，预测准确率提高了13.32%。更多还原

【Abstract】 During major public health events,predicting the number of infections of infectious diseases is beneficial for determining the development trend of the epidemic and formulating epidemic prevention measures.However,traditional methods for predicting the number of infections are not highly accurate in long-term predictions.To address this issue,this article proposes the SEIDR-BiGRU-Informer model.Firstly,the infectious disease dynamics are combined with the recurrent neural network model to construct the SEIDR-BiGRU model.The training steps of the model are adjusted,and multiple-step predictions are made.Subsequently,the predictions of different steps on the same date are integrated by the Informer model to obtain the final prediction for that date.Experimental results show that compared to traditional LSTM prediction models,the proposed SEIDR-BiGRU-Informer model can achieve more accurate predictions.In comparison to commonly used recurrent neural network models,the accuracy of predictions is improved by 13.32%.更多还原