【摘要】 COVID-19因各国气候、政府政策和疫苗接种人数等因素的不同而呈现不同的发展趋势，这导致COVID-19数据不稳定，传统的机理模型无法根据历史时序数据做出准确预测。因此，提出一种在深度学习LSTM网络框架下引入Self-Attention机制的改进模型。通过仿真实验，对中国、英国和意大利的COVID-19现存病例数据进行预测，并与带有非线性传染率的SIS模型、LSTM模型和ConvLSTM模型的预测结果对比，实验证明，相比于其他三种模型，LSTM-Self-Attention模型的预测精度更高。更多还原

【Abstract】 COVID-19 presents different development trends due to different climate, government policies and vaccination population in different countries, which leads to the instability of COVID-19 data. The traditional mechanism model cannot make accurate prediction based on historical time series data. Therefore, this paper proposes an improved model with self-attention mechanism in the framework of deep learning LSTM network. Through simulation experiments, the existing data of COVID-19 in China, Britain and Italy were predicted, and the prediction results were compared with those of SIS model, LSTM model and ConvLSTM model with nonlinear infection rate. Experiments show that LSTM Self-Attention model has higher prediction accuracy than the other three models.更多还原