在COVID-19疫情的防控工作中，对疫情传播过程中确诊人数的预测工作具有重要意义。在现有疫情传播预测工作中，以SEIR（Susceptible-Exposed-Infected-Recovered）模型为代表的传染病模型能反映疫情相关人群人数变化，但由于其人群均匀接触的前提假设，模型的应用具有局限性。基于时间序列分析的模型可以通过简单建模历史确诊人数的时间序列对当前确诊人数进行预测，但缺乏对传染病传播的传染性、爆发性、衰减性等固有性质的认识，对疫情发展趋势变化的预测能力受到制约。为解决上述问题，该文采用基于自增强泊松过程（Reinforced Poisson Process, RPP）的模型对疫情确诊人数进行预测，考虑病毒传染性、级联传染的自增强效应和病毒传播的时效性等三个关键因子，对疫情传播的动态过程进行建模，从而对确诊人数做出预测。实验证明，相较SEIR模型，使用RPP模型进行疫情预测不依赖人群均匀混合假设，在各尺度的地理区域都有稳定且准确的预测结果，也解决了SEIR模型在后期预测值过高的问题；对比时间序列分析模型，RPP模型能够掌握疫情发展的内在规律，对疫情发展前、中、后期的发展趋势预测误差分别减小5.29%、5.04%，0.47%，并且能准确把握疫情发展的重要阶段性变化。该文方法已应用于线上平台实时疫情预测，平均误差率小于0.5%。

In the prevention and control of COVID-19, it is of great significance to predict the number of confirmed cases in the transmission of the epidemic. In the current works of epidemic development prediction, epidemic dynamics models, such as SEIR (Susceptible-Exposed-Infected-Recovered) model, are used to estimate the epidemic situation, but the models are limited in utilization because of the homogenous mixing hypothesis. The models based on time series analysis can predict the number of confirmed cases by simply modeling the time series of historical numbers. Unfortunately, these models also have some limitations for lacking the understanding of the inherent nature of the epidemic development, such as infectivity, explosion and attenuation. In this paper, the Reinforced Poisson Process (RPP) model were used for forecasting the number of confirmed cases by modeling three key factors of epidemic development, outbreaks of infectious virus, cascading effect of infection and aging effect of infectiousness. Experiments demonstrate that, compared with traditional SEIR model, the RPP model for epidemic prediction achieves stable and accurate prediction results in geographic areas at multiple scales, without the overestimation problem suffered by SEIR model in the later period of the epidemic development. Compared with the time series analysis method, the RPP model reduces the prediction errors of epidemic trends by 5.29%, 5.04%, and 0.47% respectively in early, middle and later period of the epidemic development, and accurately forecasts the stage changes of epidemic development. The method in this paper has been applied to the real-time epidemic prediction on the online platform, with the average error rate less than 0.5%.