Abstract: The coronavirus diseases around the end of 2019 (COVID-19) has spreading within open space and the parameters of this dynamic system are time variant. As consequences, existed models are impractical for real-time forecasting of the COVID-19 trend. Therefore, this study provides a real-time trend forecasting model called extreme IR model with the propagation law. Based on SIR model, the propagation of COVID-19 is simplified to three states. Also, the model uses extreme learning machine to generate the parameters of each state. As a result, the proposed model can evaluate time-variant parameter and predict the number of confirmed cases, dead cases and recovered cases during spreading process. The experiment demonstrates that the extreme IR model achieves accurate real-time prediction. Hence, this paper provides an effective approach to analyze the trend of the diseases.