Applying the CL-transformer model in intercity passenger flow prediction under the impact of travel restriction strategies: a case study of Beijing, China

The scale of intercity mobility fluctuated greatly before and after the end of the COVID-19 pandemic and the adjustment of restrictive entry policies. This study proposes a novel prediction model, CL-Transformer, to predict ridership arriving at transport hubs in Beijing. The model comprises convolutional neural networks (CNN), LSTM and Transformer, considering the number of daily new COVID-19 cases (DNCC) and the dynamic restriction policies of Beijing during different periods, as well as advance ticket sales, weather, time, holidays, flight volume and flight punctuality rate. The impact of urban management policies and the COVID-19 pandemic was quantified and discussed. The results indicate that DNCC and restriction policies have a significant impact on the prediction of ridership. In addition, their relative importance exceeds that of weather conditions, flight punctuality rate, date and holiday, but is lower than that of time, the number of flights and advance ticket sales. Meanwhile, the accuracy of CL-Transformer for all eight hubs exceeds 80%, with approximately 90% accuracy for five major hubs. CL-Transformer’s stability in forecasting across different time periods throughout the day, compared with LSTM, Transformer, and other baseline models, is demonstrated. Relevant management suggestions for external transport hubs in the post-epidemic era were proposed to help city managers grasp the trend of arriving ridership and provide decision-making support for the continuous dispatching of urban transport services.