Pffm-se: a passenger flow forecasting model for urban rail transit based on multimodal fusion of AFC and social media sentiment under special events

Abstract

Conventional methods of rail transit passenger flow forecasting usually use general rail transit data for analysis, such as the spatial structure of the network, the distribution of stations, historical passenger flow, etc. However, these methods tend to focus on forecasting regular passenger flow and are insufficient under special events. With the widespread of social media, special events are often disclosed in advance on social media. The attitudes of citizens towards them become an important factor affecting their travel willingness and mode. Existing models usually ignore people’s sentiment, where people’s sentiment tendencies can influence travel destination choices. Particularly during special events, sentiments expressed on social media can trigger short-term sudden changes in passenger flow, which cannot be effectively achieved using traditional automatic fare collection data alone. Therefore, this paper proposes a deep learning-based forecasting model: passenger flow forecasting model for urban rail transit based on multimodal fusion under special events (PFFM-SE), aimed at improving the accuracy of short-term passenger flow forecasting by incorporating social media sentiment data under special events. PFFM-SE includes a travel sentiment analysis, a point-of-interest association, and an outbound passenger flow forecasting. By integrating long short-term memory networks, variational auto encoders, multi-head cross-attention mechanisms, and convolutional neural networks, this model achieves enhanced forecasting of passenger flows augmented with social media sentiment. The experiments used real-world special events social media sentiment and AFC datasets from two cities in China. The results demonstrate that PFFM-SE outperforms various existing advanced models in passenger flow forecasting under special events.