A Method for Predicting Real-Time Carriage-Level Alighting Flow Based on Train Weighing Data by Incorporating Correlations Among Carriages

The dynamics and randomness of boarding and alighting flow usually lead to unbalanced congestion across multiple carriages of one metro train. Train weighing sensors make it possible to acknowledge each carriage’s weights in real-time, but it is still hard to know the alighting passengers at the next station, making it difficult to organize boarding flow in advance to balance each carriage’s congestion. Moreover, the correlations among multiple carriages strengthen the complexity of alighting flow prediction. This study will adopt the convolutional long-short-term memory (ConvLSTM) model based on the multichannel features to predict the carriage alighting flow with the consideration of correlations among carriages. First, the historical carriages’ alighting flow is extracted by monitoring the variations of weights caused by the passengers getting on/off carriages based on the train weighing sensors. Then, the convolution operations are used to extract the correlation between carriages in spatial dimensions. Finally, the LSTM model captures the temporal correlations among carriages and predicts the alighting flow of each carriage. Based on the train weighing data, the model is applied and specified in Guangzhou Metro Line 14 where one metro train consists of six carriages. The results show that the proposed model has a better performance than other deep learning methods and the model that considers the adjacent carriages’ correlations performs best.

I. Introduction

With the construction of a smart metro system and the increasing travel demand, the issue of unbalanced congestion among carriages in one metro train has become a concern in precise metro operation and management. The congestion adversely impacts passenger satisfaction and can increase passengers’ perceived travel time [1], [2], [3], [4]. Predicting passenger flow information for carriages enables precise guidance for passengers to board the carriage and alleviates local congestion during boarding and alighting on the platform, thereby enhancing the travel experience. For metro operators, it enables them to evaluate the utilization of carriage resources effectively and make timely capacity adjustments. The alighting flow from carriages can reflect the remaining capacity of the carriages, making it crucial for guiding passengers to board the appropriate carriage to balance the distribution of passengers across carriages of the train. Moreover, research indicates that the carriages’ alighting flow can help to predict the train dwell time which contributes to metro operations [5]. Therefore, when a train with multiple carriages departs from a station, it is essential to accurately predict the alighting flow of each carriage at the next station in real-time as soon as possible.