PROJECT TITLE :

Online Spatio-temporal Crowd Flow Distribution Prediction for Complex Metro System

ABSTRACT:

Crowd flow prediction (CFP), which is an essential part of contemporary traffic management, contributes to the success of a wide variety of tasks involving intelligent transportation services. However, the majority of the existing methods concentrate solely on predicting the flows of people entering and leaving metro stations, which does not provide sufficient information that can be utilized for traffic management. In real-world applications, managers are eager to find a solution to the issue of obtaining potential passenger distributions in order to assist authorities in improving transportation services. This issue is referred to as a crowd flow distribution (CFD) forecast. As a result, in order to enhance the quality of transportation services, we proposed three spatiotemporal models to effectively address the network-wide CFD prediction problem using an online latent space (OLS) strategy. These models are based on the concept that more information can be found in more places at once. Our models are able to accurately predict CFD as well as entrance and exit flows because they take into account a variety of trending patterns and the influences of the climate, in addition to the inherent similarities that exist among the various stations. The training data for our online systems consists of a series of CFD snapshots taken at different times. It is possible to learn about the latent attribute evolutions of various metro stations from the trend that came before them and then make the following prediction based on the transition patterns. On three large-scale datasets from the real world, all of the empirical results demonstrate that the three models that were developed perform better than any of the other state-of-the-art approaches.