An intelligent model for the optimisation of passenger facilities in transportation stations

Abstract

To cope with rapid population growth in the 21st century, more railway systems are being constructed across the world. Thus, transportation stations play a more important role in railway systems than ever before. A transportation station is characterised by high passenger volume, short train headway and limited capacity. Therefore, the performance of the station is crucial to accommodate daily passenger volume. Currently, spatial designs of transportation stations are carried out according to the experiences of designers and relevant design guidebooks. However, the performance of the stations which were designed by using these methods may not be optimised because actual station performance involves the route choice decisions of passengers, which are highly non-linear in nature. In this research, an artificial neural network (ANN) was applied to mimic the non-linear route choice behaviour of passengers inside transportation stations from both microscopic and macroscopic points of view. The quality and quantity of the available training samples are important to the success of the system modelling by using ANN. Therefore, two surveys were conducted for the microscopic and macroscopic route choice model development. For the microscopic route choice behaviour study, we focused on only a specific area inside a station. The details of the passenger characteristics, including walking speed and density, were collected from video clips. For the macroscopic route choice behaviour study, we analysed the whole station. A 12-month survey of nine transportation stations in Hong Kong during the AM and PM peak hours was conducted to capture the general route choice behaviour of passengers. For both the microscopic and macroscopic route choice behaviour studies, the collected samples were applied to train the ANN models to mimic the route choice behaviour of the passengers inside the stations. The performances of the ANN models trained by using different approaches (e.g., leave-one-out and ensemble) were compared. The ensemble approach was adopted in this research because of its superior performance. The contributions of the input parameters of the trained ANN models to the predicted output were also investigated. The contribution of all the input was similar to that of the ANN models, which may imply that the contributions of the input parameters of the model are similar in both microscopic and macroscopic route choice predictions. One of the limitations of this study is that, by the use of the ANN models for the route choice prediction, extensive resources (i.e., labour and time) are required to collect the data for the ANN models training. In addition, the developed intelligent route choice model may not be applicable in other countries. It is because the culture and environment of the countries are different from Hong Kong. For capturing the route choice behaviour of a country, a new survey should be conducted with the same approach proposed in this study to develop a route choice model for that country. An intelligent passenger facilities design optimisation model was developed based on the macroscopic ANN route choice model with genetic algorithms. The common passenger facilities for ingress movement inside the station are automatic fare collection (AFC) gates, escalators and stairs. This novel intelligent approach is proposed to maximise the total passengers' flow rate of the escalators and/or stairs by optimising the number of AFC gates in each station. It provides an alternative tool for designers to design stations for maximum performance. Finally, an additional site survey was carried out in two stations to verify the performance of the trained ANN models. The results obtained from the trained ANN models are in good agreement with the actual data collected from the two stations. These results confirm the applicability of the ANN models in route choice prediction. Moreover, a case study with 2 situations was conducted to demonstrate and verify the optimisation model for station design. The result shows that the performance of the improved station is higher than that of the original station. To conclude, this research offers not only an effective approach for route choice prediction (i.e., with the developed ANN models) and facilitates the spatial design of transportation stations (i.e., by using the intelligent optimisation model) but also explores a new area of application of ANN.

Date of Award	2 Oct 2013
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Wai Ming LEE (Supervisor)