Title A Seat Allocation Optimization Model for Taiwan Railway
Year 2018
Degree Master
School Department of Transportation and Logistics Management National Chiao Tung University
Author
Summary The railway is one of the trunk-line public transportation systems in Taiwan with properties of high capacity, reliability, low pollution and suitability for long-distance travel. With lots of invested resource and technical improvement, the capacity of railway systems has been gradually increased. However, the passenger demand in the peak periods continues to grow, without proper seat allocation and control models, it is hard to meet passenger expectation.Based on this, this study aims to develop a dynamically-adjusted seat allocation model towards revenue maximization (i.e., seat utilization maximization) based on a stochastic seat allocation planning model and predicted passenger demands of all OD pairs under demand uncertainty. Firstly, a stochastic seat allocation model is proposed to optimally determine the numbers of seats allocated to specific OD pairs and the number of flexible seats opened to all OD pairs (first come first serve basis). Secondly, a resilient backpropagation neural network (RBNN) demand prediction model is trained by historical ticketing data. Lastly, a dynamically-adjusted seat allocation model under stochastic programming framework is then developed to dynamically adjust the allocated numbers of remaining seats inventory.To show the applicability of the proposed models, a case study on the train #438 of Eastern line of Taiwan Railway Administration (TRA) has been conducted. The RBNN model was trained by one year ticketing data (December, 1, 2016 to December 28, 2017). The training and validation error rates (MAPE) were 24.86% and 25.92%, respectively. As to the results of seat allocation and dynamic seat control, the load factor of the study train can exceed 90% and the revenue is 8.16% higher than the static deterministic model and 5.70% higher than the static stochastic model. The proposed model performs well even if passenger demand drastically varies. In terms of fairness, i.e. the variation of percentages of demand served in all O-D pairs, the proposed model performs also better than the static models.
Hashtags
View count: 213