第四届“泰迪杯”全国数据挖掘挑战赛 www.tipdm.org
The thesis title
Abstract: This paper aims to study the prediction of railway passenger traffic. A new
model for predicting railway passenger flow based on data mining and multidimensional
analysis of passenger data has been proposed. Furthermore, guiding by the established system,
the railway train operation plan has been optimized.
Current passenger data are analyzed by trips, time, stations and section by using Java,
eclipse and mySQL. Firstly, the rule of passenger volume is multi-dimensional evaluated by
passenger load factors of all train number, fluctuations of passenger flow into the section, the
daily flow peak hours of passengers, site rating and traffic flow, analysis of traffic laws. RFM
model was used for optimization of the stations, while the passenger volume for each section
was diagnosed by three special cases, e.g. local passenger volume between ZD111-01and
ZD190-01, local passenger volume between ZD190-01 and ZD111-01 and foreign passenger
volume between ZD111-01 and ZD190-01.
Secondly, prediction model of SARIMAX was set up to forecast the future passenger
flow. Besides, compared to the effect of weather, the influence of holidays was found to be
more obvious, which is modified by using Fourier fit in the model, while the effect of weather
is ignored in the model.
Lastly, multi-stage scheduling decision model was built to optimize the railway train
operation plan. However, in consideration of the complexity of the model solution, the model
was simplified to the 0-1 programming problem of mean distribution of daily load factor trips.
Then, combined with rule of passenger volume, the optimization of the railway train
operation plan was achieved by using greedy algorithm. Particularly, D11 and D18 in ZD250
is cancelled; station ZD062 is added to D16; station ZD190-02 is added to D05, D06, D07
and D16; D05 is changed from 16-grouped to 8-grouped.
Key words: RFM model,SARIMAX model,Fourier fit,0-1programming,greedy algorithm
