Crosssection passenger flow forecasting of urban rail transit based onPSOLSSVM algorithm(PDF)
长安大学学报（自然科学版）[ISSN:1006-6977/CN:61-1281/TN]
- Issue:
- 2021年6期
- Page:
- 91-102
- Research Field:
- 交通工程
- Publishing date:
Info
- Title:
- Crosssection passenger flow forecasting of urban rail transit based onPSOLSSVM algorithm
- Author(s):
- LI Yaxiang; WANG Yuanqing
- (College of Transportation Engineering, Changan University, Xian 710064, Shaanxi, China)
- Keywords:
- traffic engineering; urban rail transit; crosssection passenger flow forecasting; particle swarm optimization; least squares support vector machines; mixed kernel function
- PACS:
- -
- DOI:
- -
- Abstract:
- In order to improve the level of service and the utilization efficiency of urban rail transit, in view of the problems such as long waiting time for passengers during the offpeak period, high train loading rate during the peak period, and wasted train capacity etc., the maximum crosssection passenger flow of urban rail transit with the feature of nonlinear and random fluctuation was urgent to forecast accurately and rapidly, which could be applied to adjust the train operation plan in time. Firstly, the fullday and timesharing crosssection passenger flow of Xian Urban Rail Transit Line 2 from April to June in 2017 were collected as the data basis, which were divided into the upward passenger flow and the downward passenger flow on weekdays, weekends and holidays. Secondly, the maximum crosssection passenger flow of each phase in the first three days and the crosssection passenger flow of each phase in the day before were adopted as the input variables of the model, and the intraday maximum crosssection passenger flow were adopted as the output variables of the model. Optimal parameters of kernel functions in least squares support vector machines (LSSVM) model were constructed. Thirdly, the RBF kernel, POLY kernel and Sigmoid kernel were combined into a variety of alternative kernel functions. Taking regularization parameter, width parameter, order parameter and offset parameter as the objects, particle swarm optimization (PSO) was chosen to optimize these alternative functions and key parameters. Finally, the regression evaluation indexes of the prediction effects before and after PSO optimization and regression errors in each phase on weekdays, weekends and holidays were compared in different models. The results show that the values of mean absolute percentage error are lower than 5.0% after that the alternative functions and key parameters in LSSVM are optimized by PSO on weekdays, and the accuracy of passenger flow prediction is improved. Under the condition of faster running speed and higher accuracy, the RBF and Sigmoid mixed kernel has the best fitting effect on the upward crosssection passenger flow on weekdays, and the RBF kernel function has the best fitting effect on the downward crosssection passenger flow on weekdays. PSOLSSVM model has the optimal prediction effect on crosssection passenger flow during the peak hours on weekdays. Compared with the better prediction effect of PSOLSSVM model on weekdays, it has a relatively poor prediction effect on crosssection passenger flow on weekends and upward crosssection passenger flow on holidays. The improved kernel function LSSVM model with PSO can better explain the complex fluctuations of crosssection passenger flow on weekdays, achieve ideal regression accuracy under the small sample, and has the real practical value. 1 tab, 12 figs, 25 refs.
Last Update: 2021-12-14