01476792

Component

Prediction for short-term urban passenger rail flow is essential for effective urban rail transit operation and management. It is important for a forecasting model to capture the periodicity and nonlinearity of short-term passenger flow and to embed these characteristics into the model to enhance forecasting performance. In this research, a support vector machine global online model (SVMGOL) is first proposed by embedding the periodic characteristics via SARIMA model to capture the inherent periodicity of passenger flow. A support vector machine local online model (SVMLOL) is then proposed by embedding the nonlinear characteristics via successive passenger flow value inputs to capture the local nonlinear characteristics of the passenger flow. To take advantage of the two online models, this research then constructs a support vector machine hybrid online model (SVMHOL) based on the idea of data fusion. The model building process and its application in the prediction of short-term passenger flow at Zhujianglu Station of Nanjing Metro is discussed. Testing results show that for the one-step forecasting, the SVMHOL model outperforms the individual SARIMA or SVM model in terms of mean absolute error, mean absolute percent error and root mean square error.

This paper was sponsored by TRB committee ABJ70 Artificial Intelligence and Advanced Computing Applications.

01470560

13-4393

English

16p

2013

Transportation Research Board 92nd Annual Meeting

Digital/other

Figures; References; Tables

Information processing; Mathematical models; Neural networks; Passenger transportation; Rail transit; Time series analysis; Urban transit; Vector analysis

Nanjing (China)

Data and Information Technology; Passenger Transportation; Railroads; I72: Traffic and Transport Planning

Transportation Research Board Annual Meeting 2013 Paper #13-4393

TRIS, TRB, ATRI

Feb 5 2013 12:51PM