We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
A hybrid neural network for urban rail transit short-term flow prediction.
- Authors
Zhu, Caihua; Sun, Xiaoli; Li, Yuran; Wang, Zhenfeng; Li, Yan
- Abstract
Accurate and rapid short-term passenger flow prediction is the foundation for safe and efficient operation of urban rail transit systems. The urban rail transit passenger flow is related to the surrounding land properties and is accompanied by random interference. However, the passenger flow characteristics of stations and the role played by random disturbances in predicting passenger flow signals are not clear. A hybrid neural network named Urban Rail Transit Short-Term Flow Prediction Neural Network (URTSTFPNN) is proposed to improve the accuracy and efficiency of short-term passenger flow prediction. The network consists of three modules: feature Processing Module, Data Reconstruction Module, and Prediction Module. In this process, the urban rail transit stations are classified by dynamic time warping based on the time series attributes of entry and exit passenger flow. The noise reduction technology of wavelet transform is added to the network to increase the accuracy of the model. The analysis results of the proposed model using data from Metro Line 2 in Xi'an, Shaanxi Province, China, indicate that urban rail transit stations can be divided into commercial and official stations, high-density residential stations, low-density residential stations, and tourist and passenger transport terminal stations. The URTSTFPNN shows higher predictive accuracy in revealing the errors compared to the single Long Short-Term Memory model, the Auto-Regression and Moving Average model, and the BP neural network model. The coefficient of determination increased by 1.91 ~ 3.48%, 6.45 ~ 9.45%, and 2.72 ~ 5.69%, with a reduction of calculation time by 19.57 ~ 33.29%, 0.88 ~ 11.61%, and 27.87 ~ 36.71%, respectively. The model proposed by this research can accurately and quickly predict passenger flow, which can be used to guide various categories of urban rail stations to develop effective passenger flow management measures.
- Subjects
ARTIFICIAL neural networks; RAILROAD stations; WAVELET transforms; URBAN transit systems; NOISE control; MOVING average process
- Publication
Journal of Supercomputing, 2024, Vol 80, Issue 16, p24297
- ISSN
0920-8542
- Publication type
Article
- DOI
10.1007/s11227-024-06331-2