Accurately predicting bus bunching during the operation phase can pave the way to improving transit service quality. This paper proposes a Long Short-Term Memory (LSTM) neural network model for predicting bus headway at downstream stops based on Automatic Fare Collection (AFC) and Global Positioning System (GPS) data of buses at upstream stops. Then bus bunching can be forecasted through the predicted headway. The LSTM model incorporates an attention mechanism to focus its “memory” function on capturing crucial information that influences headway prediction. The proposed method, validated on a real-world bus route, can accurately identify 89 % of bus bunching events with the lowest prediction error compared to other well-established algorithms.
Headway-Based Bus Bunching Prediction Using LSTM with Attention
28.10.2023
1784988 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Headway-based bus bunching prediction using transit smart card data
Online Contents | 2016
|A headway-based approach to eliminate bus bunching: Systematic analysis and comparisons
Online Contents | 2009
|A headway-based approach to eliminate bus bunching: Systematic analysis and comparisons
Online Contents | 2009
|NTIS | 1965
|