Real-Time Decision Making for Train Carriage Load Prediction via Multi-stream Learning
- Publisher:
- Springer
- Publication Type:
- Conference Proceeding
- Citation:
- AI 2020: Advances in Artificial Intelligence, 2020, 12576 LNAI, pp. 29-41
- Issue Date:
- 2020-01-01
Open Access
Copyright Clearance Process
- Recently Added
- In Progress
- Open Access
This item is open access.
© 2020, Springer Nature Switzerland AG. Real-time traffic planning and scheduling optimization are critical for developing low-cost, reliable, resilient, and efficient transport systems. In this paper, we present a real-time application that uses machine learning techniques to forecast the train carriage load when a train departure from a platform. With the predicted carriage load, crew can efficiently manage passenger flow, improving the efficiency of boarding and alighting, and thereby reducing the time trains spend at stations. We developed the application in collaboration with Sydney Trains. Most data are publicly available on Open Data Hub, which is supported by the Transport for NSW. We investigated the performance of different models, features, and measured their contributions to prediction accuracy. From this we propose a novel learning strategy, called Multi-Stream Learning, which merges streams having similar concept drift patterns to boost the training data size with the aim of achieving lower generalization errors. We have summarized our solutions and hope researchers and industrial users who might be facing similar problems will benefit from our findings.
Please use this identifier to cite or link to this item: