2732. Scenario-driven Algorithm Selection for Transportation Planning Problem
Invited abstract in session TD-51: Network Design for Public Transport, stream Public Transport Optimization.
Tuesday, 14:30-16:00Room: M5 (building: 101)
Authors (first author is the speaker)
| 1. | Zhang Jiarui
|
| School of Transportation and Traffic, Beijing Jiaotong University | |
| 2. | Xiaojie Luan
|
| Beijing Jiaotong University | |
| 3. | li haiying
|
| Beijing jiaotong university | |
| 4. | Yifan Zhang
|
| ETH Zurich |
Abstract
Countless algorithms have been developed to tackle transportation planning problems. However, there is no one-size-fits-all algorithm that can ensure absolute superiority in all scenarios. To improve solving resilience across various instances and demands, we take Service Network Design Problem (SND) as a starting point. A Convolutional Neural Network-based method is proposed to distill and represent the feature of the instance and demand. Meanwhile, historical solving data are used to train a machine learning module for selecting the best candidate algorithm. This module is embedded into an algorithm framework, employing several algorithms to solve diverse instances under various solving demands. Numerical experiments reveal that the proposed method has superior solving resilience compared to individual and random algorithms utilization.
Keywords
- Algorithms
- Machine Learning
- Transportation
Status: accepted
Back to the list of papers