A reinforcement learning-based dynamic congestion pricing method for the morning commute problems
| Autor: | Toru Seo, Kimihiro Sato, Takashi Fuse |
|---|---|
| Rok vydání: | 2021 |
| Předmět: |
Scheme (programming language)
050210 logistics & transportation Mathematical optimization biology Computer science 05 social sciences 0211 other engineering and technologies 02 engineering and technology Congestion pricing Value of time Bottleneck Zero (linguistics) Traffic congestion Toll 021105 building & construction 0502 economics and business biology.protein Reinforcement learning computer computer.programming_language |
| Zdroj: | Transportation Research Procedia. 52:347-355 |
| ISSN: | 2352-1465 |
| Popis: | A reinforcement learning-based dynamic congestion pricing method for morning commute problems is proposed. In this method, tolls are iteratively updated day by day based on observable information such as traffic volume. The advantage of the method is two fold. First, the method does not require travelers personal preferences such as value of time. Second, the method does not require manually designed toll update scheme; instead, the method automatically find the most efficient toll by a data-driven manner. Results of numerical experiments show that the proposed method properly decreased traffic congestion in various types of networks. Especially, in the experiment with a single bottleneck model, total waiting time decreases to nearly zero in about 200 days. One of future issues is more stable coordination among the time zones. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|