# Dynamic network congestion pricing based on deep reinforcement learning

arXiv: https://arxiv.org/abs/2206.12188 · HTML: https://mlsanity.com/paper/2206.12188/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 5 of its 18 references that are in the map.
- First posted on arXiv 24 Jun 2022.

## Builds on (most cited first, 5 of 5)

- [Continuous control with deep reinforcement learning](https://mlsanity.com/paper/1509.02971/) (2015)
- [Network-wide traffic signal control optimization using a multi-agent deep reinforcement learning](https://mlsanity.com/paper/2104.09936/) (2021)
- [Deep Reinforcement Learning Algorithm for Dynamic Pricing of Express Lanes with Multiple Access Locations](https://mlsanity.com/paper/1909.04760/) (2019)
- [Day-to-Day Dynamic Traffic Assignment with Imperfect Information, Bounded Rationality and Information Sharing](https://mlsanity.com/paper/1811.10729/) (2018)
- [Dynamic optimal congestion pricing in multi-region urban networks by application of a Multi-Layer-Neural network](https://mlsanity.com/paper/2107.14585/) (2021)

## Abstract (arXiv)

Traffic congestion is a serious problem in urban areas. Dynamic congestion pricing is one of the useful schemes to eliminate traffic congestion in strategic scale. However, in the reality, an optimal dynamic congestion pricing is very difficult or impossible to determine theoretically, because road networks are usually large and complicated, and behavior of road users is uncertain. To account for this challenge, this work proposes a dynamic congestion pricing method using deep reinforcement learning (DRL). It is designed to eliminate traffic congestion based on observable data in general large-scale road networks, by leveraging the data-driven nature of deep reinforcement learning. One of the novel elements of the proposed method is the distributed and cooperative learning scheme. Specifically, the DRL is implemented by a spatial-temporally distributed manner, and cooperation among DRL agents is established by novel techniques we call spatially shared reward and temporally switching learning. It enables fast and computationally efficient learning in large-scale networks. The numerical experiments using Sioux Falls Network showed that the proposed method works well thanks to the novel learning scheme.