01698299

Component

Adaptive traffic signal control systems are deployed to accommodate real-time traffic conditions. Yet needs and behaviors of the individual vehicles might be overseen by their model-based decision making and aggregated input traffic data. Recently, several pioneering studies employed deep reinforcement learning to come up with model-free control algorithms based on information regarding individual vehicles. However, those studies are limited to isolated intersections and their effectiveness was only evaluated in ideal simulated traffic conditions by hypothetical benchmark signal control algorithms. To fill the gap, this study proposes a decentralized network level adaptive signal control algorithm using one of the famous deep reinforcement learning techniques, double dueling deep Q network. The algorithm was evaluated by the real-world coordinated actuated signals in a simulated suburban traffic corridor in Seminole County, Florida. The evaluation results showed that the algorithm outperforms the benchmark signal control system. It is able to reduce 10.27% of travel time and 46.46% of total delay.

This paper was sponsored by TRB committee AHB25 Standing Committee on Traffic Signal Systems.

19-03479

English

5p

2019

Transportation Research Board 98th Annual Meeting

Digital/other

Figures; Maps; References

Adaptive control; Algorithms; Machine learning; Signalization

Seminole County (Florida)

Data and Information Technology; Highways; Operations and Traffic Management

Transportation Research Board Annual Meeting 2019 Paper #19-03479

TRIS, TRB, ATRI

Dec 7 2018 9:51AM