Deep reinforcement learning using neural network configuration for traffic light signal control
- Authors
-
-
Youcef Hassani
University of Sidi Bel Abbes
-
Bilal Tolbi
University of Sidi Bel Abbes
-
Hicham Zatla
University of Sidi Bel Abbes
-
- Abstract
- The traffic jams in Algerian cities often pose major challenges, particularly in terms of total waiting time and user frustration. Deep reinforcement learning (DRL), especially through the adoption of a model-free approach, could eventually help to optimize traffic congestion by making intelligent decisions in order to avoid congested areas. In this paper, a novel intelligent sequential methodology, known as Self-Reaction Deep Q Network (SR_DQN) is introduced, taking account critical states (tram arrival situation, numbers of vehicles, random pedestrians) within the traffic environment and irrelevant features distant from intersection, the real experiments were carried out and the obtained results showed the best adaptability to various traffic patterns and represent a significant breakthrough that could potentially mitigate congestion and improve the efficiency of large urban transportation systems, the method’s performance is justified by measuring some specific metrics like: total waiting time, average queue length, vehicle average speed, and Co2 emission by comparing with the existing experimental method of the real environment and other DRL methods: Deep Q Network (DQN) and gated recurrent unit deep Q networks (GRU-DQN). This research contributes to notably advancing the field of various single-agent intelligent transportation systems by highlighting the importance of DRL in addressing urban traffic challenges.
- References
- Cover Image
-
- Downloads
- Published
- 2026-09-29
- Issue
- Vol. 28 No. 3 (2026)
- Section
- Articles
- License
-
Copyright (c) 2026 Journal of Control Engineering and Applied Informatics
This work is licensed under a Creative Commons Attribution 4.0 International License.
