Simulation-Based Model for Adaptive Traffic Signal Control Using Forecasting and Reinforcement Learning

Abstract

This paper presents a simulation-based model for adaptive traffic signal control that integrates computer vision, LSTM-based forecasting, and reinforcement learning. The proposed approach enables proactive traffic management by incorporating predicted traffic states into the decision-making process of the RL agent. The model is evaluated in a simulated environment using real-time traffic parameters obtained from video-based detection. Experimental results demonstrate significant improvements compared to fixed-time control, including a reduction in average delay (32%), queue length (30%), and waiting time (35%), as well as an increase in traffic throughput (20%). The results confirm that combining prediction and control within a unified framework enhances system adaptability and efficiency. The proposed model satisfies real-time constraints and shows scalability for practical intelligent transportation system applications.