Scholarlink

<p>Rapid urbanization and increasing traffic density are leading to growing complexity in urban traffic networks, making accurate traffic flow analysis and collision prediction an urgent challenge for modern cities. Traditional traffic management systems, which rely primarily on rule-based heuristics or static models, often fail to capture the nonlinear dynamics and real-time variability of traffic conditions. In this paper, we propose a two-dimensional speed–density model for traffic flow, solved using Physics-Informed Neural Networks (PINNs). The results obtained are compared with those produced by Agrawal’s model. These results are then used to simulate traffic and generate vehicle data using the SUMO traffic simulator. The collected dataset is used to train several machine learning classifiers, including Logistic Regression, Support Vector Machines (SVM), Gradient Boosting, and AdaBoost. Our findings show that Gradient Boosting achieves the highest prediction accuracy (0.97) and demonstrates strong robustness for collision prediction. Finally, we explore the integration of the proposed model with a two-dimensional Intelligent Driver Model (IDM) to enhance collision control mechanisms in traffic simulations. Incorporating IDM significantly improves all performance metrics, including accuracy, precision, recall, F1-score, and AUC-ROC, across all evaluated classifiers.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Physics-Informed Deep Learning and Machine Learning Integration for Collision Prediction in 2D Traffic Systems

  • 期刊论文

  • 发表日期：

    2026年6月15日

查看全文
  • Mourad Haddioui,
  • Youssef Qaraai,
  • El Arbi Abdellaoui Alaoui,
  • Cédric Stéphane Tekouabou Koumetio

摘要

Rapid urbanization and increasing traffic density are leading to growing complexity in urban traffic networks, making accurate traffic flow analysis and collision prediction an urgent challenge for modern cities. Traditional traffic management systems, which rely primarily on rule-based heuristics or static models, often fail to capture the nonlinear dynamics and real-time variability of traffic conditions. In this paper, we propose a two-dimensional speed–density model for traffic flow, solved using Physics-Informed Neural Networks (PINNs). The results obtained are compared with those produced by Agrawal’s model. These results are then used to simulate traffic and generate vehicle data using the SUMO traffic simulator. The collected dataset is used to train several machine learning classifiers, including Logistic Regression, Support Vector Machines (SVM), Gradient Boosting, and AdaBoost. Our findings show that Gradient Boosting achieves the highest prediction accuracy (0.97) and demonstrates strong robustness for collision prediction. Finally, we explore the integration of the proposed model with a two-dimensional Intelligent Driver Model (IDM) to enhance collision control mechanisms in traffic simulations. Incorporating IDM significantly improves all performance metrics, including accuracy, precision, recall, F1-score, and AUC-ROC, across all evaluated classifiers.