Scholarlink

<p>Traffic congestion results in substantial costs due to time loss, reduced productivity, and economic inefficiency, which underscores the need for accurate forecasting frameworks. This study developed a digital twin (DT) for Tehran integrating real-time traffic data from Mapbox with advanced machine learning models. To validate the reliability of the input data, the Mapbox travel-time estimates were compared with 200 independent GPS and Google Maps samples. The results showed strong agreement (mean absolute error (MAE) = 0.821&#xa0;min, root mean square error (RMSE) = 1.034&#xa0;min, mean bias error (MBE) = 0.367&#xa0;min, <i>r</i> = 0.991, <i>p</i> &lt; 0.001). Three approaches were evaluated: (1) Adaptive Neuro-Fuzzy Inference System (ANFIS), (2) Feedforward Neural Network (FFNN), and (3) a hybrid FFNN + XGBoost model. The test results showed that the ANFIS approach achieved only moderate accuracy (R² = 0.63, RMSE = 0.1000, and MAPE = 9.4%), whereas the FFNN approach performed more strongly (R² = 0.89, RMSE = 0.0649, and MAPE = 7.8%). The hybrid model outperformed both approaches. It achieved an R² of 0.94, an RMSE of 0.0421, a MAE of 0.0256, and a MAPE of 5.0%. Its residuals were narrow and centered, indicating robust generalization across temporal segments. Beyond prediction, the study introduced the Congestion Severity Index (CSI) and a Value of Time (VOT)-based Economic Impact Index (EII) to quantify congestion costs. The results showed that during peak hours, when there were (20–25) minute delays and a + 40% increase in traffic volume, the EII exceeded 400&#xa0;min (approximately USD 83,000 per peak hour), while off-peak periods remained below 50&#xa0;min. Sensitivity analysis showed that while absolute values varied with occupancy and VOT assumptions, peak–off-peak differences remained consistent. The dual-track framework, integrating hybrid AI forecasting with economic impact assessment, thus proves both technically feasible and policy-relevant as a preliminary basis for data-driven traffic management in the studied Tehran corridor, with potential scalability pending further validation.</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

Traffic assessments and economic outcomes: real-time optimization through digital twins and machine learning

  • 期刊论文

  • 发表日期：

    2025年11月25日

查看全文
  • Zahra Rezaei,
  • Hossein Aghamohammadi,
  • Mohammad H. Vahidnia

摘要

Traffic congestion results in substantial costs due to time loss, reduced productivity, and economic inefficiency, which underscores the need for accurate forecasting frameworks. This study developed a digital twin (DT) for Tehran integrating real-time traffic data from Mapbox with advanced machine learning models. To validate the reliability of the input data, the Mapbox travel-time estimates were compared with 200 independent GPS and Google Maps samples. The results showed strong agreement (mean absolute error (MAE) = 0.821 min, root mean square error (RMSE) = 1.034 min, mean bias error (MBE) = 0.367 min, r = 0.991, p < 0.001). Three approaches were evaluated: (1) Adaptive Neuro-Fuzzy Inference System (ANFIS), (2) Feedforward Neural Network (FFNN), and (3) a hybrid FFNN + XGBoost model. The test results showed that the ANFIS approach achieved only moderate accuracy (R² = 0.63, RMSE = 0.1000, and MAPE = 9.4%), whereas the FFNN approach performed more strongly (R² = 0.89, RMSE = 0.0649, and MAPE = 7.8%). The hybrid model outperformed both approaches. It achieved an R² of 0.94, an RMSE of 0.0421, a MAE of 0.0256, and a MAPE of 5.0%. Its residuals were narrow and centered, indicating robust generalization across temporal segments. Beyond prediction, the study introduced the Congestion Severity Index (CSI) and a Value of Time (VOT)-based Economic Impact Index (EII) to quantify congestion costs. The results showed that during peak hours, when there were (20–25) minute delays and a + 40% increase in traffic volume, the EII exceeded 400 min (approximately USD 83,000 per peak hour), while off-peak periods remained below 50 min. Sensitivity analysis showed that while absolute values varied with occupancy and VOT assumptions, peak–off-peak differences remained consistent. The dual-track framework, integrating hybrid AI forecasting with economic impact assessment, thus proves both technically feasible and policy-relevant as a preliminary basis for data-driven traffic management in the studied Tehran corridor, with potential scalability pending further validation.