OPTIMIZATION OF CONSECUTIVE SIGNALIZED INTERSECTIONS BASED ON COMBINED ALGORITHMS – COMPARING RESULTS WITH MICROSIMULATION

Afandizadeh Zargari, Shahriar; Tajaddini, Atousa; Khalilzadeh, Mohammadreza · 2015 · Crossref

DOI: 10.7708/ijtte.2015.5(4).07

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study addresses the optimization of signal timing for consecutive signalized intersections, a critical challenge in urban traffic management driven by increasing traffic density. Traditional optimization methods often rely on simplified mathematical models or trial-and-error simulation approaches that lack analytical rigor and automation capabilities. To overcome these limitations, the authors propose and compare two combined algorithmic approaches: Genetic Programming (GP) coupled with Genetic Algorithms (GA), and Neural Networks (NN) coupled with GA. The primary objective is to minimize total network delay by using these meta-heuristic methods to predict intersection delays and subsequently optimize signal timing parameters. The methodology involves a case study of two adjacent intersections in Tehran, Iran (Dadman-Farahzadi and Dadman-Darya). Traffic data, including origin-destination matrices and peak-hour volumes, were collected and modeled using Synchro software. The researchers generated 162 different signal timing scenarios by varying cycle lengths (150–200 seconds), green time ratios, and phasing modes. These scenarios served as training data for both the NN and GP models. The NN was implemented in MATLAB using a multilayer perceptron with Levenberg-Marquardt backpropagation, while the GP utilized Multigene Symbolic Regression to evolve mathematical formulas for delay prediction. In both cases, the GA was applied to optimize the predictive models by searching for signal timing variables that minimized the estimated total delay. The optimized timings were then validated using Aimsun microsimulation software. The results indicate that the Neural Network model demonstrated higher efficiency and lower error rates in predicting intersection delays compared to Genetic Programming. Specifically, the NN achieved a mean squared error of 133, whereas the GP model showed higher root mean squared errors during training and testing. However, the study highlights a distinct advantage of GP: its ability to produce explicit mathematical formulas, which enhances interpretability and applicability in certain contexts compared to the "black box" nature of neural networks. The validation via microsimulation confirmed the accuracy of the optimized timings derived from both combined algorithms. The significance of this research lies in demonstrating the effectiveness of combining intelligent learning methods (NN, GP) with evolutionary optimization (GA) for traffic signal control. The findings suggest that while NN-GA combinations offer superior predictive accuracy for delay minimization, GP-GA combinations provide valuable analytical insights through formula generation. This approach offers a robust, automated alternative to traditional manual scheduling, potentially improving traffic flow efficiency in complex urban networks without requiring extensive infrastructure changes.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• gap acceptance