ARTIFICIAL INTELLIGENCE IN DRIVEN DIGITAL TWIN FOR REAL-TIME TRAFFIC SIGNAL OPTIMIZATION AND TRANSPORTATION PLANNING

Abstract

This study examined the effectiveness of an artificial intelligence–driven digital twin system for real-time traffic signal optimization and transportation planning using a quantitative simulation-based approach. The digital twin replicated a congested urban corridor consisting of six signalized intersections and evaluated performance under baseline and AI-controlled conditions across 48 experimental scenarios covering peak, off-peak, and incident periods. A total of 192 simulation replications were analyzed to ensure statistical robustness. Descriptive results showed that the AI-based controller reduced average delay from 68.4 seconds per vehicle to 29.5 seconds, representing a 56.9% improvement. Average queue length declined from 19.6 vehicles to 8.4 vehicles, a 57.1% reduction, while throughput increased from 12.7 to 19.8 vehicles per minute, reflecting a 55.9% gain. Travel time reliability improved as the reliability index rose from 74.3% to 91.4%, indicating a more stable operating environment. Correlation analysis showed that the relationship between demand intensity and congestion severity weakened under AI control, with the demand-delay correlation decreasing from 0.74 to 0.48. Reliability and validity checks confirmed strong internal consistency, with simulation replications producing variability below 0.30 units across all key metrics. Criterion validity assessment showed that simulated baseline values differed from field observations by less than 7%. Collinearity diagnostics confirmed model suitability, with all VIF values below 2.3, ensuring stable regression estimates. Regression analysis showed that AI control significantly predicted improvements in delay (β = –0.62, p < .001), queue length (β = –0.58, p < .001), throughput (β = 0.49, p < .001), and reliability (β = 0.57, p < .001). All hypotheses were supported. The study demonstrated that integrating digital twin technology with reinforcement learning produced substantial quantitative benefits, establishing a strong empirical foundation for adopting AI-enhanced signal control in advanced transportation management systems.