Research on Dynamic Train Scheduling for Shanghai Metro Based on Enhanced PageRank and Data Fusion

This paper addresses the issues of idle capacity and peak congestion in Shanghai Metro caused by uneven spatiotemporal passenger flow distribution. It proposes a dynamic train scheduling model integrating an improved PageRank algorithm. The study pioneers a three-part data fusion architecture combining "section flow data, timetable data, and transfer station observations." By employing the Expectation-Maximization (EM) algorithm to back-propagate dynamic origin-destination (OD) matrices from publicly available section flow data, it effectively resolves the challenge of obtaining complete passenger flow data. At the algorithmic level, an innovative dual-mechanism PageRank model is designed: a time-sensitive factor (λ=0.7) quantifies passenger flow variation rates, while capacity-adaptive weights (β=1.2) reflect station load states, precisely capturing the dynamic characteristics of station importance. Based on this, a multi-objective spatiotemporal network optimization model is constructed, embedding station importance parameters into decision variable constraints such as headway intervals and train formation quantities. Experimental results demonstrate: Compared to fixed timetables, this model reduces average passenger waiting time during peak hours by 22.4% (measured from 158s to 123s), decreases train empty-run rates during off-peak periods by 31.7%, and achieves a 17.3% improvement in multi-objective Pareto solution sets over the genetic algorithm benchmark. Field validation at People's Square Station (sample size: 300 passengers) demonstrated a waiting time prediction error of only 3.2 seconds (MAPE=2.1%) and an overall reliability of 92.7%, providing an innovative solution for dynamic scheduling in urban rail transit.