Abstract
This article proposes a stochastic differential equation (SDE) to model the temporal evolution of queue length dynamics at signalized intersections. Motivated by the observed quasiperiodicity, long-range dependence, and self-similar characteristics of queue length time series, the proposed model incorporates these three features into the SDE through: (i) a periodic mean-reversion term to capture quasiperiodicity, (ii) a multiplicative noise term, and (iii) a fractional Brownian motion (fBm) to capture self-similarity and long-range dependence. Model parameters are estimated directly from data using a particle-filter–based likelihood framework. Using the extracted parameters, the model is applied to one-step-ahead prediction of queue lengths at four signalized intersections along a traffic corridor in Orlando, Florida. The proposed gray-box approach achieves error metrics that are better than those obtained with an LSTM-based predictor, despite using substantially fewer training samples. In addition to prediction accuracy, the model reproduces key statistical properties of the observed data, including a $1/f$ -type power spectral density and closely matching probability distributions. The results demonstrate that the proposed SDE and particle-filtering framework provides a data-efficient, interpretable approach for modeling and predicting queue dynamics at signalized intersections.
Funding Information
  • Arvind Singh was supported in part by the U.S. National Science Foundation (NSF-EAR-2425747)

This publication has 63 references indexed in Scilit: