EUVIMEDEUVIMED entdecken
Crossref

An Adaptive Simultaneous Perturbation Stochastic Approximation for Dynamic Parameter Optimization in Ride-Sourcing Systems

Ye Tao, Lu Hu, Jun Zhang, Qiujie Dai

IEEE Transactions on Intelligent Transportation Systems · 2026-08 · 27 (8) · 9676-9689

DOI: 10.1109/tits.2026.3685641 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Intelligent Transportation Systems, 2026-08. Verfasst von Ye Tao und 3 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 33 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Sichuan Province Department of Science and Technology.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Dynamic optimization of matching intervals, radii, and relocation thresholds plays a critical role in determining the operational efficiency and service quality of ride-sourcing systems. Given the complex interplay between ride-sourcing operations and traffic congestion, coupled with tight time budgets for system optimization, we adopted a validated simulator that ensures both accuracy and efficiency as a rapid evaluation module. On this basis, we further proposed a novel Adaptive Simultaneous Perturbation Stochastic Approximation (ASPSA) algorithm. The ASPSA algorithm is embedded within a rolling-horizon framework to enable dynamic decision-making. A key innovation of ASPSA lies in integrating a tailored Deep Q-Network (DQN) to adaptively tune the search step sizes, accelerating convergence. More importantly, we incorporate an experience replay buffer and an exponential smoothing technique for gradient estimates, enhancing algorithmic robustness. We benchmark the proposed ASPSA against standard SPSA, two variants of ASPSA, and the state-of-the-art Heteroscedastic and Evolutionary Bayesian Optimisation (HEBO) algorithm on a complex road network of Berlin, Germany. The results demonstrate that ASPSA achieves a 55.3–77.8% improvement in vehicle relocation failure rates and enhances solution stability by 25%–32% relative to the second-best algorithm. Furthermore, across all 30 independent trials on the high-dimensional multimodal Ackley function, ASPSA consistently converged within a very small error margin of the global optimum, whereas competing algorithms get trapped in poor solutions.

Förderung

  • National Natural Science Foundation of China
  • Sichuan Province Department of Science and Technology

Zitieren

APA

Tao, Y., Hu, L., Zhang, J., & Dai, Q. (2026). An Adaptive Simultaneous Perturbation Stochastic Approximation for Dynamic Parameter Optimization in Ride-Sourcing Systems. IEEE Transactions on Intelligent Transportation Systems, 27(8), 9676-9689. https://doi.org/10.1109/tits.2026.3685641

Vancouver

Tao Y, Hu L, Zhang J, Dai Q. An Adaptive Simultaneous Perturbation Stochastic Approximation for Dynamic Parameter Optimization in Ride-Sourcing Systems. IEEE Transactions on Intelligent Transportation Systems. 2026;27(8):9676-9689. doi:10.1109/tits.2026.3685641

BibTeX
@article{Tao2026,
  author = {Tao, Ye and Hu, Lu and Zhang, Jun and Dai, Qiujie},
  title = {An Adaptive Simultaneous Perturbation Stochastic Approximation for Dynamic Parameter Optimization in Ride-Sourcing Systems},
  journal = {IEEE Transactions on Intelligent Transportation Systems},
  year = {2026},
  volume = {27},
  number = {8},
  pages = {9676--9689},
  doi = {10.1109/tits.2026.3685641}
}
Kostenlos starten →