HLSR: Hybrid Live Forecast Selective Dynamic Vehicle Rerouting for Real-Time Congestion Avoidance

TL;DR

HLSR integrates real-time edge speeds with short-term forecasts, using dual-threshold congestion detection and driver personalization, significantly improving congestion mitigation efficiency.

cs.AI 🔴 Advanced 2026-08-19 80 views
Xiao Wang Shun Ren Yang Hui Nien Hung
traffic optimization dynamic routing short-term prediction hybrid real-time and forecast selective vehicle rerouting

Key Findings

Methodology

The proposed HLSR framework combines dual-threshold congestion detection, calibrated upstream vehicle selection, and driver-specific travel time prediction. It employs the LSTAN_GERPE model for multi-horizon edge speed forecasting and a hybrid cost function that blends live and predicted speeds based on horizon. The system performs selective vehicle rerouting within limited intervention scopes, utilizing multi-cost route allocation algorithms that incorporate travel-time-weighted k-shortest-path generation. Experimental validation on a SUMO-based Tainan scenario demonstrated that at 8000 vehicles, HLSR reduced average travel time from 438.1 seconds (live-only control) to 380.6 seconds, outperforming full-network Dijkstra (408.1 seconds) while intervening on fewer vehicles. The approach scales effectively to larger vehicle counts, maintaining superior performance.

Key Results

  • At 8000 vehicles, HLSR achieved an average travel time of 380.6 seconds, outperforming both live-only control (438.1 seconds) and network-wide live Dijkstra (408.1 seconds). The intervention scope was limited, involving only a subset of vehicles, which demonstrates high efficiency in congestion alleviation.
  • In larger scenarios with 16000 and 20000 vehicles, HLSR maintained its advantage, with average travel times of 895.7 and 971.7 seconds respectively, showing excellent scalability and robustness.
  • The integration of short-term prediction with hybrid costing mechanisms effectively mitigated the limitations of purely reactive or predictive strategies, leading to more accurate route ranking and better congestion management.

Significance

This work addresses the longstanding challenge of balancing traffic control effectiveness with operational costs. By limiting interventions to congestion-related vehicles and leveraging short-term forecasts, HLSR offers a practical, scalable solution for real-time urban traffic management. Its hybrid approach enhances the responsiveness and accuracy of routing decisions, reducing travel delays, emissions, and fuel consumption. The methodology paves the way for deploying intelligent, personalized traffic control systems in large cities, contributing to smarter, greener urban mobility. The research also demonstrates how combining data-driven prediction models with traditional shortest-path algorithms can yield significant improvements over existing methods.

Technical Contribution

The key technical innovation lies in the hybridization of live edge speeds with short-term forecasts within a selective rerouting framework, supported by dual-threshold congestion detection and calibrated upstream vehicle selection. The development of the travel-time-weighted k-shortest-path generator and a horizon-dependent hybrid cost function enables more accurate route ranking and efficient intervention. The integration of a network-level spatio-temporal forecaster (LSTAN_GERPE) with driver-specific personalization modules enhances prediction accuracy and adaptability. Additionally, the proposed ranking-aware fine-tuning procedure ensures that route orderings remain consistent with predicted travel times, improving the overall effectiveness of the rerouting process. These advancements collectively push the state-of-the-art in scalable, intelligent traffic management.

Novelty

This study is the first to combine dual-threshold congestion detection, calibrated upstream vehicle selection, and short-term edge speed forecasting into a unified, selective rerouting framework. Unlike prior work that relies solely on real-time data or long-term predictions, HLSR dynamically balances between immediate sensing and short-horizon forecasts, optimizing intervention scope and effectiveness. Its multi-objective, multi-cost route generation and driver personalization introduce a new level of granularity and adaptability in traffic control, setting a novel benchmark for scalable, intelligent rerouting systems.

Limitations

  • The approach heavily depends on high-quality, real-time sensing and short-term prediction accuracy. Any sensor failures or forecast errors could impair the system’s effectiveness.
  • In extreme scenarios such as accidents or sudden road closures, the dual-threshold detection might not promptly identify congestion, leading to suboptimal rerouting.
  • Computational complexity remains a concern, especially for large-scale networks, necessitating further optimization for real-time deployment.

Future Work

Future research will focus on enhancing the robustness of congestion detection under uncertain conditions, integrating multi-source heterogeneous data for improved prediction accuracy, and developing more computationally efficient algorithms. Additionally, expanding the framework to multi-agent, multi-modal transportation systems and exploring reinforcement learning-based adaptive control strategies could further improve scalability and resilience. Real-world pilot deployments are also planned to validate the system’s practical effectiveness and facilitate adaptive learning in dynamic urban environments.

AI Executive Summary

Urban traffic congestion continues to challenge cities worldwide, impeding productivity, increasing emissions, and elevating travel costs. Traditional navigation systems like Google Maps and TomTom rely on static or infrastructure-based traffic data, which are insufficient for real-time, network-wide congestion management. These systems often fail to replan routes for vehicles already en route, leading to suboptimal traffic flow and bottlenecks.

Recent advances in intelligent transportation systems have introduced connected sensing, edge computing, and data-driven control strategies to address these issues. However, existing rerouting approaches typically require comprehensive network interventions, which are costly and impractical at scale. The need for a more efficient, targeted, and adaptive solution has driven research into selective vehicle rerouting mechanisms.

This paper introduces HLSR (Hybrid Live–Forecast Selective Rerouting), a novel framework that combines real-time edge speed sensing with short-term travel time forecasts. The system employs dual-threshold congestion detection to identify bottlenecks proactively, calibrated upstream vehicle selection to limit intervention scope, and driver-specific travel time prediction modules for personalized routing. The core of HLSR is a hybrid cost function that blends live and forecasted speeds, dynamically adjusting based on the prediction horizon. This approach enables the generation of multiple candidate routes using a travel-time-weighted k-shortest-path algorithm, facilitating multi-objective route allocation.

Experimental validation on a SUMO-based simulation of the city of Tainan demonstrated that HLSR significantly outperforms traditional methods. At 8000 vehicles, it reduced average travel time from 438.1 seconds (live-only control) to 380.6 seconds, while intervening on a smaller subset of vehicles. The system maintained superior performance at larger scales, with 16000 and 20000 vehicles, achieving average travel times of 895.7 and 971.7 seconds respectively. These results highlight the effectiveness of combining short-term prediction with selective, targeted rerouting, offering a scalable solution for real-world urban traffic management.

The significance of this work lies in its ability to balance congestion alleviation with operational efficiency. By limiting interventions to congestion-related vehicles and leveraging short-term forecasts, HLSR reduces unnecessary rerouting, minimizes driver inconvenience, and lowers system costs. Its innovative integration of multiple algorithms and data sources sets a new benchmark for intelligent transportation control.

Looking ahead, future research will focus on improving the robustness of congestion detection, integrating multi-source data, and optimizing computational efficiency for real-time deployment. The potential for broader application in multi-modal, multi-agent transportation systems promises to transform urban mobility into a smarter, greener, and more resilient ecosystem.

Deep Dive

Abstract

Urban traffic congestion reduces productivity and increases travel cost and emissions. Network-wide live travel-time shortest-path rerouting can be highly effective in simulation, but assumes that essentially every on-road vehicle is replanned every decision period. We propose HLSR, a selective hybrid live--forecast vehicle rerouting framework that fuses live edge speeds with short-horizon forecasts under limited intervention scope. Building on dual-threshold congestion detection, calibrated upstream selection, and driver-tailored travel-time prediction, HLSR further introduces approaching-vehicle expansion, travel-time-weighted k-shortest-path generation, and a horizon-dependent hybrid live--forecast segment speed used in multi-cost route allocation.

cs.AI

References (17)

Short-Term Traffic Prediction for Edge Computing-Enhanced Autonomous and Connected Cars

Shun-Ren Yang, Yuxuan Su, Yao-Yuan Chang et al.

2019 48 citations ⭐ Influential

SUMO - Simulation of Urban MObility An Overview

M. Behrisch, Laura Bieker, J. Erdmann et al.

2011 1976 citations ⭐ Influential

Lightweight Spatio-Temporal Attention Network with Graph Embedding and Rotational Position Encoding for Traffic Forecasting

Xiao Wang, Shun Yang

2025 2 citations ⭐ Influential View Analysis →

Recent Development and Applications of SUMO - Simulation of Urban MObility

Daniel Krajzewicz, J. Erdmann, M. Behrisch et al.

2012 2358 citations ⭐ Influential

Short paper: Distributed vehicular traffic congestion detection algorithm for urban environments

Milos Milojevic, V. Rakocevic

2013 23 citations

Google Maps

Josh Harris

2014 1180 citations

A Unified Framework for Vehicle Rerouting and Traffic Light Control to Reduce Traffic Congestion

Zhiguang Cao, Siwei Jiang, Jie Zhang et al.

2017 172 citations

Next Road Rerouting: A Multiagent System for Mitigating Unexpected Urban Traffic Congestion

Shen Wang, S. Djahel, Zonghua Zhang et al.

2016 127 citations

Finding the K Shortest Loopless Paths in a Network

J. Yen, Jin Y. YENt

2007 1709 citations

Proactive Vehicle Re-routing Strategies for Congestion Avoidance

S. Pan, Mohammad A. Khan, I. S. Popa et al.

2012 51 citations

Dynamic urban traffic rerouting with fog‐cloud reinforcement learning

Runjia Du, Sikai Chen, Jiqian Dong et al.

2023 27 citations

Scorpion: A Solution Using Cooperative Rerouting to Prevent Congestion and Improve Traffic Condition

A. M. Souza, R. S. Yokoyama, L. C. Botega et al.

2015 42 citations

Increased Traffic Flow Through Node-Based Bottleneck Prediction and V2X Communication

Christian Backfrieder, G. Ostermayer, C. Mecklenbräuker

2017 79 citations

ReFOCUS+: Multi-Layers Real-Time Intelligent Route Guidance System With Congestion Detection and Avoidance

Mahboobe Rezaei, Hamed Noori, M. M. Razlighi et al.

2019 47 citations

Information Theory and an Extension of the Maximum Likelihood Principle

Hirotogu Akaike

1973 23413 citations

ReFOCUS: A hybrid fog-cloud based intelligent traffic re-routing system

Mahboobe Rezaei, Hamed Noori, Dadmehr Rahbari et al.

2017 18 citations

UAV-assisted traffic rerouting in disaster scenarios via grammar-guided genetic programming: Effects of limited smart-vehicle adoption and cross-segment generalization

Damian Wlodarczyk, Takfarinas Saber

2026 1 citations