Predicting the travel time of arterial traffic using particle filter with speed matrix

Qiangrong, Yang; Qi, Peng · 2018 · DOAJ

DOI: 10.1051/matecconf/201818910004

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the challenge of accurate short-term travel time prediction for urban arterial roads, a domain historically under-researched compared to freeways due to the complexity of traffic patterns and lower sensor coverage. Accurate travel time data is critical for intelligent transportation systems, enabling efficient traffic management, dynamic control strategies, and informed route decisions for travelers. To tackle this, the authors propose a novel travel time prediction algorithm based on a particle filter that utilizes historical floating car data and introduces the concept of a "speed matrix" to capture the spatiotemporal properties of arterial traffic. The methodology replaces traditional state-transition functions with a particle-based approach where each particle represents a historical traffic pattern. The core innovation is the speed matrix, which aggregates speed data from the target road segment and its neighboring segments (within a distance of less than 3) across specific time slots. Particle weights are determined by calculating the similarity between the speed matrix of each particle and the current real-time traffic pattern, using linear correlation coefficients with weighted importance assigned to the target road versus neighboring segments. To address the degeneracy problem common in particle filters, the authors implement a partial Sampling Importance Resampling (SIR) process that replaces low-weight particles with historical data exhibiting high similarity to current conditions. The algorithm predicts future travel times by shifting particles along historical data sequences and computing a weighted average of the resulting travel times. The algorithm was evaluated using a real-world dataset from Beijing, comprising GPS trajectories from 10,357 taxis over three months (September to December 2013), totaling 15 million GPS points. The proposed method was compared against two Kalman filter variants and a K-Nearest Neighbors (KNN) algorithm across prediction horizons of 10 to 60 minutes. Results demonstrated that the particle filter approach consistently outperformed the other methods in terms of Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Root Mean Square Error (RMSE). For instance, at a 60-minute horizon, the proposed method achieved a MAPE of 16.52%, significantly lower than the KNN (23.76%) and Kalman filters (33.73%–35.78%). Sensitivity analysis identified optimal parameters: a data sequence length of 4 time slots (40 minutes), 100 particles, and a resampling ratio of 20%. The study concludes that the proposed particle filter with speed matrix effectively models the complex spatiotemporal trends of arterial traffic without relying on macroscopic physical models. While the performance is superior to existing methods, it remains less accurate than freeway predictions due to the inherent complexity of arterial networks. The findings suggest that future research should focus on more precise representations of arterial traffic spatiotemporal properties to further improve prediction accuracy.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• telematics crash prediction