Kisa Donem Uzam-Zamansal Trafik Tahmini

TL;DR

This paper presents a spatio-temporal traffic speed forecasting method that uses sparse matrices to improve accuracy and reduce computation time by selecting the most informative data points from historical traffic data.

Contribution

It introduces a novel algorithm that leverages sparse matrices for efficient and accurate short-term traffic speed prediction using multi-point historical data.

Findings

Improved forecasting accuracy over existing methods.

Significantly reduced computation times.

Effective selection of informative data points.

Abstract

The studies carried out with the objective of minimizing the effects of congestion, delay and environment problems on the transportation network have gained increasing importance in the last years. Among these studies, short-term traffic flow and average vehicle speed forecasting methods have come into prominence due to their easy implementations, efficient usage on different areas and cost-effectiveness. A large number of studies have reported that these methods, in which the expected future values of link flows and average speeds are forecasted in desired points, can reduce the traffic congestion by anticipating the problems in traffic management. In this paper, a spatio-temporal approach accounted for historical traffic characteristics data collected from a large number of points is presented for average speed forecasts in a given link. The proposed approach includes an algorithm that enables to take into account the most informative data in an input set by determining them for each stage. It is aimed to increase the forecasting accuracy by using sparse matrices in the algorithm while decreasing the calculation times significantly compared to the similar methods presented in the literature.