Скачать или смотреть A Predictive Time-Dependent Routing Model forImproving Urban Navigation Beyond Google Maps

Real-time navigation applications typically compute the
shortest path based on current travel times. Although effective
for immediate recommendations, this approach ignores the fact
that traffic conditions fluctuate during the trip. Rush hours,
accidents, and recurrent congestion patterns often degrade the
initial estimate. As a result, a route estimated as 20 minutes
at departure may evolve into a 40-minute trip, while an
alternative 30-minute route with more stable speeds could have
been preferable.
This paper addresses this limitation by incorporating the
temporal dimension into routing decisions. Rather than optimizing only the instant of departure, the system considers how
the travel time of each road segment evolves throughout the
expected traversal interval.
Our contribution is a hybrid approach combining:
• Time-dependent shortest path theory, where edge weights
vary over time.
• Traffic forecasting models, particularly graph-based deep
learning architectures.
• Stochastic reliability concepts to evaluate route stability.
This approach seeks to produce robust future-aware routes
that outperform classical real-time estimators such as Google
Maps.