DSETA: A Dual-Stage Continual Learning Framework for Travel Time Prediction in Dynamic Traffic Environments

2026-07-31

AI Summary

Existing arrival time prediction methods struggle to simultaneously handle sudden traffic congestion and long-term shifts in traffic patterns, leading to degraded accuracy. This work proposes the DSETA framework, which uniquely decouples continual learning into intra-day and inter-day stages: the intra-day stage dynamically responds to short-term disturbances using real-time data, while the inter-day stage captures long-term trends by aggregating recent historical data through a sliding window. To mitigate catastrophic forgetting, DSETA incorporates a historical traffic knowledge consolidation mechanism. Evaluated via real-world A/B tests on DiDi’s platform across Beijing, Wuhan, and Xi’an, the method reduces mean absolute error (MAE) by 6.62%, 0.73%, and 2.40%, respectively. The system has been deployed in production, handling hundreds of millions of requests daily.

Abstract

Estimated Time of Arrival (ETA) prediction is a core component of intelligent transportation systems. As traffic congestion patterns become increasingly dynamic in large cities, maintaining high prediction accuracy poses a major challenge for ride-hailing platforms. Existing methods either fail to adapt to irregular traffic patterns and sudden congestion, or suffer from new distributions without disentangling long-term trends from short-term fluctuations, thereby degrading model performance in real-world scenarios. To address this challenge, we propose DSETA, an incrementally updated Dual-Stage ETA prediction framework. Specifically, the continual learning process is divided into \textit{inter-day} and \textit{intra-day} stages. We first design the \textit{intra-day} learning stage, which relies entirely on real-time data to enable dynamic adaptation to short-term traffic patterns caused by events like holidays or accidents. Next, we develop the \textit{inter-day} learning stage, which leverages aggregated historical data from a short time window to capture knowledge of long-term distribution shifts, such as seasonal trends and traffic network evolution. Subsequently, to prevent catastrophic forgetting and preserve knowledge of regular patterns, we explore a \textit{Historical Traffic Knowledge Consolidation} module. Finally, we validate DSETA's effectiveness and robustness through extensive offline and online experiments conducted on real-world datasets from DiDi's platform. Online A/B tests across three major cities including Beijing, Wuhan, and Xi'an consistently demonstrated performance gains, achieving MAE reductions of 6.62\%, 0.73\%, and 2.40\% respectively. This framework has been successfully deployed in DiDi's production environment, processing hundreds of millions of daily requests and validating its strong performance in industrial applications.

Problem

• ETA prediction
• dynamic traffic environments
• continual learning
• distribution shift
• catastrophic forgetting

Innovation

• Dual-Stage Continual Learning
• Travel Time Prediction
• Dynamic Traffic Adaptation
• Historical Knowledge Consolidation
• ETA Prediction