Short-term bus travel time prediction for transfer synchronization with intelligent uncertainty handling

This paper presents two novel approaches for uncertainty estimation adapted and extended the multi-link bus travel time problem. The is modeled directly as part of recurrent artificial neural networks, but using fundamentally different approaches: one based on Deep Quantile Regression other Bayesian network. Both use a network to predict multiple steps into future, handle time-dependent differently. We present sampling technique in order aggregate quantile estimates link level yield distribution needed vehicle from its current position specific downstream stop point or transfer site. To motivate relevance uncertainty-aware models domain, we focus connection protection application case study: An expert system determine whether driver should hold wait connecting service, thus ensuring connection, break reduce own delay. Our results show that proposed method performs overall best 80%, 90% 95% prediction intervals, both 15 min horizon future (t+1), also 30 45 (t+2 t+3), with constant, very small underestimation interval (1–4 pp.). However, show, model still can outperform DQR cases. Lastly, demonstrate how simple decision support take advantage our prioritize difference holding at strategic points, reducing introduced delay application.