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

This repository consist of the source code used for producing the results in the paper:

Petersen, N. C., Parslov, A., & Rodrigues, F. (2020). Short-term bus travel time prediction for transfer synchronization with intelligent uncertainty handling. Submitted to Expert System with Applications.

Abstract

This paper presents two novel approaches for uncertainty estimation adapted and extended for the multi-link bus travel time problem. The uncertainty is modeled directly as part of recurrent artificial neural networks, but using two fundamentally different approaches: one based on Deep Quantile Regression (DQR) and the other on Baysian Recurrent Neural Network (BRNN). Both models predict multiple time steps into the future but handle the time-dependent uncertainty estimation differently. We present a sampling technique in order to aggregate quantile estimates for link level travel time to yield the multi-link travel time distribution needed for a vehicle to travel from its current position to a specific downstream stop point or transfer site.

To motivate the relevance of uncertainty-aware models in the domain, we focus on the connection assurance application as a case study: An expert system to determine whether a bus driver should hold and wait for a connecting service, or break the connection and reduce its own delay. Our results show that the DQR-model performs overall best for the 80%, 90% and 95% prediction intervals, both for a 15 minute time horizon into the future (t + 1), but also for the 30 and 45 minutes time horizon (t + 2 and t + 3), with a constant, but very small underestimation of the uncertainty interval (1-4 pp.). But we also show, that the BRNN model still can outperform the DQR for specific cases. Finally, we demonstrate how a simple decision support system can take advantage of our uncertainty-aware travel time models to prioritize the difference in travel time uncertainty for bus holding at strategic points, reducing the introduced delay for the connection assurance application.

Data

The raw data for the 200S and 300S dataset is stored in data/200S.csv.gz and data/300S.csv.gz. We also include pre-processed datasets (raw, descaled and centered) for the 15min regular time series, devided in train/validation/test in 200S_15min_(...).csv.gz and 300S_15min_(...).csv.gz files.

DQR model files

After running the DQR model, result files (e.g. predicted quantiles, drawn samples, model check points) will be stored in the dqr folder.

• dqr-param-search-joint.py: Performs hyper parameter tuning for DQR model.
• dqr-joint-200S.ipynb / dqr-joint-300S.ipynb: Notebooks containing train/predict for the 200S and 300S dataset
• dqr-draw-samples-joint-200S.py / dqr-draw-samples-joint-300S.py: Draw samples using the curve-fitting approach described in the paper for the 200S and 300S test dataset
• dqr-results-joint-300S.ipynb: Creates tables with results presented in the paper

BRNN model files

After running the BRNN model, result files (e.g. drawn samples) will be stored in the brnn folder.

• brnn-param-search-blitz.py Performs hyper parameter tuning for BRNN model.
• brrn-blitz-raw-200S / brnn-blitz-raw-300S: Notebooks containing train/predict for the 200S and 300S dataset. Results are included in the bottom.

Other files

• baseline-kalman.ipynb: Baseline with linear Kalman filter.
• plots-*.ipynb: Create all plots used in the paper.