Dealing with uncertainty in railway traffic management and - PowerPoint PPT Presentation

Algorithmic Methods for Optimization in Public Transport Schloss Dagstuhl, April 24-29, 2016 Dealing with uncertainty … in railway traffic management and disruption management April 26, 2016 Dr. Rob M.P. Goverde Department of Transport and Planning Delft University of Technology r.m.p.goverde@tudelft.nl 1 Delft University of Technology Challenge the future

Introduction Uncertainty in railway operations • Traffic management of small disturbances  Errors in traffic prediction and conflict detection  Impact on conflict resolution policy? • Disruption management  Uncertain disruption length  Impact of disruption length to contingency plan  Impact of error in disruption length prediction?  What to tell to the traffic controllers and passengers? 2

Outline Dealing with uncertainty… • Introduction • Railway traffic management  Traffic prediction & conflict detection  Dealing with uncertainty • Disruption management  Prediction of disruption length  Dealing with uncertainty • Conclusions • References 3

Railway traffic management ON-TIME real-time traffic management framework TMS Track Train Real-Time Interlocking Driver Advice Traffic Plan Traffic State Train operations Monitoring Traffic State Real-Time Traffic Plan (RTTP)* Prediction • Successive routes of trains • Train orders over routes Conflict Detection • Route setting times Conflict Resolution *Quaglietta et al. (2016) 4

Traffic prediction & conflict detection Data driven approach* • Acyclic precedence graph based on RTTP and timetable • Nodes: (microscopic) train events • Arcs: precedence relations (run, dwell, transfer, signal headway, …)  Arc weights estimated from historical track occupation data conditional on actual circumstances* *Kecman and Goverde (2015ab) 5

Traffic prediction & conflict detection Prediction at 7:13 Inbound route conflict Rtd Track conflict Schiedam Inbound route conflict Conflict resolution Slow down IC9216 • before Sdm Slow down IC 2127 • before Rtd *Kecman and Goverde (2015b) 6

Traffic prediction & conflict detection Realization (after 8:20) Inbound route conflict Rtd Track conflict Schiedam Inbound route conflict Inbound route conflict Conflict resolution Slow down IC9216 • before Sdm Slow down IC 2127 • before Sdm *Kecman and Goverde (2015b) 7

Traffic prediction & conflict detection Prediction at 7:13 and realization Realized (black) Predicted (colour) *Kecman and Goverde (2015b) 8

Traffic prediction & conflict detection Adaptive prediction *Kecman and Goverde (2015b) 9

Traffic prediction & conflict detection Prediction errors *Kecman and Goverde (2015b) 10

Dealing with uncertainty … in traffic management of disturbances • Adaptive prediction of train paths • Model-based predictive control* *Quaglietta et al. (2013) 11

Disruption management Bathtub model Traffic intensity Failure 1 st phase 2 nd phase 3 rd phase Time Disruption length 1 st phase : Situational awareness, contingency plan, transition 2 nd phase : Operate to contingency plan, prediction disruption length 3 rd phase : Return transition to normal timetable 12

Prediction of disruption length Copula Bayesian Network (BN) of track circuit failure Mean +/- st.dev. Prediction by mean Example 1 July 2014 in Sloterdijk 1) 104 min (initial) • Disruption 51+101=152 min • *Zilko et al. (2016) 13

Prediction of disruption length Conditionalized BN after situational information Prediction by mean Example 1 July 2014 in Sloterdijk 1) 104 min (initial) • Disruption 51+101=152 min • 2) 134 min (siuational) *Zilko et al. (2016) 14

Prediction of disruption length Conditionalized BN after contractor diagnosis Prediction by mean Example 1 July 2014 in Sloterdijk 1) 104 min (initial) • Disruption 51+101=152 min • 2) 134 min (situational) 3) 150 min (post diagnosis) *Zilko et al. (2016) 15

Dealing with uncertainty … in disruption management When prediction gives a wide distribution • What to tell to the traffic controllers and passengers?  Entire distribution  Mean and standard deviation  Mean, median or mode (most likely prediction)  Low percentile (optimistic prediction)  High percentile (pessimistic prediction) • Optimization of traffic control measures during disruption  Scenario analysis • Get more and better data to decrease variance  Improve registration of disruption details  What exact part failed, how was it repaired? 16

Conclusions Dealing with uncertainty… • Traffic management of small disturbances  Errors in traffic prediction and conflict detection  Stochasticty, rare cases, trains or routes without data, …  Adaptive prediction  Model-based predictive control • Disruption management  Uncertain disruption length  Scenario analysis  Pessimistic, most likely, and optimistic predictions 17

References Cited in presentation 1. Quaglietta, E., Corman, F., Goverde, R.M.P. (2013). Stability analysis of railway dispatching plans in a stochastic and dynamic environment. Journal of Rail Transport Planning and Management , 3(4), 137 – 149. 2. Quaglietta, E., Pellegrini, P., Goverde, R.M.P., Albrecht, T., Jaekel, B., Marlière, G., Rodriguez, J., Dollevoet, T., Ambrogio, B., Carcasole, D., Giaroli, M., Nicholson, G. (2016). The ON-TIME real-time railway traffic management framework: A proof-of- concept using a scalable standardised data communication architecture. Transportation Research Part C: Emerging Technologies , 63, 23-50. 3. Kecman, P., Goverde, R.M.P. (2015a). Predictive modelling of running and dwell times in railway traffic. Public Transport , 7(3), 295-319. 4. Kecman, P., Goverde, R.M.P. (2015b). Online data-driven adaptive prediction of train event times. IEEE Transactions on Intelligent Transportation Systems , 16(1), 465-474. 5. Zilko, A.A., Kurowicka, D., Goverde, R.M.P. (2016). Modelling railway disruption lengths with Copula Bayesian Networks. Transportation Research Part C: Emerging Technologies. 18