predicting for the adaptive transport system …and other ...€¦ · transport system …and other...
TRANSCRIPT
![Page 1: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/1.jpg)
Predicting for the adaptive transport system
…and other necessary ingredients for resilient urban mobility
Filipe Rodrigues / Francisco [email protected] / [email protected]
![Page 2: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/2.jpg)
Some “future mobility” visions…Dynamic
Public TransportMobility as
a Service (MaaS)
Shared modes
Dynamicpricing
Autonomous Mobility on Demand (AMoD)
![Page 3: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/3.jpg)
Some “future mobility” visions…
![Page 4: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/4.jpg)
And some possible nightmares...
![Page 5: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/5.jpg)
And some possible nightmares...
![Page 6: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/6.jpg)
The prediction-optimization pipeline
![Page 7: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/7.jpg)
Wrong balancingWrong pricing
In non-recurrent conditions…
![Page 8: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/8.jpg)
Wrong routingWrong scheduling
In non-recurrent conditions…
![Page 9: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/9.jpg)
Transport services of the (near!) future
Attention to stress scenarios!!Large events
IncidentsSystem breakdowns
…
![Page 10: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/10.jpg)
No “future mobility” will be adaptive without VERY robust prediction capability…
…especially in non-recurrent conditions!
![Page 11: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/11.jpg)
Focus of this talk
![Page 12: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/12.jpg)
Non-recurrent scenarios
Demand Supply
Expectable
Unexpectable
![Page 13: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/13.jpg)
Non-recurrent scenarios
Demand Supply
special events, demonstrations,
holidays Expectable
Unexpectable
![Page 14: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/14.jpg)
Non-recurrent scenarios
Demand Supply
special events, demonstrations,
holidays
road works, closures, mega
events Expectable
Unexpectable
![Page 15: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/15.jpg)
Non-recurrent scenarios
Demand Supply
special events, demonstrations,
holidays
road works, closures, mega
events Expectable
crisis situations, unknown high
fluctuations Unexpectable
![Page 16: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/16.jpg)
Non-recurrent scenarios
Demand Supply
special events, demonstrations,
holidays
road works, closures, mega
events Expectable
crisis situations, unknown high
fluctuations
incidents, weather, crisis situations
Unexpectable
![Page 17: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/17.jpg)
Non-recurrent scenarios
Demand Supply
special events, demonstrations,
holidays
road works, closures, mega
events Expectable
crisis situations, unknown high
fluctuations
incidents, weather, crisis situations
Unexpectable
![Page 18: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/18.jpg)
Expectable demand
![Page 19: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/19.jpg)
Event information is usually online
● Event homepages● Event listings● Social media (e.g. twitter, facebook)● News feeds
…
Lots of potential sources, but a lot in free form text
![Page 20: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/20.jpg)
natural language processing Machine-
interpretable features
Accounting for event information
![Page 21: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/21.jpg)
Accounting for event information
Topic modeling
Data p
repara
tion
Bayesian additive model (BAM)
Demand prediction
Rodrigues, F., Borysov, S., Ribeiro, B. and Pereira, F., 2016. A Bayesian additive model for understanding public transport usage in special events. IEEE transactions on pattern analysis and machine intelligence
![Page 22: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/22.jpg)
Bayesian additive model
Gaussian process that models routine component
Gaussian process that models the effect of events
![Page 23: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/23.jpg)
Public transport arrivals in Singapore
Data:
● 5 months of smartcard data (bus, metro, light rail)● 2 study areas
○ Singapore Indoor Stadium○ Singapore Expo
![Page 24: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/24.jpg)
large event
Observed arrivalsModel WITHOUT events informationModel WITH events information
An example
![Page 25: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/25.jpg)
Public transport arrivals in Singapore
RAE: relative absolute errorCorrCoef: correlation coefficientR2: coefficient of determination
![Page 26: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/26.jpg)
Exploiting the model’s additive structure
![Page 27: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/27.jpg)
Predicting taxi demand in event areas
Goal: Given the taxi demand in the last L days, predict taxi demand for the next day
Focus: special event areas
![Page 28: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/28.jpg)
A deep learning approach
Recent work currently under review!
![Page 29: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/29.jpg)
Predicting taxi demand in NYC
Data:
● All taxi trips from 2013 to 2016● 2 study areas
○ Barclays Center○ Terminal 5
![Page 30: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/30.jpg)
Predicting taxi demand in NYC
L = time-series lags onlyW = weatherE = event informationT = event text
MAE: mean absolute errorRMSE: root mean sq. errorMAPE: mean absolute percentage error
![Page 31: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/31.jpg)
Non-recurrent scenarios
Demand Supply
special events, demonstrations,
holidays
road works, closures, mega
events Expectable
crisis situations, unknown high
fluctuations
incidents, weather, crisis situations
Unexpectable
![Page 32: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/32.jpg)
Unexpectable demand
![Page 33: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/33.jpg)
Step 1: anomaly detection
Markou, I., Rodrigues, F. and Pereira, F.C., 2017. Use of Taxi-Trip Data in Analysis of Demand Patterns for Detection and Explanation of Anomalies. Transportation Research Record: Journal of the Transportation Research Board
![Page 34: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/34.jpg)
Step 2: propagating information
Correlated models
![Page 35: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/35.jpg)
Preliminary results
Percentage of improvement
![Page 36: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/36.jpg)
Non-recurrent scenarios
Demand Supply
special events, demonstrations,
holidays
road works, closures, mega
events Expectable
crisis situations, unknown high
fluctuations
incidents, weather, crisis situations
Unexpectable
![Page 37: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/37.jpg)
Unexpectable supply
![Page 38: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/38.jpg)
Step 1: anomaly detection
![Page 39: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/39.jpg)
Running application in Denmark
Extraordinary queueing with Inrix data
![Page 40: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/40.jpg)
Step 2: robust prediction
![Page 41: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/41.jpg)
Step 2: robust prediction
![Page 42: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/42.jpg)
Accounting for uncertainty
Why should we care?
● We use prediction models to make decisions!
![Page 43: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/43.jpg)
Accounting for uncertainty
Heteroscedastic models
● Observation noise is non-constant (e.g. time-dependent)
● Jointly model signal mean and variance
![Page 44: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/44.jpg)
Accounting for uncertainty
Heteroscedastic Gaussian processes (GPs)
● GP models the mean of the signal
● Another GP models the variance of the signal
Recent work currently under review!
![Page 45: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/45.jpg)
Aggregated speeds as measured by mobile devices
Google data
![Page 46: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/46.jpg)
HeteroscedasticMore accurate predictions and more precise confidence intervals!
![Page 47: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/47.jpg)
More uncertain during night periods
More confident during day periods
Properly handling uncertainty
![Page 48: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/48.jpg)
Some results
Evaluation of the prediction means
MAE: mean absolute errorRAE: relative absolute errorR2: coeff. of determination
![Page 49: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/49.jpg)
Some results
Evaluation of the prediction intervals
NLPD: negative log predictive densityICP: interval coverage percentageMIL: mean interval length
![Page 50: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/50.jpg)
Bus travel time prediction
We can exploit the network structure and spatio-temporal correlations to get robust joint prediction models!
![Page 51: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/51.jpg)
Robust bus travel time prediction
Deep learning architecture
Recent work currently under review!
![Page 52: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/52.jpg)
Movia data
● Real-time AVL-system● 1,2M travel time observations ● 4A bus line in Copenhagen● May to October 2017
![Page 53: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/53.jpg)
Results
RMSE: root mean squared errorMAE: mean absolute errorMAPE: mean absolute percentage error
![Page 54: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/54.jpg)
Summary
● Accounting for the effect of external factors○ E.g. events, weather, etc.
● Anomaly detection● Properly handling uncertainty
○ Account for observation noise○ Produce confidence intervals for predictions
● Robust joint prediction models○ Exploit spatio-temporal correlations
![Page 55: Predicting for the adaptive transport system …and other ...€¦ · transport system …and other necessary ingredients for resilient urban mobility Filipe Rodrigues / Francisco](https://reader034.vdocuments.mx/reader034/viewer/2022050610/5fb163c53509d5189550119f/html5/thumbnails/55.jpg)
Predicting for the adaptive transport system
…and other necessary ingredients for resilient urban mobility
Filipe Rodrigues / Francisco [email protected] / [email protected]