Bayesian Travel Time Reliability Feng Guo, Associate Professor Department of Statistics, Virginia...

33
Bayesian Travel Time Reliability Feng Guo, Associate Professor Department of Statistics, Virginia Tech Virginia Tech Transportation Institute Dengfeng Zhang, Department

Transcript of Bayesian Travel Time Reliability Feng Guo, Associate Professor Department of Statistics, Virginia...

Page 1: Bayesian Travel Time Reliability Feng Guo, Associate Professor Department of Statistics, Virginia Tech Virginia Tech Transportation Institute Dengfeng.

Bayesian Travel Time Reliability

Feng Guo, Associate ProfessorDepartment of Statistics, Virginia TechVirginia Tech Transportation InstituteDengfeng Zhang, Department of Statistics, Virginia Tech

Page 2: Bayesian Travel Time Reliability Feng Guo, Associate Professor Department of Statistics, Virginia Tech Virginia Tech Transportation Institute Dengfeng.

Travel Time Reliability

• Travel time is random in nature. • Effects to quantify the uncertainty

– Percentage Variation– Misery index– Buffer time index– Distribution

• Normal • Log-normal distribution• …

Page 3: Bayesian Travel Time Reliability Feng Guo, Associate Professor Department of Statistics, Virginia Tech Virginia Tech Transportation Institute Dengfeng.

Multi-State Travel Time Reliability Models

• Better fitting for the data• Easy for interpretation

and prediction, similar to weather forecasting:– The probability of

encountering congestion – The estimated travel time

IF congestion

Guo et al 2010

Page 4: Bayesian Travel Time Reliability Feng Guo, Associate Professor Department of Statistics, Virginia Tech Virginia Tech Transportation Institute Dengfeng.

Multi-State Travel Time Reliability Models

• Direct link with underline traffic condition and fundamental diagram

• Can be extended to skewed component distributions such as log-normal

Park et al 2010; Guo et al 2012

Page 5: Bayesian Travel Time Reliability Feng Guo, Associate Professor Department of Statistics, Virginia Tech Virginia Tech Transportation Institute Dengfeng.

Model Specification

– : distribution under free-flow condition – : distribution under congested condition– is the proportion of the free-flow

component.

Page 6: Bayesian Travel Time Reliability Feng Guo, Associate Professor Department of Statistics, Virginia Tech Virginia Tech Transportation Institute Dengfeng.

Model Parameters Vary by Time of Day

Mean Variance

Probability in congested state

90th Percentile

What is the root cause of this fluctuation?

Page 7: Bayesian Travel Time Reliability Feng Guo, Associate Professor Department of Statistics, Virginia Tech Virginia Tech Transportation Institute Dengfeng.

Bayesian Multi-State Travel Time Models

• The fluctuation by time-of-day most like due to traffic volume

• How to incorporate this into the model?

Page 8: Bayesian Travel Time Reliability Feng Guo, Associate Professor Department of Statistics, Virginia Tech Virginia Tech Transportation Institute Dengfeng.

Model Specification: Model 1

– : distribution under free-flow condition – : distribution under congested condition– is the proportion of the free-flow

component.

Link probability of travel time state with covariates

Link mean travel time of congested state with covariates

Page 9: Bayesian Travel Time Reliability Feng Guo, Associate Professor Department of Statistics, Virginia Tech Virginia Tech Transportation Institute Dengfeng.

Bayesian Model Setup

• Inference based on posterior distribution

• Using non-informative priors: let data dominate results.

• Developed Markov China Monte Carlo (MCMC) to simulate posterior distributions

Page 10: Bayesian Travel Time Reliability Feng Guo, Associate Professor Department of Statistics, Virginia Tech Virginia Tech Transportation Institute Dengfeng.

Issues with Model 1

• When traffic volume is low, the two component distribution can be very similar to each other

• The mixture proportion estimation is not stable

Page 11: Bayesian Travel Time Reliability Feng Guo, Associate Professor Department of Statistics, Virginia Tech Virginia Tech Transportation Institute Dengfeng.

Model Specification: Model 2

– : distribution under free-flow condition – : distribution under congested condition– is the proportion of the free-flow

component.

where is a predefined scale parameter: How large the minimum value of comparing to

Page 12: Bayesian Travel Time Reliability Feng Guo, Associate Professor Department of Statistics, Virginia Tech Virginia Tech Transportation Institute Dengfeng.

Comparing Model 1 and 2

• =1: the minimum value of congested state is the same as free flow

• =1.5: congested state is at least 50% higher than free flow

-1

Page 13: Bayesian Travel Time Reliability Feng Guo, Associate Professor Department of Statistics, Virginia Tech Virginia Tech Transportation Institute Dengfeng.

Simulation Study

• To evaluate the performance of models• Based on two metrics

– Average of posterior mean– Coverage probability

1.Set n=Number of simulations we plan to run.2.For (i in 1:n){

Generate dataDo{

Markov Chain Monte Carlo}While convergenceRecord if the 95% credible intervals cover the true values

}

Page 14: Bayesian Travel Time Reliability Feng Guo, Associate Professor Department of Statistics, Virginia Tech Virginia Tech Transportation Institute Dengfeng.

Simulation Study: Data Generation

• : Observed traffic volume at time interval i on day j

• : Average Traffic volume at time interval i (e.g. 8:00-9:00)

Page 15: Bayesian Travel Time Reliability Feng Guo, Associate Professor Department of Statistics, Virginia Tech Virginia Tech Transportation Institute Dengfeng.

Model 1 VS Model 2: Posterior Means

Page 16: Bayesian Travel Time Reliability Feng Guo, Associate Professor Department of Statistics, Virginia Tech Virginia Tech Transportation Institute Dengfeng.

Model 1 VS Model 2: Coverage Probability

Setting 3: when both and are small

Page 17: Bayesian Travel Time Reliability Feng Guo, Associate Professor Department of Statistics, Virginia Tech Virginia Tech Transportation Institute Dengfeng.

Robustness

• What if…– the true value of is unknown– The two components are too close

• We showed that the overall estimation are quite stable, even if the tuning parameter is misspecified

• When the two components are too close, by selecting a misspecified tuning parameter could improve the coverage probabilities of some parameters

Page 18: Bayesian Travel Time Reliability Feng Guo, Associate Professor Department of Statistics, Virginia Tech Virginia Tech Transportation Institute Dengfeng.

Robustness

Page 19: Bayesian Travel Time Reliability Feng Guo, Associate Professor Department of Statistics, Virginia Tech Virginia Tech Transportation Institute Dengfeng.

Robustness

Page 20: Bayesian Travel Time Reliability Feng Guo, Associate Professor Department of Statistics, Virginia Tech Virginia Tech Transportation Institute Dengfeng.

Robustness: Coverage Probabilities

Page 21: Bayesian Travel Time Reliability Feng Guo, Associate Professor Department of Statistics, Virginia Tech Virginia Tech Transportation Institute Dengfeng.

Data

• The data set contains 4306 observations• A section of the I-35 freeway in San Antonio,

Texas.• Vehicles were tagged by a radio frequency

device• High precision

Page 22: Bayesian Travel Time Reliability Feng Guo, Associate Professor Department of Statistics, Virginia Tech Virginia Tech Transportation Institute Dengfeng.

Study Corridor

Page 23: Bayesian Travel Time Reliability Feng Guo, Associate Professor Department of Statistics, Virginia Tech Virginia Tech Transportation Institute Dengfeng.

Modeling Results

Page 24: Bayesian Travel Time Reliability Feng Guo, Associate Professor Department of Statistics, Virginia Tech Virginia Tech Transportation Institute Dengfeng.

Real Data Analysis

Page 25: Bayesian Travel Time Reliability Feng Guo, Associate Professor Department of Statistics, Virginia Tech Virginia Tech Transportation Institute Dengfeng.

Probability of Congested State as A Function of Traffic Volume

Page 26: Bayesian Travel Time Reliability Feng Guo, Associate Professor Department of Statistics, Virginia Tech Virginia Tech Transportation Institute Dengfeng.

Next Step…

• Apply the model to a large dataset Any available data are welcome!• Hidden Markov Model

Page 27: Bayesian Travel Time Reliability Feng Guo, Associate Professor Department of Statistics, Virginia Tech Virginia Tech Transportation Institute Dengfeng.

HMM

• The models discussed are based on the assumption that all the observations are independent. Is it realistic?

0 10 20 30 40

0.0

0.2

0.4

0.6

0.8

1.0

Lag

AC

F

Simulated TravelTime

0 5 10 15 20 25 30 35

0.0

0.2

0.4

0.6

0.8

1.0

Lag

AC

F

Observed Travel Time

Page 28: Bayesian Travel Time Reliability Feng Guo, Associate Professor Department of Statistics, Virginia Tech Virginia Tech Transportation Institute Dengfeng.

Hidden Markov Model

• Hidden Markov model is able to incorporate the dependency structure of the data.

• Markov chain is a sequence satisfies:

• In hidden Markov Chain, the state is not visible (i.e. latent) but the output is determined by

Page 29: Bayesian Travel Time Reliability Feng Guo, Associate Professor Department of Statistics, Virginia Tech Virginia Tech Transportation Institute Dengfeng.

Hidden Markov Model

• Latent state:

• Distribution of travel time:

• and satisfy Markov property:

• If { are independent, this is exactly the traditional mixture Gaussian model we have discussed.

Page 30: Bayesian Travel Time Reliability Feng Guo, Associate Professor Department of Statistics, Virginia Tech Virginia Tech Transportation Institute Dengfeng.

Model Specification

• Transition Probability: • E.g. is the probability that the travel time is jumping

from free-flow state to congested state.• We use logit link function to model the transition

probabilities with traffic volume:

Page 31: Bayesian Travel Time Reliability Feng Guo, Associate Professor Department of Statistics, Virginia Tech Virginia Tech Transportation Institute Dengfeng.

Preliminary Results

• When the traffic volume is higher, the congested state will be more likely to stay and free-flow state will be more likely to make a jump.

• The mean travel time of the two states are 578.8 and 972.6 seconds.

• If we calculate the stationary distribution, the proportion of congested state is around 11.3%.

• AIC indicates that hidden Markov model is superior to traditional mixture Gaussian model.

Page 32: Bayesian Travel Time Reliability Feng Guo, Associate Professor Department of Statistics, Virginia Tech Virginia Tech Transportation Institute Dengfeng.

Simulation Study

0 200 400 600 800 1000

-22000

-21000

-20000

-19000

-18000

Dots: Hidden Markov Lines: Traditional Mixture

Data Set ID

Log-lik

elih

ood

Page 33: Bayesian Travel Time Reliability Feng Guo, Associate Professor Department of Statistics, Virginia Tech Virginia Tech Transportation Institute Dengfeng.

• Questions?• …• Thanks!