Predicting road travel times Luis Campos (with Dan Elliott) 10 th - - PowerPoint PPT Presentation

Predicting road travel times

Luis Campos (with Dan Elliott) 10th INFRADAY Conference, Berlin, 07 Oct 2011

2 Frontier Economics

Predict road travel time to prioritise investments

Road network investments Road management strategies Road delays Measure the impact of different changes to road conditions/layout

• n average delays

Different factors affect travel time Help prioritise future investments

The Highways Agency was interested in measuring the relative contribution of each

• f the main factors affecting

road conditions to travel time. The Highways Agency wanted assess the impact of past road network investments and road management strategies on average delays.

3 Frontier Economics

An empirical model to predict road travel times

50 100 150 200 250 300 350 400 1 21 41 61 81 101 121 141 161 181 201 221 241 261 281 301 321 341 361 381 401 421 441 461 481 501 521 541 561 581 601 621 641 661

Average travel time (in seconds) 15 minute intervals over one week period

At any moment, travel time is affected by a variety of factors

?

Different models, of “specific” changes to “representative” roads Road conditions are constantly changing across the network. Most empirical analysis relied on simple “before and after” comparisons.

• Comprehensive model that reflects the particular features of and

the actual traffic flow behaviour on the HA road network. Bespoke model

4 Frontier Economics

Conceptual challenge – finding a simple descriptive formula

Impact of accidents on journey time

(simulation using model parameters)

100 200 300 400 500 600 700 800 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90

15 minute intervals Average journey time (in seconds)

Accident

Congested road Non congested d

Impact of a sudden increase in journey time

(additional 200 seconds)

50 100 150 200 250 300 350 400 450 500 1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 81

15 minute intervals Average journey time (in seconds)

Road section Upstream road section

Illustrative example: interaction between flow and speed Illustrative example: impact of an accident depends on initial traffic conditions Illustrative example: disruptions in one section affect travel times across the network

• the large number of factors affecting travel

times and:

• how these factors vary across the

network and over time;

• how these factors interact; and
• how road delays are likely to persist
• ver time and propagate over the

network. A simple formula needs to capture the complex dynamics of traffic behaviour, given:

5 Frontier Economics

Weather conditions, incidents, road works, etc.

Time-series non-linear descriptive formula

1 6 5 4 3 1 2 1

* ) (

  

      

it it i it i it i it i it i n it i i it

J R b I b W b F b N b P b a J

Density

6 Frontier Economics

Practical challenge – parameter estimation

• despite the richness of this dataset,

we found gaps in the data for a significant number of road sections and so:

• we were unable to directly

estimate the average impact of changes to road conditions on travel time in these road sections.

• To construct a fully dynamic model we

had to find an indirect way of

• btaining these estimates.

We collated an impressive dataset of more than 190 million records, but:

Data for > 2,500 sections, over

2 years, 15 min. frequency

1st stage

Road section time-series data

3rd stage

Weighted average

2nd stage

Network level cross- section data parameter estimates parameter estimates traffic conditions over time road features across sections

We used what we knew about “data rich” road sections to infer what the likely parameters for the “data poor” road sections was

• We used this approach to improve the estimates on all the road

sections, even on “data rich” road sections.

• The objective was to improve the forecast power of the model, which

can be different from maximising the “in-sample” fit. More generally

7 Frontier Economics

Practical challenge – prediction power

• A statistical model is, by its very

nature inaccurate.

• Inaccurate parameters, when

combined with strong dynamic features, can result in very unstable predictions.

• Minor errors can spread over the

network and persist over time, leading to large prediction errors in subsequent periods. Using a statistical model to make predictions is not straightforward

Illustrative example: unstable predictions

• 3000.0
• 2500.0
• 2000.0
• 1500.0
• 1000.0
• 500.0

0.0 500.0 1000.0 1500.0 1 2 3 4 5 6 7 8 Average journey time (in seconds) Hours Actual journey time Predicted journey time

Distribution of estimated parameter across road sections Predicted vs. actual journey time

• We constrained the estimation to

ensure the model generated stable predictions, e.g.:

• used log-linear model
• restricted the parameters

associated with , ,

1  it

J

1  it

N

n it

P 

• 3000.0
• 2500.0
• 2000.0
• 1500.0
• 1000.0
• 500.0

0.0 500.0 1000.0 1500.0 1 2 3 4 5 6 7 8 Average journey time (in seconds) Hours Actual journey time Predicted journey time

Predicted vs. actual journey time (after constraints)

8 Frontier Economics

The National Delay Model

How is travel time expected to change after several incidents? What is the relative contribution of different factors for delays? How long is the impact on an accident expected to last? How do changes in road conditions in one section affect average travel time across the network? What is the expected impact

• f adding a new

lane on a road to the average travel time?

The NDM should answer…

9 Frontier Economics

Outputs

An easy-to-use tool of travel time prediction…

The NDM measures the impact on road delays of changes to the road conditions and/or road layout by comparing the average journey time on a selected set of road sections before and after those changes, i.e. comparing a base case with a scenario case.

• Choose the base case:
• period considered; and
• roads considered.
• Insert the changes to road

conditions and/or road layout to include in the scenario. Inputs

• travel time (average, relative contributions,

profile over time)

• reliability (number of periods, vehicle-hours)

10 Frontier Economics

…to help the HA prioritise its investments

Reduce the impact of accidents on road delays Reduce the impact of rain on road delays

• Strong impact on travel time.
• Relatively rare.
• Low impact on travel time.
• Relatively frequent (in England!).

The quantification of the relative impacts of different incidents/events over time and across the network has obvious implications on the type of intervention the HA should be investing on to reduce delays.

• r

11 Frontier Economics Frontier Economics Limited in Europe is a member of the Frontier Economics network, which consists of separate companies based in Europe (Brussels, Cologne, London and Madrid) and Australia (Melbourne & Sydney). The companies are independently owned, and legal commitments entered into by any one company do not impose any obligations on other companies in the network. All views expressed in this document are the views of Frontier Economics Limited.

12 Frontier Economics FRONTIER ECONOMICS EUROPE LTD. BRUSSELS | COLOGNE | LONDON | MADRID Frontier Economics Ltd, 71 High Holborn, London, WC1V 6DA

• Tel. +44 (0)20 7031 7000 Fax. +44 (0)20 7031 7001 www.frontier-economics.com