Junction modelling in strategic models

7 December 2016 1

Junction Modelling in Strategic Models

___

James Parrott

AITPM 2017 National Conference

Junction Modelling for Strategic Models 2

Junction Modelling

The majority of network delay is at intersections

Junction Modelling

• Experience in using Strategic, Meso, Micro and Single Point

• Tiers of modelling have differing purposes

• Everything below strategic accounts for junction modelling

Question: Should we include junction modelling in strategic models

• Not common in Australia

• No definitive answer

• Investigate reliability, ability to model upgrades and attempt to minimise model noise


Cube Junction Modelling

It’s already in use……

Focus of review was Cube Voyager’s junction modelling

• Over 10 years experience of use

• Conducted detailed analysis of possible methodologies

Junction Model Types


Cube Junction Modelling

Some parameters are more important than others

Junction modelling has developed over the years with numerous additions to the capability. However there are several key inputs:


Impact and Limitations

No model is correct, nor perfect. As modellers, it is our job to use these tools to the best of our ability

Inclusion of junction modelling has two key impacts

• Must reduce speed flow delays to remove junction delay

• Use volume averaging for assignment

Limitations include

• Simplified phasing

• Max of four arms

• No lane lengths

• Intergreen only total


Reliability

In order to use something, we must have confidence that it is fit for purposeSix intersections tested


Performance Comparison

Understanding strengths and weaknesses is a core ingredient to the successful use of any tool

Results validated well with some exceptions

• Signalised junctions for controlled movements were generally similar

• Optimisation of phasing tended to favour major movements

• Left turns (yield or continuous) were not modelled well

• Unsignalised intersections are much harder to model and vary more between software packages


Sensitivity and Model Noise

Model noise is not new, but we still grapple with how to best account for it.

Series of options to test model sensitivity and whether proposed approaches could minimise junction noise

• Options: Actuated, Fixed and No Junction Modelling

• Use of warm-start

• Simple project case involving upgrade of single intersection

• Approach focused on isolating parameters for testing


Junction Modelling Approach

We must have confidence in the results, but not blind faith Impact of junction upgrade


Junction Modelling Approach

Some tools are better than others at providing guidance for certain problems Comparison of junction upgrade vs non-junction upgrade

• Link (lane) amendment to reflect change

• Time savings were negligible

• Junction savings of 18 seconds per veh for 110% of base demand


Mitigating Model Noise

Isolating and improving the accuracy of project benefits must be at the forefront for us to improve our business model and create more successful projects

Attempts to reduce the impact of noise throughout the network tested:

• Combination of assessing actuated/fixed and use of warm-starts

• Review of results at sector level for Simple junction upgrade

• Noise persisted in all options

• Warm-starts converged much faster

• Size of network modification irrelevant

• No. of iterations critical in the assessment of benefits – what is equilibrium?


Conclusions

Some answers, more questions. Let the debate continue…..

Set out to review reliability, ability to assist in assessing network impacts and ability to isolate changes:

• Signalised junctions are reliable – must account for left turns

• Junction Modelling is reactive to network changes with intuitive results

• Model noise is not easily addressed

• Warm starts should be considered to improve speed of model (given slower convergence with junction modelling)

• Assessing benefits – consider local area, review iterations completed

• Continue the research…….

Junction modelling in strategic models

Education