Estimating Signal Timings for Assignment Models from Junction Flow and Delay Data



Estimating Signal Timings for Assignment Models from Junction Flow and Delay Data

Authors

Tom Van Vuren, Mott MacDonald, Matt Millard, Mott MacDonald

Description

In most networks, vehicle actuated control means that obtaining representative green times for a peak period is impossible. We present a methodology that estimates green times for input to junction-based assignment models that make use of readily available data (flows at counters and delay data from GPS sources ) and the software's green time optimiser.

Abstract

Traffic assignment models increasingly have the capacity to include explicitly junction operations and control. For many years this has been recognised as a desirable feature in, particularly, congested urban situations – see for example the advice in the UK Department for Transport’s WebTAG guidance item3.19. Whereas in the eighties and nineties this functionality was only available in SATURN, more recently other packages such as CUBE and VISUM have also developed junction modelling capabilities.

Obtaining signal timings in large scale urban networks is surprisingly difficult. Although in many cases the stage diagrams can be obtained, extracting average green times for modelled time periods, such as the AM or PM peak, is time consuming and in some cases impossible, as vehicle actuation leads to a large number of actual timings in a model hour, which also do not tend to be stored.. Increased functionality but lack of data availability – an undesirable situation.

However, we have relevant data available, for at least part of the network: traffic flow observations from sensors on most signal approaches. And the software packages contain signal optimisers, mimicking the more complex algorithms deployed on-street. The two tools combined allow us to proxy signal timings in a network model that lead to consistent link flows and resulting delays.

We use the software signal optimiser to estimate the signal timings that would have resulted from the observed flows (as long as the optimiser in the software replicates reasonably well the optimiser on-street). Where input flow data was incomplete (for example where not all approaches are equipped with sensors or where the flow data was out-of-date) a matrix estimation from counts approach (at junction level) was used to provide up to date and consistent input flows.

To validate the approach, we had delay data available from GPS data sources. This enabled us to compare the delays calculated in the signal optimiser against the observed values. As delays are key drivers of route choice, a good fit between calculated delays in the signal optimiser and those observed in reality, gives confidence in the junction green time representation for base year assignment.

Do not confuse this approach with running a
base year assignment with the junction signal optimiser switched on. Rather than relying on assignment flows from a network model, our approach uses observed flows as a starting point for a signal optimisation, and on a junction by junction level rather than network-wide. The network is not optimised – the green times at individual junctions are optimised for the observed flows, as tends to happen in reality.

We provide a number of examples for individual test junctions and then present the results of a full-scale application in a network model of the greater Birmingham area in the UK, with approx. 400 signalised junctions.

Note that the proposed methodology is particularly useful for assignment models that include junction simulation; in microsimulation models signal timings are usually optimised within the application so knowledge of the stage diagrams generally suffices.

Publisher

Association for European Transport