Using Activity Based Modelling to Implement a Peak Spreading Model in a Practical Multi-modal Context.

Using Activity Based Modelling to Implement a Peak Spreading Model in a Practical Multi-modal Context.


P Clarke, R Culley, P Davidson, Peter Davidson Consultancy, UK


Using activity based modelling to implement a peak spreading model in a practical multi-modal context.


When forecasting 10, 20 or more years ahead it is usual to find that levels of travel demand cannot be accommodated on the transport networks ? even after accounting for switch to other modes and destinations. When faced with the levels of congestion associated with such high forecast demand, many potential travellers would retime their trip to avoid the most congested period.

With current technology, peak spreading models require data about people?s travel which is not normally available, data from assignment models which they cannot produce and current theory relates predominantly to road traffic. It is therefore not surprising that most models duck the issue.

However activity based modelling provides us with a new set of tools to tackle this lack of peak spreading realism. With peak spreading becoming an increasingly important feature of forecasting, we set out to produce a practical methodology for peak spreading in a multi-modal context suitable for application to a City or County which draws upon theory originally developed by Vickery, further developed by Small and encapsulated under Equilibrium Scheduling Theory (EST) as discussed by Hyman (1997) and others.

We enhanced the assignment models to use time-dependent queuing for highway modelling and timetable assignment for public transport so as to get realistic paths, travel times etc for each 15-minute time interval before, during and after the peaks. The assignment model was sensitive to the travel demand so the more the peaks spread, the slower the pre and post peak travel times became.

The preferred-arrival times and other segmentation data were taken from available data. The path builder was modified so as to build paths to arrive at the destination at a particular preferred arrival time rather than to depart from the origin at a particular preferred departure time.

An activity based model was built to micro-simulate every individuals arrival or departure time choice depending upon their own particular circumstances, and the model micro-simulated their own decision of which 15-minute time interval to take. They were each assigned to the transport networks to the 15-minute path used to define their choice. For public transport, they were assigned to their chosen public transport route and service, which meant that the time interval choice was sensitive to the actual service selected.

The methodology was implemented in the Gloucestershire sub-Regional multi-mode model and used to forecast different land use and transport scenarios for the 2026 Regional Spatial Strategy. The paper describes the implementation, results and lessons learned.

There are many innovations in this paper. One of the key weaknesses of discrete choice models is in their poor interface with aggregate assignment models. This paper uses a new assignment algorithm to derivate sensitive, realistic travel times for each 15 minute time interval. There is innovation in using activity based modelling to micro-simulate the peak spreading decisions of each member of the population, in applying the model in a real situation and in using it to develop more realistic highway and public transport demand forecast travel volumes.


Association for European Transport