Agent-Based Models in Action
Case studies that demonstrate the power of advanced modelling techniques
Claire is the Senior Product Manager for Arup’s City Modelling Lab
At Arup’s City Modelling Lab, we are developing capabilities that can be used on a broad set of questions that can help make our cities work better- for all of us. We have been working with a few partners and clients to advance and apply our research in simulating cities, regions, and nations.
Here we share three case studies that describe our work with Transport for London, Transport Infrastructure Ireland, and the Ministry of Transport in New Zealand.
There are similarities in the challenge facing each of these organisations. In general, local and national transport authorities are trying to:
- Monitor and improve links between people and key activities (e.g. schools, jobs and public services like hospitals.)
- Understand the impact of any change in the cost of travel (e.g. public transport fares or road pricing.)
- Anticipate the interactions between transport modes (e.g. the impact of ride-sharing on public transport ridership, or the introduction of scooters and bikes on congestion.)
- Identify and understand the greatest network pinch-points in more detail (e.g. what time is the network most stressed and for how long, or which routes serve essential workers who are unlikely to have access to alternative transport options.)
How would we traditionally tackle these challenges? Existing methods to tackle these questions typically make use of aggregate data. This can provide high-level, sensible estimates for transport service performance. But we are limited in how we explore, slice, and integrate the data with other models (like economic planning models, or even health models). As we try to answer new questions like, how would an increase in telecommuting impact my transport network? — it becomes even more challenging to rely on traditional approaches.
With aggregated data, we miss the underlying information about what drives behaviour.
“Ultimately it is not a transport model. It is a social model. There is real opportunity outside of transport if we build a framework where people’s behaviours are better understood.”
Dan Jenkins. New Zealand, Ministry of Transport.
Below is more about the agent based models we are building with Transport for London, Transport Infrastructure Ireland and the New Zealand Ministry of Transport.
(For more background on what an agent based model is, try our post: Def: City Modelling.)
Transport for London
We worked collaboratively with TfL to simulate where and when roads become congested; where and when demand for public transit increases; and where and when people spend time at work, school, or home. This was a part of a joint-research initiative with TfL to answer the question, “What does a city-wide agent based model for London require?”
Together, we built a city-wide alpha simulation based on bottom-up interactions of a simulated (or ‘synthetic’ — in transport speak) population. TfL developed the synthetic population with initial activity patterns. These activity patterns describe what people are trying to do each day (For example a person might start from home, drop children off at school, travel to work, and then return home.)
Building with this synthetic population, we simulated behaviours of ~10% of London’s residential population, including some non-London residents, and freight vehicles. The result was a behaviour model for ~1 million unique agents, taking ~750 actions each, over the course of one day. We could see where and when people were likely to travel to work, and which routes people traveled on. Simulating what we observe in the real world is the foundation for testing different “what if” scenarios.
In order to simulate London’s population, we needed to build an entire UK transportation network, including National Rail services and the motorways. Within London itself, we represented all public transport and roads.
This work demonstrated that agent based models are suitable and technically feasible to tackle a range of questions they may be planning for. Agent based models would allow TfL to simulate the impact of policy or infrastructure changes, at a local-level and assess the equity consequences of any change. Agent based models would also allow TfL to simulate the effect of technology or cultural changes, which we don’t have any precedent for, like connected autonomous vehicles or new remote-working patterns.
You can find more detail about this project in this paper.
New Zealand Ministry of Transport
The Ministry has partnered with Arup to build a national simulation capability. This research project will explore how an agent based model could be used to evaluate a variety of policy and infrastructure options. Our modelling work with the Ministry will support an evidence-base for transport interventions and outcomes over the next 5–50 years. For example, understanding the impact of new infrastructure or road pricing on travel behaviour, and importantly, assessing the social and distributional impacts of such changes.
Together, we are co-creating an agent based model. We built the model foundation, calibrating the model with 10% of the New Zealand population, pursuing individual plans across the country.
Moving forward, we will be expanding the model to encompass New Zealand’s wide mix of transport services (including ferries and airplanes.) We will use a large sample of the population (>15%) and we will be focusing on assessing the impact of tolling on the transport network.
Transport Infrastructure Ireland
To support our work with Transport Infrastructure Ireland in planning for a more equitable and sustainable road network, we will be simulating a variety of potential transport changes using an agent based model. We have built the foundations for a country wide model, which will allow us to assess complex interactions across the network.
We have calibrated our model using a range of historic data, providing decision makers with confidence in the model inputs. This historic data, enables us to test changes and report against observed conditions. Using this information, we will look for innovative ways to pilot new solutions for the Irish public, promoting the EU principles of User Pays and Polluter Pays more fairly. For example, if a road toll changes, how will that affect communities differently? Who might be most vulnerable to policy changes? And what new travel behaviours might be formed as a result.
The next stage of this work will expand the model to simulate a range of potential changes to the network — so that Transport Infrastructure Ireland can assess the consequences of potential transport polices. This will support a broad set of outcomes, like business continuity, greater equity, reductions in carbon emissions, and service reliability for people.
