Can we predict Crowd Behaviour at Mass Gatherings?

Can we predict Crowd Behaviour at Mass Gatherings?

By Paul Townsend and Dan JefferyCrowd Dynamics International.
A mass gathering is a coming together of people for some reason – maybe a concert, sports event or protest.

However, the mass of people that attend, whether they are called visitors, spectators, protestors or any other name are, to a crowd modeller at least, just a new type of crowd that can be understood.

To most people, crowds appear unpredictable. The people in the crowd appear mindless and seem to be carried along within the crowd, so how can we predict what people are going to do? The answer to this question is complicated, but is being revealed through the work in the LETSCROWD project.

How can we plan for the safety of crowds?

Before a mass gathering, it is imperative that crowds who will attend are understood in their movement patterns, likely behaviours. This means forecasting the number of attendees, how they arrive, how to process them and manage them throughout the mass gathering and how to allow them to leave; not to mention how to manage emergency situations. This complicated process is helped in many situations by crowd modelling.

It is the responsibility of the organiser to carry out this planning, but Law Enforcement Agencies (LEAs) need to assess this information, test their own contingency planning, or even organise parts themselves, when the mass gathering has no particular responsible organiser.

What can the LETSCROWD models do?

LETSCROWD plan to help this planning process with an advanced crowd modelling toolkit (called the ‘Crowd Modelling and Planning Tool’, that can assess everything from the capacity of an area, to evacuation scenarios, to predicting crowd behaviour in reaction to a suspected bomb, or in reaction to LEA tactics (like police dogs, communication procedures or simply testing moving crowds away from a danger are).

Can we simulate crowd behaviour during a live event?

These simulations will be able to help in training LEA staff (decision makers and frontline) by recreating potential scenarios at a mass gathering, but also during operations. A real-time approach to crowd modelling is being trialled where LEAs can assess a situation, quickly trial the proposed response and get feedback on the consequences before implementing it. This has been shown to be possible for evacuation scenarios, and the possibility of the other features used by LEAs during their operations is going to be demonstrated and tested live.

What are the models/simulations made up of?

The crowd models use a combination of modelling scales from macroscopic, mesoscopic and microscopic agent based models to capture the requisite behaviour across different areas. For example, use simpler macro models to forecast crowd movements during evacuation over the entire mass gathering for multiple scenarios, and use detailed microscopic models (individual people who make their own choices in the model) to determine the effect of a particular crowd management/control strategy on the crowd itself. For example: What is the resulting crowd density?; How many staff are required to be effective?; Is the proposed communication effective?

With a basis on the FP7 funded project eVACUATE ( , the real-time aspect of the model is extremely fast at computing crowd movement over many different possible routes. The detailed model is based on the social force approach for movement, but with a behavioural component that allows modelling of behavioural reactions to objects, staff, communication with the crowd etc. The models are all visualised in 3d for ease of understanding, and more work is being done on summary outputs that are easy to understand.

How can the predictions be trusted?

A validation process for the models is in place, many basics of the models have already been validated such as the movement of people by comparing against other commercial simulations, data sets and observed phenomenon. The behavioural implementation will be validated alongside the 7 LEAs within the project. They are helping to calibrate them using their knowledge of crowd behaviours and they will use the tools during the proposed practical demonstrations to provide feedback for incorporation. This user driven human factors approach should overcome any issues due to trust.