Crowd Science

Module 4, lesson 1 · 35 minutes

Why model, and the modelling ladder

Why engineers simulate crowds, how models are classified, and what verification and validation mean.

You’ll be able to

  • give three reasons to model a crowd: design, planning and forensics;
  • classify a model as macroscopic or microscopic, discrete or continuous, force-, rule- or velocity-based;
  • define verification and validation, and name ISO 20414.

Introduction

A design team shows a committee an animation: thousands of dots stream out of a new stadium and the clock stops at 6 minutes 40 seconds. Everyone relaxes. Nobody asks what the dots were told to do, which measured crowd they were tuned to, or whether the same software has ever been checked against a real evacuation. This module teaches you to ask.

Three reasons to model

  • Design. Before anything is built, will the space cope? Keith Still frames the questions as capacity, throughput and risk, and advises planning for 50%, 100% and 500% of the maximum capacity, each with a separate plan.
  • Planning and operations. Which gates open when, how long will queues be, where will density peak? “Risk is dynamic. Risks can change with time”, so the answer is a timeline, not one number.
  • Forensics. After a disaster, a model can reconstruct conditions nobody measured. The police and National Forensic Service simulation of Itaewon estimated 10.74 p/m² at 22:25.

The modelling ladder

Helbing and Molnár named three classes of model: microscopic, mesoscopic and macroscopic. Macroscopic models describe a crowd by fields such as density and flow; microscopic models track every person. Within microscopic models, space can be continuous or a grid of cells, and motion can come from forces, rules or chosen velocities. Each rung of the ladder below adds detail and cost.

The modelling ladderFrom the cheapest description to the most detailed.
RungClassWhat it representsLesson
Fundamental diagramMacroscopicSpeed and flow as functions of density1.3
Cumulative curvesMacroscopicCounts past a point over time; queues and delays4.4
Continuum (Hughes 2002)MacroscopicThe crowd as a density field that flows4.4
Floor-field cellular automatonMicroscopic, discrete, rule-basedPeople hopping between 40 cm cells4.3
Social forceMicroscopic, continuous, force-basedPeople as particles pushed by forces4.2
Velocity-based (ORCA, Optimal Steps)Microscopic, continuousPeople choosing collision-free velocities or steps–

Every model is calibrated to a fundamental diagram somewhere. Helbing’s social force parameters were tuned so that 0.73 persons per second pass a 1 m door at a desired speed of about 0.8 m/s; Burstedde’s cellular automaton fixes a 40 cm cell and a 0.3 s step, which is a walking speed of about 1.3 m/s. Find that anchor and you know what kind of crowd the model can speak for.

Verification and validation

Verification asks whether the software solves its own equations correctly. A simple test: one person walking 10 m at 1.34 m/s should take 7.46 s (our arithmetic). Validation asks whether the model reproduces what real crowds do: measured flows, densities, evacuation times. A model can pass the first and fail the second.

The international protocol is ISO 20414:2020, “Verification and validation protocol for building fire evacuation models”. It was first published in November 2020 and confirmed in 2026; it is aimed mainly at agent-based models, can be adapted to macroscopic ones, and excludes trains and ships. Still reviewed pedestrian and evacuation simulations in 2007; lesson 5 turns these questions into a checklist.

A vendor says its model “has been validated”. What two questions do you ask first?

Validated against what data (which crowd, which geometry, which measured quantity)? And does that resemble your problem? Then: which verification tests does it pass, and how large was the difference between model and measurement?
  • “More detail means more accurate.” A microscopic model has more parameters to get wrong. For 50,000 runners a macroscopic model is often the better tool.
  • “Validated means correct.” It means it matched particular data under particular conditions.