Crowd Dynamics Ltd - Micro v Macro Simulation techniques

Micro/Macroscopic simulations

(updated 31st August 2007)

Keywords: simulating people, simulating crowds, simulating crowd dynamics

Myriad is a successful hybrid of a number of different modelling systems. It includes elements of microscopic and macroscopic techniques. It is important that you use the right tool for the right job and we shall now highlight the pitfalls and advantages of both techniques. One or the other may be the best tools for a specific study/analysis but be careful about relying on only ONE approach
"Someone who builds a mathematical model can get carried away with all the clever things that can be done with it. The model becomes a safe little world, free from anxiety, free from office politics, rewarding in its own right. This often results in a very clever model that has little to do with reality. Resist the temptation to be too smart in building a model. Take small model building steps interspersed with healthy doses of reality." Sam. L Savage. Author of Insight.XLA - Analysis software.
"When all you have is a hammer - every problem look like a nail."

MICROSCOPIC MODELLING

The advantages of microscopic modelling

Microscopic techniques provide insight to the behaviour of a system under a wide range of conditions provided you do not become too ambitions with the scope of the technique. It is an excellent discovery tool and can provide valuable information over a wide range of behavioural inputs. Psychologists have the ability to experiment with human factors in the relative safety of the laboratory. When flow through a specific geometry is unknown a microscopic technique should be applied as part of the discovery process. This allows operators to test sections of their environment for behavioural dynamics. We would refer to this process as "crowd simulation".

Problems with microscopic modelling

This type of modelling technique resolves down to individual parameters (eg: aggression, speed of movement, size, affinity, family, social position, disability etc, etc) and involves a much higher degree of validation. One of the main criticism for microscopic techniques is the problems of analytical intractability which is best described by the "little old lady" effect.

When next in a busy underground station, try and observe the clustering of groups behind slower moving individuals (the little old lady) and you will notice that the location, speed and direction of the little old lady can have considerable effect on the the following crowd. To be sure you have captured the little old lady effect in a microscopic simulation you need to test a much wider range of behaviours (such as frustration to the fact that you're stuck behind a little old lady). This can be a serious problem in that a given egress behaviour may be significantly effected by this factor. How many models of little old ladies do you need to build? How many different positions may the little old lady occupy in a group of 1000 individuals (ie: is she at the front, the back or somewhere in the middle - and how many people does she affect?)

This is only one factor of analytical intractability - it gets much worse when you need to consider the time related effects. Take an egress scenario as a typical example of this problem with microscopic simulations. Imagine you are in a room, filling with smoke. The room has 3 doors and it is occupied with 99 other people (including 10 little old ladies).

Which door you choose is a function of your visual acuity (which doors you can see), the familiarity of the areas behind the doors, the nearest door, the congestion at your chosen door, the position of the little old lady, the frustration of the people behind the little old lady, the rising threat. How many microscopic simulations do you need to build to be sure you have captured the dynamics of this kind of environment ?

If there is any doubts surrounding the coefficients in a model, how can we have any confidence in any answers it may produce?

This kind of model building is very expensive (both in time and expertise to interpret the results. A vast array of data input and output can be produced and, as analytical data is concerned, it can provide a blanket behavioural test required for a complete understanding of a system. However, one would rarely know if there was a problem in the model, a problem in the code that produces the model or the behaviour of the model if a small change in the environment were suddenly introduced. One example where these kind of models fail completely would be the sudden change in the environment (say heavy rain for example) where the position of the little old lady will now have a considerable effect on the crowd dynamics.

A similar problem exists when you try and build microscopic models for complex spaces - you never know which element may hold the critical factor for the accurate prediction of the system as a whole. Be very careful of any claims that a microscopic technique promises to deliver. Unless used by experts this type of modelling can be very misleading.

One client remarked that the fancier the graphics the less he trusted the results. Never a truer word was spoken in jest.

At all times - ask for validation of the process, the technique, the modelling inputs, the skills required of the operator. Reliance on a model alone can be a very dangerous situation for a organisation to undertake. Always use a range of models and modelling techniques. If they produce different results - ask why. We were asked why Simulex and Steps produce different results - the answer is they make different underlying assumptions about the dynamics of individuals. Both are very useful tools, both can be applied to a wide range of potential situations - the skill in in knowing which to use and when. To know the limitations of a system requires intimate knowledge of a wide range of techniques that underpin the various commercially available tools.

We are happy to provide independent expert advice on which modelling technique is most appropriate and to provide evaluation services for both modelling inputs and outputs.

MICROSCOPIC MODELLING

The advantages of macroscopic modelling

Easy to apply, conforms with accepted data and field observations. Quick to get reliable answers (example click here for queueing models). Macroscopic modelling techniques form the basis of of the existing, tried and tested, codes of practice around the world.

The problems with macroscopic modelling.

There is a wide body of science associate with macroscopic modelling - fluid dynamics (tried, tested, proven and applied) uses macroscopic model - you do not model the interaction of every particle in a fluid - but rely on the behaviour of the fluid as a large scale interactive system. Thus frictional forces, flow, and pressure/density are the macroscopic terms associated with fluidic motion.

A crowd can be successfully modelled using flow, level of service, probability of conflict, queueing models and shockwaves - however one needs to rely on trustworthy data and an understanding of the systems behaviour over a wide range of inputs. Could one every know the behaviour of a system of 50,000 people (surely not if you use a microscopic modelling technique - the time to test ALL of the possible interactions is huge). The "Pedestrian and Evacuation Dynamics" conferences held in 2001 and again in 2003 ((ISBN 3-540-42690-6 and 1-904521-08-8) illustrate the wide range of modelling techniques available to site design and pedestrian and evacuation planning.

To model a crowd using macroscopic models one needs to state the underlying assumptions and work in the domain of statistics and probability (most of us have a problem understanding probabilities). So the tools used to create macroscopic models need to be based on solid, reliable, tried and tested results. There should be peer reviewed publications for ANY and ALL simulations systems that are applied to places of public assembly.

The bulk of pedestrian planning and design, evacuation and contingency planning and architectural/building codes of practice are macroscopic models. They are applied because, by and large, they are proven to work. There is always room for improvement and the wide range of modelling tools currently available shows that the industry is applying some of the best minds to the on-going problems of crowd dynamics.

MULTI-SCALAR MODELLING

The multi-scalar approach.

During Dr. Still's PhD research he uncovered a wide range of differences in the various codes around the world and explained the differences using a range microscopic models. These are now integrated to a more useful multi-scalar modelling technique.

Please note that the Myriad concept is "use MANY tools"

A good engineer will use a range of tools to do the job. Testing one result against another and, where differences arise, evaluating which results are appropriate, which can be trusted and which are problems with a specific approach to the problem. In many cases it is the skill of the model builder that makes the difference. As the saying goes: "Any fool can drive a car - but only the best can win the grand prix." We run a series of workshops to train engineers on how to build and how to recognise good model building techniques from bad ones.

One final note of caution: Evaluate the model builder as much as the model.

Building a Myriad model to assess the "wear and tear" and under utilised space in a complex environment.

Step 1. From a CAD file remove all the information (text etc.) that does NOT relate to the areas that the pedestrians will occupy. This is easily achieved by using the appropriate layers in autocad. Then you "export" as a bitmap.

Step 2. Save as a screen image and import to Myriad - Myriad converts these files to a grey/black scale for processing. Once the bitmap is saved from from autocad you click on "import bitmap" and use the free space buttons to define the pedestrian areas.

Step 3. Run a series of diffusion maps for each Origin-Destination pair. At this stage (a few minutes of user time) the system displays the areas of highest usage (wear and tear). the red area (left) shows the wear and tear for JUST the escalator bank at the bottom left. Given the Origins from all the other ingress areas. The process utilises a mathematical theorem which produces the overall probability of conflict maps.

Step 4. Once the Diffusion Maps are stored for each Origin-Destination pair (called and O-D matrix) the superposition of ALL the O-D pairs produces a Level of Service map. Red areas have the highest utilisation and therefore at a given ingress/egress rate the highest Level of Service.

Step 5. The areas of LOWEST space utilisation indicate those areas that could be reclaimed - it is this map (wasted space) that allows the user to determine the value engineering (how to develop cost effective spaces) can be justified.