Transcript Chapter 2

Chapter 8
Geocomputation Part A:
Cellular Automata (CA) & Agent-based
modelling (ABM)
www.spatialanalysisonline.com
Geocomputation
“the art and science of solving complex spatial
problems with computers”
www.geocomputation.org
Key new areas of geocomputation:
Presentation 8A: Geosimulation (CA and ABM)
Presentation 8B: Artificial Neural Networks (ANNs); &
Evolutionary computing (EC)
3rd edition
www.spatialanalysisonline.com
2
Geocomputation
Many other, well-established areas:
Automated zoning/re-districting (e.g. AZP)
Cluster hunting (e.g. GAM/K)
Interactive data mining tools (e.g. brushing and linking,
cross-tabbed attribute mapping)
Visualisation tools (e.g. 3D and 4D visualisation,
immersive systems… some also very new!)
Advanced raster processing (e.g. ACS/distance
transforms, visibility analysis, image processing etc.)
Heuristic and metaheuristic spatial optimisation, …. and
more!
3rd edition
www.spatialanalysisonline.com
3
Geocomputation: Geosimulation
For the purposes of this discussion:
Geosimulation includes
Cellular automata (CA)
Agent-based modelling (ABM)
Geosimulation is particularly concerned with
Researching processes
Identifying and understanding emergent
behaviours and outcomes
Spatio-temporal modelling
3rd edition
www.spatialanalysisonline.com
4
Geocomputation: ANNs
In the next presentation on geocomputation:
ANNs discussed include
Multi-level perceptrons (MLPs)
Radial basis function neural networks (RBFNNs)
Self organising feature maps (SOFMs)
ANNs are particularly concerned with
Function approximation and interpolation
Image analysis and classification
Spatial interaction modelling
3rd edition
www.spatialanalysisonline.com
5
Geocomputation: Evolutionary computing
In the next presentation on geocomputation:
EC elements discussed include
Genetic algorithms (GAs)
Genetic programming (GP)
EC is particularly concerned with
Complex problem solving using GAs
Model design using GP methods
3rd edition
www.spatialanalysisonline.com
6
Cellular automata (CA)
 CA are computer based simulations that use a
static cell framework or lattice as the environment
(model of space)
 Each cells has a well-defined state at every
specific discrete point in time
 Cell states may change over time according to
state transition rules
 Transition rules that are applied to cells depend
upon their neighbourhoods (i.e. the states of
adjacent cells typically)
3rd edition
www.spatialanalysisonline.com
7
Cellular automata
State variables
 typically binary (e.g. alive/dead), but can be more complex
 may have fixed (captured) states
Spatial framework
 typically a regular lattice, but could be irregular
 boundary issues and edge wrapping options
Neighbourhood structure
 Typically Moore (8-way) or von Neumann (4-way)
 Typically lag=1 but lag=2 .. and alternatives are possible
Transition rules
 Typically deterministic but may be more complex
 Time treated as discrete steps and all operations are
synchronous (parallel not sequential changes)
3rd edition
www.spatialanalysisonline.com
8
Cellular automata
Neighbourhood structure
 Typically Moore (8-way) or von Neumann (4-way)
 Typically lag=1 but lag=2 .. and alternatives are possible
3rd edition
www.spatialanalysisonline.com
9
Cellular automata
Example 1 – Game of life
 State variables: cells contain a 1 or a 0 (alive or dead)
 Spatial framework: operates over a rectangular lattice
(with square cells)
 Neighbourhood structure: 4 adjacent (rook’s move) cells
 State transition rules: time tntn+1
1. Survival: if state=1 and in neighbourhood 2 or 3 cells
have state=1 then state  1 else state  0
2. Reproduction: if state=0 but state=3 or 4 in neighbouring
cells then state  1
3. Death (loneliness or overcrowding): if state=1 but
state<>2 or 3 in neighbourhood then state  0
3rd edition
www.spatialanalysisonline.com
10
Cellular automata
Life (ABM framework): Click image to run model (Internet access required)
3rd edition
t0 35% cell occupancy
tn – evolved pattern
Randomly assigned
(still evolving – to density 4%)
www.spatialanalysisonline.com
11
Cellular automata
Example 2 – Heatbugs
 State variables:
 Cells may be occupied by bugs or not
 Cells have an ambient temperature value 0
 Bugs have an ideal heat (min and max rates settable) – i.e.
a state of ‘happiness’
 State transition rules: time tntn+1
1. Bugs can move, but only to an adjacent cell that does not
have a bug on it
2. Bugs move if they are ‘unhappy’ – too hot or too cold (if
they can move to a better adjacent cell)
3. Bugs emit heat (min and max rates settable)
4. Heat diffuses slowly through the grid and some is lost to
‘evaporation’
3rd edition
www.spatialanalysisonline.com
12
Cellular automata
Heatbugs (ABM framework): Click image to run model (Internet access required)
3rd edition
www.spatialanalysisonline.com
13
Cellular automata
 Example geospatial modelling applications:
Bushfires
Deforestation
Earthquakes
Rainforest dynamics
Urban systems
 But..
Not very flexible
Difficult to adequately model mobile entities (e.g.
pedestrians, vehicles)… interest in ABM
3rd edition
www.spatialanalysisonline.com
14
Agent-based modelling
 Dynamic systems of multiple interacting agents
 Agents are complex ‘individuals’ with various
primary characteristics, e.g.
Autonomy, Mobility, Reactive or pro-active behaviour,
Vision, Communications capabilities, Learning
capabilities
 Operate within a model or simulation
environment
 Time treated synchronously or asynchronously
 CA can be modelling using ABM, but reverse
may be difficult
 Bottom-up rather than top-down modelling
3rd edition
www.spatialanalysisonline.com
15
Agent-based modelling
 Sample applications:
Archaeological reconstruction
Biological models of infectious diseases
Modelling economic processes
Modelling political processes
Traffic simulations
Analysis of social networks
Pedestrian modelling (crowds behaviour,
evacuation modelling etc.) …
3rd edition
www.spatialanalysisonline.com
16
Agent-based modelling
 Example 1: Schelling segregation model
Actually a CA model implemented here in an ABM framework.
Agents represent people; agent interactions model a social
process
Spatial framework: Cell based
State variables: grey – cell unoccupied; red – occupied
by red group; black – occupied by black group
Neighbourhood structure (Moore)
State transition rules:
 If proportion of neighbours of the same colour x% then
stay where you are, else
 If proportion of neighbours of the same colour <x% then
move to an unoccupied cell or leave entirely
3rd edition
www.spatialanalysisonline.com
17
Agent-based modelling
Schelling (ABM framework): Click image to run model (Internet access required)
3rd edition
www.spatialanalysisonline.com
18
Agent-based modelling
 Example 2: Pedestrian movement
Realistic spatial framework
Multiple passengers arriving and departing
Multiple targets – ticket machines, ticket booths,
subway platforms, mainline platforms, shop,
exits …
Free movement with obstacle avoidance
3rd edition
www.spatialanalysisonline.com
19
Agent-based modelling
Pedestrian movement: Click image to run model (Internet access required)
3rd edition
www.spatialanalysisonline.com
20
Agent-based modelling
 Advantages of ABM
Captures emergent phenomena
 Interactions can be complicated, non-linear,
discontinuous or discrete
 Populations can be heterogeneous, have differential
learning patterns, different levels of rationality etc
Provides a natural environment for study
 Spatial framework can be complex and realistic
Flexible
 Can handle multiple scales, distance-related components,
directional components, agent complexity etc
3rd edition
www.spatialanalysisonline.com
21
Agent-based modelling
 Disadvantages of/issues for ABM
What is the real ‘purpose’ of model?
What is the appropriate scale for research?
How are the results to be interpreted?
How robust is the model?
Can the model be replicated?
Can the results be validated?
Are behaviours/patterns observed likely to occur in the
real world?
How much is the outcome dependent on the model
implementation (design, toolset, parameters etc.)?
3rd edition
www.spatialanalysisonline.com
22
Agent-based modelling
 Choosing a simulation/modelling system
Ease of development
Size of user community
Availability of support
Availability of demonstration/template models
Availability of ‘how-to’ materials and
documentation
Licensing policy (open source,
shareware/freeware, proprietary)
3rd edition
www.spatialanalysisonline.com
23
Agent-based modelling
 Choosing a simulation/modelling system
Key features
Number of agents that can be modelled
Degree of agent-agent interaction supported
Model environments (and scale) supported (network,
raster, vector)
Multi-level support (agent hierarchies)
Spatial relationships support
Event scheduling/sequencing facilities
3rd edition
www.spatialanalysisonline.com
24
Agent-based modelling
 Major simulation/modelling systems
open source: SWARM, MASON, Repast
shareware/freeware: StarLogo, NetLogo, OBEUS)
proprietary systems: AgentSheets, AnyLogic
3rd edition
www.spatialanalysisonline.com
25