Cities and Regions as Complex Evolving Systems: Spatial MultiAgent Models

1
Cities and Regions as Complex Evolving Systems
Spatial Multi-Agent Models

• Peter M. Allen
• Presentation for
• WEHIA Conference, Essex University,
• 13-15th June, 2005

2
Origins of Complexity Science - The B-Z Reaction
The Brusselator!!!
A -gt X B X -gt Y D 2X Y -gt 3X X -gt F
A simple non-linear chemical reaction where
spatial and functional structure is triggered by
fluctuations of local density and parameters
The agents are spatially distributed molecules
With feedback interactions. Dialogue with
fluctuations.
3
But a brusselator model of a REAL ecosystem
FAILS!!!!

• It is a description of an ecosystem not an
  explanation, a description in conflict with
  reality.

4
The co-evolution of heterogeneous organisms leads
to NATURAL organization

• Dialogue of system with fluctuations and also
  with variant individuals leads to Ecosystem
  structure. The agents are organisms and we find
  a spatial co-evolution
• Change occurs when a system cannot restrain its
  own fluctuations and experiments
• New behaviours that stay, stay, while those that
  dont, dont. Over time we get an incredible
  assemblage.
• This leads to a miraculous diversity
  maximisation rather than to an efficient,
  functional design.

5
Social Science can we do more than describe it
post-hoc? (for Professor Sunder)
Structure??? Structural Change Structural
Stability Stationarity
NO LEARNING!!
Dictionary
Take out the detail the non-average
Decontextualise
Complexity
Post Modernists
Strategy
Freedom
Constraints
Operations
Simplicity
1 2 3
4 5
Boundary Classification average types
average events
6
Economics what drives Market Organization?

• What market structures of qualities, and price
  strategies emerge? Is the INVISIBLE HAND really
  there?
• Multi-agent models consider how the exploration
  of strategy space leads to an ecology of agents
  (Allen, 1978)
• There is no BEST STRATEGY for an agent, but
  there can be good and bad ecologies to co-evolve
  in.
• Selection operates through the choices of buyers,
  but for agents this is both about competition and
  co-operation or niche division among them

Strategy Space
Changing Demand
6 interacting firms/agents Evolving supply
7
Agents seek complementary strategies How?

• Base version Darwinian market. (death and
  replacement) Firms/Agents are launched with
  random strategies. They do not modify this, but
  if they die they are replaced other with random
  strategies
• Agents start with a loss and cannot calculate
  expected profits. They BELIEVE.
• Or, agents imitate winning strategies, or
  explore strategy space actively, or have
  response rules depending on what happens.
• Free Marketeers do not mind which operates.

Market Structure Diversity
Selection Emergent structure depends on Luck
8
What evolves in P/Q Strategy Space?
6 Darwinian Firms random re-launch
6 Firms as before, but Firm 1 is an imitator.
Model 1
Model 2
6 Firms but 5 imitators.
All imitators...
Model 3
Model 4
Firm 1 can Learn
All Firms learn
Model 5
Model 6
Mixed Strategies
Model 7
9
Overall sector value depends on chance4
different strategies show mix of learning
imitation is best.
Market structure and even the overall
profitability of a sector depends on the
non-linear parameters, and on LUCK.
10
Multi Agent, Spatial Dynamic Models 1980!!!

• Different types of actor at each zone, with
  characteristic needs to be fulfilled
• These characteristic needs are stable, but the
  behaviour of actors depends on the changing
  circumstances
• The spatial distributions of the different types
  of job and different kinds of people affect each
  other as the potential for housing demands,
  commercial activities and for travel affect and
  are affected by transportation and land-use.

11
The Interaction mechanisms of Brussaville..
Different types of agent distributed over
spatial zones
NETWORKS Transport Networks Road, Rail,
Buses, trams, walking Endogenous Flows on all
links of all networks as dynamic output of the
model. Impacts of changed infrastructure,
with feedbacks..
Each Agent Type is a distribution of origins,
with others as destinations
But they share the different transport networks
12
Multi-Agent Urban or Regional Model

• Each type of Economic Agent has characteristic
• labour requirements
• Revenues/m2
• Economies of scale
• Input patterns and costs
• Distributed customers
• Output patterns and costs
• Each Residential Agent type has characteristic
• Income
• Consumer preferences
• Patterns of job and other destinations
• Locational preferences
• Demographic parameters
• Each located agent is BOTH an origin and a
  destination as they decide
• on the network path and travel mode for people
  and goods.

13
Emergent Brussaville spatial co-evolution.
Spatially distributed agents interact and
simultaneously create structures and flows of
communications, people and goods.
14
Evolving Brussaville.
Onwards with existing structure channelling the
future
15
Retail Strategies.
History and timing matter equilibrium doesnt do
it!
Launching 40 units at t10 SUCCEEDS.. Launching
40 units at t 20 FAILS Launching 50 units at
t20 SUCCEEDS Launching 40 units at
a Different location at t 20 SUCCEEDS
16
Changing Transport Infrastructure A New Metro
System
Without the New Metro
With the New Metro
Changed distribution of tertiary services, and of
residents. Changed house and land prices,
commuting patterns, traffic flows, Congestion and
pollution.
17
Shopping Centres..
Simultaneous Development At 4 sites
No Intervention
Successive Development of Same 4 sites
Maximum Possible..
18
Self-Organizing, dynamic spatial models

• Taylor and Francis, ISBN -9056990705 and
  9056990713 A Great Read!!!
• 1975- 1995 Urban, Regional models linking
  land-use and transportation
• Spatially distributed interacting heterogeneous
  agents in the 1980s
• Also, links to environmental factors (air, water,
  land, etc.)
• Unfortunately, still not used in the UK.

19
Example Consequences of improved Transport
Infrastructure for West Bengal (2003)

• Work with Asian Development Bank for West Bengal
• Survey by Halcrow Consultants of the current
  flows of goods on the roads
• Transport Infrastructure projects effects on
  transport costs.
• Economic gains lead to increased consumption and
  production. Spatial multipliers on jobs created
  allows calculation of the impact on poverty
  where and how much extra employment and wealth
  created

20
2025 Jobs Created Poverty Reduction.
Jobs in Agriculture, Industry and Services
Savings made by Poor, Medium and Rich
21
Urban and Regional Spatial models

• The equations used in these models can also be
  written using cellular automata, which allow very
  detailed models.
• They can find spatial rules of interaction that
  generate the current patterns and then can be run
  into the future, or they can be related to the
  rules of multiple, spatially distributed agents.
  (White and Engelen, 2000)

22
Model Interactions
System Interactions
Transport Network
23
LocalConstrained Cellular Automata

• 250 m resolution
• 17 identical and coupled CA models1 per Region
• Neighbourhood 8 cell-radius, 196 cells
• Overall growth of each function is determined at
  the Regional level

• 18 land-use classes
• 8 Function states
• 8 Feature states
• 2 Vacant states

This model calculates on a yearly basis the
changing land use for 225,000 cells (250 m
resolution, 18 land use categories)
24
Recent Complex Spatial dynamics, RIKS
Puerto Rico
Mutual interdependence of multi-actor system,
flows of goods, services, energy, water.. Can
calculate emissions, traffic, Patterns of
service demand etc
Dublin
25
DTI Foresight Intelligent Infrastructure Study

• The objectives are
• Consider the Gateshead transport model, and
  develop a Smart trading scheme for traffic slot
  allocation (Essex and Newcastle)
• For this area consider the changing patterns of
  locations of both residential, leisure and
  employment locations and their generation of
  travel demand
• Consider the effects of intelligence and
  communications on the patterns of demand, in
  order to see how patterns of demand may change.
  New definition of work and workplace, of
  internet shopping, virtual businesses etc.
• Include the effects of better travel management
  systems such as smart market multi-agent models
  on these patterns of demand, and inversely, of
  the changing patterns of demand on the
  intelligent management systems proposed.

26
Multi-agents within multi-agents
Longer term change in travel demand impacts of
ICT, of demographics, of Work and workplace
Multi-agent distributed responses
Slow Dynamic
Origins Destinations Travel timing and choices
Rapid Dynamic
Daily use of smart market to get slots on routes
and Experience the evolution of costs
27
The Gateshead Zones and Road Network

• Chopwell Rowlands
• Crawcrook Green side
• Ryton
• Winlaton
• Blaydon
• Whickham South
• Whickham North
• Dunstan
• Lamesley
• Teams
• Bensham
• Saltwell
• Low Fell
• Chowdene
• Bede
• Deckham
• High Fell
• Birtley
• Felling

3
5
15
2
19
21
11
7
8
16
4
12
22
10
17
13
14
20
6
1
9
18
28
NOISE Affects the results !
Conclusions

• Complex systems evolve structurally and generate
  emergent, co-operative organization, with
  multiple scales.
• Complex, multi-agent models can be used to help
  imagine how a system may evolve in future, and
  repeated runs with noise and different
  scenarios can can indicate and the stability of
  different possible trajectories
• This provides the basis for a new type of
  integrated model for policy exploration linking
• Economic activities
• Households and population
• Transport and travel demand and flows
• Environmental impacts
• Public service provision and social implications
  

Comments p.m.allen_at_cranfield.ac.uk