You are on page 1of 11

Journal of Simulation (2010) 111

r 2010 Operational Research Society Ltd. All rights reserved. 1747-7778/10


www.palgrave-journals.com/jos/

The Hybrid War Model: a complex adaptive


model of complex urban conict
James Moffat1*, Michael Bathe2 and Lorna Frewer2
1

Defence Science and Technology Laboratory, Portsdown West, UK; 2Craneld University, Craneld, UK

This paper rstly discusses the modelling of Peace Support Operations (PSO) within the defence simulation modelling
context. It then provides a summary background of the current relevant approaches in such modelling, taking account
of the increasing complexity of the strategic environment, and the relevance of ideas from Complex Adaptive Systems
theory. It goes on to describe the details of two agent-based models spanning the problem domain, which capture the
key ideas of complexity, within a PSO context, taking account of the complex interactions between peacekeepers,
civilians, insurgents and non-governmental organisations involved.
Journal of Simulation advance online publication, 26 March 2010; doi:10.1057/jos.2010.2
Keywords: conict analysis; defence studies; military

1. Introduction
Over the 60 years and more of defence operational research
(OR) as a recognised discipline, the nature of the operations
under examination have changed signicantly several times:
the battles of World War II; the early, nuclear-focused Cold
War, together with sporadic smaller operations, in the 1950s
and 1960s; the mature Cold War of the 1970s and 1980s; and
then the rapid change to the present far more complex
operational environment. Looking to the future, we need to
consider a very wide spectrum of possible operational
contexts. At one end of the spectrum are major hostilities,
although on a lesser scale than the NATO-Warsaw Pact
cataclysm which was the focus of planning during the Cold
War. While serious conict in Europe cannot be entirely
ruled out, the Gulf Wars of 1991 and 2003 are seen as much
more representative of this end of the spectrum in the near
term. Towards the other end of the spectrum we have a
range of operations including peacekeeping, peace enforcement, protected evacuations and the provision of humanitarian relief. We bundle all of these together under the term
Peace Support Operations (PSO). Increasingly, military
means are used only in close conjunction with other means,
such as diplomacy and economic reconstruction. Following
the events of 11 September 2001, counter-terrorism is now
also given prominence.
In the more global change which we are now experiencing,
transitioning from the industrial age to the information age,
defence is not immune, and the rise of concepts such as
network-centric operations (Alberts et al, 2003), also known
as network-enabled capability in the UK and elsewhere in
*Correspondence: J Moffat, Policy and Capability Studies Department,
Defence Science and Technology Laboratory (Dstl), Portsdown West,
PO17 6AD, UK.

Europe, are signicantly changing the kinds of tools and


models which we need to develop as analysts (Moffat, 2002,
2003, 2007; Atkinson and Moffat, 2005). In OR modelling
terms, the transition is from models of the Cold War with
attrition at their core to models of current and future
complex operations with information and command decision making at their core. The importance of information is
also seen as going beyond providing the basis on which
better and faster decisions can be made. The analyst must
also be able to assess the effectiveness of information
operations (designed to inuence an adversarys decision
making in a favourable direction), which now play an
increasingly important role in achieving the effect traditionally delivered by force (Ministry of Defence, 2002). It will be
seen, therefore, that the modelling of military operations in
the post-Cold War strategic environment presents many
intellectual and practical challenges, but lies at the heart of
most applications of OR in the defence domain.
Within this broad future context, closed form simulation
modelling of PSO is a relatively new and developing focus.
Table 1, for example, based on NATO (2002) and Alberts
and Hayes (1995) illustrates the key differences between PSO
and conventional warghting operations, as perceived by
that NATO working group in 2002.
NATO (2002) also notes that in symmetric conventional
warfare, the mission tends to be relatively stable, there is
clear focus on the enemy, and the military has a common
understanding and commitment. Conversely, in PSO the
objective is often more dynamic. In many of the operations
in question there is no enemy. This is obviously true for
operations such as humanitarian assistance and disaster
relief. In addition, peacekeeping activities involve multiple
actors who must be treated even-handedly if the operation

2 Journal of Simulation Vol. , No. 0

Table 1 Differences between warghting and PSO operations, as perceived in 2002


Factor

Conventional warghting

PSO

Mission/operation
Stability
Focus
Commitment

Relatively stable
Enemy
Common (military)

May be more dynamic


No traditional opponent
Uncertain (political/military)

Principles
Unity
Decision making
Operations

Of command
Hierarchical
Surprise, secrecy

Of purpose
Consensus
Transparency

Information
Nature of the problem
Key question
Focus
Situation awareness
Information database

Known unknowns
How to get information
Enemy military
Common: air-land-sea
Very large, well structured

Unknown unknowns
What information to get
Military, political, economic, social factors
Limited dissemination; more complex
Larger, less structured

Analysis and modelling


approaches
Unit of resolution
Ease of integration
Focus
Approach

Battalion level entity


Relatively easy
Military (systems, organisations)
Traditional operational analysis (hard OR)

Behavioural aspects
Very difcult
Political, military and societal
Softer analysis (including soft OR).

is to be successful. In PSO operations, politicalmilitary


ambiguities can result in uncertain understanding of the
goals and objectives of the mission and a limited commitment (Starr et al, 1997).
Analyses of PSO often require consideration of individual
behaviour. The issue is compounded by the many factors
that have to be considered in the analysis process (eg
military, political, economic, social). This has led, at least in
initial efforts in this area, to the application of softer
analytic approaches, such as extensive reliance on expert
elicitation.

2. Development of simulation models of PSO


Hartley (http://home.comcast.net/Bdshartley3/NRL/NRL
.htm, accessed 30 July 2009) gives a comprehensive review
of the trajectory of developments in the US. He divides
the development of such models into three periods,
the Academic period, the Early DoD (US Dept of
Defense) period, and the Current period. The Academic
period covers the early 1990s with little serious model
development. In the Early DoD period (the late 1990s), there
were attempts to apply existing simulation approaches to the
PSO problem. It was not until about 2001 and later that
there were signicant attempts to create simulation models
focussed on the PSO domain from rst principles. Paralleling these US activities, in the UK Defence Science and
Technology Laboratory, a signicant step forward was made
with the development of the DIplomatic And Military
Operations in a Non-warghting Domain (DIAMOND)

closed form simulation model of Diplomatic and Military


Operations (Prins, 1998)another term for PSO, emphasising the meshing of political and military aspects. This was a
high risk but ultimately successful development of a closed
form simulation of PSO operations at a whole of country
(campaign) level which captures, in a quantied sense, some
of the key issues of PSO, including the interaction of
multiple actors, including civilians, and the active role of
non-government organisations (NGOs) such as the Red
Cross. This model is now in regular use to support evidencebased decision making within the UK Ministry of Defence,
and has also been given to a number of other countries. We
will describe DIAMOND in more detail later in the paper.
Following the success of DIAMOND, a number of
complementary models of PSO have been put in place.
Peace Support Operations Model (PSOM) is a high-level
human in the loop gaming environment representing a
range of civilian and military activities in PSO. It includes
both a High Level Game, in which key regional and
international actors play a role in determining operational
outcomes, and an Operational Level Game which considers
the actions and effects created by units and factions within
the area of operations. PSOM is supported by a more
resolved gaming environment at the tactical level, named
Stabilisation Operational Analysis Tool (STOAT). Development of a closed form simulation version of such a tactical
level PSO environment is currently underway, as a high risk,
high payoff research initiative by combining together the
STOAT approach and the Hybrid War Model (HWM)
developments. The HWM is an agent-based closed form
simulation of a complex tactical PSO environment, including

J Moffat et alThe Hybrid War Model 3

multiple actors, and details of the HWM development will


be given later in the paper.
Examples of other simulation developments include the
PAX model. This was created in Germany by EADS
Dornier GmbH. The paper (Schwarz and Bertsche, 2003)
describes the background to the model and how it was
developed. They wished to model a simple food distribution
scenario using an agent-based model but found existing
agent-based approaches to be inadequate. As a result a new
model, PAX, was developed using the behavioural model
Physical Conditions, Emotional State, Cognitive Capabilities and Social Status (PECS) as a basis, for application to
small scale situations where the individual entities and their
changing personalities are an important part of the analysis.

3. The shift from complicated to complex operations


In Moffat (2003), we identied a number of dening
characteristics of a Complex Adaptive System (CAS), as
summarised below.
Non-linear interaction. The interactions between agents are
not linear. Thus the outcome is not, in general, proportional
to the input. This can give rise to surprising behaviour. Such
interactions are local, either in the sense of physical
proximity, or local in the sense of having a close linkage in
a communications or information network.
Decentralised control. There is no centralised control
determining the evolution in time of the ensemble of agents.
Many natural systems, such as the co-evolution of an
ecosystem, or the movement of a uid front through a
crystalline structure, are not controlled centrally. The overall
system behaviour is thus generated as a result of all of the
local agent to agent interactions.
Self-organisation. The system tends to a particular state or
objective, without the need for guidance from outside the
system.
Non-equilibrium order. Patterns emerge across the system
of agents, which are a result of the local agent to agent
interactions, rather than being imposed from above.

Adaptation. The local interactions result in adaptation of


the agents through a learning process.
Collectivist dynamics. The ability of agents to locally
inuence each other, and for these effects to then ripple or
cascade through the system, allows continual feedback
between the evolving states of the elements of the system.
In Chapter 2 of Moffat (2003), we apply these characteristics to an Information Age force, as in Table 2.
It turns out that these characteristics of a CAS are very
applicable to the complex environment of coalition PSO
operations. In the US, such complex coalition environments
have been termed complex endeavours (Alberts and Hayes,
2007). We rst characterise these, and then tease out the
similarities to a CAS, as described in Table 2. The term
complex endeavours refers to undertakings that have one
or more of the following characteristics:
The number and diversity of the participants is such that
there are multiple interdependent chains of command, the
objectives and goals of the participants may conict with one
another, and the participants perceptions of the situation
may differ in important ways. Moreover, the effects resulting
from agent actions span both physical and psychological
consequences (thus including the effects of inuence and
consent, for example, as well as physical destruction or
rebuilding). As a result, there is a lack of understanding of
networked cause-and-effect relationships and an inability to
predict the ultimate effects that are likely to arise from
alternative courses of action.
The above characteristics of a complex endeavour embody
many of the thoughts contained in Table 2, characterising
such endeavours as examples of CAS, as we now tease these
out.
The number and diversity of participants results in a
correspondingly large number of degrees of freedom that,
in turn, can generate a large number of different ways in
which participants could interact. The interactions that can
and are likely to take place between participants are affected
by the nature of the perspectives of the individual
participants, the amount of information that is shared, their

Table 2 Relation between CAS and information age warfare


CAS concept

Information age force

Non-linear interaction

Combat forces composed of a large number of non-linearly interacting parts.

Decentralised control

There is no master oracle dictating the actions of each and every combatant.

Self-organisation

Local action, which often appears chaotic, induces long-range order.

Non-equilibrium order

Military conicts, by their nature, proceed far from equilibrium. Correlation of local effects is key.

Adaptation

Combat forces must continually adapt and co-evolve in a changing environment.

Collectivist dynamics

There is a continual feedback between the behaviour of combatants and the command structure.

4 Journal of Simulation Vol. , No. 0

individual qualities of awareness, and the extent of their


shared awareness. Given the large number of factors that
inuence the nature of each interaction, it is reasonable to
assume that these interactions will not be linear (thus small
differences in initial conditions may lead to large changes in
outcome).
Multiple, interdependent chains of command mean
that there is no single person in command, hence no
master oracle dictating the actions of each and every
combatant.
An agile command capability means that there is continual
feedback between the behaviour of the combatants (the
circumstances and context) and the command approach
adopted.
Networked cause-and-effect relationships are likely to
result in cascades of effects (Smith, 2003). Our ability
to predict these cascades and the resulting effects is, at best,
limited. However, we may be able to bound the range
of values that could occur, or are likely to occur (Smith,
2006).

4. The OR challenge of such complex endeavours


These complex endeavours present a series of major
challenges that affect both the problem formulation and
solution phases of OR, including:
 an incoherent objective function (ie, there is no single set
of dependent variables to be maximised or minimised);
and
 an inability to predict outcomes as a function of particular
agent actions.
Without either the ability to nd feasible regions of the
effects space that represent solutions or the ability to easily
link agent actions to specic outcomes, the situation is not
easily amenable to deductive analysisthat is, the philosophic approach that involves breaking a problem into a
number of parts in order to understand the whole. The
problem domain is thus complex, rather than merely
complicated.
A key concept in Table 2 involves the nature of the
interactions between entities and between entities and the
environment. These interactions involve both entities and
effects. They can be either direct or indirect. A direct
interaction requires that the entities or effects have no
intervening entity or effect interposed (they are like adjacent
links of a chain). An indirect interaction involves one or
more intermediaries, as in a cascade of effects or a chain of
command. Thus, during a physical interaction, the entities
need to be in close proximity to one another and these
interactions are usually one-on-one. For interactions related
to information exchange, physical proximity is not relevant
and there can be a one-to-many relationship. Consider those
who are directly affected by a media report. They may be

widely distributed geographically and may number in the


thousands if not the millions.
If we think of these interactions as parts and consider the
behaviour that emerges, in complicated problem spaces, the
behaviour of the whole can be deduced or understood from
the sum of individual behaviours (the largely independent
behaviours of the parts). This is not true for complex
endeavours, where one may easily understand individual
behaviours (decision rules associated with agents, for
example) but not be able to understand or predict overall
behaviour. The same argument holds true for interactions
across a range of outcome spaces.

5. Complex adaptive models for complex environments


Having now set out the dimensions and nature of the
complex space within which we may operate in the 21st
century, we clearly need analytical models which can capture
this dimensionality, while also being tractable enough to
allow us to do practical analysis. There is a tension here
between attempting to capture the complexity of the
problem explicitly in our model (hence a complex model of
a complex problem) and the need for simplicity and
transparency in order to allow practical analysis and
understanding. The Project Albert initiative of the US
marine corps was an early attempt to develop agent
simulation models of such complex problem spaces, leading
to the development (in order of both time of development
and complexity) of the ISAAC, MANA, and PYTHAGORAS agent-based simulations (Lauren and Stephen,
2002; Illachinski, 2004). Lauren has taken the lead in
developing the MANA simulation, and in applying it to a
range of both warghting- and PSO-related issues in support
of the New Zealand Defence forces (Lauren, 2003, Lauren
et al, 2003). MANA (Lauren and Stephen, 2002) is
essentially a model of warghting (with two sides representedBlue and Red) at the tactical level which puts the
emphasis on the complexity of the warghting agents and
their interactions, in order to create emergent behaviours at
the force level, corresponding, to some extent, to the more
complex problem spaces described earlier.
MANA uses a simple approach to agent decision-making
based on attraction or repulsion from other perceived
agents. (An agent in this case is a small tactical force unit
such as a section or a squad.) The general parameters in
MANA are used to set the size of the battleeld, the position
of the physical objectives for each force, and the size and
initial location of the two sides. In order to program the
objective for each side, waypoints are used. Alternative
waypoints can also be set so that the agents have a choice of
route. For each agent in MANA, the personality parameters of the agent are used to determine its movement.
There are a number of parameters (13 in total), each giving
an attraction to (or repulsion from) specic other factors

J Moffat et alThe Hybrid War Model 5

such as alive friends, alive enemy, injured friends, distant


friends and alive neutrals. There are also attractions towards
the next waypoint and alternate waypoints. These 13 values
are set between (100) and ( 100). A positive value means
the agent is attracted towards the factor, a negative value
indicates the agent will move away from it. A value of zero
means the agent is neither attracted nor repelled. The values
themselves are not important; it is the relative sizes that
matter. MANA also has range parameters. These put
constraints on the movement of an agent as well as
determining its effectiveness in battle. The personality and
range parameters are together dened for each side (Blue
and Red) and for each state the agents can be in. The HWM
builds on the ideas of MANA and extends them into more
complex problem spaces including multiple types of actor
(rather than just Red and Blue), and multiple types of
agent to agent interaction.

6. The DIAMOND model


DIAMOND represents PSO, and usually involves scenarios
that include military factions and civilian populations along
with peacekeeping forces. The DIAMOND model thus
represents these peace support and stability operations at a
joint level (including the effects of non-military entities such
as refugees, or aid agencies) and exploits the mission-based
approach to command, developed as part of our earlier
research (Moffat, 2002, 2007).
DIAMOND is a closed form simulation. A simple node
and arc network provides a graphical representation of the
region and environment, with representation of key areas of
interest, areas of sea or lake, and the airspace above.
Facilities, such as airports and civilian shelters can also be
represented. The model represents the main actors and
contributors to PSO by the use of agents known as entities.
These represent the capabilities and behaviours of military
units, civilians, non-military organisations (NMOs) and the
leaders or commanders for each. Entities interact with each
other and the environment and exchange or consume key
commodities such as food, fuel, and ammunition. The
simulation incorporates a mechanism to organise entities
into common parties that represent specic organisations
or common groups within a scenario. These parties have an
appropriate command structure and communications network to allow for the allocation of missions and ow of
intelligence throughout the party, and these parties have
relationships with one another which dene their interactions (ranging from friendly to hostile).
Using our mission-based approach to the representation
of Command and Control, the simulation includes a
mechanism to represent each partys Concept of Operations
by nesting objectives in a series of plans and for those
objectives to consist of a series of missions that entities can
carry out during a campaign. Commanders within a party
allocate resources to achieve their objectives in line with the

sequence of plans, and the simulation completes when a set


number of parties achieve their end state conditions or when
a predetermined period of time has elapsed.
As a simulation run progresses, each of the entities gains
information about their environment and other entities
through sensing, interactions, and communication. This
information is organised into a local picture on the basis of
which, entities make informed decisions on how they should
carry out their missions. DIAMOND also includes a
mechanism (referred to as negotiation) to obtain access
(through a checkpoint or roadblock, for example) to an area
denied to one party by another, and to allow multi-party
co-operation to achieve aims and objectives without having
to rely entirely on their own resources.

7. Representing the physical environment


As already noted, the physical environment in DIAMOND
is represented by a node and arc network. Nodes represent
areas of operational interest, population centres and the
locations of key infrastructure and terrain features. Arcs
represent the routes between these nodes. An example of a
NodeArc network for DIAMOND is shown in Figure 1.
Nodes can, depending on the nature of the scenario,
represent whole cities or individual districts or regions within
a city. They can be used to represent individual villages but a
more appropriate aggregation level would be a collection of
local villages. Nodes are also used to mark areas of deep
water, points along an air corridor, strategic junctions, and
key terrain features. In this way, the model is able to capture
the full air, maritime, and land context of the operation.
Arcs represent the routes between the nodes and each one
has several channels which can include ground routes (road,
rail, and cross-country links), air corridors, inland waterways (canals, rivers, lake crossings), close to shore waterways, and deep waterways. They can also be used to
represent important infrastructure networks such as oil or

Villages 2
Village 1
Valley pass
High ground
overlooking city

Seaport

City (N)

City (E)

City (W)

City (S)

Figure 1 Example node and arc network representation of the


physical environment.

6 Journal of Simulation Vol. , No. 0

water pipelines and electricity cables. The anticipated length


of each arc is around 1030 km, although this can be much
shorter where areas of interest are close to one another (eg,
the districts of a city). Nodes and arcs both have a terrain
type (called culture) which inuences a variety of calculations
within the simulation such as: the effectiveness of sensors;
the rate of attrition between two units engaged in combat;
and movement rate. Weather is also represented to the
extent of having a local, temporary effect. At each node it is
possible to dene facilities, which are key attributes of that
area that any entity can interact with, including Hospitals,
Shelter, Water, Targets, Food, Airports, and Seaports. Each
facility is represented in terms of its ability to sustain
damage, and its ability to continue, given such damage. They
also have a local ability to repair themselves.

8. The agents (entities) represented in the model


The entities in the model can be considered to fall broadly
into the four categories below:
Intervention forces: These are the peacekeeping and peace
enforcement forces with entities representing land, air,
maritime, and special forces units operating under a UN
or other international mandate. Supplementary police forces
to assist a failed state are also covered under this category.
Factions: The military and paramilitary forces of belligerent
or warring factions who are not part of the peacekeeping or
peace enforcement forces. The host nations forces are also
covered under the heading of factions. The entities include
land, air, maritime, and special forces units.
NMOs: NMOs include monitors and observers, commercial
companies, governmental and international humanitarian
agencies, and NGOs.
Civilians: Civilians, including neutral civilians and those
associated with individual factions, internally displaced
persons, refugees, and evacuees.
Although various types of commander are specied it is
implicit for entities, including civilians that they can make
their own decisions if they have no direction from a superior.
They have their own local picture and are capable of making
decisions for their own survival and to achieve their
missions. The higher level commanders take into account
broader considerations, such as deciding which stage of a
campaign plan should be followed, allocating resources to
missions or directing a number of subordinate entities to
work together to achieve a common goal. Commanders
here represent military headquarters, local government,
individuals and in some cases the intangible collective
actions of a set of common entities (eg, refugees) as
appropriate to the group being considered.

All information received by an entity (whether through


direct sensing or sharing of information) is assimilated into
its local picture. The representation of a local picture, and
perceptions based upon it, are important aspects of DIAMOND, as all entities decide what to do in the simulation on
the basis of the information available to them. If this
information is incomplete or out of date the entitys actions
may be different, compared to their actions based on
complete and current information. The local picture in
DIAMOND is an aggregation of all the information made
available to that entity. Each piece of information in this local
picture has an assessed resolution level, a degree of credibility
and a record of when it was collected. DIAMOND also
represents a number of communications networks allowing
entities to share information. Some of these communication
networks (such as military networks) are party based, whereas
others (such as commercial news stations) are global.
Messages communicated include, for example, orders, status
reports, requests for assistance, intelligence, local picture
information, and media broadcasts.

9. Missions and decision making


The activities of entities within the environment are
governed by two criteria: Firstly, the missions represented
in the model that entities are able to perform and secondly,
the decision-making processes in each party that determine
how and when those missions should be carried out.
Following the mission-based approach to command,
(Moffat 2002, 2007) there are 12 discrete missions dened
in the model. They are: Transport, Intelligence, Move,
Engineering, Defend, Reserve, Evacuate, Escort, Presence,
Strike, Secure, and Deny Movement. The majority of these
missions cover general tasks that any entity in the simulation
could undertake (Transport, Intelligence, Move, Engineering, Defend, and Reserve). The other missions are those that
are likely to be specic to either the peacekeeping forces
(Evacuate, Escort, Presence, and Strike) or to the belligerent
factions (Secure and Deny Movement). This is not to
prevent the missions being interchangeable between the
different parties within DIAMOND but to indicate that the
design has focused on providing specic tasks associated
with the principal actors involved in PSO. Each of the
missions is interpreted by the entities that perform them as a
series of activities. For example, the transport mission
consists of the sequence: plan, move, commodity exchange
(ie, load), move, commodity exchange (ie, unload), reserve
(ie, become available for a new task) and communicate (ie,
report to superior commander that the entity is now
available for new missions).
The missions themselves are organised into concurrent
and sequential strings, referred to as plans. For example, a
plan may include a mission to secure an area after which
several transport and presence missions may occur concurrently. The entities undertaking the missions within the plan

J Moffat et alThe Hybrid War Model 7

Plan

Objective

BLUE PARTY
CONCEPT OF OPERATIONS
A group of missions linked by logical initiation conditions.

PLAN A
Mission

A set sequence of activities.

Activity

The smallest divisible action any entity can perform.

Figure 2 Relationship between plans, objectives, missions, and


activities.

report at regular intervals on whether they are succeeding or


failing and their superiors may allocate additional resources
(if they have them) to move failing missions back towards
success. For DIAMOND, the relationship between plans,
missions, and activities is shown in Figure 2.
Monitoring the overall progress of the plan is the Joint
Theatre Commander or his non-military equivalent. His/her
perceptions include a Campaign State Vector indicating
whether the plan is succeeding or failing. Each plan has an
associated set of initiation conditions and end conditions,
which may be time dependent and/or success dependent. If a
plan is failing (or has completed) the commander (or civilian
equivalent) will decide which is the next most appropriate
plan to follow. This sequence of plans forms the partys
Concept of Operations, as shown in Figure 3. The choice of
the sequence of plans adapts as the simulation progresses,
and may, as indicated in Figure 3, end in success or failure.

10. Relationships
In models of warghting, typically only two sides are
represented. This is a suitable assumption for most conventional battles. However, in non-warghting operations this
assumption is not valid, as there are often a large number of
participants, none of whom can be classied purely as hostile
to each other. For example, in the context of Bosnia there
were three main armed factions, their respective civilian
populations and the peacekeeping forces. Very quickly it
becomes obvious that any successful attempt to model nonwarghting operations requires a multi-sided approach. It was
decided that each side in the simulation would be identied
as a separate party and that the relationships between those
parties would be used to describe their afliations, rather than
aggregating like-minded parties into distinct sides.
In accepting that a multi-sided model is required it is
necessary to identify the relationships that will be required to
describe the afliations of each party. Again, in modelling of
warghting, only one type of relationship is modelled, that
of hostility between parties. A greater range of relationships
is required in non-warghting models. A ve-point scale is
used to represent the range: hostile, uncooperative, neutral,
sympathetic (co-operative), and friendly. It was also
recognised that a relationship between two parties does not
have to be symmetric. For example, a NMO may consider its

Figure 3

PLAN B

PLAN C

PLAN F

PLAN D

SUCCESS

PLAN G

PLAN I

FAILURE

Example of a partys Concept of Operations.

Figure 4 Example of combat and the impact of Rules of


Engagement.

relationship with a belligerent faction as neutral whereas that


faction may adopt an uncooperative or even hostile stance in
return. In DIAMOND, we currently assume that a party will
always at least know the stance of other parties towards
them, even if it is an asymmetric relationship, and these
relationships can change within a given simulation run.
Figure 4 shows an example of how combat works within
DIAMOND. The Red armoured units advance into the
node, attacking the civilians and industrial facilities that are
there. They do not attack the medical facilities as their Rules
of Engagement do not permit them to do this. The
relationship between the Red and Blue forces is such that
normally they would not engage each other. However, the
Rules of Engagement for the Blue forces allow them to go
the defence of the civilian population and hence start to
attack the Red forces. As a result of this, the Red forces
switch their attention to the Blue forces since they present
the biggest threat. The combat will end when either of the
forces withdraws. If it is the Blue force that withdraws then
Red will return to attacking the civilians and industry.
The previous example could be modied such that the
Red forces only attacked the industrial facilities. The Rules

8 Journal of Simulation Vol. , No. 0

of Engagement for Blue would not, in this case, allow them


to engage Red as the Red forces were not attacking civilians.

agent is also noted with the coordinates xPrev and yPrev.


Finally, we have the alive indicator that is changed if an
agent is killed by either gunre or a suicide bomb.

11. The HWM


Finally we turn to the HWM development. This closed form
agent simulation model has been developed as part of PhD
research at the University of Craneld (Frewer, 2007). It
takes some of its ideas from both the DIAMOND model
and the MANA model, and represents a tactical urban
environment in which four classes of actor interact in a
number of complex ways. The model is aimed explicitly at
taking account of CAS principles. These four actors are:
Blue peacekeepers; Red insurgents; NGOsrepresented
as one collective actor type; and civilians, also represented as
one actor type. Each of these actors has a broad intent. The
peacekeepers aim to create security and stability through
attacking any insurgency, protecting the civilians and
helping to x supply failures (water and electricity supplies).
The insurgents aim is to disrupt the intent of the peacekeepers. The NGO intent is to provide aid to the civilians
(including xing supply failures), and the civilians aim is to
survive, by avoiding conict areas, and moving to areas
where the basic supplies (water and electricity) are working.
Basic agent properties. Each of the four different types of
agent object has a set of parameters. These differ slightly but
the general structure is the same. First, we have the basic
properties. There is the agent type which is given by a
number between one and four: one represents peacekeeper,
two is NGO, three is insurgent and nally four is civilian.
We then have the number of the squad the agent belongs to.
General parameters for the initial positions of the squad are
also given, these are the x and y coordinates of the home
location, and a radius which determines how spread out the
agents are, as shown in Figure 5.
The current location of the agent is given by the
coordinates xPos and yPos, the previous location for the

Ranges and constraints. The ranges dene a distance away


from the agent in terms of the number of cells. This concept
is illustrated in Figure 6; this example shows cells that are
one cell away and those that are two or three cells away from
an agent. A one step neighbour of the agent cell is thus one
of the eight cells shown. All the agents have a sensor range
which denes the square of cells an agent can detect. For
example, if an agent has a sensor range of one it can only see
the eight cells that surround its current location. In addition,
the peacekeeper and insurgent agents have ranges relating to
gunre. There is a single shot kill probability, SSKP, and the
ring range which is the maximum number of cells away a
target agent can be.
So far we have many parameter types which are similar to
the combat model MANA, but the next set of parameters
differs. The insurgent agents have a probability of unprovoked re and a probability they will set off a suicide bomb.
There is also a radius of damage for the bomb; this species
how much damage a bomb will create if set off. For
example, if there is a bomb radius of two, all occupants of
cells at most two cells away from the bomb site will be killed.
The peacekeepers and NGOs also have probabilities related
to their ability to x the water and electricity supplies,
reecting their technical capabilities.
Movement. The approach used to dene movement of the
agent is based on similar ideas in the MANA model. The
MANA personality parameters or weights which we
referred to earlier, would not have been sufcient, since this
is not a purely combat model, therefore the method had to
be modied.
As shown in Figure 7, when an agent at site i considers
whether to move to site j, it calculates the utility for both of

Key:

Agent

2
One cell away
Two cells away
Three cells away

Figure 5

Example initial location of a squad.

Figure 6

Example cell distances for ranges.

J Moffat et alThe Hybrid War Model 9

Sensor range

Figure 7

Add up
weights within
shaded area for
both i and j
cells

Calculating the utility of a move.

these sites and will prefer to move to the site with the higher
utility. Broadly speaking, the agent will look at each site
within sensor range and add the weights corresponding to the
contents of each site, multiplied by a distance factor. It does
this for both site i and j and moves towards the site where this
utility calculation is higher. Thus although the agent course of
action is set to move, it may decide to stay in the same cell, if
this has higher utility than any other cell it could move to.
In calculating this utility, each factor in the calculation is
assigned a weight by the original agent, with a positive
weight meaning attraction, and a negative weight a
repulsion. Each weight has an integer value between
(100) and ( 100); the higher the value, the more the
agent wants to move towards that factor. These weights
depend on the actor type of the agent (ie, whether it
represents a peacekeeper squad, NGO group, etc).
The following factors are included:
1.
2.
3.
4.
5.
6.
7.
8.

PK peacekeeper;
NG NGO;
IN insurgent;
CI civilian;
CO site containing combat;
CN site containing a civilian in need;
EL site without electricity;
WA site without water.

12. Choice of course of action by the actors


A square grid of cells is constructed, and is split into a
specied number of equally sized square sectors. These are
used to indicate the city blocks. Initially these have a given
supply of water and electricity. When either supply fails in a
sector it fails at all the individual cells in that region. The
initial positions for the agents are then calculated using a
random number generator in conjunction with the relevant
squad home position and radius. The agents then begin to
move and interact with each other, based on their local or
shared perceptions of the locations of other agents, and their

local perceptions of the utility of moving to another location.


The emergent behaviour of the simulation is thus based on
this self-synchronisation (if locations of other agents are
shared) or self-organisation (if locations of other agents are
not shared).
Peacekeepers: The peacekeepers rst priority is to protect
themselves and thus if they are under attack they defend
themselves. Firstly, if there have been any shots red to the
cell in the last timestep, their course of action is set to
combat. Next, if there have been any bomb blasts within
sensor range in the last timestep, their course of action is set
to combat. Their next priority is to help the civilian
population. If repairs to the water or electricity supply are
needed at the cell then the course of action is set to repair,
and if not then the course of action is set to move so that
the peacekeeper squad can head to where it may be needed.
The agent then calculates the utility of cells within its
movement range.
NGO: In the rst instance, the NGO would move away from
danger since they are unarmed. Thus if there have been any
shots red to the cell they occupied in the previous timestep
their course of action is set to move. Next, if repairs to the
water or electricity supply are needed at the cell then their
course of action is set to repair. Otherwise, the agent is set
to move so they can best help the civilians.
Insurgent: First, the insurgents would defend themselves
against any enemy re, so if there have been any shots red
to the cell they occupied in the last timestep then their course
of action is set to combat. If they are not in immediate
danger they would decide whether or not to start conict
according to their suicide bomb and shot probabilities. If the
insurgents in the cell are not involved in combat their course
of action will be set to move.
Civilians: Since the civilians do not carry out repairs and are
unarmed, their course of action is always set to move
(although a move may result in staying in the same cell,
dependent on the relative utility of moving versus staying).
Failure of water or electricity: To determine whether the
water or electricity supply in each sector will fail we use a
random number generator and the probabilities WATERFAIL and ELECFAIL during each timestep of the
simulation. The city block in question may lose one or
other of these supplies, for a period, before being repaired.

13. Example simulation results


As the simulation progresses, the actors evolve their locations
through adaptive and short range local interactions. Figure 8
is a screenshot of actor locations part way through one of the

10 Journal of Simulation Vol. , No. 0

Figure 8 Screenshot of the Hybrid War Model part way


through a simulation.

Figure 9 Statistics of distribution of size of skirmishes for the


most complex scenario (log-log plot with straight line of best t).

simulations. As time progresses, we would expect the peacekeepers and insurgents to self-organise and form clusters of
interaction, which we call skirmishes.

14. Skirmishes
We dene a skirmish as a continuing series of interactions
between peacekeepers and insurgents. A skirmish is started
by the insurgents through a single shot or bomb. This can
result in anything from no response to an exchange of re
involving the whole set of peacekeepers. A skirmish
continues until it peters out with no interaction from either
side for at least one timestep. Two metrics are used; the
number of shots and bombs employed (the size of the
skirmish), and the length of the skirmish in model timesteps
(the duration of the skirmish).
A number of scenarios of increasing complexity were
modelled. In each case, the scenario is run 50 times, and the
results pooled as a cloud. (This is standard practice in
complexity-based analysis.) In our case this is justied
because the skirmishes occur at random throughout the
runs, and can be seen as independent events. For each of
these, statistics were accumulated throughout the 50 runs of
the model, in terms of the distribution of skirmish sizes, and
skirmish durations.
Figures 9 and 10 show the results for the most complex
scenario modelled. These are plotted on a log-log scale to
test the hypothesis that they approximate to a powerlaw
distribution. (The reasons for expecting such powerlaw
behaviour, based on data from real conicts, are discussed
below.) A powerlaw distribution would form a straight line
on such a plot, and the best t straight line is shown on
each plot.
These results (and the others obtained from the model)
indicate a wide spectrum of skirmish sizes and durations

Figure 10 Statistics of distribution of duration of skirmishes


for the most complex scenario (log-log plot with straight line of
best t).

ranging from small to very long durations of interaction.


However, the evidence for emergent powerlaw behaviour is
still inconclusive.
Real-world data appears to show this powerlaw effect
strongly. An analysis was carried out (Johnson and Spagat,
2005) of casualty data for conicts in Colombia and Iraq.
For Colombia, each data point consists of an event in which
casualties occur. For each such event, all conict-related
killings plus injuries are counted, and represent the severity
of that event. For Iraq, each data point again consists of an
event in which casualties occur. In this plot, the severity of
each such event is represented by the total number of civilian
deaths. A powerlaw relationship is shown to be a good t
to the data, with the exponent of the powerlaw adapting
over time. Dobias and Sprague (2009) have also shown that

J Moffat et alThe Hybrid War Model 11

casualty data have such powerlaw statistical behaviour for


such complex PSO problem spaces.

15. Summary
In this paper, we have discussed the ideas of complex and
adaptive systems, and how these ideas relate to the
increasing complex environment of PSO. Having reviewed
the evolution of closed form simulation modelling ideas in
this environment, we described in detail two agent-based
approaches spanning the problem domainthe DIAMOND model at the whole of country perspective, and
the HWM at the local tactical level. Further development
of the HWM is underway, with the aim of producing a
closed form simulation of PSO at the tactical level, which
incorporates the ideas being developed in the STOAT
tactical gaming environment.
AcknowledgementsThe lead author wishes to acknowledge discussion and exchange of ideas on complex endeavours with their
originators: Dr David Alberts (Ofce of the Secretary of Defense,
US DoD, Washington DC, USA) and Dr Richard Hayes (President,
Evidence Based Research, Washington DC, USA).

References
Alberts D and Hayes R (1995). Command Arrangements for Peace
Operations. US DoD CCRP: Washington DC, USA.
Alberts D and Hayes R (2007). Planning; Complex Endeavours. US
DoD CCRP: Washington DC, USA.
Alberts D, Garstka J and Stein F (2003). Network Centric Warfare.
US DoD CCRP: Washington DC, USA.
Atkinson S and Moffat J (2005). The Agile Organisation: From
Informal Networks to Complex Effects and Agility. US DoD
CCRP: Washington DC, USA.
Dobias P and Sprague K (2009). Intermittency of casualties in
asymmetric warfare. J Battleeld Technol 12(1).
Frewer L (2007). The development of an agent based model to
investigate possible power law relationships in peacekeeping
operations PhD Thesis, Craneld University.

Illachinski A (2004). Articial War: Multi-agent Based Simulation of


Combat. World Scientic Publishing Co. Pte. Ltd: Singapore.
Johnson N and Spagat M (2005). From old wars to new wars and
global terrorism Cornell University Library Archive Reference
arXiv:Physics/0506213.
Lauren M (2003). On the temporal distribution of casualties and
determination of medical logistical requirements New Zealand
DTA Report 187.
Lauren M and Stephen R (2002). Map-aware non-uniform
automata (MANA): A New Zealand approach to scenario
modelling. J Battleeld Technol 5(1).
Lauren M, Stephen R and Hore N (2003). Application of the
MANA model to operations in East Timor New Zealand DTA
Report 194.
Ministry of Defence (2002). The Strategic Defence Review: A New
Chapter Cmd 5566. The Stationery Ofce: London, UK.
Moffat J (2002). Command and Control in the Information Age
Representing its Impact. The Stationery Ofce: London, UK.
Moffat J (2003). Complexity Theory and Network Centric Warfare.
US DoD CCRP: Washington DC, USA.
Moffat J (2007). Modelling human decision-making in simulation
models of conict. The CCRP International C2 Journal 1(1): 3160.
NATO (2002). NATO Code of Best Practice for C2 Assessment. US
DoD CCRP: Washington DC, USA.
Prins G (1998). Strategy, Force Planning and Diplomatic/Military
Operations. The Royal Institute of International Affairs:
London, UK.
Schwarz G and Bertsche K (2003). Agent based simulation of (de-)
escalation in Peace Support Operations. In Proceedings of 20th
International Symposium on Military Operational Research (20
ISMOR), UK.
Smith E (2003). Effects Based Operations. US DoD CCRP:
Washington DC, USA.
Smith E (2006). Complexity, Networking and Effects-based Approaches to Operations. US DoD CCRP: Washington DC, USA.
Starr S, Haut D and Hughes W (1997). Developing intellectual
tools to support C4ISR analyses for operations other than war.
In Proceedings of 3rd ICCRTS. National Defense University;
Fort McNair, Washington DC, USA.

Received 28 January 2009;


accepted 3 November 2009 after one revision

You might also like