2010-06 Plans validation using DES and agent-based simulation · 2010-06 Plans validation using DES...

Calhoun: The NPS Institutional Archive

Faculty and Researcher Publications Faculty and Researcher Publications

2010-06

Plans validation using DES and

agent-based simulation

Soo, Andy Ong Kim

Proceedings of the 15th International Command and Control Research and Technology

Symposium (ICCRTS) 2010 "The Evolution of C2".

http://hdl.handle.net/10945/36611

Plans Validation Using DES and

Agent-Based Simulation

Andy Ong, Teck Hwee WongDefence Science and Technology Agency (DSTA)

Christian J. Darkens, Arnold H. BussNaval Postgraduate School (NPS)

15th ICCRTS

Agenda

Introduction Methodology

Overall Design Agent-based Plan Generation DES-based Simulator Design

Prototype Implementation DES-based Simulator Agent-based Strike Plan Generation

Experiment & Results Future Work

Introduction

Military plans validation is typically a long drawn process.

Require planners to validate plans using

Anticipated scenariosMilitary exercises

Image Source: http://www.ibiblio.org/hyperwar/USMC/V/maps/USMC-V-16.jpg

Basic Concepts

Discrete Event Simulation Engine

Agent-Based Adversary Model

Red Plan

(DES Model)

Blue Plan(DES Model)

Blue Plan/Capability

Interactions

Outcome

Methodology

Main Building Blocks

Discrete Event Simulation Engine

Agent-Based Adversary Model

rike Plan

Model)

Air Defence Plan & Cap

Messaging & Event Handler

Display Component

Air Defence Plan & Cap,Air Strike Cap

Notifications Relevant to Aircraft Agents

All Events

Agent-Based Model Architecture

Central Tasking Agent

Evaluation Agent

Executing Agent

Database

Action / Reaction

-aluation

Level 1

Level 2

Level 3

AircraftAircraftAircraftAircraft

Air Formation(Evaluation Agent)

Air Formation(Evaluation Agent)

Main Formations Controller(Central Tasking Agent)

Air Formation(Evaluation Agent)

Air Formation(Evaluation Agent)

Aircraft(Executing Agent)

Aircraft(Executing Agent)

Aircraft(Executing Agent)

Aircraft(Executing Agent)

Agent Architecture

DES-Based Simulator Design

ener nt ph ects

Prototype Implementation

DES Engine

Modeling SAM Sensor

SAM Sensor Event Graph

SAM System Event Graph

SAM Site Event Graph

FireMissile

CheckRange WithinRangeMissileSelfDestruct Adjudicate

Chase

SamSite If missile is within max range

If missile is out of range

tc

treload

If missile launcher is

empty and a reload is required

If missile launcher is not

empty

If missile launcher is not

empty

Reload

Gun Site and Gun Event Graph

Agent Model

Agent-based Model

gent model approach on air strike:Plan GenerationApproach vector generationA route generation

Weighted Map (Threat Map) Cell Based Decomposition - Grid A-Star optimize search algorithm

Plan Adjustment Individual aircraft as agent

Employ maneuvers

Approach Vector

Approach Vector ComputationDistance over the Air Defense coveragesExpose time over the composite Air Defense

Coverage'sSpeed of AircraftBased on scoring

Route Generation

Cell-based decomposition or real world abstractionReal World area is 40km by 40kmEach cell is represented as a pixel = 200 meters real worldTotal cells abstraction = 40000 cells

Agent Behavior

Agent behavior mechanism Input messages from simulator affecting the state of the agent

environmentAgent responses to incoming messages from simulator

Alert agent aircraft of Radar lock onDAR LOCK ON

Evasive ActionAlert agent aircraft of Anti-Airgunfiring

TI-AIR GUN ING

Evasive ActionAlert agent aircraft of Incoming surface-to-air missile

OMING SA SILE

Alert agent aircraft of Radar lock offDAR LOCK OFF

Update Agent Status (Aircraft status & Environment changes) Strike Action (Depending on Situation)

Positional updates of the agent in the simulation environment

SITIONAL

Agent Action Messageut Messages to Strike Aircraft Agent

Experiment & Results

Scenario

Air strike on protected site Attacker has knowledge

of the defence layout Attack plan generated

by Agent based tool Operational Analysis

Question: “How sensitive is our

attack plan to variation in the weapon systems?”

Design of Experiment

15 potential main effects65 design points NOLH was used50 replicas for each design pointOverall 65X50 = 3250 runs

For comparison, Full factorial 2 level design

2^15 = 32768. Overall 32768 * 50 = 1,638,400

runs Most factors are continuous, or

discrete with more than 2 levels …

100

200250

SAM

Loa

ding

Tim

e

1357

Mis

sile

Per

Laun

cher

5

7

9

SAM

Min

Ran

ge

4070

100130

SAM

Max

Ran

ge

5.5

6.5

7.5

SAM

Max

Spee

d

2

57

SAM

Enga

gem

ent T

ime

50

6575

SAM

SSK

P

4

6Gun

Rea

ctio

n Ti

me

25354555

Gun

Loa

ding

Tim

e

1

3

Gun

Min

Ran

ge

29323538

Gun

Max

Ran

ge

500700900

1100

Gun

Rat

eof

Fire

300Gun

agaz

ine

Size

Scatterplot Matrix

Partition Tree

SquareSquare Adjoot Mean Square Errorean of Responsebservations (or Sum Wgts)

0.8479530.8014080.1382963.836615

65

Summary of Fit

Regression Model

Analysis

The gun parameters has minimal affect on the outcome. It has reflected the scenario pretty well (recall that the agent generated plan avoids the AA gun).

The SAM Max Range and SAM Reaction Time are the two key main effects.

Analysis (Cont’d)

The mission should be reconsidered if the attacker is concerned about aircraft attrition and there is great uncertainty about the following:

SAM Max Range (preferably less than 109.4) SAM Reaction Time (preferably greater than 37.7)

Future Work

Future Work

Discrete Event SimulatorSensors model can be refined to reflect more

realistic characteristics Enhancements of sensor footprint of irregular

shapes Modeling sensor detection/undetection time

using the glimpse modelAir defense model can be made more

complex

Future Work (Cont’d)

Agent modelEnhance Route generation by adding

additional cost factors such as duration of exposure to threat

Implementation of a dynamic area of operation for individual air formation

Provides a realistic terrain model such as DTED map or vegetation information

Individual agent can be enhance further to include a Neural Net or a Bayesian network

Thank You