Transcript Using Semi-Automated Decision Algorithms in Military Operations

TRW
General Purpose
Decision Aids
Using Semi-Automated
Decision Algorithms in
Military Operations
By
Dennis R. Ellis
[email protected]
TRW Systems
1
PHOENIX CHALLENGE 2002
February 20-22, 2002
New Mexico State University,
Las Cruces, NM
General Purpose
Decision Aids
TRW
General Purpose Decision Aids Overview
• The goal is a set of configurable, flexible,
semi-automated decision support tools.
• Support for multi-echelon, joint services
command simulation at the operational and
strategic levels for the IO/IW, AOC, NMD,
TMD, and Strategic Deterence missions.
• Based on Joint Operations Planning and
Execution System (JOPES) concepts.
Decision Makers
2
Decision
Algorithms
Operational
Databases
• Designed for maximum scalability and
parallelism as a distributed system.
Doctrine
• Supports parallelism, shared memory, and
multiple processors.
Coordinated
CONOPS
• Includes Data Mining, Dempster-Shafer
Belief Network, Genetic, Optimal Policy,
and Case Based Planning algorithms.
General Purpose
Decision Aids
TRW
Spectrum of Options for Approach
Our focus has been in this area
Practical
Implementation
at AOCs
Basis for multi-mission synergy
TADIL-J
is a start
Message
passing
Operational
Cells
Co-dependent
Missions
Smart
Collaboration
Collaborative
Agents
Degree of Integration
Tight
System of
Systems
Loose
NMD/TMD/AOC
Wargames
Joint
Exercises
Operator
Dialog
Passive
Observance
Single
System
3
Common
Decisions
Non-Interacting
Systems
General Purpose
Decision Aids
JOPES-based Deliberative Processes
TRW
Joint Operations Planning & Execution System
Mission
Course of
Detailed
Situation
Repeat
Execution
Action Selection
Planning
Assessment
if necessary
- Manage Forces
- Force Survivable
- Weather
- Target Viability
- Threats
- Identify Goals
- Filter Preplans
- Selects COAs
- Defensive Strategies
- Offensive Strategies
- Integrate/Deconflict
- Weapon Allocation
- Asset Tasking
- Joint Scheduling
- Define C2 timeline
- Satisfy Constraints
- Optimize Plan
Typical Mission
4
- Adjust Status
- Set Engagement
- Assign Assets
- Monitor Strikes
- Enforce Timeline
- Assess Mission
- Plan Update
- Redirect Forces
- Reconstitute
TRW
Decision Centered Approach
General Purpose
Decision Aids
• Decision = State Transition
Current
State
Legend: known
Plan
NEW!
Assess
unknown
Desired
State
Transition
- Current State, Transition (Plan & Assess) , and Desired State may not be known
- This paradigm has been shown to neatly sort decisions, algorithms, and displays
Representation:
5
Decision Categories:
Military Equivalent:
Clarify Current State
Situation Analysis
Clarify Desired State
Determine Objective
Define Alternatives
Develop COAs
Execute/Assess Plan
Combat Assessment
Decision Tools
General Purpose
Decision Aids
TRW
• Our decision tools span the full spectrum of military decision making,
including Effects Based Operations assessment feedback.
• Background Tasks
C&C
Fog-of-War
Mission
Simulator
Decision Tools
Controller
• Foreground Tasks - Decision Toolkit
6
Deliberate Planning
• Case-based Planner
• Planning Horizon
• Terrain Reasoning
• Target Designation
Mission Planning
• Mission Previewer
• Engagement Planner
• Optimize Effectiveness
• Schedule Repair
Problem Recognition
• Profile Generation
• Clustering
• Rule Induction
Element Management
• Force Management
• Force Survival
• Target Viability
Execution Monitoring
• Timeline Enforcer
• Rule-Based Execution
• Effectiveness Metric
• Plan Update Trigger
Assessment
• Dempster-Shafer
Belief Network
General Purpose
Decision Aids
Interleaved Planning & Execution
Legend:
Planning Thread:
TRW
Function
Technology
Determine
Course of Action
Assess Mission
Readiness
Optimize
Plan
Preview
Mission
Case-Based
Reasoning
Data
Mining
Genetic
Algorithm
Map-Based
Visualizer
Execution Thread:
7
Assess Mission
Effectiveness
Perturb
Environment
Monitor
Timeline
Update
Plan
Belief
Network
Fog-of-War
Module
Timeline
Visualizer
Optimal
Policy
General Purpose
Decision Aids
Situation Diagnosis
• Dempster-Shafer Networks
TRW
It is crucial that situations are properly
diagnosed because military decisions that
adversely affect lives, property, and the
environment are based on the assessment
of the situation, making the decision
inherently risky.
• Information Operations
• Anti-Terrorism Operations
• Combat Assessment
• Weather Forecast
• Space & Terrestrial Weather
• Time Critical Target Assessment
• Foreign Launch Assessment
Evidence
Task
to
Strategy
Belief in
Hypotheses
Time
8
Under intense time pressure, people often:
1) Refuse to decide
2) “Jump to conclusions”
3) Ignore potentially important evidence
4) Have difficulty understanding data
correlation
The data:
- Is usually uncertain, incomplete, ambiguous
- Possibly conflicting
- Arrives asynchronously
General Purpose
Decision Aids
Evidence about
associated hypothesis
9
Information Ops - Operator View
TRW
Belief in evidence supporting associated hypothesis below blue line.
Disbelief in evidence supporting associated hypothesis above red line.
Amount of ignorance about associated hypothesis between lines.
General Purpose
Decision Aids
• Data Mining
• Weka
• Subdue
Pattern Discovery
TRW
Given a large, dynamically changing
database, a decision-maker must
understand relationships among data
fields, guard against corrupt data, and
be alerted to changes in the
underlying structure of the data.
• Weka uses Quinlan’s C5.0 Classifier
to find underlying structure and build
a rule tree that identifies “outliers”.
• Inconsistencies, omissions, and
conflicts are explicitly shown.
• Weka uses statistical techniques for
data extraction:
Our
Efforts
10
- Rule Induction
- Neural Networks
- Regression Modeling
- K Nearest Neighbor Clustering
• Subdue uses a Minimum Description
Length (MDL) algorithm to compress
and hierarchically cluster.
General Purpose
Decision Aids
TRW
WEKA - Rule Induction View
Rules that determine if a
threat Surface-to-Air site
is operational or not
Legend:
OPR -> Operational
NOP -> Non-Operational
Lat > 39.4
59 OPR
2 NOP
Lat > 39.1
Lon < 127.2
31 OPR
2 NOP
Lat < 39.1
Lon < 127.2
30 NOP SA-2s
Lat > 38.5
Lon < 127.2
1 OPR SA-3
30 NOP SA-3s
11
Lat < 39.4
Lon > 127.2
2 OPR
56 NOP
General Purpose
Decision Aids
Strategy Determination/Detailed Planning
TRW
• Course-of-Action Development
The characteristics of this decision type are:
1) A decision-maker is given broad objectives that result in an ill-posed problem,
2) Exponential explosion of possible solutions,
3) Many dimensions and/or criteria, and
4) Difficulty in quantitatively rating the worth of a strategy.
When faced with these circumstances, people often attempt to:
1) Recall a course-of-action that worked in previous similar circumstances,
2) Caucus to arrive at a consensus (i.e., group-think), or
3) Identify a dominant criteria and assemble a strategy around it.
• Detailed Plan Development
Because of the complexity and size of the planning space and the number of
possible combinations of factors and constraints that must be considered during
the planning phase, manual planning methods, and most current automated
methods, are too narrowly focused, inflexible, time consuming, and not scalable.
12
ActiveOptions
Detailed Plan
ActiveTargets
Option, Target
...
Name
Weapon
Arena
Type
Plan
Optimization
JCS
Order
Name
Locale
Location
Type
Value
Hardness
Mobile
Constraints
Selection Criteria
COA
Planner
COA
Case Base
13
TRW
COA/Detailed Planning Overview
General Purpose
Decision Aids
Target
Assessment
TargetStatus
Name
Type
Arena
Weapon
COA Name
Keywords
Viable Area
Targets
ResponseOptions
Dynamic Status
Target Selection Process
General Purpose
Decision Aids
Strategy Determination - COA View
TRW
This screen displays the “goodness-of-match” of the input Planning Order against
previously stored cases to aid the analyst in generating a new Course-of-Action.
14
General Purpose
Decision Aids
Detailed Planning - Operator View
TRW
This screen allows the analyst to evaluate the effectiveness of additional plans and
instantly see the diversity and details of each plan.
15
Summary
General Purpose
Decision Aids
TRW
We have refined, tailored, and demonstrated innovative decision tools that span
the full spectrum of JOPES command and control decision-making in these areas:
• Situation diagnosis using a Dempster-Shafer Belief algorithm
• Pattern discovery using Data Mining algorithms
• Strategy determination using a Case-Based Reasoning algorithm
• Detailed planning using a Genetic Algorithm
We have demonstrated the applicability of these algorithms and support tools in
multiple mission domains:
• IO/IW Defensive Counter-Information
• Aerospace Operations Center (AOC)
• Anti-terrorism
• Strategic Deterrent Forces (SDF) and National Missile Defense (NMD)
16
For more information, contact the author at [email protected]