Decision Support Systems

1
Decision Support Systems

• Real World Applications

2
The abstract problem

• Control personal has to manage a complex system
• Identify problems
• Understand the problems
• Classify
• Explain
• Evaluate problems
• Anticipate consequences
• Solve the problems
• Generate a plan
• Take actions

3
Why Agents?!

• Agents design advantages for control systems
• Easy design - Each agent corresponds to some role
  in the system (very self explaining)
• Abstraction
• Functions ? object ? agents
• Task oriented
• Basic and compound methods.
• Social methods.
• Knowledge based
• The expertise model can be improved
• Reuse Same role at different environment

4
Why Agents?!

• Decision Support Systems interact/replace human
  beings
• Decisions must be understandable to human,
  therefore using agents will yield
• better understanding of each role in the system
• Each role supports the humans
• At any level of expertise
• better understanding of the Logic and
  interactions among the components
• There already is a control structure
• Agents replace the existing structure

5
Problems Characteristics

• A lot of input
• Background work
• Human decision maker at the end
• Task oriented
• Examples
• Energy management
• Traffic management

6
Energy Management

• Power plants generate electricity
• Final consumption takes place far away
• Many things can go wrong in the middle
• Unpredictable problems
• Equipment damage
• Disasters (winds, lightning)
• Predictable problems
• Temperature changes
• Overall demand changes.
• Some damages effect quality while others deny the
  service

7
The Architecture

• Based on a network of a company in Spain
• Networks are managed from a control room
• Information is sent to the control room
• Protection equipment can be remotely operated
• Field engineer operate in the field
• The network consists of substations, and each
  substation consists of
• Lines
• Breakers switches
• May fire automatically, sending alarm messages

8
The Goal

• Main Problem
• Usually caused by short circuits in the lines
• Malfunctioning equipment may cause a chain
  reaction that extends the area of effect
• Solution
• Isolating the effected area usually solves the
  problem
• The goal
• Minimize the disconnected area
• restore supply as soon as possible

9
The electricity transport management problem

• Control personal has to manage a complex system -
  control the switches and breakers
• Identify malfunctioning in switches and breakers
• Understand the problems
• Classify - Diagnose the problem
• Explain the alarm messages according to the
  diagnosis
• Evaluate problems
• Anticipate consequences that may cause expansion
  of the area of effect
• Solve the problems
• Generate a switching plan that isolates the area
  of effect and restore supply to maximum number of
  customers

10
The Multi-Agent Architecture

• Constraints
• Existing expert systems
• Existing configuration of the data transmission
• Two formats
• Non chronological alarm messages NAM
• Chronological alarm messages CAM
• Existing control structure

11
The Multi-Agent Architecture

• Alarm Analysis Agents
• Replaces an existing expert system
• Methods
• Reads messages
• Detects faults
• Establishes hypotheses regarding the
  malfunctioning equipment
• Basic methods compound methods
• Rule based

12
The Multi-Agent Architecture

• Control System Interface Agent
• constitutes the applications front end to the
  user
• Basic methods
• Acquires and distributes network data to other
  agents (formats the message for use by other
  agents)
• Done using a hard-wired algorithm
• Calculates the power distribution, given a
  certain state
• Done using a numerical simulator
• A compound method which is used when a certain
  set of messages arrive
• A social method which generates classification
  with the help of the alarm analysis agents
• This agent wraps existing functionality

13
Example of TMST
CSI
Information Model
Messages
Disturbance Detection
Classify Situation
Alarm Classification
Alarm Detection
Acquire Data (direct algorithm)
Coordinate classification
Alarm Analysis Agent
Alarm Analysis Agent
14
Additional Agents

• Blackout Area Identifier
• Determines the results of a given scenario
  (network state and faults)
• Rule based
• Service Restoration Agent
• Proposes a switching plan given alarm messages
  and the results of the diagnosis
• User Interface Agent
• Serves as an interface between the multi-agent
  system and the users for presenting data
• Browse through the lists of alarms
• Display results of diagnosis along with
  explanations
• Sets up guidelines for the other agents
• Simulates the effect of a restoration plan

15
Coordination

• Can be done with an acquaintance model
• Frames that contain the methods that the other
  agents can perform including
• The types of the methods
• The competence with which the method can be
  applied

16
Summary

• The energy transport problem is very suitable for
  DSS
• Every agent decision may be explained to the
  responsible engineer using the trace of the
  reasoning methods
• Problem definition fits into the abstract problem
  definition
• The multi-agent system managed to cope with the
  existing constraints

17
Road Traffic Management

• Traffic flows on public roads increase at high
  rate
• Number of vehicles increase
• Roads infrastructure cannot be expanded
• Significant economic loses
• Traffic Control Centers (TCC)
• In charge of managing urban transport

18
Available Information

• Messages from human observers
• Gal-Galatz
• Policemen
• Devices
• TV cameras
• Cellular phone
• Sensors
• Loop detectors -Installed on strategic channels
• Speed - mean velocity of the passing vehicles
• Flow - average number of vehicles per unit of
  time
• Occupancy - average time that vehicles are
  spotted

19
Available Control Devices

• Variable Message Sings (VMS)
• Installed above the road
• (like those on the way to Tel-Aviv)
• Traffic signs (closed road sign)
• Arbitrary message signs
• Traffic lights
• Parameters of the traffic light can be modified
• Relative amount of green time
• Overall length of a cycle
• Order of traffic lights

20
The Urban Highway Traffic Control Problem

• system Control the traffic lights and VMSs
• Identify and locate problematic situation
• Understand the problems
• Classify the cause of the problem
  (congestion/accident)
• Explain the problem in terms of traffic flows
• Evaluate problems
• Anticipate consequences due to chain reactions of
  the congestion
• Solve the problems
• Generate a legal sign plan and/or traffic lights
  handling plan, in order to eliminate or alleviate
  the congestion

21
The Multi-Agent Architecture

• The structure of the system was dictated by the
  way human operators worked
• Problem areas topology
• All agents share the same architecture and the
  same reasoning structure
• Their knowledge however, was based on the
  specific problem area in their responsibility

22
Basic Methods of the Agents

• Data abstraction
• Determines qualitative measure for different
  variables
• Problem Type identification
• Takes the data generated by the data abstraction
  method and classifies the underlying problem
• Done by matching the data against problem
  scenario frames
• Demand estimation
• Calculate the normal demand for a section of
  the network
• Based on temporal pattern (hour, day of week,
  events...)
• Effect estimation
• Anticipates the effect of flows on a certain
  problem
• The state of the control devices
• Contribution of certain routes to the problem
• Signal plan selection
• Short term prediction estimation
• Calculates the effect of change in traffic flows

23
Compound Methods

• Heuristic classification
• Problem solving method
• Acquires relevant information
• Problems type are matches upon the information
• The problems are integrated and refined
• Contributor differentiation
• Determines how much a set of causes contributes
  to a problem
• Identifies possible contributors
• Estimates each contributor

24
Compound Methods

• Generate Test
• Evaluates proposals generated by the basic method
  until an adequate plan is found
• Depends on outside constraints (coordination)
• Local management
• Manages the network by integrating all the
  methods
• Identifies traffic problem
• Diagnoses its causes
• Generate a proper plan to overcome it.

25
Coordination

• Problem areas are not disjoint
• Physical conflicts
• Logical conflicts
• Two coordination solutions
• Coordinator agent
• Peer-to-peer communication
• Acquaintance model
• Does not represent information concerning method
  of other agents
• Describes the resources that acquaintances
  require and which effects they may have (on
  sections in the agents problem area)
• Local plans are sent to the relevant agents
• The agent with the most severe problem takes
  precedence

26
Summary

• Once again a DSS is a very suitable solution
• The traffic management problem fits the abstract
  DSS problem
• The DSS had to be based on existing control
  engineers understanding of a towns traffic
  behavior

27
Additional Potential Examples

• Intelligence Word
• Medicine
• Every other problem that fits that abstract
  problem definition