Intelligent Agents for Supply Chain Coordination

1
Intelligent Agents for Supply Chain Coordination

• Professor Abhijit Deshmukh
• FARMS Laboratory
• Department of Mechanical Industrial Engineering
  
• University of Massachusetts
• Amherst, MA 01003
• Funded in part by NASA Ames, AFOSR/WL and GM
  Powertrain

2
Supply Chain Characteristics

• Competing goals and objectives
• Dynamic operating conditions
• Need for cost reduction and on-time delivery
• Emphasis on responsiveness to design changes

3
Distributed Solution Search

• Online bidding for airline tickets
• Also serves as a yield management tool for
  airlines
• Purchase price based on perceived utility
• Utility functions may not be explicitly known

4
Communicating Autonomous Agent Architecture

• Agents represent each entity in CA3
• Decisions are based on interactions between
  agents
• Agents have local information
• Search for mutually acceptable solution

5
CA3 in Action

• Customer demand generates the top level request
• Request for bids propagates downwards
• Costs are resolved bottom-up
• Suppliers are found using YP/WP services
• Virtual negotiation table serves as a venue for
  auctions

6
Supply Chain Agent
7
Agent Representation

• Agents map external environment variables on to
  an internal representation of the world
• Internal representation and goals define the
  operating characteristics of an agent at a point
  in time
• Internal objective functions may not be
  consistent with the overall system objectives

• Prices
• Order quantity
• Delivery dates
• Quality
• Past performance
• Products and services

• Objective function
• Costs
• Commitments
• Requirements
• Past performance
• Variable constraints

External
Internal
8
Auctions

• Provide a flexible negotiation framework
• Provide matching and price setting schemes for
  buyers and sellers
• Promote automated negotiation
• Closed or open
• Currency or barter
• Standardized or individualized evaluation
  criteria
• Performance of certain single stage auctions is
  known (Bertsekas, 1992)
• Iterative auctions have been studied in game
  theoretic context (Aumann, 1995)

9
Bid Construction Processes
max Ui(Pi(t))
Pi(t) s.t Pi(t)xi(t) gt
maxCi(t)xi(t) , Ii(t)xi(t) Ui() a multi
attribute utility function for agent i,
defining its goals xi(t) is the demand
vector for goods at time t, Pi(t) agent is
estimated price vector at time t, Ci(t )
cost of producing goods at time t Ii(t)
cost of not producing goods at time t
10
Auction Rules

• In single buyer case, select the highest utility
  bid
• In multiple buyer case, match the highest seller
  to the lowest buyer
• Trading price (highest sellerlowest buyer)/2

Seller
Buyer
11
Continuous Cost Recovery Schemes

• Resource leveling using price incentives
• Inventory and overtime cost reduction passed on
  to customers
• Reduces variability amplification in supply
  chains (a.k.a. bullwhip phenomenon)

12
Price Incentive Functions
Supplier Price exp(0.1Deviation) Custome
r Price 2atan(exp(0.1Deviation))
13
Strategic Alliance Formation
Emergent alliance formation between supply chain
entities

• supplier alliance
• customer alliance
• hybrid alliance

14
Cooperative Negotiation
15
Performance of Negotiation Schemes

• SWARM simulation results
• Model 1 Hierarchical control
• Model 2 Agents using a myopic policy
• Model 3 Multi-attribute utility model for agents
• Model 4 Alliance formation in agents

16
Dynamics of Supply Chains

• Centralized systems
• high variability
• latency in control
• Agent based systems
• reduced lag time
• convergence not guaranteed

17
Summary

• Agents based systems offer a potential solution
  for complex supply network coordination
• An important advantage is the reduction of
  repetitive tasks
• Convergence characteristics of agent negotiation
  mechanisms need further investigation
• Potential for integrating design and supply chain
  activities over the internet