270 likes | 412 Views
Automated Negotiation in Supply Chain Management Using Multi-Agent System. Masabumi Furuhata University of Western Sydney Computing and Information Technology 22.08.2005. Outline. Research Objectives Research Overview Motivation Prospected Advantages of the Research Model
E N D
Automated Negotiation in Supply Chain Management Using Multi-Agent System Masabumi Furuhata University of Western Sydney Computing and Information Technology 22.08.2005
Outline • Research Objectives • Research Overview • Motivation • Prospected Advantages of the Research Model • Industrial Benefits • Research Model • Conclusion • Future Works • Relevant Research
Research Objectives • Internal and external automated negotiation model and algorithm establishment in supply chain management area using multi-agent system techniques. • We suppose that organizational issues and understanding market equilibriums in automated negotiation market are most important things to realize the model.
Negotiations in Supply Chains • External negotiation: inter-corporate negotiation • Internal negotiation: inner-corporate negotiation, among agent clusters and between agents in the same agent cluster Customer A Competitor x Supplier 1 CORPORATION Customer B Supplier 2 Customer C Competitor y Supplier 3 Legend Company Agent cluster Agent
Motivation • Information technology coverage of current best practices in supply chain management is limited. It is no doubt that there are requirements to extend the area of IT coverage. • Some of planning and transaction is automated. • For strategy development, target setting and action plan development, IT used as support tools or decision support systems. • We spend too much time on solving problems for the exceptional situations. • Generally speaking, staffs spend their 80% of their time for correspondence to the 20% of irregular situation, and they spend the rest of the time for 80% of the normal situation. • Are we ready to connect to a real-time market? Strategy Target Action Plan Planning and Transaction IT coverage
Prospected Advantages of the Research Model • Reduction of the manual negotiation among planners and back-office staffs. • Agility in response to market environmental change. • Deficit planned inventory and automatic adjustment, dynamic pricing, etc. • More solid planning with putting off the planning confirmation deadline. • Agents run with assuming that the prior planning results contain probability. Moreover, unlike the centralized model, the distributed model runs with limited information. Therefore, a successive agent dose not always require results from a prior agent. • Quick entrance to new location and product area. • Reduction of planners and executers learning time.
Industrial Benefits • Using our platform, we can analyze the market behaviors of e-trading market: • Dynamic pricing • Choosing competitors’ strategies • Changing market conditions, such as interest rate, demand, number of competitors, BOM, production lead-time, distribution lead-time, storage cost, and etc.
Research Model • Agent Definition Level • Basic Behavior of Agent • Agent, Organization, KPI, and KGI • External Negotiation Process • Internal Negotiation Architecture
Agent Definition Level Department A Department B • In the research, we define agents as small particles. • For example, the level of sales agents is equivalent to the multiplied dimension of (product) x (customer) x (distribution channel). • Compared to centralized agents, we have more transactions among agents, but there are many advantages. • Distributed agents are able to map to many different type of actual organizations easily. • Unlike the centralized system, we do not need the global supply chain parameter settings by super planners. This type of the people do not exist in the most companies. Actual Organization Organizational Unit Agent cluster IT model Agent
Basic Behavior of Agent • Functions of agents are event driven. • When agents are kicked by an event, each agent gets datum, common knowledge, from the blackboard to comprehend the situation. Here, all datum that are able to share among other agents are saved on the blackboard. • To determine the preference among decisional options, each agent gets KGI (key goal indicators, ex. sales, resource utilization, etc.) from their belonging organization and KPI (key performance indicators, ex. order fill rate, inventory turn over, etc. ). • Agents make decisions according to common knowledge, KGI and KPI. Get common knowledge from blackboard Start of event Get KGI (Key Goal Indicator) Execute plan or transaction End of event Get KPI (Key Performance Indicator)
Agent, Organization, KPI, and KGI KPI (Key Performance Indicators) Department A Department B • Each agent belongs to one organizational unit. • Each organizational unit has some key goal indicators. • Each agent gets some KGIs from its belonging organizational unit. • Some KPIs cover different departments, therefore they are effective to different agents clusters. • Agents’ autonomous behaviors are based on KGIs, and coordinating behaviors are based on KPIs. • If autonomous decision makings are not feasible, then agents make reasonable decision with coordination rules. Key Goal Indicator Organizational Unit Agent cluster Agent
External Negotiation Process Customer Supplier Demand forecast Request for proposal Sales offer Purchase order Purchase order (update) Available-to-promise Available-to-promise (update) Advanced ship notification Advanced ship notification (update) External negotiation Delivery Payment
Internal Negotiation Architecture KPI Sales Department Logistics Department Sales agent cluster Transportation Department Transportation agent cluster Logistics agent cluster Production agent cluster Purchase agent cluster Production Department Purchase Department Legend Agent cluster Agent Organizational unit KGI KPI
Functional Example – Sales Offer Generation - Sales agent RFP receiving start Get common knowledge from blackboard Generate offer RFP receiving end Offer sending start <RFP> ID Customer Product Quantity Due Date Price Penalty <RFP> <Offer> <Sales> <Inventory> <Market Data> <Forecast> <Offer> ID Customer Product Quantity Due Date Price Penalty <Offer> ID Customer Product Quantity Due Date Price Penalty Get KGI (Key Goal Indicator) <KGI> Get KPI (Key Performance Indicator) <KPI>
Functional Example – Purchase Delinquency Recovery - Logistics agent Purchase parts delinquency info receiving start Get common knowledge from blackboard Check parts inventory allocation options Purchase parts delinquency info receiving end <Inventory Allocation Option> Option Date Stock Point Part Quantity <PO> ID Supplier Ship-to Product Quantity (original) Quantity (new) Due Date (original) Due Date (new) Price Penalty Get KGI Get KPI Production agent Option Plans Negotiation Get common knowledge from blackboard Check production plan options <Production Plan Option> Option Date Stock Point Product Quantity Get KGI Get KPI Sales agent Get common knowledge from blackboard Determine sales order preferences <Sales Order Option> ID Customer Product Quantity Due Date Get KGI Get KPI
Future Works • Algorithm development to comprehend KGI and KPI. • Mathematical representation. • Generalization. • Agent coordination mechanism development. • Especially for the case that some agents have to concede their benefit to satisfy the constraints. • Internal negotiation model development. • General model. • Industry specific model. • Assembly industry. • Chemical industry. • Automotive industry. • External negotiation model development.
Future Works • Simulation analysis on external market conditions and reasonable market behavior. • Competitiveness: number of competitors, and market share. • Lead-time pressure: delivery lead-time, production lead-time, customer expected lead-time. • Interest rates: bank interest rates. • Demand fluctuation: average demand, variance, and probability distribution function.
Relevant Researches • MASCOT (Multi Agent Supply Chain COordination Tool): • N. Sadeh, “MASCOT: An Agent Architecture for Multi-Level Mixed Initiative Supply Chain Coordination,” Internal Report, Intelligent Coordination and Logistics Laboratory, Carnegie Mellon University, 1996 • ANTS (Agent Network for Task Scheduling): • J. Sauter, H. Parunak, and J. Goic, “ANTS in the Supply Chain,” the Workshop on Agents for Electronic Commerce at Agents '99, Seattle, WA, May 1-5, 1999 • ISCM (Integrated Supply Chain Management): • M. Barbuceanu and M. S. Fox, "Coordinating Multiple Agents in the Supply Chain", Proceedings of Fifth Workshop on Enabling Technologies: Infrastructure for Collaborative Enterprises, Stanford, CA, IEEE Computer Society Press, pp 134-142. 1996
MASCOT • MASCOT (Multi Agent Supply Chain COordination Tool) • Blackboard architecture: Knowledge Sources (KS) and Blackboard • Functionalities: • Coordination • Integration with heterogeneous plans and scheduling module • Mixed-initiative decision support • Alternative problem instances and solutions • Selective problem definition • Controller of the module visualization:
ANTS • ANTS (Agent Network for Task Scheduling) • Unit Process Broker (UPB): • Part Broker (PB): • Resource agent: • Supplier agent: • Customer agent: • Market architecture:
ISCM • ISCM (Integrated Supply Chain Management) • Function agents: • Order fulfillment • Logistic resource management • Transportation resource management • Production resource management • Dispatching • Scheduling • Information agents: • Central communication • Knowledge management • Conflict solving • Coordination support