A new approach of coordination of decisions in a multi site system is proposed. It is based this approach on a multi-agent concept and on the principle of distributed network of enterprises. For this purpose, each enterprise is defined as autonomous and performs simultaneously at the local and global levels. The basic component of our approach is a so-called Virtual Enterprise Node (VEN), where the enterprise network is represented as a set of tiers (like in a product breakdown structure). Within the network, each partner constitutes a VEN, which is in contact with several customers and suppliers. Exchanges between the VENs ensure the autonomy of decision, and guarantiee the consistency of information and material flows. Only two complementary VEN agents are necessary: one for external interactions, the Negotiator Agent (NA) and one for the planning of internal decisions, the Planner Agent (PA). If supply problems occur in the network, two other agents are defined: the Tier Negotiator Agent (TNA) working at the tier level only and the Supply Chain Mediator Agent (SCMA) working at the level of the enterprise network. These two agents are only active when the perturbation occurs. Otherwise, the VENs process the flow of information alone. With this new approach, managing enterprise network becomes much more transparent and looks like managing a simple enterprise in the network. The use of a Multi-Agent System (MAS) allows physical distribution of the decisional system, and procures a heterarchical organization structure with a decentralized control that guaranties the autonomy of each entity and the flexibility of the network.
Franco Guidi-Polanco, Claudio Cubillos, Giuseppe Menga
This work presents our agent-based architecture for the development of Global Automation Systems. These systems consist of software applications that manage all the processes in a network of enterprises, in distributed, decentralized and autonomous way. These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.
We study a two-level system having N local systems in the lower level subordinate to a central system in the higher one, such that both central and local systems have decision-making units. The central system is a coordinating agency and the local ones are semi-autonomous operating devisions. The basic principle of planning for this organization is that the central system allocates resources so as to optimize its own objective, while the local ones optimize their own objectives using the given resources. A local objective function, fn, is a function of the lower level decision variable vector x=(x1,・・・, xN) and the higher level one a=(a1,・・・, aN), where an is a resource vector allocated to the local system n. Since the functions ■ are mutually independent, the lower level composes a multi-objective system, in which the lower level decision-makers minimize a vector objective function f =(f1,・・・,fN) with respect to x in cooperation with each other. Thus, the lower level generates a set of noninferior (i.e. Pareto optimal) solutions ■(a) being parametric with respect to a. The central decision-maker, then, chooses the optimal resource allocation a⁰ and the best noninferior solution ■⁰ corresponding to a⁰ from among a set of ■(a). The above problem becomes a decentralized two-level optimization, when the local system contains only its own variables (xn, an). Several theorems and iterative algorithms for the formulated problems are obtained by use of mathematical programming techniques.