Agent Based Modelling and Simulation of Intelligent Distributed Scheduling Systems

MILAGROS ROLÓN; MERCEDES CANAVESIO; ERNESTO MARTINEZ

Computer-Aided Chemical Engineering

Elsevier B.V.

Lugar: Amsterdam; Año: 2009; p. 985 - 990

Keywords: Agent-based modeling, Distributed scheduling, Interaction mechanisms, Simulation. Problem statement- The advent of e-business on a global scale, virtual markets and enterprise networking put forward the development of decentralized planning and scheduling methodologies tailored to multiple decision-making agents coordinating and controlling inputs and outputs from thousands of tasks and cooperating to minimize conflicts in accessing hundreds of resources. For responsiveness and agility disruptive events must be managed locally to avoid propagating the effects along the value chain. Contrasting with the traditional top-down approach, intelligent distributed scheduling highlights a bottom-up emergence of interactions between autonomous agents which mainly communicate through actions taken, following a strategy which is not fully unveiled to other agents. Methodology and Prototype implementation- The design of the mechanism for distributed scheduling is based on generative modeling and simulation of agent-based interactions and asyncronous decision-making. Agents defined in the batch processing system have different roles with the objective of reflecting the real manufacturing setting (resources and products). The so-called order manager, order agent and resource agent are inspired in the PROSA reference architecture, with some changes added to facilitate mechanism design. One desired characteristic of the distributed scheduling mechanism is its ability to detect changes and disruptive events that occur in the manufacturing system along with the use of learning algorithms to promote efficient decision-making by knowledge accumulation. To exemplify the proposed approach, a distributed scheduling mechanism for batch processes was modeled in Netlogo, a software environment specifically desig ed for Agent Based Modeling and Simulation (ABMS). The example involves distributed production of +100 dairy products in ten different resources. Each order has different attributes such size, type of product and due date whereas arrival times, process times and machine failure rate are stochastic. In the "distributed scheduling world" interactions and events are perceived in a dynamic Gantt chart whereas performance indicators are used as guidelines for mechanism design. In this simulation world, order managers and their respective order agents (clients) negotiate with resource agents (servers) the provision of required resources, and the global scheduling thus emerges as a result of such interactions. Handling of events without a centralized decision making structure is a feature of the designed mechanism. Significance- The complexity of global competition is forcing many Small and Medium Enterprises to cooperate through virtual networking without loosing their identities and core competencies. Based on the case study, analysis and design of an optimized mechanism for distributed scheduling will be presented. For achieving this, distributed scheduling systems acting as the electronic interface between firms is a key technology to achieve the level of coordination and communication required. The emergent behavior from interactions among heterogeneous agents with different objectives will highlight the importance of resorting to the generative modeling approach of ABMS which allows designing meta-rules that change the interaction rules in distributed decision-making. In this regard, it is expected that the work may motivate others to explore more deeply the paradigm shift that ABMS is producing in the discovery of strategic solutions to complex management problems, such as distributed scheduling and supplier selection.