A Multi-Agent Architecture for Process Management Accommodates Unexpected Performance
John Debenham
University of Technology, Sydney
Computing Sciences, UTS
PO Box 123 Broadway, NSW 2007, Australia
+61—2—9514—1837
debenham@socs.uts.edu.au
ABSTRACT
A generic process agent architecture is used to build systems for business process management. The architecture is a three-layer BDI, hybrid architecture. The agents in this system attempt to make choices in line with corporate priorities. Choices are made on the basis of estimates of the values of particular parameters. These estimates are accepted as long as they are statistically stable. If variations in these estimates lie outside specified limits then the system attempts to deduce why this is so. If it is unable to deduce a reason then it asks for one. If the reason for an unexpected measurement is quantifiable then the reason is used to revise subsequent measurements while that reason remains significant. The architecture has been trialed on emergent process management in a university administrative context.
Keywords
Multi-agent systems, business process management.
1. INTRODUCTION
A multi-agent architecture is designed for business process applications. The growth of activity in agent research [1] and of industrial interest in the applications of agent technology is well-documented [2]. The number of deployed multi-agent systems as reported in the research literature is comparatively low [3] where the term ‘agent’ has a wide range of meaning [4]. The term ‘agent’ is used here, following [3] in the sense that "an agent is a computer system, situated in some environment, that is capable of flexible autonomous action in order to meet its design objectives" and "the term ‘multi-agent systems’ ... is now used to refer to all types of systems composed of multiple (semi-) autonomous components". [5] distinguishes between three forms of agent: information agents, cooperation agents, and transaction agents. Here a generic agent architecture is used to build both cooperation agents and transaction agents.
Business process management is an established application area for agent research [6]. The term business process is taken here to refer to processes that are not pre-defined, that are not of a routine nature and that may rely on some level of initiative from the system to bring them to a conclusion [7]. One practical feature of business process management as an application area is that ‘real’ experiments may be performed with the cooperation of local administrators; a business process management system for dealing with applications from potential research students was trialed in this way [8].
2. Agent Architecture
A variety of architectures have been described for autonomous agents [3]. A fundamental distinction is the extent to which an architecture exhibits deliberative (feed forward, planning) reasoning and reactive (feed back) reasoning. Architectures that combine these two forms of reasoning are called hybrid architectures. One well reported class of hybrid architectures is the three-layer, BDI agent architectures. One member of this class is the InteRRaP architecture [9], which has its origins in the work of [10]. An architecture similar to InteRRaP has been applied successfully to business process management [8].
The role of