Dynamic Reconfiguration Of Multi-Agent Systems And Its Applications

Multi-agent systems are well suited to engineering complex software systems and construct artificial systems. But how to dynamically reconfigure an agent system based on different tasks and changing environment is a key issue which remains unsolved. This is crucial for complex software systems which solve complex problems.Actually, dynamic reconfiguration is listed as one of the key challenges in support of agent adaptation to environments, which is called 'run-time reconfiguration and redesign'. Further research indicates that dynamic reconfiguration or self configuration is also a key issue requiring urgent solution in Grid Computing as well as Autonomic Computing. In Grid Computing, such an issue is called 'virtual organization formation and management'. And in Autonomic Computing, it is called 'self-management of computing system'. The emphasis of this paper is on the dynamic reconfiguration with different tasks and environment changes.Though a lot of work has been done on this issue, there is no efficient solution reported to date about dynamic reconfiguration with different tasks and environment changes. This paper will focus on three key issues: modelling the abstract concept of dynamic reconfiguration appropriately for agent-based systems, producing techniques to efficiently work on the model of dynamic reconfiguration, and studying how to apply dynamic reconfiguration techniques for agent-based systems to supply chain management.The contributions of this paper are in the following three aspects:1. Modelling dynamic reconfiguration of agent-based systemsAn emerging computational paradigm-Autonomy Oriented Computing (AOC as short) has been used to character dynamic reconfiguration of agent-based systems; a model of dynamic reconfiguration of multi-agent systems has been proposed; the key elements of this artificial model has been clearly defined; the behavioral rules of autonomous entities (here are agents) have been presented. This research will, on one hand, produce new dynamic reconfiguration techniques for agent systems; on the other hand, it will also contribute to a new computing paradigm - autonomy oriented computing.2. Designing algorithms for dynamic reconfigurationThe multi-Agent System Dynamic Reconfiguration (ASDR) algorithm was developed to control the dynamic reconfiguration process. And other two algorithms called IERA (Improved ERA) and HIERA (Heuristic Improved ERA) which improved upon ERA (Environment-Reactive rules-Agents) algorithm have been proposed to support the organization formation behavior, which is essential in dynamic reconfiguration. We do some experiments on the convergence speed and parameter settings of the two algorithms. Experiments show that these two algorithms can convergent efficiently, and can be used to support not only the ADRM model, but also the other similar distributed search problems.3. Conducting research on real dynamic reconfiguration problem in Supply Chain Management SystemThe International Trading Agents Competition for Supply Chain Management (http://www.sics.se/tac) (TAC SCM) was selected as the scenario of this paper. The ADRM Model has been extended and tailored to the scenario. A supply chain management system which named DR-SCM with dynamic reconfiguration ability has been developed by JAVA and TAC-SCM toolkit. The experiment on DR-SCM demonstrates that our research can benefit real dynamic reconfiguration problem in Supply Chain Management System a lot. It also shows that with dynamic reconfiguration, agent-based systems can opt to different tasks, and be adaptive to the environment changes. Many complex problems can be solved in a shorter time frame and resulting in higher quality solutions.