A reconfigurable distributed multiagent system optimized for scalability

Abstract

A RECONFIGURABLE DISTRIBUTED MULTIAGENT SYSTEM OPTIMIZED FOR SCALABILITY by Summiya Moheuddin This thesis proposes a novel solution for optimizing the size and communication overhead of a distributed multiagent system without compromising the performance. The proposed approach addresses the challenges of scalability especially when the multiagent system is large. A modified spectral clustering technique is used to partition a large network into logically related clusters. Agents are assigned to monitor dedicated clusters rather than monitor each device or node. The proposed scalable multiagent system is implemented using JADE (Java Agent Development Environment) for a large power system. The performance of the proposed topology-independent decentralized multiagent system and the scalable multiagent system is compared by comprehensively simulating different fault scenarios. The time taken for reconfiguration, the overall computational complexity, and the communication overhead incurred are computed. The results of these simulations show that the proposed scalable multiagent system uses fewer agents efficiently, makes faster decisions to reconfigure when a fault occurs, and incurs significantly less communication overhead. The proposed scalable multiagent system has been coupled with a scalable reconfiguration algorithm for an electric power system attempting to minimize the number of switch combination explored for reconfiguration. The reconfiguration algorithm reconfigures a power system while maintaining bus voltages within limits specified by constraints.

Decentralized control and management systems for power industry via multiagent systems technology

“…Especially for large scale power systems with hundreds of components. The size of the multiagent system used in addressing the problem and interactions among agents introduce challenges due to scalability [20].…”