Multi-agent fuzzy inference control system for intelligent environments using Jade

Abstract

In this paper, a novel physical testbed for intelligent environments and its software based multi-agent control system are presented. In the physical testbed, a fair amount of embedded sensors and actuators are interconnected in three types of physical networks, namely the LonWorks network, RS-485 network and IP network. Universal plug and play (UPnP) is introduced in the system architecture to provide unique control interface between high level multi-agent control system and low level devices on different physical networks. The multi-agent control system is developed on an existing agent platform, JAVA agent development framework (JADE). Fuzzy inference learning is implemented with multiple fuzzy inference agents, each models the human behaviour associated with a predefined group of devices in forms of fuzzy rules. Corresponding fuzzy logic controller agents can be initiated to provide user preferred control actions according to the fuzzy rule bases. A comparative analysis shows that our control system achieves a great improvement in both control accuracy and computational efficiency compared to other offline control systems. Online adaptive learning, automatic device group formation and advanced wireless device networks are within the scope of our system architecture