Multi-agent System with Hybrid Intelligence Using Neural Network and Fuzzy Inference Techniques

Abstract

In this paper, a novel multi-agent control system incorporating hybrid intelligence and its physical testbed are presented. The physical testbed is equipped with a large number of embedded devices interconnected by three types of physical networks. It mimics a ubiquitous intelligent environment and allows real-time data collection and online system evaluation. Human control behaviours for different physical devices are analysed and classified into three categories. Physical devices are grouped based on their relevance and each group is assigned to a particular behaviour category. Each device group is independently modelled by either fuzzy inference or neural network agents according to the behaviour category. Comparative analysis shows that the proposed multi-agent control system with hybrid intelligence achieves significant improvement in control accuracy compared to other offline control systems.