多智能体协同故障诊断方法研究及其在煤矿瓦斯监测中的应用

Due to the development of network and communication technologies, multi-agent system is playing a very important role in many fields. Compared to the traditional system, the faults in the multi-agent system are much more complicated, including homologous fault (all agents are affected by the same fault) and heterolous fault (each agent is affected by different faults). Both of the two kinds of faults may exist in multi-agent systems at the same time, so how to design an observer that estimates these two kinds of faults simultaneously become an important problem. This project mainly aims at the homologous and heterolous fault diagnosis in multi-agent system. First, an observer is designed to diagnose homologous fault. Then, we study the fault diagnosis scheme for both homologous and heterolous faults. This part will be divided into two parts: these two kinds of faults can and/or cannot be decoupled. The expected theory results will be tested in coal mine gas monitoring. In the process of coal production, when the gas concentration in the mine exceeds a certain index, it will seriously impact the production schedule and even personal security. If several gas detection devices can be interconnected to form a multi-agent system, one can get a more accurate gas detection result. Therefore, this project has important theoretical value and broad application prospect.

随着网络和通讯技术的发展，多智能体系统在很多领域发挥着越来越重要的作用。相较于单一系统，多智能体受到的故障要复杂的多，例如同源故障（各智能体受到相同故障的影响）和非同源故障（各智能体受到不同故障的影响）。上述两类故障有可能同时存在于多智能体系统中，因此如何设计观测器使其可以同时估计同源和非同源故障成为一个重要课题。本项目主要针对多智能体系统的同源/非同源故障诊断进行研究。首先为多智能体设计同源故障的诊断方案；以此为基础，设计同源/非同源故障的同时诊断方案；分别研究同源故障和非同源故障可解耦和不可解耦两种情况。预期理论成果将进一步在煤矿瓦斯监测方面进行应用验证。当矿井内瓦斯浓度超过一定指标后，轻则停止生产，重则发生瓦斯爆炸。如果将各瓦斯监测装置互联构成一个多智能体系统，就可以更精确的获得当前瓦斯浓度，进而避免一些生产事故。因此，研究课题具有重要的理论意义和广阔的应用前景。

随着网络和通讯技术的发展，多智能体系统在很多领域发挥着越来越重要的作用。因此如何设计观测器使其可以同时估计同源和非同源故障成为一个重要课题。本项目主要针对多智能体系统的同源/非同源故障诊断进行研究。首先项目组以传统的卡尔曼滤波器为基础，利用多智能体中各个智能体可以信息共享的特点，适当变换卡尔曼滤波器形式，引入邻居智能体残差信息，设计出分布式观测器，推导出多智能体系统同源故障诊断条件，进而设计相应增益矩阵，设计出多智能体同源故障诊断方案。项目组将同源故障与不同源故障进行解耦，再单独的估计每一部分故障，设计两类残差信息，进而利用上述卡尔曼滤波器的相关理论和条件极值的有关理论得到两类故障可解耦的条件，从而实现对同源故障和非同源故障的最小方差无偏估计，设计出多智能体系统同源故障与非同源故障可解耦时的故障诊断方案。项目组考虑到常常会出现两类故障不可解耦的情况发生，利用H∞控制理论和方法，使得设计出的两种残差信号分别与一类故障间的相关程度尽极大，与另一类故障中的相关程度极小，进而设计观测器实现同时估计两类故障。