【作者】 熊伟；

【导师】 韩璞； 张丽静；

【作者基本信息】 华北电力大学 ， 控制理论与控制工程， 2014， 博士

【摘要】 近年来智能故障诊断技术成为故障诊断领域的研究热点,该技术以复杂系统(设备、元件)的管理、监测和故障诊断为主要内容,以建立故障预防和系统维修体系为目标,在诸多生产实践领域得到应用。智能技术的应用提高了系统故障诊断的科学化、智能化水平,在实践中得到了充分检验,极大丰富了人们有关故障的领域知识。智能故障诊断技术主要包括专家系统、模糊故障诊断、神经网络、信息融合和Petri网等。尽管当前的智能故障诊断技术采用了许多成熟的数学方法和基于专家系统的推理算法,依然有许多关键问题需要进一步研究,包括知识表达和推理、知识学习、智能识别和信息融合等。智能故障诊断系统需要将诊断理论、专家经验知识有效融合,通过计算机软件编程加以实现。故障诊断专家往往不是诊断系统的具体实现者,而软件编程人员亦非有着丰富领域知识的专家,能否使诊断系统真正体现专家的意图和决策,决定了系统的优劣；能否使诊断过程清晰、高效,决定了系统是否可用、可推广。Petri网的诸多特性使其能够简洁方便的表示专家知识、诊断组织结构以及诊断推理过程,能够对专家知识及推理过程进行优化,能够很容易地融合模糊推理、神经网络和面向对象编程等技术。应用Petri网技术,可以使得诊断系统专家、系统开发人员、系统使用者紧密关联,体现了故障诊断从理论到智能再到应用的全过程。论文包括以下几方面内容：1、对Petri网技术的研究。以Petri网理论为基础,结合模糊产生式规则推理,研究模糊故障Petri网及建模方法；研究基于Petri网模型图的推理方法；重点研究、分析并设计了基于矩阵及其运算的推理方法；综合运用极值法、求和法实现正反向混合推理。2、对智能Petri网技术的研究。研究了加权模糊Petri网、自学习(自适应)模糊Petri网,分析其利用单一BP (Back Propagation)神经网络算法训练Petri网中的权值、阈值和置信度等参数的优缺点；研究了神经网络集成学习的理论,提出动态选择性集成与故障诊断Petri网融合方法。3、对PSO(Particle Swarm Optimization)算法的研究。在PSO算法研究的基础上,提出了改进的增强型PSO算法；研究了PSO算法与神经网络的融合技术,提出了基于改进的PSO算法的神经网络集成学习方法,并应用于智能Petri网的参数学习和训练。4、设计并实现了基于Petri网的远程智能故障诊断系统,应用于电厂汽轮机组故障诊断实例。深入研究并应用数据整形与压缩技术、网络通信技术、服务器推送技术等解决远程诊断过程中所面临的实际应用问题。更多还原

【Abstract】 In recent years, the intelligent fault diagnosis technology has become a hot spot in research of fault diagnosis. This technique mainly includes complex system (equipment, components) management, state monitoring and fault diagnosis. The goal of this technique is to build up the fault prevention and maintenance system, and it has been used widely in various fields. Intelligent technology makes the fault diagnosis more scientific, more reasonable, and more accurate. It greatly enriches the knowledge in the field of failure mechanisms. The intelligent fault diagnosis technology consists of expert system, fuzzy fault diagnosis, neural network, information fusion and petri net, etc.Currently, many mathematical methods and some reasoning method in expert system have reached a very high level in intelligent fault diagnosis technology, but there are some key issues need to be addressed, such as knowledge representation and reasoning, knowledge learning, intelligent identification, information fusion,etc. Intelligent fault diagnosis system combines the diagnosis theory and expert experience knowledge, and it is realized by the computer software programming. In the realization process of intelligent fault diagnosis system, the fault diagnosis expert is often not the developer, and the excellent software programmer is not fault diagnosis expert. So that, whether or not the intelligent diagnosis system reflects real expert intelligence, it determines the pros and cons of the system. And the promotion of the system, the application scope and effect are dependent on the clear and efficient diagnosis. Petri net has many characteristics, so that it not only can simply and conveniently represent expert knowledge, the organizational structure and the process of diagnosis, but also can optimize the expert knowledge and the reasoning rules. In addition, petri net can be easily combined with other techniques and theories, such as fuzzy reasoning, neural network and object-oriented programming etc. In short, petri net makes the fault diagnosis expert, the system developers and the users of the system closely related. It reflects the whole process from theory to intelligence, and then to application. This paper mainly studied the following aspects.Firstly, researches on the technology of petri net. The modeling method of petri net is researched based on the theory of petri net and the fuzzy production rules. Focused on the fault diagnosis reasoning strategy, such as graphics-based reasoning approach and matrix-operation-based reasoning approach. Integrated uses the extreme value method and the summation method to realize the application of graphics-based reasoning approach (graphics-based reverse reasoning approach) in fault diagnosis.Secondly, researches on the technology of intelligent petri net. Researched on the weighted fuzzy petri net and self-learning fuzzy petri net, analyzed the advantages and disadvantages, such as using the single BP neural network to train weights, thresholds and other parameters. The theory of neural network ensemble was introduced, then put forward a method combined with dynamic selective ensemble method and fault diagnosis petri net.Thirdly, researches on the optimization of PSO algorithm. Based on the optimization of PSO algorithm put forward a new improved PSO optimization algorithm. Researched on the fusion technology of PSO algorithm and neural network, then put forward a new integrated learning optimized method of neural network based on the improved PSO algorithm, and this method is applied to intelligent petri net.Lastly, designed and realized the remote intelligent fault diagnosis system based on petri net.It was used for fault diagnosis of steam turbine unit. Researched on some computer network technologies, such as data-shaping technology, data compression technology, data transmission technology and server-push technology, etc. These computer network technologies were used to solve practical problems, which are usually encountered during remote diagnosis.更多还原