【作者】 刘忠；

【导师】 张勇传； 周建中；

【作者基本信息】 华中科技大学 ， 水利水电工程， 2007， 博士

【摘要】 国民经济的快速发展对能源需求的不断增大和电力市场的自由竞争的引入,使得水电生产企业在提高生产效率的同时,也要重视降低生产成本,尤其是减少由于机组检修或者故障停机导致的无法预计的经济损失和社会影响。故障诊断的研究领域应该朝着预知方向发展,研究如何在故障发生之前发现各种故障发生的可能性,如何让机器避免故障的发生、“知道”故障何时发生和发生轻微故障后如何防止进一步扩散、传播,做到防患于未然。为了提高水电机组故障诊断的准确性、实时性及鲁棒性,一方面需要对现有方法进行改进优化,另一方面需要加强新方法的研究,特别是基于生物智能的新方法研究。人工免疫系统是借鉴生物免疫系统的结构特征和工作机理而建立起来的、解决某方面工程技术问题的系统形式或者算法结构的统称。本文在全面总结现有智能诊断技术和人工免疫系统发展的基础上,针对水电机组状态检修体制下智能诊断中存在的问题,将人工免疫系统理论与智能诊断技术有机地结合,研究出了一系列的免疫诊断方法,并应用于水电机组的故障诊断、状态预测和检修决策中。主要从以下几个方面开展研究并取得了如下成果:首先明确了本文研究的问题、目的和意义。阐述了“状态检修为主、多种检修方式并存”的检修体制下智能诊断概念的内涵和重要意义;介绍了水电机组故障的特点;综述了智能诊断方法的研究现状,指出了目前存在的不足;对人工免疫系统的理论研究与应用研究现状进行了阐述。在此基础上,给出了本文研究的总体思路和主要内容。系统地对生物免疫系统及其运行机制、免疫系统的主要特点作了描述;介绍了人工免疫系统的基本原理,包括形态空间模型和免疫细胞模型;讨论了几种典型的人工免疫算法;着重对人工免疫系统在智能诊断领域中的应用潜力进行了剖析。针对水电机组状态监测中特征提取方法的局限与不足,提出了故障诊断中的复合特征概念,并建立了集成小波分析、模糊理论和径向基神经网络(Radial Basis Function Neural Network,RBFNN)的水电机组故障诊断模型。重点讨论了采用小波分析方法提取稳定性状态信号能量特征、运用模糊理论提取关系型征兆以及水电机组RBFNN故障诊断模型的构建过程。工程应用结果表明,该方法能够全面准确地提取水电机组稳定性状态特征,能够诊断出机组的典型故障类型及其严重程度,具有一定的可行性和有效性。针对传统诊断方法中对样本要求高、缺乏自主学习能力等问题,根据免疫系统中抗体对抗原的识别机理,提出了基于免疫应答机制的故障诊断方法。建立了机组诊断系统与免疫系统的映射关系,详细阐述了分别对应于免疫初次应答和二次应答的诊断流程。该方法能够实现已知状态的识别,而且能够自主记忆并识别机组未知状态,具有一定的快速反应性和鲁棒性。将该方法应用于水电机组振动故障诊断,验证了其可行性和准确性。从免疫系统的视角分析了RBFNN结构设计与免疫系统机制的相似性,根据免疫系统多样性和免疫响应亲和力成熟的过程,提出一种新型的免疫优化算法。将免疫优化算法用于选择和确定RBFNN模型结构,并与递推最小二乘法相结合,形成免疫优化RBFNN混合学习算法。采用免疫优化RBFNN对某水电机组水导轴承处的振动状态进行预测,并与K均值聚类RBFNN和BPNN的预测结果进行了对比。计算结果证明,免疫优化RBFNN的预测结果更接近实测值,且精度较高,具有一定的准确性和实用性。在分析人工免疫系统与多智能体系统(Multi-Agent System,MAS)相似之处的基础上,提出了一种免疫Agent模型;阐述了相关知识的获取、组织与管理方法,建立了基于免疫Agent的水电机组状态检修系统模型,并对各功能免疫Agent作了描述;最后对该模型框架下的免疫协同诊断策略作了较为细致的阐述。基于人工免疫系统的水电机组智能诊断方法研究,是将人工免疫系统理论方法与智能诊断技术相结合,并运用于水电机组状态检修领域的一次探索。尽管取得了一些研究成果,但由于人工免疫系统的理论与应用研究起步较晚,要实现人工免疫系统理论与智能诊断技术的高度融合,实现水电机组状态检修体制下的自主诊断、高精度预测以及优化决策维护,还需做大量的工作。论文最后指出了人工免疫系统理论在水电机组智能诊断领域中需要进一步研究的问题。更多还原

【Abstract】 The demands on energies from the fast-developing national economy are increasing and the freely competing mechanism has been introduced into the electric power market. Hydropower enterprises have to try to improve their producing efficiency and reduce the production cost, especially the unpredictable economical costs and social effects, which results from the breakdown of Hydroelectric Generating Unit (HGU) due to casualties or maintenance actions. The researches on fault diagnosis should advance towards the direction of prognosis, emphasizing on how to determine the probability of the faults before their occurrence, how to avoid the faults, when the faults will occur, and how to prevent their development at their primary states. In order to improve the correctness, real-timeness, and robustness of the fault diagnosis of HGU, on the one hand, the current approaches should be improved and optimized. On the other hand, new approaches should be resorted to, especially those inspired by biological intelligences.Artificial Immune System (AIS) includes the system forms or algorithm structures to solve some engineering technique problems, inspired from the structure characteristics and function mechanism of Biological Immune System (BIS).Previous research works on the modern intelligent diagnostic techniques and the theory and applications of AIS have been summarized first in the dissertation. Aiming to solve the diagnosis problems under the conditioned maintenance of HGU, diagnosis techniques are combined with AIS appropriately. Several immune diagnostic approaches are researched, and have been applied to the fault diagnosis, state prognosis and maintenance decision of HGU. The outline of this dissertation is as follows:The problems, purposes and significance of this research are firstly presented. The intension and significance of the intelligent diagnosis are discussed under the maintenance system where conditioned maintenance takes the predominant place and other maintenance forms coexist. Then the characteristics of the fault of HGU and the status in researches on intelligent diagnostic techniques are summarized, followed with the existing problems in intelligent diagnosis. After the description of the status in both theoretic research and application of AIS, the universal thought and the main content of this dissertation are presented.BIS and its function mechanism are described in details as well as its main characteristics. Then the basic principles which include shape-space model and immunocyte model and several typical AIS algorithms are present. The potentials of AIS in the intelligent diagnosis field are analyzed and emphasized on.In order to overcome the limitations of the traditional FFT in non-stationary signal processing and overdependence on the energy characteristics during the analysis and fault diagnosis for HGU, the concept of combined feature is proposed, and a novel fault diagnosis model integrating the combined feature and RBFNN is built. How to extract relative energy feature of the stability signals via wavelet transform and how to evaluate the influences of the processing parameter changes on the stability state and extract the relationship symptoms are described in details, as well as the construction of the proposed fault diagnosis model based on RBFNN. Application results show that this proposed method is feasible and efficient in the overall feature extraction of HGU as well as the appropriate evaluation of fault types and severity degrees.In order to solve the problems such as large sample demands and lacking ability of active learning in the fault diagnosis techniques, a new diagnosis method based on immune response mechanisms is proposed. It is inspired from the recognition progress of antibodies with antigens. The mapping relations between the diagnosis system of HGU and AIS are set up. The diagnostic schemes corresponding to the primary response and the second response are discussed detail. This method can not only recognize the known states, but also remember the unknown ones and recognize them with fast reactions and robustness. The proposed method was applied to the vibration diagnosis of hydroelectric generating unit. Results indicate the high diagnosis accuracy with low demands on fault samples, and the merits of this method in application as well.The analysis on similarities between the structure design and the immune mechanism is implemented at the view of the immune system. A novel immune optimization algorithm is proposed. It is used to select and determine the structure of RBFNN, and jointed with the recursive least squares (RLS) algorithm to form the hybrid learning algorithm of immune-optimized RBFNN. The prediction of the vibration state at the water guide bearing of HGU based on the above immune-optimized RBFNN is performed. The results are compared with those by K-mean clustering RBFNN and BPNN, which show the high accuracy, correctness and feasibility of the proposed immune optimization algorithm. The immune agent model is proposed after the comparison of AIS and MAS. The approaches to acquire, organize and manage relative knowledge are discussed. Then the frame of the immune agent based conditioned maintenance system of HGU is presented, followed with the detailed description of each immune agent with different functions. Finally the immune co-operation diagnostic strategy under the frame is expounded.Study on AIS based intelligent diagnosis technology of HGU integrates the theory and approaches of AIS and those of intelligent diagnosis. It is a brave try in the research field of conditioned maintenance of HGU. Some research fruits are achieved. Because the theoretic and application researches on AIS start relatively late, there remains a large quantity of work to do, to implement the highly fusion of AIS and intelligent diagnosis technology, to realize the active diagnosis, the high-accuracy prognosis, and the optimal maintenance decision making.At the end of this dissertation, the future research directions in the field of AIS based intelligent diagnosis technology are pointed out.更多还原