【作者】 陈昆亮；

【导师】 杨昆； 顾煜炯；

【作者基本信息】 华北电力大学 ， 热能工程， 2012， 博士

【摘要】 随着世界范围内工业各领域频发汽轮发电机组重大事故,不但给事发地区的社会与经济发展造成极大损失,同时也给我国大型复杂设备的安全使用敲响警钟,保障大型复杂系统安全稳定高效运行成为各行业进行智能化、自动化转型过程中的首要条件,这同时对我国大型汽轮发电机组运行状态的安全监控能力提出了更高的要求。当前,我国正面临着能源效率、运营效率和资源利用率亟待提高、环境质量迫切需要改善等挑战,同时也面临工业智能化应用的新机遇,许多理念、技术和产品也急待新的突破。本文充分认识到事物间相关性联系,从多层角度对影响和反映汽轮发电机组安全稳定特性的状态变化关系进行研究,在研究机组典型故障模式的表述及分类、故障征兆的分类及优化的基础上,重点从故障发生范围、故障属性、故障概率三个方面进行故障预警,其中包括对征兆的异动搜索、属性识别以及风险概率等关键技术：(1)基于粗糙集的故障特征征兆优化方法研究。在机组各典型故障模式分类表述的基础上,将故障征兆分类为反映故障发生范围的故障范围征兆,反映故障属性发展的故障属性征兆以及反映故障强度的故障强度征兆,并提供对故障征兆归纳分析的解决方法。利用序列模式定义,将在线、离线征兆进行统量化,并进行约简。为避免特征参数的复杂性,利用参数重要度指标进行优化约简,最终综合考虑到故障类型,提出一套有参考价值的故障特征征兆集合。(2)基于多特征征兆模式的汽轮发电机组K-均距异常搜索方法研究。在分析汽轮发电机组监测参数特征及表现的基础上,首次提出采用时间序列分割技术、时间序列管理技术、参数异动搜索技术对故障范围征兆参数的时间序列进行深入分析,利用序列子模式作为搜索规则,利用K-均距方法搜索可能由异常数据组形成的函数指标,依此建立预警机制,实现预测故障发生的范围或部位。(3)基于灰色加权-AR组合预测以及多特征状态识别的识别方法研究。在对比了典型预测方法的基础上,本文采用基于灰色加权-AR的组合预测模型,对可以反映故障属性发展的征兆参数进行预测；为了避免单一征兆预测结果对故障趋势的误判,根据状态空间理论,本文定义了自由状态空间以及基准状态空间的概念,建立了多特征识别模型,同时给出制定状态空间的法则。解决了汽轮发电机组状态监测分析过程中,对故障趋势的预判不精确的缺点,实现了对机组的故障属性质的精确判断,为机组的状态监测提供了指导依据。(4)基于辨识分类逻辑回归的汽轮发电机组故障概率研究方法。在对典型故障发展程度水平分析的基础上,利用逻辑回归原理,对反映故障发生概率对应的故障强度征兆历史样本进行综合分析,标准化特征参数表现模式并建立相应的回归模型,通过最大似然函数法求解出故障概率回归模型,最终利用当前监测获取的特征参数值,分析得出当前疑似故障类别的各个故障的可能发生的概率,还建立了故障处理措施的查询机制。最后,在上述理论指导下,利用UCML技术平台,设计开发了汽轮发电机组状态监测与故障预警系统软件平台。更多还原

【Abstract】 As major accident of the turbine generator frequently happened in all areas of world-wide industrial, not only which caused great losses to the social and economic development, but also gave a early warning to safety in producing and using large-scale complex equipments in China. And how to ensure stability and efficient operation for large-scale complex equipments becomes a primary condition in the process of industrial intelligent and automated transformation. And while it put forward higher requirements for safety monitoring capabilities of the large turbine generator in China. At present, challenges are faced in China, for example energy efficiency, system operation efficiency and resources utilization should be enhanced urgently, and environment quality needed to be improved etc. As well, some new opportunities of industrial intelligent application came out, such as many concepts and technology and product need to be improved. In this paper, relevance among things is fully recognized, and which among turbo-generator security and stability state is studied from multi-layer. Based on the study of the classification of typical failure modes and the optimization for failure symptoms, the research of failure warming can be mainly from three sides, such as failure range, failure properties and the failure probability. In this paper, it includes some key technologies as following, such as abnormal search for signs, property identification and risk probability:(1) Study on the failure characteristics symptoms optimization method based on rough set. Based up on the statements of typical failure mode, the classifications of failure symptoms are the failure range symptom which can reflect where the failure happened, the failure property which can reflect the failure development property, and the failure intensity symptoms which can reflect how frequent the failure is. And then this paper provides the resolved approach for failure symptoms analysis. In this paper, sequence pattern is defined, and both the symptoms online and offline can be quantization and reduction. In order to avoid the complexity of characteristic symptoms, with symptoms importance index, the failure characteristics symptoms can be optimized and reduced. Finally, synthetically considering the types of failures, valuable failure symptoms set can be proposed.(2) Study on the K-distance abnormal search method for turbo-generator based on multi-characteristic symptoms model. Based on the analysis of the turbo-generator monitoring parameters, it was first proposed the analysis method for time series of failure range symptom parameter, by use of time series segmentation technology, time series management techniques and time series abnormal search technology. It takes the sub-model of time series as the search rules, and with the K-distances abnormal search method it can search the function index composed of abnormal time series. Establishing the early warning mechanism and the failure range forecast can be realized.(3) Study on the identification method based on the gray weighted-AR combination forecasting method and multi-character state method. In contrast to the typical forecasting methods, this paper proposed the combination forecasting model based upon gray-weighted-AR theory, to predict the symptoms parameters that can reflect the development of failure properties. In order to avoid prediction misjudgment with a single parameter it defines the concept of a free state space and a benchmark state space based upon the state space theory. And multi-character recognition model is established, and also the law of the state space is provided. By this method, it solves the inaccuracy for failure trend forecast during turbine generator condition monitoring process, so the failure property can be determined more accurately. All these can provide the basis guiding for turbo-generator condition monitoring.(4) Study on the failure probability calculation method based on the identification classification logic regression. Based on the analysis of the development degree recognition for typical failures, by principle of logistic regression, the corresponding historical sample, which can reflect the failure probability, is comprehensively analyzed. And then it standardized expression pattern of characteristic parameters and established the corresponding regression model. By use of maximum likelihood function method, the failure probability regression model can be calculated out, finally with the characteristic parameter values obtained by the current monitoring the failure occurrence probability may come out. In this paper, it also set up the query mechanism for failure measures.Finally, in the above theory, under the guidance of UCML technology platform, software platform of turbine generator condition monitoring and failure warning system is designed and developed.更多还原