【作者】 韩晓娟；

【导师】 徐大平；

【作者基本信息】 华北电力大学（北京） ， 热能工程， 2008， 博士

【摘要】 由于火电设备日趋于高参数、大容量、复杂化,其安全经济运行对社会的影响越来越大。发电设备一旦出现故障,其损失和造成的影响是巨大的,因此需要及时监测设备的运行状况,识别故障早期征兆,对故障部位、故障程度和发展趋势作出准确判断,以提高机组的可靠性和可利用率。而热力系统是火电厂生产的主要部分,这一部分的故障诊断就显得尤为重要了。本课题依托于原国家电力公司电力行业青年促进基金项目:《多传感器信息融合技术在火电厂状态评估中的应用》。论文从如何提高故障诊断的确诊率和容错性出发,针对火电厂热力系统故障诊断中多信息的特点,将多源信息融合技术引入到设备故障的诊断中,围绕火电厂热力系统故障诊断展开了以下研究:1.针对火电厂热力设备测点多、数据间存在强相关性等特点,将主分量和神经网络相结合的融合诊断方法引入火电厂凝汽系统的故障识别。利用主分量分析实现热力设备的故障特征优选,由主分量贡献率确定神经网络的输入空间,比较分析了主分量BP网络和主分量RBF网络对凝汽系统故障诊断的优缺点。通过凝汽系统故障诊断实例,验证了该方法可以有效地简化网络结构,提高网络的分类精度。2.针对热力系统故障具有异步、离散等特点,建立故障诊断的神经Petri网模型。以信息熵作为属性约简的标准,从大量的故障征兆信息中获得最小的诊断规则,建立最优的Petri网模型。由于单纯的故障诊断Petri网缺乏自学习功能,将神经网络引入Petri网。通过对火电机组凝汽系统故障诊断研究表明:基于信息熵、神经网络和Petri网相结合的故障诊断方法改善了它们各自诊断的能力。用神经Petri网对故障诊断系统建模,增加了网的表达能力,适用性强。该方法为Petri网应用于热力系统故障诊断提供了一条有效途径。3.由于火电厂热力系统故障过程大部分属于缓变故障,从设备正常运行到出现故障征兆再到发生故障灾害是一个较慢的过程,其间设备许多状态量的变化是连续的。针对上述特点,本文提出了基于分层的混合模型融合诊断策略,将灰色理论、特征评估、神经网络有机地结合起来,完成故障子空间的识别,最后根据多属性决策对设备状态进行综合评价。通过制粉系统实际数据仿真验证了该方法的有效性和可行性。该方法有利于早期故障的准确识别,并对故障程度以及故障趋势作出准确判断。更多还原

【Abstract】 The equipments in power plants trend to be characterized by high parameters, large-capacity and complexity,its safe and economy operating has a great influence on the society. Once faults occur in the power generation equipment, the influence and loss caused by it would be enormous. Therefore, the operating status of equipments needs to be monitored in time to identify the earlier period symptoms of the faults and make an accurate judgment about the fault positions, degrees as well as fault tendency so as to raise the reliability and availability of the unit. However thermal system is the major production part in power plants and fault diagnosis of this part seems to be especially important. This topic is supported by young people promote fund project in the power industry:Application reseach of the state assessment in power plants based on multi-sensors data fusion technology.This paper begins with how to improve fault diagnosis rate and fault-tolerant, in view of the characteristics with multi-information of thermal system faults in power plants, multi-soruces information fusion technique was introduced into fault diagnosis of equipments which has a good theory significance and application values. Based on the thermal system fault diagnosis in power plant, the main work of the dissertation can be presented as follows in this paper:In term of thermal equipment for more measuring points in power plant, there is a strong correlation between datas and symptoms. A diagnostic method combined principal component analysis with neural network was introduced into the fault identification of condensate system in this paper. Firstly, principal component analysis was used to achieve the optimization features of thermal equipment, the rate of contribution gained from principal component was used to comfirm the input space of the neural network, the advangtages and weakness between principal components combined with BP neural networks and RBF neural network to carry on condensate system fault diagnosis were compared and analyzed. Finally, the effectiveness of the method was verified by diagnosis example of condensate system. The given method in this paper can simplify the structure of neural network and improve classification accuracy of the network.In view of thermal system fault with asynchronous and discrete features, a neural Petri net model for fault diagnosis was put forward and established. Information Entropy was used as attribute reduction standards, the smallest of diagnosis rules can be obtained from a large number of fault information so as to establish the optimal Petri net model. Because fault diagnosis Petri net model lacks self-learning function, neural network was introduced into Petri net.The reaserch results were shown by the fault diagnosis example of condensate system in power plant that the combination of diagnosis method based on information entropy, neural network and Petri net can improve the ability of their respective diagnosis.It can increase the expression ability of net using neural Petri net to model for fault diagnosis system. The method provides an effective way for Petri net to the fault diagnosis of thermal system.Because fault processes of thermal system in power plant partly belong to soft fault, from equipment normal to fault symptoms and then to fault disaster is a slower process in fault during the status changes of many equipments are continous.In response to these characteristics, a hybrid model based on hierarchical integration strategy was provided in this paper. The grey theory, characteristic assessment and neural network were organically integrated to complete the fault subspace identification, according to the multi-attributes decision-making model to achieve comprehensive evaluation of the equipment state. The validity and feasibility of the method was verified through actual data simulation of pulverizing system which is advantageous to get accurate identification to early failure, fault degree as well as fault tendency.更多还原