【作者】 宋凯；

【导师】 刘润华；

【作者基本信息】 中国石油大学 ， 控制科学与工程， 2010， 硕士

【摘要】 电力系统的可靠性和供电质量关系着国民经济的发展和人们日常生活水平的提高。当配电网发生故障尤其当保护和断路器发生误动或拒动以及通信装置故障造成信号改变或丢失时,会形成变异的故障模式,从而造成警报模式复杂化,给配电网故障诊断造成更多的困难。现有的研究方法在处理变异模式时难以保证故障定位的高容错性,会出现错判或漏判,其实用价值受到一定限制。因此如何在配电网系统发生事故时能尽快判定故障,为故障解列和恢复供电提供依据,减少停电损失,已成为配电网故障诊断研究的重要课题。本文针对单一数据挖掘方法在配电网故障诊断中存在的容错性差、诊断速度不高等缺点,采用了将粗糙集方法和神经网络方法相结合的混合数据挖掘方法对配电网故障进行诊断。首先对各种应用于配电网故障诊断的人工智能方法进行研究,分析它们在配电网故障诊断中的特征以及存在的主要问题,然后以保护、断路器作为条件属性,配电网的故障区域作为决策属性,考察各种故障情况并建立决策表,并利用布尔逻辑与可辨识矩阵相结合的属性约简算法对算例构造的决策表进行约简,删除其中不必要的属性,再利用人工神经网络调用最简规则集进行学习训练,这样既减少了神经网络的学习训练时间,又提高了故障诊断的准确度。此方法充分利用了粗糙集理论的知识约简能力和人工神经网络的容错学习能力,提高了系统的泛化能力,弥补了各自的不足。本文最后采用了VB语言作为开发工具,调用Matlab神经网络工具箱建立了一个简化的配电网故障诊断系统,并通过配电网实例验证了该方法的正确性和有效性。更多还原

【Abstract】 The reliability and power quality of power system is directly related to national economic development and the level of people daily lives elevation. When the distribution network failures, especially when the protection and circuit breaker tripping or malfunction occurs and signal changes or loss caused by communications device failure, it will form variation failure mode which result in complication of alarm mode and bring more difficulties to the distribution system fault diagnosis. Practicality of existing research methods is subject to certain restrictions because when it deal with variation mode, it is difficult to guarantee the high fault tolerance and there will be miscarriage of justice. So it has become a very meaningful research topic to determine fault,providing a basis to disconnection and restoration of power supply and reducing the outage cost when the distribution network system failures.In this article, i use hybrid data mining methods that combine rough set approach and neural network in order to overcome the defect of single data mining method in fault tolerance and low speed that exist in distribution system fault diagnosis. Based on the analysis and study predecessors for the distribution network fault diagnosis, i research on a variety of artificial intelligence methods used in distribution network fault diagnosis and analyse the characteristics and the main problems existence in distribution network fault diagnosis. Using hybrid data mining method in the distribution network fault diagnosis, firstly seeing protecting the circuit breaker as a condition attribute and the regional distribution network as a decision attribute.At the same time, examine a variety of fault conditions and the establishment of decision-making table the failure to study a variety of fault conditions and the establish decision-making table.Then, using attribute reduction algorithm that combinaed with Boolean logic and discernibility matrixcan to reduction an example decision table constructed and delete unnecessary attributes. Using artificial neural network that called the most simple set of rules for learning and training, which is not only reduces the time that neural network training learn, but also improve diagnostic accuracy. This method takes advantage of knowledge reduction capacity of rough set theory and fault-tolerant ability of artificial neural networks. It improves the ability of generalization and make up for their shortcomings.At last, this article use VB language as development tool and using Matlab neural network tool box to establish a simplified distribution network fault diagnosis system. It is verified through the distribution network examples that the method is correctness and effectiveness.更多还原