【作者】 孙向飞；

【导师】 束洪春；

【作者基本信息】 昆明理工大学 ， 水利水电工程， 2003， 硕士

【摘要】 软计算工具主要包括人工神经网络、模糊集理论、进化算法和粗糙集理论等。粗糙集理论是一种较新的软计算方法，它能有效地分析和处理不精确、不一致、不完整等各种不完备信息，并从中发现隐含的知识，揭示潜在的规律，是一个强大的数据分析工具，具有良好的容错性能。粗糙集理论在某些方面可以弥补其它软计算方法的不足，其在电力系统中的研究和应用还非常少，尚处在起步、探索阶段。本文利用粗糙集理论在处理不完备信息方面的独特性能，以及强大的数据分析能力和容错性能，对其在电力系统中的应用，进行探索研究。主要内容有： 1．基于粗糙集理论的配电网故障诊断研究 配电网一旦发生事故，如何快速、准确地对故障发生区位进行诊断并有效地隔离，是提高供电可靠性的关键问题，也是实现配电自动化的重要研究课题之一。近年来，各级调度中心引入使用的SCADA系统，能够将配电网中的实时信息及时提供给调度员，为配电自动化水平的提高奠定了基础。然而，现有的SCADA系统在配电网发生故障时提供给调度员的警报信息有如下局限性：(1)面向现场的户外馈线终端FTU运行环境恶劣，承受强的电磁、雷电干扰，并且由于继电器节点故障、FTU元器件损坏等因素的存在，使得配电网信息受干扰、畸变的可能性较高；(2)在信号传输及变换过程中，通讯装置的故障等原因，也可能导致信号出错甚至丢失；(3)保护装置和断路器本身也有可能误动或拒动。以上原因使故障后的系统响应复杂化，产生不完备警报信息，给故障诊断工作造成很大困难。 针对配电网故障模式存在不完备信息的情况，本文运用粗糙集理论探索了一种配电网故障诊断的新方法。其思路是把保护和断路器的信号作为对故障分类的条件属性集，考虑各种可能发生的故障情况建立决策表，然后进行决策表化简，最后抽取出诊断规则。这种方法揭示了故障信息集合内在的冗余性，能够区分关键信号和非关键信号，可以在不精确、不完备的警报信息模式下达到正确诊断的目的。 2．基于气相色谱分析和粗糙集理论的电力变压器故障诊断专家系统方案设计 变压器内部故障的检测有多种手段，IEEE认为对变压器内部故障早期诊断最有效的方法是油中气体色谱分析法。由于目前对于变压器的故障机理尚未清楚，以及监测手段存在一定的局限性，要建立故障现象与故障原因之间的精确数学模型是十分困昆明理工大学硕士学位论文难的。专家系统可以有效地模拟专家的决策过程，被国内外电力系统工作者广泛探索用于变压器故障诊断，但难以获取完备知识的瓶颈问题一直制约着专家系统的发展。 本文提出了一种基于气相色谱分析和粗糙集理论的电力变压器故障诊断专家系统设计方案，并着重研究了变压器故障诊断专家系统知识库建立和维护的粗糙集方法。该方法从历史故障数据所形成决策表的约简出发，通过计算规则的粗糙隶属度，形成不同简化层次上符合置信度要求的节点网络规则集。随着故障样本的增多，重新计算每个节点的规则的粗糙隶属度，错误样本将被众多正确样本“淹没”，从而实现知识库的维护和自适应能力。用变压器故障信息与知识库中相应节点的规则集进行匹配，即使在气相色谱分析数据不完备的情况下，也能得到正确的诊断结果。更多还原

【Abstract】 Soft computing includes Artificial Neural Network, Fuzzy Logic, Evolutionary Algorithms, Rough Set (RS) Theory, etc. As a new soft computing, Rough Set can analyze and handle imprecise, inconsistent and incomplete data efficiently. In addition, connotative knowledge and latent rules will be discovered by using Rough Set Theory. Therefore, Rough Set is a powerful tool for analyzing data and is tolerant for faults. Hence, Rough Set can overcome shortages of other soft computing in some aspects. But the researching and application of RS in Power System are very rare. So, using the advantages of RS, the paper does some pilot study on Power System. The majority of work is reported as follows:1. Study on fault diagnosis for distribution network based on Rough Set TheoryThe quick and precise fault diagnosis is the key problem to enhance the reliability of power supply and is one of the most important problems in distribution automation. Nowadays, the real time information of distribution network can be afforded to operators through Supervisory Control and Data Acquisition (SCADA). But, the alarm signals from SCADA have disadvantages as follows: (1) The open-air Feeder Terminal Units (FTU) are interfered by strong electromagnetic and thunder. Besides, the failure of relays and the fault of FTU can also lead to that fault information is interfered and aberrant. (2) In the process of transmission and commutation, the error in the communication equipment may result in the signals err or lose. (3) The malfunction or failing operation of protective relays and circuit breakers may lead to imperfect alarm signals. Since these factors, the alarm signals are imperfect, and make fault diagnosis more difficult.Based on Rough Set Theory, the paper proposes a new distribution network fault diagnosis approach to deal with the imperfect alarm signals. Firstly, a decision table including all kinds of fault cases is established by considering the signals of protective relays and circuit breakers. Then, diagnostic rules are extracted by reducing the decision table. Using the reductions of decision table, diagnosis rules can be obtained directly from fault samples which have been established. The method can tell indispensable fault signals from dispensable ones and discover the inherent redundancy of alarm signal set. In a word, the method can realize effective fault diagnosis when the alarm signals are imperfect.2. Design the scheme of Expert System for transformer fault diagnosis based on DGA and Rough Set TheoryIEEE considers Dissolved Gas Analysis (DGA) is the most efficient method to examine the interior fault of transformer. Because the mechanism of transformer fault is yet not clear and the instruments have some limitation, it is difficult to establish the math model of fault phenomena and causes. Therefore, Expert System, which can simulate the decision-making of experts, has been abroad applied to diagnose the fault of transformers. However, obtaining complete knowledge, as a bottleneck problem, always restrict the development of Expert System.Based on DGA and Rough Set Theory, the paper designs a scheme of expert system for fault diagnosis of power transformer. In order to resolve the bottleneck problem of obtaining complete knowledge of Expert System, a Rough Set approach is mainly proposed to build and maintain knowledge base for transformer fault diagnosis. From the reductions of decision table defined by history fault data, a series of nodes network rule sets, with suitable belief degree under different reductive levels, is developed by calculating the rough subjection degree of every rule. With the increasing of fault samples, the error samples will be submerged by right samples through computing the rough subjection degree of every rule. In this way, the knowledge base is maintained. When the fault information of transformer is given, one can match the information to the rule sets of relative nodes. Even when the data of DGA are imperfect, the diagnosis results are correct.更多还原