【作者】 杨健维；

【导师】 何正友；

【作者基本信息】 西南交通大学 ， 电力系统及其自动化， 2011， 博士

【摘要】 电网规模的不断扩大,联网程度的增强可以提高电网的可靠性与供电质量,但同时也带来了潜在的危险。为了降低或避免事故所带来的影响,需要及时检测电网的各种故障,快速准确地分类故障。同时,智能电网的大力发展,也迫切的要求及时发现、准确诊断并消除故障隐患,提升系统的安全可靠性。因此,研究和建立起一套高效的电网故障智能诊断系统,对于及时认知电网中设备事故,处理电网故障是十分必要的。目前该领域已有不少研究工作,但警报信息丢失、虚假信息或信号不完备等现场不可避免的因素使故障诊断结果的准确性难以保证,阻碍了电网故障诊断方法的实用化。Petri网方法可以清晰、直观、准确地描述和研究电网中各个元件在故障状态下的离散动态行为,具有良好的容错性。随着Petri网的广泛应用,众多专家学者又进一步研究了模糊Petri网,它更符合人类的思维和认知方式。本论文主要采用模糊Petri网方法,研究了自适应模糊Petri网、方向性加权模糊Petri网、计及时序信息的模糊Petri网以及混合模糊Petri网,并将其应用于电网故障元件的诊断和故障类型的识别。论文首先研究了自适应模糊Petri网及其在电网故障诊断中的应用。为了避免Petri网应用过程中由于人为主观因素造成的误差,并充分考虑保护、断路器信息对故障诊断的影响程度,提出了基于自适应模糊Petri网的电网故障诊断方法。根据SCADA上传的保护、断路器信息,应用BP神经网络对自适应模糊Petri网的权值进行调整。在应用过程中,利用由统计规律得到的保护动作正确率来表示保护和断路器动作的真实值,使推理过程更令人信服,解释更有现实依据。利用220kV电网模型对该方法进行验证,结果表明在保护误动、拒动以及存在信息丢失等情况下,该方法都能够给出正确的诊断结果。针对传统Petri网对网络拓扑结构适应性差的缺点,提出了基于方向性加权模糊Petri网的电网故障诊断新方法。在诊断过程中对元件的每个扩展方向分别建立模糊Petri网模型。网络拓扑变化时,该方法不需要重新建模,具有较好的结构适应性。以14节点电网模型为例,对存在不确定信息情况下的单一故障和复杂故障进行了仿真分析。仿真结果表明该方法具有较好的结构适应性和容错性,为电网故障诊断提供了新的思路。研究了计及信息时序属性的警报信息处理方法。针对现有的故障诊断方法没有充分运用警报信息时序特性的缺陷,在不考虑算法结构适应性的情况下,定义了时间约束通路的概念,直观地表达了警报信息之间的时序特性,分析了警报差错信息的识别算法。以220kV电网模型为例,对不考虑时间约束和考虑时间约束两种情况下的故障诊断结果进行了对比分析。分析结果表明考虑警报信息的时序属性,能够更加准确地描述警报信息间的时间约束关系,使诊断结果更加精确。在考虑算法结构适应性的情况下,研究了基于时序模糊Petri网的电网故障诊断方法。应用该算法对14节点电网模型的部分算例进行了测试,并对适应电网拓扑结构变化的快速修正方法进行了研究。仿真结果表明该方法充分地利用了信息的时序属性,不仅能够给出更加精确的诊断结果,并且对电网拓扑变化具有较好的适应能力,适用于大规模复杂电网的故障诊断,具有较好的应用前景。研究了基于混合模糊Petri网的高压输电线路故障类型识别方法。针对单一的Petri网无法完成故障类型识别的缺点,研究了基于混合模糊Petri网的高压输电线路故障类型识别方法。混合模糊Petri网方法将小波变换的特征提取、模糊逻辑相结合作为模糊Petri网输入向量的前端处理模块,最终给出各种故障发生的可信度,可望为工作人员提供良好的辅助决策。以500kV高压输电线路模型为例,在不同故障类型、故障电压初始角、故障过渡电阻和故障位置等工况下验证故障类型识别的效果。并对该方法在不同小波基、不同数据窗、不同故障工况、不同噪声和不同线路参数等情况下的适应性进行了分析。仿真结果表明混合模糊Petri网方法在故障类型识别中取得了理想的分类效果,并且能够较方便地移植到其他网络中。最后,在Windows的操作平台上,以Matlab的图形用户界面(Graphical User Interfaces, GUI)为工具,设计并开发了一套电网故障元件与故障类型集中诊断的故障诊断系统。该系统主要包括故障元件诊断和故障类型识别两个主要功能。本论文是国家自然科学基金项目《基于信息理论的多信源电网故障诊断方法及应用》(No.50877068,2009-2011)成果的组成部分。更多还原

【Abstract】 The reliability and power quality of the power network have been improved with continuous expansion of the power network and enhancement of the network connection. However, the potential risk of power system malfunction is also increased. To reduce or avoid the effect of power system malfunction, quick various power system faults detection and accurate different fault type classification are necessary. Meantime, as the development of the smart grid, quick detecting, accurate diagnosis and eliminating the potential fault are required to improve the system safety and reliability. Therefore, research and foundation of a high performance and intelligent power network fault diagnosis system is essential for artificial cognition and treatment of the power system malfunction.Currently, many research results have already issued at the related field. However, the greatest obstacle, which hinders the application of the fault diagnosis system, is the unavoidable on-site factors like losing the alarm information, false information and incomplete signal, etc. These factors will result in aberration of the fault characteristic and unassured fault diagnosis. Petri nets can clearly, visually and precisely depict and research the discrete dynamic event in fault state of each component. It has shown good fault tolerance capability. With wide application of Petri nets, fuzzy Petri nets which imitates human intelligent cognition was proposed by experts and researchers. Fuzzy Petri nets, adaptive fuzzy Petri nets, directional weighted fuzzy Petri nets, temporal order fuzzy Petri nets and hybrid fuzzy Petri nets are presented in this paper. Methodology of fuzzy Petri nets is applied in power system fault component diagnosis and fault type recognition.Adaptive fuzzy Petri nets and its application in power system fault diagnosis are studied. This method not only avoids errors caused by subjective factors in certain degree, but also fully considerate the influence caused by protective relays and circuit breakers. The diagnosis knowledge is adjusted by BP neural network. Diagnosis precision is improved by applying statistic relay correct action rate to represent the action value of protective relays and circuit breakers. Inference procedure is based on reality and more convinced. A220kV power system network model is used as verification. Results show that correct diagnosis can still be obtained even with protection malfunction, protection rejection and protection information loss.A new power system fault diagnosis method based on directional weighted fuzzy Petri nets is proposed. The traditional fuzzy Petri nets for power system fault diagnosis is unable to adapt to topology changes. In directional weighted fuzzy Petri nets fault diagnosis method, fuzzy Petri nets model is build on each fault spread direction. The method can adapt to topology changes automatically without remodeling and obtain good structure adaptability. A14nodes power network is chosen as an example for single fault and complex fault diagnosis under uncertain information. Simulation results show good structure adaptation and excellent fault tolerance capability. It provides new thoughts in power network fault diagnosis.Alarm information processing method with temporal order is proposed in this paper. Aiming at solving the problem of insufficient consideration of temporal order characteristic of the alarm information, time constraint route is presented without the algorithm structure adaptability. Temporal order characteristic is directly expressed and the distinguished algorithm for alarm error information is analyzed. Comparison of fault diagnosis results between two scenarios:with time constraint and without time constraint are made to a220kV power network as an example. After the consideration of alarm information’s temporal order, time constraint relationship between alarm information is depicted in detail and more accurate results are obtained. Fault diagnosis method based on fuzzy Petri nets involing temporal order information is proposed taking the algorithm structure adaptability into account. A fast model revision algorithm for power network topology variation has been studied to improve the remodeling problem when the network topology changes. Some examples of a14-node power network are used as verification. Simulation results show that the novel method can obtain better diagnosis results and be adapted to the network topology change for large power system fault diagnosis with the consideration of temporal order characteristic.Because single Petri nets can not be used to recognize fault type, a new hybrid fuzzy Petri nets based method for recognizing fault type in high voltage transmission line is proposed. The feature extraction of wavelet transform and fuzzy logic are integrated as pre-processing for input vectors of fuzzy Petri nets. Through deduction of fuzzy Petri nets, statistic results of various fault happening reliability rate are given to staff for assisting decision-making. Take a500kV HVDC transmission line as an example, various simulation scenarios, like different fault types, initial angles of fault voltage, transition resistances and fault locations, etc. are chosen for simulation verification. Results show that this method has good adaptability in different wavelets, data windows, fault conditions, noises and line parameters. Successful experiment results are obtained with hybrid fuzzy Petri nets application in power system fault diagnosis. Moreover, this method can be easily migrated to other network.At last, an integrated fault diagnosis platform with power system fault component and type recognition function is developed based on Matlab graphical user interfaces (GUI). The platform has two subsystems:intelligent fault component diagnosis and intelligent fault type recognition.The thesis is supported by National Natural Science Foundation of China:’Information theory based power network fault diagnosis of multi-sourced signal’(No.50877068,2009-2011).更多还原