基于自适应指纹识别的电力系统复杂原发故障诊断方法
发布时间：2018-09-09 10:41
【摘要】：根据“8·14”大停电等典型电力系统故障的演变规律,针对电力系统的故障一般都可以划分为缓慢的相继开断、快速的相继开断、短暂的振荡、大面积雪崩和漫长的恢复等5个阶段的特点,本文把研究目标确定为实现对电力系统的复杂原发故障的快速准确诊断,希望设计一种能够自适应电网变化的实时的电力系统故障诊断方法,在故障发生的初期就可以快速分析出故障节点,从而有效辅助调度员进行故障处理,阻止故障的进一步扩大。通过对国内外电力系统故障诊断方法的研究现状的分析,本文发现目前提出的各种故障诊断方法都存在着两个共性问题：1)电网运行方式的复杂多变性对故障诊断的影响问题：电网网架结构及运行方式的变化很频繁,导致这些故障诊断方法的模型、规则、算法均难以适应,从而影响了这些方法的实用化；2)开关保护信号的不准确性和不完备性对故障诊断的影响问题：开关保护装置的误动拒动及动作信号的误报漏报导致目前这些故障诊断方法的诊断结果的错误率较高,也影响了这些方法的实用性；对于电网运行方式的复杂多变性问题,运行方式的频繁变化对故障诊断的影响主要体现在故障时对保护装置以及备自投装置的动作行为逻辑的分析上。本文借鉴复杂适应系统(Complex Adaptive System,CAS)理论的研究方法,结合本文故障诊断方法的具体需求,提出了一种对保护及备自投装置的模型及其动作逻辑的自适应建模分析算法,给电网各设备赋予了动态自适应电网变化的能力,在此基础上提出了一种电网复杂原发故障的预想故障集自适应分析搜索方法,实现不用人工干预自动分析出当前电网的所有可能发生的预想故障,而且当电网发生变化时,可以基于变化后的电网运行方式重新自动更新这些预想故障。预想故障集的构建是本文自适应指纹识别故障诊断方法的基础。针对开关保护信号的不确定性问题,本文提出了一种综合利用电网故障时的动作信号信息和电网各支路潮流等量测信息作为诊断依据的新思想,通过提取不同类型故障发生时电网各支路潮流的不同变化特征作为区别每种故障的“潮流指纹”,并提取故障发生时开关保护动作信号之间的动作逻辑关系形成用来量化表征各种电网故障的“动作信号指纹”,借鉴实际人体指纹识别系统的成熟经验,采用模式识别的方法进行故障的识别诊断。为了实现该思想,本文重点研究了对这两种指纹的提取方法：1)通过对电网故障时开关保护动作信息的深入挖掘分析,提出了从动作情况、动作之间的时序约束关系、拓扑约束关系、相关量测变化的约束关系等方面来量化电网故障的动作信号指纹的自适应提取方法。该方法一方面继承了已有的专家系统知识库的诊断知识,另一方面通过这种把离散型信号转化为连续性变量的量化处理,可以在很大程度上避免专家系统推理机会受信号准确性的影响而导致推理错误问题,而且量化后的结果可以方便的实现与其他故障诊断方法的结合,如下文的基于潮流指纹的故障诊断。2)提出了一种基于主成分分析方法的潮流指纹特征向量提取方法：通过对原始特征的在均方差最小意义上的线性变换,剔除具有相关性和重叠性的变量,在不损失原始数据的主要信息的前提下,实现了以较少的综合性正交变量来替代原始变量,构建出维数较低的故障潮流指纹主成分特征向量。实践证明,基于该空间进行故障诊断具有更高的效率和精度。在此基础上,本文根据潮流指纹和动作信号指纹的特点分别设计了相应的故障识别策略,并在分析了这两种方法各自的诊断盲区后,进一步提出了一种基于动作信号指纹和潮流指纹的故障诊断融合识别策略。该策略基于D-S证据理论的证据组合方法,把潮流指纹信息与动作信号指纹信息相融合,很大程度上解决了由于上述两问题对故障诊断的影响,从而提高本文自适应故障识别诊断方法的准确性和实用性。最后,本文用一个具体的地区电网故障实例,对本文提出的复杂原发故障自适应指纹识别诊断方法进行实际应用分析,证明了本方法在电力系统故障诊断中的有效性和实用性。
[Abstract]:According to the evolution law of typical power system faults such as "8.14" blackout, power system faults can be generally divided into five stages: slow successive interruption, fast successive interruption, short oscillation, large area avalanche and long recovery. The research goal is to realize the complex origin of power system. In order to diagnose faults quickly and accurately, we hope to design a real-time fault diagnosis method which can adapt to the changes of power grid. In the early stage of faults, fault nodes can be quickly analyzed, which can effectively assist dispatchers to deal with faults and prevent further expansion of faults. Based on the analysis of the research status of the fault diagnosis methods, this paper finds that there are two common problems in all kinds of fault diagnosis methods: 1) the influence of the complex and changeable operation modes of the power grid on the fault diagnosis: the frequent changes of the grid structure and operation modes lead to the difficulty of the models, rules and algorithms of these fault diagnosis methods. 2) The influence of the inaccuracy and incompleteness of the switching protection signal on the fault diagnosis: the misoperation of the switching protection device and the false alarm and missed alarm of the action signal lead to the high error rate of the diagnosis results of the current fault diagnosis methods, and also affect the reality of these methods. For the complex and changeable operation mode of power grid, the influence of frequent change of operation mode on fault diagnosis is mainly reflected in the analysis of action logic of protective devices and automatic switching devices. In order to meet the specific requirements of fault diagnosis methods, an adaptive modeling and analysis algorithm for the model and operation logic of protective and standby automatic switching devices is proposed, which endows each device with the ability of dynamic self-adapting to the changes of power grid. On this basis, an adaptive analysis and search method for the expected fault set of complex primary faults in power grid is proposed. All the possible expected faults in the current power grid can be automatically analyzed without manual intervention, and when the power grid changes, these expected faults can be automatically updated based on the changed operation mode of the power grid. In this paper, a new idea is proposed, which integrates the action signal information of power system faults and the measurement information of power flow of each branch of the power system as the diagnostic basis. Taking the action logic relation between the action signals of the switching protection when the fault occurs to form the "action signal fingerprint" which is used to quantify and characterize various power network faults, and referring to the mature experience of the actual human fingerprint identification system, the method of pattern recognition is used to identify and diagnose the faults. In order to realize this idea, this paper focuses on the two methods. This paper presents an adaptive fingerprint extraction method to quantify the fingerprint of power system fault action signals from the following aspects: action condition, time sequence constraint relation between actions, topological constraint relation, constraint relation of correlation measurement change, etc. On the one hand, it inherits the diagnostic knowledge of the existing expert system knowledge base, on the other hand, by quantifying the discrete signals into continuous variables, it can largely avoid the inference error caused by the influence of the accuracy of the signals on the inference opportunities of the expert system, and the quantified results can be easily realized. Combining with other fault diagnosis methods, the following is the fault diagnosis based on power flow fingerprint. 2) A power flow fingerprint feature vector extraction method based on principal component analysis is proposed. By linear transformation of the original features in the sense of minimum mean square deviation, the variables with correlation and overlap are eliminated without losing the original features. On the premise of the main information of the data, the principal component eigenvector of fault power flow fingerprint with lower dimension is constructed by replacing the original variables with fewer comprehensive orthogonal variables. Practice proves that fault diagnosis based on this space has higher efficiency and accuracy. After analyzing the blind areas of the two methods, a fault diagnosis fusion identification strategy based on action signal fingerprint and power flow fingerprint is proposed. The strategy is based on the combination of D-S evidence theory and power flow fingerprint. The fusion of fingerprint information solves the influence of the above two problems on fault diagnosis to a great extent, so as to improve the accuracy and practicability of the adaptive fault diagnosis method in this paper. Application analysis shows that this method is effective and practical in power system fault diagnosis.
【学位授予单位】：华北电力大学(北京)
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TM732
本文编号：2232146
[Abstract]:According to the evolution law of typical power system faults such as "8.14" blackout, power system faults can be generally divided into five stages: slow successive interruption, fast successive interruption, short oscillation, large area avalanche and long recovery. The research goal is to realize the complex origin of power system. In order to diagnose faults quickly and accurately, we hope to design a real-time fault diagnosis method which can adapt to the changes of power grid. In the early stage of faults, fault nodes can be quickly analyzed, which can effectively assist dispatchers to deal with faults and prevent further expansion of faults. Based on the analysis of the research status of the fault diagnosis methods, this paper finds that there are two common problems in all kinds of fault diagnosis methods: 1) the influence of the complex and changeable operation modes of the power grid on the fault diagnosis: the frequent changes of the grid structure and operation modes lead to the difficulty of the models, rules and algorithms of these fault diagnosis methods. 2) The influence of the inaccuracy and incompleteness of the switching protection signal on the fault diagnosis: the misoperation of the switching protection device and the false alarm and missed alarm of the action signal lead to the high error rate of the diagnosis results of the current fault diagnosis methods, and also affect the reality of these methods. For the complex and changeable operation mode of power grid, the influence of frequent change of operation mode on fault diagnosis is mainly reflected in the analysis of action logic of protective devices and automatic switching devices. In order to meet the specific requirements of fault diagnosis methods, an adaptive modeling and analysis algorithm for the model and operation logic of protective and standby automatic switching devices is proposed, which endows each device with the ability of dynamic self-adapting to the changes of power grid. On this basis, an adaptive analysis and search method for the expected fault set of complex primary faults in power grid is proposed. All the possible expected faults in the current power grid can be automatically analyzed without manual intervention, and when the power grid changes, these expected faults can be automatically updated based on the changed operation mode of the power grid. In this paper, a new idea is proposed, which integrates the action signal information of power system faults and the measurement information of power flow of each branch of the power system as the diagnostic basis. Taking the action logic relation between the action signals of the switching protection when the fault occurs to form the "action signal fingerprint" which is used to quantify and characterize various power network faults, and referring to the mature experience of the actual human fingerprint identification system, the method of pattern recognition is used to identify and diagnose the faults. In order to realize this idea, this paper focuses on the two methods. This paper presents an adaptive fingerprint extraction method to quantify the fingerprint of power system fault action signals from the following aspects: action condition, time sequence constraint relation between actions, topological constraint relation, constraint relation of correlation measurement change, etc. On the one hand, it inherits the diagnostic knowledge of the existing expert system knowledge base, on the other hand, by quantifying the discrete signals into continuous variables, it can largely avoid the inference error caused by the influence of the accuracy of the signals on the inference opportunities of the expert system, and the quantified results can be easily realized. Combining with other fault diagnosis methods, the following is the fault diagnosis based on power flow fingerprint. 2) A power flow fingerprint feature vector extraction method based on principal component analysis is proposed. By linear transformation of the original features in the sense of minimum mean square deviation, the variables with correlation and overlap are eliminated without losing the original features. On the premise of the main information of the data, the principal component eigenvector of fault power flow fingerprint with lower dimension is constructed by replacing the original variables with fewer comprehensive orthogonal variables. Practice proves that fault diagnosis based on this space has higher efficiency and accuracy. After analyzing the blind areas of the two methods, a fault diagnosis fusion identification strategy based on action signal fingerprint and power flow fingerprint is proposed. The strategy is based on the combination of D-S evidence theory and power flow fingerprint. The fusion of fingerprint information solves the influence of the above two problems on fault diagnosis to a great extent, so as to improve the accuracy and practicability of the adaptive fault diagnosis method in this paper. Application analysis shows that this method is effective and practical in power system fault diagnosis.
【学位授予单位】：华北电力大学(北京)
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TM732
【相似文献】
相关博士学位论文 前1条
1 刘延乐;基于自适应指纹识别的电力系统复杂原发故障诊断方法[D];华北电力大学(北京);2014年
，本文编号：2232146
本文链接：https://www.wllwen.com/kejilunwen/dianlilw/2232146.html
