WO2020133735A1 - 基于类人概念学习的配电网早期故障检测方法 - Google Patents

基于类人概念学习的配电网早期故障检测方法 Download PDF

Info

Publication number
WO2020133735A1
WO2020133735A1 PCT/CN2019/078575 CN2019078575W WO2020133735A1 WO 2020133735 A1 WO2020133735 A1 WO 2020133735A1 CN 2019078575 W CN2019078575 W CN 2019078575W WO 2020133735 A1 WO2020133735 A1 WO 2020133735A1
Authority
WO
WIPO (PCT)
Prior art keywords
distortion
primitives
waveform
distribution network
humanoid
Prior art date
Application number
PCT/CN2019/078575
Other languages
English (en)
French (fr)
Inventor
刘亚东
熊思衡
丛子涵
罗林根
江秀臣
Original Assignee
上海交通大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 上海交通大学 filed Critical 上海交通大学
Priority to JP2019538613A priority Critical patent/JP7107580B2/ja
Priority to GB1912002.1A priority patent/GB2582676B/en
Priority to US16/674,233 priority patent/US11143686B2/en
Publication of WO2020133735A1 publication Critical patent/WO2020133735A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
    • G01R31/08—Locating faults in cables, transmission lines, or networks

Definitions

  • the invention relates to the technical field of early fault detection of a distribution network, in particular to a method for early fault detection of a distribution network based on humanoid concept learning.
  • Power supply reliability is the most important evaluation index of distribution network. Due to the large number of distribution network equipment and wide area, its operation and maintenance work is mainly focused on the post-fault treatment, such as fault location, fault isolation, and fault recovery. However, with the increase in the reliability requirements of power supply in the country and the electricity sales market, the troubleshooting of the distribution network must not only pay attention to the restoration of power supply after the failure, but also pay attention to the early warning of the equipment before the failure. "Inspection” has been changed to "active early warning”, which eliminates permanent faults before they occur, greatly reducing the probability of power outages caused by equipment failures.
  • Early fault detection as a way of power state detection, provides new ideas for distribution network operation and maintenance, so that defective equipment can be replaced in advance, improving power supply reliability. At the same time, it reduces the operation and maintenance workload and saves costs.
  • Model-driven is often limited to a single model and cannot adapt to complex actual situations. Therefore, data-driven methods are often used. Traditional data-driven methods require large amounts of data and often only target individual scenarios, so improvements are needed.
  • the concept of humanoid learns to decompose waveforms by analogy with the process of human observation of waveforms, reconstruct the waveform generation process according to the decomposition results, and realize the waveform recognition by learning this process.
  • how to apply humanoid concept learning to early fault diagnosis of distribution networks is still a problem that needs to be solved urgently in this field.
  • the purpose of the present invention is to introduce the relevant theories and methods of humanoid concept learning into early fault diagnosis of the distribution network, and to propose a method for early fault detection of the distribution network through theoretical analysis Distribution network early fault detection method based on human-like concept learning, and verify the rationality of the method.
  • the concept of humanoid learning simulates the process of humans recognizing waveforms, decomposes the waveform into a superposition of different components, and recognizes the waveform by learning the components and the time relationship between the components.
  • the early fault detection method of distribution network based on humanoid concept learning proposed by the present invention has the characteristics of introducing a priori knowledge, requiring fewer samples, and having higher accuracy.
  • the present invention is achieved by the following technical solutions.
  • An early fault detection method for distribution network based on humanoid concept learning includes the following steps:
  • the wavelet transform is used to decompose the waveform into an approximate part and a detail part.
  • the approximate part is called a rough shape primitive, and the detail part is called a distortion primitive;
  • the wavelet transform function is selected as a 5-layer Meyer function
  • the approximate shape is taken as a 5 coefficient
  • the distortion is taken as the difference between the original waveform and the a 5 coefficient.
  • the distortion is decomposed, the curve is split into multiple segments according to the extreme points in the distortion curve, and each segment is combined with adjacent segments to form pulses, harmonics and Three other primitive distortions.
  • the rule for combining each segment with adjacent segments is:
  • the amplitude threshold is set to 0.5 times the fundamental wave amplitude; the time length threshold is set to 0.25 times the fundamental wave cycle.
  • feature extraction of primitives includes the following extraction principles:
  • a o extract the amplitude of each cycle, the length of time T o and a DC component A oft; for harmonic z h, extract the amplitude A h, the frequency f h and the length of the total time t h; for pulse z p , extract amplitude A p , pulse width t p ; for other distortions z other , do not extract features.
  • the time relationship between the primitives includes the time relationship between the approximate shape and distortion and the time relationship between distortion and distortion; wherein:
  • the time relationship between the approximate shape and the distortion is called the relative fundamental wave position Po , and the relative fundamental wave position Po describes the position on the approximate shape at the initial moment of distortion, and this position is expressed by the phase angle;
  • the temporal relationship between distortion and distortion includes:
  • the single-phase element pair PP uni describes two distortions that are the same or close at the initial moment in the voltage or current waveform of the same phase
  • the three-phase element pair PP tri describes three distortions in the three-phase voltage or current waveform that are the same or close at the initial moment.
  • the time relationship related to other distortions is ignored.
  • the constructed waveform probability distribution formula is:
  • ⁇ w is a known waveform type
  • the noise follows a normal distribution S N ⁇ N( ⁇ , ⁇ 2 )
  • the number of primitives is ⁇
  • the primitive feature parameter is p
  • the time relationship between primitives is R;
  • the neutral point current is summed to produce 7 different types of waveforms (I A , I B , I C , I N , U A , U B , U C ),
  • the probability distribution formula in different types of waveforms to obtain abnormal events is:
  • ⁇ E is a known kind of event
  • the waveform w ⁇ I A, I B , I C, I N, U A, U B, U C ⁇ .
  • the type of abnormal event is determined according to the probability distribution of the abnormal event in different types of waveforms, that is, the comparison In the size of different types of waveforms, the waveform type corresponding to the maximum value is taken to obtain the waveform judgment result.
  • the present invention has the following beneficial effects:
  • the present invention provides an early fault detection method for distribution network based on humanoid concept learning. It introduces related theories and methods of humanoid concept learning into early fault detection of distribution network. Fault detection algorithm and verify the rationality of the algorithm. As a kind of visual concept, voltage and current waveforms are decomposed into rough shapes and various distortions. By calculating the probability distribution of each component, the probability distribution of the entire waveform can be obtained to determine the type of waveform.
  • the early fault detection method for distribution network based on humanoid concept learning provided by the present invention is much better than traditional detection in terms of required data volume and accuracy, and is of great significance for the detection and processing of early faults in distribution network.
  • FIG. 1 is a schematic diagram of waveform decomposition provided by an embodiment of the present invention.
  • FIG. 2 is a schematic diagram of distortion decomposition provided by an embodiment of the present invention.
  • FIG. 3 is a schematic diagram of the definition of the time relationship between primitives provided by an embodiment of the present invention. wherein, (a) is a schematic diagram of the definition of the time relationship between the general shape and distortion; (b) is the time relationship between the distortion and distortion Define the schematic diagram;
  • FIG. 4 is a schematic diagram of a waveform generation process provided by an embodiment of the present invention.
  • FIG. 5 is a working flowchart of an early fault detection method for a distribution network based on humanoid concept learning provided by an embodiment of the present invention.
  • this embodiment provides an early fault detection method for a distribution network based on humanoid concept learning, including the following steps:
  • Step S1 The wavelet transform is used to decompose the waveform into an approximate part and a detail part, wherein the two parts, the approximate part and the detail part, are called a rough shape primitive and a distortion primitive, respectively.
  • Step S2 According to the extreme point, the distortion (detail part) is divided into three elements: harmonics, pulses, and other distortions.
  • Step S3 Extract the features of the primitives and the time relationship between the primitives.
  • Step S4 Construct the probability distribution of the waveform according to the characteristics of the primitives and the time relationship between the primitives.
  • Step S5 Obtain the judgment result of the waveform according to the probability distribution of different types of waveforms.
  • the wavelet function is selected as a 5-layer Meyer function
  • the approximate shape is taken as a 5 coefficient
  • the distortion is taken as the difference between the original waveform and the a 5 coefficient. Decompose the results.
  • split the curve into small segments according to the extreme points in the distortion curve as shown in the first step of Figure 2, combine each segment with the surrounding small segments to form a pulse , Harmonics, and other distortions are shown in the second step shown in Figure 2.
  • the combination rule is as follows: 1) If the adjacent small segments have opposite monotonicity, the amplitude and time length are small, and there are more than three segments, they constitute harmonics (as shown in the combination of 1, 2, 3, and 4 segments in Figure 2) ; 2) If the two adjacent sections have opposite monotonicity, the amplitude exceeds the threshold (here set to 0.5 times the fundamental amplitude), and the length of time is less than the threshold (here set to 0.25 times the fundamental frequency), then constitute a pulse (as shown in the figure The combination of 5 and 6 segments shown in 2); 3) The small segments that cannot form harmonics and pulses are called other distortions (7 and 8 segments shown in Figure 2).
  • the general shape, harmonics, pulses, and other distortions are collectively referred to as primitives.
  • the primitive features are extracted as follows: for the general shape z o , the amplitude A o of each cycle, the time length T o and the DC component A oft are extracted; for the harmonic z h , the amplitude A h , the frequency f h and the total time are extracted Length t h ; for pulse z p , amplitude A p and pulse width t p are extracted; for other distortion z other , since z other is not related to early faults, its characteristics are not extracted. There is also a time relationship between primitives. Primitives can be divided into two types: approximate shape and distortion (harmonics, pulses, other distortions), and the time relationship between primitives can also be divided into approximate shape and distortion.
  • Time relationship between distortion and distortion Similarly, the time relationship related to z other is not discussed.
  • Time relationship between the shape of a substantially distortion fundamental primitives called relative position P o, P o is the position described distortion primitive initial moment in the general shape, the position indicated by the phase angle.
  • the time relationship between the distortion primitives is as follows: interval time t int , single-phase primitive pair PP uni and three-phase primitive pair PP tri .
  • t int describes the time interval between two adjacent distortion primitives.
  • Single-phase primitives describe PP uni as two distortion primitives with the same initial voltage and close to each other in the voltage and current waveforms of the same phase.
  • the three-phase element pair PP tri describes three distortion element elements that are the same at the initial moment in the three-phase voltage/current waveform and are close to them.
  • PP uni and PP tri describe a correlation.
  • n uni and n tri are used to denote the number of the two.
  • the schematic diagram of the above time relationship is shown in Figures 3(a) and (b).
  • a waveform has been uniquely decomposed into a combination of the above primitives, and primitive features and temporal relationships are used to describe this combination.
  • a waveform generation process is proposed based on the above decomposition, as shown in FIG. 4. This process is described as follows: First, select some primitives in the primitive library (the primitive type is 4, the number of each primitive is arbitrary), each primitive has its own characteristic parameters; combine these primitives in order Formed into a waveform, these sequences constitute the time relationship between the primitives. The above belongs to the category level. The type of waveform is determined by the cause of the fault, the faulty device, and the faulty location.
  • the waveforms of the same type are affected by line parameters, grid structure, load conditions, sensor parameters, noise, etc., and will appear as different cases.
  • the characteristic parameters and time relationship of the primitive will change to some extent, and noise will be introduced. The above belongs to the case level.
  • the probability distribution formula of the waveform can be derived:
  • ⁇ w is a known waveform type
  • the noise follows a normal distribution S N ⁇ N( ⁇ , ⁇ 2 )
  • the number of primitives is ⁇
  • the primitive feature parameter is p
  • the time relationship between primitives is R.
  • the corresponding three-phase voltage and three-phase current are often recorded.
  • the neutral point current can be obtained by summing the three-phase current, so that 7 different types of waveforms (I A , I B , I C , I N , U A , U B , U C ). Therefore, the probability distribution of an abnormal event in different types of waveforms can be written as:
  • ⁇ E is a known kind of event
  • the waveform w ⁇ I A, I B , I C, I N, U A, U B, U C ⁇ .
  • the type of abnormal events can be determined, that is, the comparison
  • the comparison For the size of different types of waveforms, take the waveform type corresponding to the maximum value.

Landscapes

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Locating Faults (AREA)
  • Remote Monitoring And Control Of Power-Distribution Networks (AREA)
  • Complex Calculations (AREA)

Abstract

一种基于类人概念学习的配电网早期故障检测方法，包括：利用小波变换分解波形为近似部分和细节部分，其中近似部分称为大致形状基元，细节部分称为畸变基元（S1）；根据极值点将畸变基元拆分为谐波、脉冲以及其他畸变三个基元（S2）；提取基元的特征以及基元间的时间关系（S3）；根据基元的特征以及基元间的时间关系，构造波形的概率分布（S4）；根据不同种类波形的概率分布，得到波形的判断结果（S5）。该方法将电压、电流波形作为视觉概念的一种，被分解为大致形状和各种畸变，通过计算各个成分的概率分布，可以获得波形整体的概率分布，从而判断波形种类。该方法在需求数据量和准确性上大大优于传统检测，对配电网早期故障的检测处理具有重要意义。

Description

基于类人概念学习的配电网早期故障检测方法 技术领域
本发明涉及配电网早期故障检测技术领域，具体是一种基于类人概念学习的配电网早期故障检测方法。
背景技术
供电可靠性是配电网最重要的评价指标。由于配电网设备多，区域广，其运维工作主要集中在故障后的处理方面，例如故障定位、故障隔离、故障恢复。但是随着国家和售电市场对供电可靠性要求的提高，配网的故障处理工作不仅要关注故障后的供电恢复还需关注故障前的设备预警，将故障处理的工作方式由“事后抢修巡检”转变成“事前预警主动处理”，在永久性故障发生之前将之消灭，大幅降低因设备故障导致停电事故发生的概率。
在设备故障之前，往往会出现一些异常的前兆性信号，这些信号被称为早期故障。早期故障检测作为电力状态检测的一种方式，为配网运维提供了新的思路，使得缺陷设备能够被提前更换，提高供电可靠性。同时降低运维工作量，节约了成本。
早期故障往往表现为持续时间短、重复发生。这类自恢复性故障往往伴随电弧，对绝缘和导体造成损坏。进一步地，绝缘受损会带来更多的故障。所以往往这类故障会反复发生直到发展成永久性故障。引起早期故障的原因与设备种类有关。在电缆中，绝缘老化是早期故障的主要原因。在架空线中，各种非电力因素如刮风、动物触线、树枝触线等往往会引起早期故障。在其他电力设备中，绝缘缺陷和接触不良也会引起早期故障。
早期故障检测主要分为模型驱动和数据驱动两种，模型驱动往往局限于单一模型，无法适应复杂的实际情况，因此往往采用数据驱动的方法。传统数据驱动方法需要大量的数据且往往只针对个别情景，因此需要改进。类人概念学习类比人类观察波形的过程对波形进行分解，根据分解的结果重构波形的生成过程，通过学习这一过程实现波形的识别。但是，如何将类人概念学习应用于配电网早期故障诊断中，还是本领域亟待解决的问题。
目前没有发现同本发明类似技术的说明或报道，也尚未收集到国内外类似的资料。
发明内容
针对现有技术中存在的上述不足，本发明的目的是，将类人概念学习的相关理论和方法引入到配电网早期故障诊断中，通过理论分析提出一种针对配电网早期故障检测的基于类人概念学习的配电网早期故障检测方法，并验证方法的合理性。类人概念学习模拟人类识别波形的过程，将波形分解为不同成分的叠加，通过学习成分及成分间的时间关系进行波形识别。本发明提出的基于类人概念学习的配电网早期故障检测方法，相较于传统算法具有可引入先验知识、所需样本少、准确率高等特点。
本发明是通过以下技术方案实现的。
一种基于类人概念学习的配电网早期故障检测方法，包括如下步骤：
S1：利用小波变换分解波形为近似部分和细节部分，其中近似部分称为大致形状基元，细节部分称为畸变基元；
S2：根据极值点将畸变基元拆分为谐波、脉冲以及其他畸变三个基元；
S3：提取基元的特征以及基元间的时间关系；
S4：根据基元的特征以及基元间的时间关系，构造波形的概率分布；
S5：根据不同种类波形的概率分布，得到波形的判断结果。
优选地，所述S1中，小波变换函数选取为5层Meyer函数，大致形状取a 5系数，畸变取原始波形与a 5系数的差。
优选地，所述S2中，对畸变进行分解，根据畸变曲线中的极值点将曲线拆分为多个分段，将每一分段与相邻分段进行组合，构成脉冲、谐波以及其他畸变三个基元。
优选地，将每一分段与相邻分段进行组合的规则为：
-如果相邻分段单调性相反，幅值与时间长度差异在0.8倍到1.2倍之内且存在三段以上，则构成谐波；
-如果相邻两段分段单调性相反，幅值超过阈值，时间长度小于阈值，则构成脉冲；
-无法构成谐波和脉冲的分段则为其他畸变。
优选地，幅值阈值设置为0.5倍基波幅值；时间长度阈值设置为0.25倍基波周波。
优选地，所述S3中，基元的特征提取，包括如下提取原则：
对于大致形状z o，提取每个周波的幅值A o、时间长度T o以及直流分量A oft；对于谐波z h，提取幅值A h、频率f h以及总时间长度t h；对于脉冲z p，提取幅值A p、脉宽t p；对于其他畸变z other，不提取特征。
优选地，所述S3中，基元间的时间关系包括大致形状和畸变之间的时间关系以及畸变与畸变之间的时间关系；其中：
所述大致形状与畸变之间的时间关系称为相对基波位置P o，相对基波位置P o描述畸变初始时刻在大致形状上的位置，这一位置采用相角表示；
畸变与畸变之间的时间关系包括：
-间隔时间t int，描述两个相邻畸变之间的时间间隔；
-单相基元对PP uni，描述同一相电压或电流波形中初始时刻相同或接近的两个畸变；
-三相基元对PP tri描述三相电压或电流波形中初始时刻相同或接近的三个畸变。
优选地，所述大致形状和畸变之间的时间关系以及畸变与畸变之间的时间关系中，忽略与其他畸变相关的时间关系。
优选地，所述S4中，构造的波形概率分布公式为：
Figure PCTCN2019078575-appb-000001
式中，
Figure PCTCN2019078575-appb-000002
为未知波形个例，ψ w为已知波形种类，噪声服从正态分布S N～N(μ,σ 2)，基元数为κ，基元类型为z＝{z o,z h,z p,z other}，基元特征参数为p，基元间的时间关系为R；
根据配网异常事件中记录的三相电流，求和得到中性点电流，产生7个不同种类的波形(I A,I B,I C,I N,U A,U B,U C)，得到异常事件的在不同种类波形中的概 率分布公式为：
Figure PCTCN2019078575-appb-000003
式中，
Figure PCTCN2019078575-appb-000004
为未知事件个例，ψ E为已知事件种类，波形w＝{I A,I B,I C,I N,U A,U B,U C}。
优选地，所述S5中，根据异常事件在不同种类波形中的概率分布，判断异常事件的种类，即比较
Figure PCTCN2019078575-appb-000005
在不同种类波形中的大小，取最大值对应的波形种类，得到波形的判断结果。
与现有技术相比，本发明具有如下有益效果：
本发明所提供的一种基于类人概念学习的配电网早期故障检测方法，将类人概念学习的相关理论和方法引入到配电网早期故障检测中，通过理论分析提出针对配电网早期故障的检测算法，并验证算法的合理性。电压、电流波形作为视觉概念的一种，被分解为大致形状和各种畸变，通过计算各个成分的概率分布，可以获得波形整体的概率分布，从而判断波形种类。本发明所提供的基于类人概念学习的配电网早期故障检测方法，在需求数据量和准确性上大大优于传统检测，对配电网早期故障的检测处理具有重要意义。
附图说明
通过阅读参照以下附图对非限制性实施例所作的详细描述，本发明的其它特征、目的和优点将会变得更明显：
图1为本发明一实施例所提供的波形分解示意图；
图2为本发明一实施例所提供的畸变分解示意图；
图3为本发明一实施例所提供的基元间的时间关系定义示意图；其中，(a)为大致形状与畸变之间的时间关系定义示意图；(b)为畸变与畸变之间的时间关系定义示意图；
图4为本发明一实施例所提供的波形生成过程示意图；
图5为本发明一实施例所提供的基于类人概念学习的配电网早期故障检测方法工作流程图。
具体实施方式
下面对本发明的实施例作详细说明：本实施例在以本发明技术方案为前提下进行实施，给出了详细的实施方式和具体的操作过程。应当指出的是，对本领域的普通技术人员来说，在不脱离本发明构思的前提下，还可以做出若干变形和改进，这些都属于本发明的保护范围。
实施例
如图5所示，本实施例提供了一种基于类人概念学习的配电网早期故障检测方法，包括如下步骤：
步骤S1：利用小波变换分解波形为近似部分和细节部分，，其中，将近似部分和细节部分这两部分分别称为大致形状基元和畸变基元。
步骤S2：根据极值点将畸变(细节部分)拆分为谐波、脉冲以及其他畸变三种基元。
步骤S3：提取出基元的特征以及基元间的时间关系。
步骤S4：根据基元的特征以及基元间的时间关系构造波形的概率分布。
步骤S5：根据不同种类波形的概率分布得到波形的判断结果。
下面结合附图，对本发明上述实施例的技术方案进一步详细描述。
如图1所示，为对原始波形进行小波分解的波形分解示意图，小波函数选取为5层Meyer函数，大致形状取a 5系数，畸变取原始波形与a 5系数的差，得到图1所示分解结果。对于图1中的畸变(细节部分)继续进行分解，根据畸变曲线中的极值点将曲线拆分为小段，如图2所示的第一步，将每一段与周围小段进行组合，构成脉冲、谐波、其他畸变三种基元，如图2所示的第二步。组合规则如下：1)如果相邻小段单调性相反、幅值与时间长度差异较小、且存在三段以上，则构成谐波(如图2所示的1、2、3、4段组合)；2)如果相邻的两段单调性相反、幅值超过阈值(这里设置为0.5倍基波幅值)、时间长度小于阈值(这里设置为0.25倍基波周波)，则构成脉冲(如图2所示的5、6段组合)；3)无法构成谐波和脉冲的小段称为其他畸变(如图2所示的7、8段)。这里将大致形状、谐波、脉冲、其他畸变统称为基元。
基元特征提取如下：对于大致形状z o，提取每个周波的幅值A o、时间长度T o以及直流分量A oft；对于谐波z h，提取幅值A h、频率f h、总时间长度t h；对于脉 冲z p，提取幅值A p、脉宽t p；对于其他畸变z other，由于z other和早期故障关系不大，所以不提取其特征。基元间同样存在时间关系，基元可分为两种：大致形状和畸变(谐波、脉冲、其他畸变)，基元间的时间关系也可分为大致形状和畸变之间的时间关系以及畸变与畸变之间的时间关系。同样地，不讨论与z other相关的时间关系。大致形状与畸变基元之间的时间关系称为相对基波位置P o，P o描述的是畸变基元初始时刻在大致形状上的位置，这一位置用相角表示。畸变基元间的时间关系有以下几种：间隔时间t int，单相基元对PP uni和三相基元对PP tri。t int描述的是两个相邻畸变基元间的时间间隔。单相基元对PP uni描述的是同一相电压、电流波形中初始时刻相同/及其接近的两个畸变基元。三相基元对PP tri描述的是三相电压/电流波形中初始时刻相同/及其接近的三个畸变基元。PP uni和PP tri描述的是一种相关关系，这里用n uni,n tri分别表示两者的数量。以上时间关系的示意图如图3(a)和(b)所示。
至此，一个波形被唯一的分解为以上基元的组合，并且基元特征、时间关系被用于描述这种组合。接下来，基于上述分解提出波形的一个生成过程，如图4所示。这一过程描述如下：首先在基元库中选定一些基元(基元种类为4，每种基元数量任意)，每个基元有各自的特征参数；将这些基元按先后顺序组合成波形，这些先后顺序构成了基元间的时间关系。以上属于种类层次。波形种类由故障原因、故障设备、故障部位决定。实际情况中，同种类波形受线路参数、网架结构、负荷情况、传感器参数、噪声等影响，会表现为不同的个例。表现在基元层面即为：基元的特征参数以及时间关系会发生一定的变化，同时会引入噪声。以上属于个例层次。
根据以上生成过程，可以推导出波形的概率分布公式：
Figure PCTCN2019078575-appb-000006
式中
Figure PCTCN2019078575-appb-000007
为未知波形个例，ψ w为已知波形种类，噪声服从正态分布S N～N(μ,σ 2)，基元数为κ，基元类型为z＝{z o,z h,z p,z other}，基元特征参数为p，基元间时间关系为R。
对于配网中的一起异常事件，往往记录有其对应的三相电压、三相电流，通过对三相电流求和可以得到中性点电流，这样就产生了7个不同种类的波形(I A,I B,I C,I N,U A,U B,U C)。因此一起异常事件在不同种类波形中的概率分布可写为：
Figure PCTCN2019078575-appb-000008
式中，
Figure PCTCN2019078575-appb-000009
为未知事件个例，ψ E为已知事件种类，波形w＝{I A,I B,I C,I N,U A,U B,U C}。
根据异常事件在不同种类波形中的概率分布可以判断出异常事件的种类，即比较
Figure PCTCN2019078575-appb-000010
在不同种类波形中的大小，取最大值对应的波形种类。
取100个已知样本进行训练，事件种类分别为单相单周波早期故障、单相多周波早期故障、相间短路早期故障、暂态干扰以及永久性故障(分别标号为1,2,3,4,5)。用另外200个未知样本进行测试，实验结果如表1所示。可以看出本发明上述实施例所提供的基于类人概念学习的配电网早期故障检测方法，准确率极高，且所需数据量较少。
表1
Figure PCTCN2019078575-appb-000011
以上对本发明的具体实施例进行了描述。需要理解的是，本发明并不局限于上述特定实施方式，本领域技术人员可以在权利要求的范围内做出各种变形或修改，这并不影响本发明的实质内容

Claims (10)

  1. 一种基于类人概念学习的配电网早期故障检测方法，其特征在于，包括如下步骤：
    S1：利用小波变换分解波形为近似部分和细节部分，其中近似部分称为大致形状基元，细节部分称为畸变基元；
    S2：根据极值点将畸变基元拆分为谐波、脉冲以及其他畸变三个基元；
    S3：提取基元的特征以及基元间的时间关系；
    S4：根据基元的特征以及基元间的时间关系，构造波形的概率分布；
    S5：根据不同种类波形的概率分布，得到波形的判断结果。
  2. 根据权利要求1所述的基于类人概念学习的配电网早期故障检测方法，其特征在于，所述S1中，小波变换函数选取为5层Meyer函数，大致形状取a 5系数，畸变取原始波形与a 5系数的差。
  3. 根据权利要求1所述的基于类人概念学习的配电网早期故障检测方法，其特征在于，所述S2中，对畸变进行分解，根据畸变曲线中的极值点将曲线拆分为多个分段，将每一分段与相邻分段进行组合，构成脉冲、谐波以及其他畸变三个基元。
  4. 根据权利要求3所述的基于类人概念学习的配电网早期故障检测方法，其特征在于，将每一分段与相邻分段进行组合的规则为：
    -如果相邻分段单调性相反，幅值与时间长度差异在0.8倍到1.2倍之内且存在三段以上，则构成谐波；
    -如果相邻两段分段单调性相反，幅值超过阈值，时间长度小于阈值，则构成脉冲；
    -无法构成谐波和脉冲的分段则为其他畸变。
  5. 根据权利要求4所述的基于类人概念学习的配电网早期故障检测方法，其特征在于，幅值阈值设置为0.5倍基波幅值；时间长度阈值设置为0.25倍基波周波。
  6. 根据权利要求1所述的基于类人概念学习的配电网早期故障检测方法，其特征在于，所述S3中，基元的特征提取，包括如下提取原则：
    对于大致形状z o，提取每个周波的幅值A o、时间长度T o以及直流分量A oft；对于谐波z h，提取幅值A h、频率f h以及总时间长度t h；对于脉冲z p，提取幅值A p、脉宽t p；对于其他畸变z other，不提取特征。
  7. 根据权利要求1所述的基于类人概念学习的配电网早期故障检测方法，其特征在于，所述S3中，基元间的时间关系包括大致形状和畸变之间的时间关系以及畸变与畸变之间的时间关系；其中：
    所述大致形状与畸变之间的时间关系称为相对基波位置P o，相对基波位置P o描述畸变初始时刻在大致形状上的位置，这一位置采用相角表示；
    畸变与畸变之间的时间关系包括：
    -间隔时间t int，描述两个相邻畸变之间的时间间隔；
    -单相基元对PP uni，描述同一相电压或电流波形中初始时刻相同或接近的两个畸变；
    -三相基元对PP tri描述三相电压或电流波形中初始时刻相同或接近的三个畸变。
  8. 根据权利要求7所述的基于类人概念学习的配电网早期故障检测方法，其特征在于，所述大致形状和畸变之间的时间关系以及畸变与畸变之间的时间关系中，忽略与其他畸变相关的时间关系。
  9. 根据权利要求1所述的基于类人概念学习的配电网早期故障检测方法，其特征在于，所述S4中，构造的波形概率分布公式为：
    Figure PCTCN2019078575-appb-100001
    式中，
    Figure PCTCN2019078575-appb-100002
    为未知波形个例，ψ w为已知波形种类，噪声服从正态分布S N～N(μ,σ 2)，基元数为κ，基元类型为z＝{z o,z h,z p,z other}，基元特征参数为p，基元间的时间关系为R；
    根据配网异常事件中记录的三相电流，求和得到中性点电流，产生7个不同种类的波形(I A,I B,I C,I N,U A,U B,U C)，得到异常事件的在不同种类波形中的概率分布公式为：
    Figure PCTCN2019078575-appb-100003
    式中，
    Figure PCTCN2019078575-appb-100004
    为未知事件个例，ψ E为已知事件种类，波形w＝{I A,I B,I C,I N,U A,U B,U C}。
  10. 根据权利要求9所述的基于类人概念学习的配电网早期故障检测方法，其特征在于，所述S5中，根据异常事件在不同种类波形中的概率分布，判断异常事件的种类，即比较
    Figure PCTCN2019078575-appb-100005
    在不同种类波形中的大小，取最大值对应的波形种类，得到波形的判断结果。
PCT/CN2019/078575 2018-12-27 2019-03-19 基于类人概念学习的配电网早期故障检测方法 WO2020133735A1 (zh)

Priority Applications (3)

Application Number Priority Date Filing Date Title
JP2019538613A JP7107580B2 (ja) 2018-12-27 2019-03-19 配電網の早期故障の検出方法
GB1912002.1A GB2582676B (en) 2018-12-27 2019-03-19 Early Failure Detection Method for Power Distribution Network Equipment Based On Human-Level Concept Learning
US16/674,233 US11143686B2 (en) 2018-12-27 2019-11-05 Method for detecting power distribution network early failure

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201811607507.8A CN109782126B (zh) 2018-12-27 2018-12-27 基于类人概念学习的配电网早期故障检测方法
CN201811607507.8 2018-12-27

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US16/674,233 Continuation US11143686B2 (en) 2018-12-27 2019-11-05 Method for detecting power distribution network early failure

Publications (1)

Publication Number Publication Date
WO2020133735A1 true WO2020133735A1 (zh) 2020-07-02

Family

ID=66497717

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/078575 WO2020133735A1 (zh) 2018-12-27 2019-03-19 基于类人概念学习的配电网早期故障检测方法

Country Status (3)

Country Link
JP (1) JP7107580B2 (zh)
CN (1) CN109782126B (zh)
WO (1) WO2020133735A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117454096A (zh) * 2023-12-25 2024-01-26 西安高商智能科技有限责任公司 一种电机生产质量检测方法及系统

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113884803A (zh) * 2021-08-27 2022-01-04 国网山东省电力公司日照供电公司 配网故障研判方法
CN113985733B (zh) * 2021-10-26 2023-11-17 云南电网有限责任公司电力科学研究院 一种基于自适应概率学习的配电网故障辨识方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103091096A (zh) * 2013-01-23 2013-05-08 北京信息科技大学 基于eemd和小波包变换的早期故障敏感特征提取方法
CN104122486A (zh) * 2014-07-30 2014-10-29 浙江群力电气有限公司 一种电缆早期故障检测方法及装置
CN107025433A (zh) * 2017-03-03 2017-08-08 深圳大学 视频事件类人概念学习方法及装置
CN107329049A (zh) * 2017-08-21 2017-11-07 集美大学 一种基于卡尔曼滤波器的输电线路早期故障检测方法
CN107340456A (zh) * 2017-05-25 2017-11-10 国家电网公司 基于多特征分析的配电网工况智能识别方法

Family Cites Families (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN100546143C (zh) * 2007-09-26 2009-09-30 东北大学 一种小电流接地故障检测与定位的装置及方法
CN101726690B (zh) * 2009-11-30 2012-01-11 上海电力学院 电力系统短路全电流各分量分解方法
CN102129013B (zh) * 2011-01-21 2013-11-27 昆明理工大学 一种利用自然频率和人工神经网络的配网故障测距方法
US9508612B2 (en) * 2012-03-15 2016-11-29 Applied Materials, Inc. Method to detect wafer arcing in semiconductor manufacturing equipment
KR101308003B1 (ko) 2012-07-31 2013-09-12 서울과학기술대학교 산학협력단 웨이블릿 기반 아크 판별방법
CN104597344A (zh) * 2015-01-08 2015-05-06 上海交通大学 基于小波一层高频分量相关性的故障电弧在线检测方法
CN106501667B (zh) * 2016-03-16 2019-06-21 国网山东省电力公司济宁供电公司 一种含分布式电源配电网单相断线故障选线方法
CN106501668B (zh) * 2016-03-16 2019-06-28 国网山东省电力公司济宁供电公司 一种传统配电网单相断线故障选线方法
CN108279364B (zh) * 2018-01-30 2020-01-14 福州大学 基于卷积神经网络的配电网单相接地故障选线方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103091096A (zh) * 2013-01-23 2013-05-08 北京信息科技大学 基于eemd和小波包变换的早期故障敏感特征提取方法
CN104122486A (zh) * 2014-07-30 2014-10-29 浙江群力电气有限公司 一种电缆早期故障检测方法及装置
CN107025433A (zh) * 2017-03-03 2017-08-08 深圳大学 视频事件类人概念学习方法及装置
CN107340456A (zh) * 2017-05-25 2017-11-10 国家电网公司 基于多特征分析的配电网工况智能识别方法
CN107329049A (zh) * 2017-08-21 2017-11-07 集美大学 一种基于卡尔曼滤波器的输电线路早期故障检测方法

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117454096A (zh) * 2023-12-25 2024-01-26 西安高商智能科技有限责任公司 一种电机生产质量检测方法及系统
CN117454096B (zh) * 2023-12-25 2024-03-01 西安高商智能科技有限责任公司 一种电机生产质量检测方法及系统

Also Published As

Publication number Publication date
CN109782126A (zh) 2019-05-21
JP2021512377A (ja) 2021-05-13
JP7107580B2 (ja) 2022-07-27
CN109782126B (zh) 2021-01-01

Similar Documents

Publication Publication Date Title
WO2020133735A1 (zh) 基于类人概念学习的配电网早期故障检测方法
US11143686B2 (en) Method for detecting power distribution network early failure
CN109444667B (zh) 基于卷积神经网络的配电网早期故障分类方法及装置
CN103713237B (zh) 一种电力系统输电线路短路故障诊断方法
CN109406949B (zh) 基于支持向量机的配电网早期故障检测方法及装置
CN106330095B (zh) 一种大型光伏电站汇集系统故障定位方法
CN104280644B (zh) 一种直流输电工程典型暂态故障识别方法
CN101499659B (zh) 基于基尔霍夫电流定律的变电站分布式状态估计方法
CN112345968A (zh) 一种直流系统接地故障综合检测方法
Qu et al. Model-free underground cable incipient fault location using two-terminal zero-sequence measurements
Wang et al. Fault detection of the power system based on the chaotic neural network and wavelet transform
CN109324268B (zh) 基于贝叶斯推理的配电网早期故障检测方法及装置
CN113933647A (zh) 一种基于首半波功率方向的小电流接地系统故障选线方法
Lopes et al. A transient based approach to diagnose high impedance faults on smart distribution networks
CN105629144B (zh) 基于指纹库的高压开关设备局部放电诊断方法及系统
CN108646145B (zh) 一种输电线路闪络塔定位方法及系统
CN111339624B (zh) 基于psasp和emtp/atp短路电流直流分量计算方法
CN116660679A (zh) 一种基于网络拓扑的配电网故障分析方法
CN115980515A (zh) 一种单相接地故障定位方法
Fahim et al. An agreement based dynamic routing method for fault diagnosis in power network with enhanced noise immunity
CN111257687B (zh) 一种基于改进的meemd的单相接地故障选线方法
Cui et al. High impedance fault detection method based on Data divergence in the distribution network
Datta et al. Line to ground and line to line fault analysis in ieee standard 9 BUS system
Yuan et al. Faulty feeder detection based on sparse waveform encoding and simple convolutional neural network with multiscale filters and one layer of convolution
Sheng et al. An overview of fault identification techniques in power distribution networks: methods and models

Legal Events

Date Code Title Description
ENP Entry into the national phase

Ref document number: 2019538613

Country of ref document: JP

Kind code of ref document: A

ENP Entry into the national phase

Ref document number: 201912002

Country of ref document: GB

Kind code of ref document: A

Free format text: PCT FILING DATE = 20190319

121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19903499

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 22.10.2021)

122 Ep: pct application non-entry in european phase

Ref document number: 19903499

Country of ref document: EP

Kind code of ref document: A1