A distribution network fault detection accuracy calculation method based on distribution network fault simulation

Abstract

The present invention discloses a distribution network fault detection accuracy calculation method based on distribution network fault simulation, including: using simulation data to construct a distribution network fault simulation model; according to the distribution network fault simulation model, using a twin timing diagram network to distinguish faults in real time; comparing the judgment result with the input simulation fault data, outputting the comparison success rate, and completing the calculation of the distribution network fault detection accuracy. Small sample learning is a key issue in early fault identification of distribution networks. In addition to building an actual waveform database, the reasonable use of simulation data can improve the accuracy of fault identification. Using simulation data and a small amount of actual data to make preliminary judgments first, and then specifically verifying the classification results, and then adding the verified results to the database, greatly improving the automation and intelligence of the distribution network.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution network fault detection, and in particular to a distribution network fault detection accuracy calculation method based on distribution network fault simulation. Background Art

[0002] The research on early fault diagnosis methods of distribution networks is generally divided into two categories: physical models based on mechanisms and mathematical models based on data. The former mainly identifies by building a circuit model of the fault line. For example, first, the main mode of the fault waveform is obtained by modal analysis, and then the transient circuit of the line is built using the main mode; the arc fault model and the non-arc fault model are derived, and an arc fault identification algorithm is given based on these two circuit models. The latter mainly derives the mathematical relationship between the waveform and the event category through data, for example, a line early fault judgment criterion based on wavelet decomposition is proposed; a human-like concept learning algorithm is proposed, that is, the waveform is decomposed according to human cognition, and the probability that the decomposition result belongs to each event type is calculated. In fact, both of the above two types of research have certain limitations. The former is greatly affected by the complexity of the grid structure and the randomness of the arc, and the latter is limited by the sample size and needs to reduce the data dependence of the model.

[0003] Although actual on-site operating data is difficult to collect, simulation data is easy to obtain and can cover various operating environments of distribution networks, the difference between the two makes the model trained based on simulation data perform poorly in actual on-site applications. How to use simulation data to help the model learn general features so that it can also be applied to actual data has become the key to solving the small sample problem of distribution network fault identification. Summary of the invention

[0004] The purpose of this section is to summarize some aspects of embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the specification abstract and the invention title of this application to avoid blurring the purpose of this section, the specification abstract and the invention title, and such simplifications or omissions cannot be used to limit the scope of the present invention.

[0005] In view of the above existing problems, the present invention is proposed.

[0006] Therefore, the technical problem solved by the present invention is that it is difficult to collect actual on-site operation data, and existing fault identification technologies have certain limitations.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions: construct a distribution network fault simulation model using simulation data; according to the distribution network fault simulation model, use a twin timing diagram network to perform real-time fault identification; compare the identification result with the input simulation fault data, output the comparison success rate, and complete the calculation of the distribution network fault detection accuracy.

[0008] As a preferred scheme of the distribution network fault detection accuracy calculation method based on distribution network fault simulation described in the present invention, wherein: the simulation data includes, fault starting angle: the value range is 0 to 360; fault distance: the ratio of the length of the fault location to the area where it is located is in the range of 0 to 1; fault arc parameters; load parameters: the active power and reactive power change multiples are in the range of 0.8 to 1.2; line parameters: the unit length resistance, inductance and capacitance change multiples are in the range of 0.8 to 1.2.

[0009] As a preferred solution of the distribution network fault detection accuracy calculation method based on distribution network fault simulation described in the present invention, wherein: the fault arc parameters include, using the Kizilcay arc model, that is:

[0010]

[0011]

[0012] Where g(t) is the arc conductance, i f (t) is the arc current, u f (t) is the arc voltage, τ is the arc time constant, r o is the arc characteristic resistance, u o is the arc characteristic voltage, the above parameters have the following value ranges: τ = 0.2 ~ 0.4ms, u o =300~4000V, r o =0.01~0.015Ω.

[0013] As a preferred solution of the distribution network fault detection accuracy calculation method based on distribution network fault simulation described in the present invention, wherein: the real-time fault identification based on the twin timing graph network includes: converting the distribution network simulation operation data into waveform data; converting the waveform data into a timing graph network format; using the defined timing graph network convolution to perform feature extraction and mapping; comparing the similarity of two groups of feature vectors in the feature space, judging whether the two groups of input waveforms belong to the same category, and performing fault identification based on the judgment result.

[0014] As a preferred solution of the distribution network fault detection accuracy calculation method based on distribution network fault simulation described in the present invention, it also includes defining a single-cycle graph convolutional network:

[0015] With f in Represents the input feature with the number of channels c, after the convolution kernel with size K×K, the single-channel output feature f at position x out It can be expressed as:

[0016]

[0017] Among them, the sampling function P is used to obtain the eigenvalues ​​of the h×w range around the position x, and the weight W is used as the inner product with the c-dimensional vector obtained by sampling to obtain the output result. At the same time, the weight W is independent of the position x.

[0018] As a preferred solution of the method for calculating the accuracy of distribution network fault detection based on distribution network fault simulation described in the present invention, it also includes defining a graph convolutional network of multiple cycles:

[0019]

[0020] Among them, Γ is the temporal convolution kernel scale.

[0021] As a preferred solution of the distribution network fault detection accuracy calculation method based on distribution network fault simulation described in the present invention, the error function of the twin timing graph network is:

[0022]

[0023] Where N is the number of samples, d = |f1-f2| is the Euclidean distance between two feature vectors, the indicator variable y(f1,f2) is equal to 1 when the feature vectors f1 and f2 belong to the same class, otherwise it is equal to 0, and m is the distance threshold between samples of different classes, that is, when the feature vector distance between two samples of different classes exceeds m, the loss is 0, otherwise the loss is (md) 2 .

[0024] As a preferred solution of the distribution network fault detection accuracy calculation method based on distribution network fault simulation described in the present invention, wherein: the twin time series graph network is used to directly predict the input category, and for the unknown category input w x , calculate its eigenvector f x The average distance to each class, grouping it into the class with the closest distance:

[0025]

[0026]

[0027] Among them, n i represents the number of known samples in category i, k is w x The category prediction results.

[0028] As a preferred scheme of the distribution network fault detection accuracy calculation method based on distribution network fault simulation described in the present invention, wherein: comparing the discrimination result with the input simulation fault data includes defining that each type of simulation fault data has its unique code; converting the discrimination result into a corresponding fault code, and matching the fault code with the simulation input fault code; if the match is successful, the fault detection is accurate.

[0029] Beneficial effects of the present invention: Small sample learning is a key issue in early fault identification in distribution networks. In addition to building an actual waveform database, the rational use of simulation data can improve the accuracy of fault identification. Simulation data and a small amount of actual data are used to make preliminary judgments, and then the classification results are verified in a targeted manner. The verified results are then added to the database, which greatly improves the automation and intelligence of the distribution network. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them:

[0031] Figure 1 A schematic diagram of a timing diagram network construction of a distribution network fault detection accuracy calculation method based on distribution network fault simulation provided by an embodiment of the present invention;

[0032] Figure 2 A schematic diagram of a twin timing diagram network structure of a distribution network fault detection accuracy calculation method based on distribution network fault simulation provided by one embodiment of the present invention;

[0033] Figure 3 A schematic diagram of a single-phase single-arcing single-cycle fault of a distribution network fault detection accuracy calculation method based on distribution network fault simulation provided by an embodiment of the present invention;

[0034] Figure 4 A schematic diagram of a single-phase single arcing multi-cycle fault of a distribution network fault detection accuracy calculation method based on distribution network fault simulation provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0035] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0036] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0037] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0038] The present invention is described in detail with reference to schematic diagrams. When describing the embodiments of the present invention, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0039] At the same time, in the description of the present invention, it should be noted that the directions or positional relationships indicated by the terms "upper, lower, inner and outer" are based on the directions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the present invention. In addition, the terms "first, second or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0040] In the present invention, unless otherwise clearly specified and limited, the terms "install, connect, connect" should be understood in a broad sense, for example: it can be a fixed connection, a detachable connection or an integral connection; it can also be a mechanical connection, an electrical connection or a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0041] Example 1

[0042] Reference Figures 1-2 , is an embodiment of the present invention, and provides a method for calculating distribution network fault detection accuracy based on distribution network fault simulation, comprising:

[0043] S1: Use simulation data to build a distribution network fault simulation model.

[0044] It should be noted that the simulation data includes:

[0045] Fault starting angle: the value range is 0 to 360;

[0046] Fault distance: The ratio of the fault location to the length of the area where it is located ranges from 0 to 1;

[0047] Fault arc parameters;

[0048] Load parameters: The range of active power and reactive power change multiples is 0.8 to 1.2;

[0049] Line parameters: The range of the resistance, inductance and capacitance change multiples per unit length is 0.8 to 1.2.

[0050] Furthermore, the fault arc parameters include:

[0051] The Kizilcay arc model is adopted, namely:

[0052]

[0053]

[0054] Where g(t) is the arc conductance, i f (t) is the arc current, u f (t) is the arc voltage, τ is the arc time constant, r o is the arc characteristic resistance, u o is the arc characteristic voltage, the above parameters have the following value ranges: τ = 0.2 ~ 0.4ms, u o =300~4000V, r o =0.01~0.015Ω.

[0055] S2: According to the distribution network fault simulation model, the fault is identified in real time using the twin timing graph network.

[0056] It should be noted that real-time fault identification based on the twin timing graph network includes:

[0057] Convert distribution network simulation operation data into waveform data;

[0058] Convert waveform data into timing diagram network format;

[0059] Use the defined temporal graph network convolution for feature extraction and mapping;

[0060] The similarity of two groups of feature vectors is compared in the feature space to determine whether the two groups of input waveforms belong to the same category, and fault identification is performed based on the judgment result.

[0061] The waveform sequence is composed of waveforms of each cycle. This embodiment uses a temporal graph network to describe this sequence hierarchically. That is, for a waveform sequence with a sampling length of N and a number of cycles T, an undirected temporal graph G = (V, E) is used to describe the connections between different sampling points in the same cycle and between different cycles of the same sampling point. In this graph network, the vertex set V = {v ti|t=1,…,T,i=1,…,N} contains all the points in the waveform sequence. For each vertex, its eigenvector F(v ti ) is the waveform amplitude at this moment, and the timing diagram network is constructed as follows Figure 1 As shown in the figure: First, all sampling points in the same period are connected in sequence, and then the same sampling points in different periods are connected in sequence. This structural design is completely based on the relationship between waveform sampling points and is applicable to waveform sequences with different sampling frequencies and number of periods. The edge set E contains two types of subsets, namely, the connections between different points in the period E S = {v ti v t(i+1)} and the connection between different cycles of the same point E F = {v ti v (t+1)i}, E F The essence is the change of the same point in different weeks.

[0062] Furthermore, we first discuss the graph convolutional network with a single cycle, that is, for a cycle τ, there are N vertices in the graph network, and the connection between the vertices is the edge set E S (τ)={v τi v τ(i+1)}, for the convolution operation of the two-dimensional matrix, by selecting the appropriate step size and padding method, the size of the feature map can be kept unchanged, that is, the size of the output feature is equal to that of the input feature, with f in Represents the input feature with the number of channels c, after the convolution kernel with size K×K, the single-channel output feature f at position x out It can be expressed as:

[0063]

[0064] Among them, the sampling function P is used to obtain the eigenvalues ​​in the range h×w around the position x, and the weight W is taken as the inner product with the c-dimensional vector obtained by sampling to obtain the output result. At the same time, the weight W is independent of the position x, that is, the weight is shared at different positions. This is the core feature of the convolution operation.

[0065] Therefore, the extension of two-dimensional matrix convolution to graph network convolution can start from two aspects: sampling function and weight matrix. First, the sampling function. Referring to the definition of adjacent positions in the two-dimensional matrix, we can define the adjacent nodes B (v ti )={v tj |d(v tj ,v ti )<D}, that is, v ti The adjacent nodes are nodes v whose distance from it does not exceed D. tj , the sampling function P can be written as, where the optimal D value can be obtained by cross-validation in the experiment:

[0066] P(vti ,D)=v tj

[0067] The key to defining the weight matrix W is how to make it universal. Compared with a two-dimensional matrix, the number of adjacent nodes of different nodes in a graph network is not consistent. How to achieve weight sharing becomes a design difficulty. The present invention adopts a grouping strategy, that is, the set B(v ti ) is divided into fixed K groups, and the weights of the elements in each group are consistent, which can be written as:

[0068] W(v ti ,v tj )=W′(l ti (v tj ))

[0069] Where W′∈R K is the weight vector after grouping, l ti (v tj ) is the grouping function, which is v tj is divided into one of the K groups, and these K groups together form v ti The set of adjacent nodes.

[0070] Combining the above two points, we can get the graph network convolution equation:

[0071]

[0072] Among them, Z ti (v tj ) is a regular term, which is related to v tj The number of points divided into the same group is used to balance the grouping of different numbers, that is:

[0073] Z ti (v tj )=|{v tk |l ti (v tk )=l ti (v tj )}|

[0074] Combined P(v ti ,D)=v tj Japanese style W(v ti ,v tj )=W′(l ti (v tj )), the above formula is finally written as:

[0075]

[0076] It can be noted that if the two-dimensional matrix is ​​written in the form of a graph network, the formula f out (v ti) can be derived from formula f out (x).

[0077] The above content completes the definition of graph network convolution for a single cycle. The following discusses the case of multiple cycles, that is, the convolution of the sequential graph network. Since the sequential graph network increases the connection of sampling points between different cycles compared to the graph network, in addition to considering the nodes in the same cycle, the adjacent nodes also need to increase the nodes between different cycles, that is:

[0078]

[0079] Among them, Γ is the scale of the temporal convolution kernel, which controls the period difference between adjacent nodes and the central node. At this time, the sampling function P will collect adjacent nodes between different periods.

[0080] Furthermore, the extension of the weight matrix W in the time series graph network is relatively simple. In a single cycle, adjacent nodes are divided into K groups, each group corresponds to a weight value, that is, W′∈R K , now there are Γ cycles, just increase the weight matrix dimension W′∈R K×Γ , and modify the grouping function:

[0081]

[0082] Among them, v tj v qj At the corresponding point in period t, l ti (v tj ) is a single-period grouping function.

[0083] Compared with other models, the twin network is less affected by the number of samples and erroneous samples, and is suitable for small sample learning. Its internal feature extraction network can be replaced with any type, and it has strong versatility. In this embodiment, the internal feature extraction network is replaced with a timing graph network. The twin network is designed to compare the similarity between two input variables and continuously optimize the feature extraction network by maximizing the similarity of feature vectors of the same category. Figure 2 The structure of the twin timing graph network is shown, that is, the two sets of original waveforms first construct their own timing graph networks, then use the timing graph network convolution to obtain two sets of feature vectors, and finally calculate the similarity between the two sets of feature vectors to determine the probability that the two sets of input waveforms belong to the same category. It should be noted that the weights of the two timing graph convolution networks are shared. It is not difficult to see that the more similar the features extracted from the two sets of original waveforms are, the greater the probability that they belong to the same category.

[0084] The error function of the twin network selects contrastive loss. Its core idea is to reduce the distance between samples of the same type and increase the distance between samples of different types in the feature space. Its calculation formula is:

[0085]

[0086] Where N is the number of samples, d = |f1-f2| is the Euclidean distance between two feature vectors, the indicator variable y(f1,f2) is equal to 1 when the feature vectors f1 and f2 belong to the same class, otherwise it is equal to 0, and m is the distance threshold between samples of different classes, that is, when the feature vector distance between two samples of different classes exceeds m, the loss is 0, otherwise the loss is (md) 2 .

[0087] Based on this threshold, it can also be determined whether two groups of inputs belong to the same category. That is, in the feature space, if the distance between the two groups of feature vectors does not exceed the threshold m, they belong to the same category, otherwise they belong to different categories. That is, if the Euclidean distance of the feature vectors is small, it indicates events of the same category under the network mapping.

[0088] In addition to determining whether two sets of inputs belong to the same category, the twin network can also directly predict the input category. x , calculate its eigenvector f x The average distance to each category is then classified into the category with the closest distance:

[0089]

[0090]

[0091] Among them, n i represents the number of known samples in category i, k is w x The category prediction results.

[0092] S3: Compare the judgment result with the input simulation fault data, output the comparison success rate, and complete the calculation of the distribution network fault detection accuracy.

[0093] It should be noted that the comparison between the judgment result and the input simulation fault data includes:

[0094] Define that each type of simulation fault data has its unique code;

[0095] Convert the discrimination result into a corresponding fault code, and match the fault code with the simulation input fault code;

[0096] If the match is successful, the fault detection is accurate.

[0097] The matching algorithm used is the Euclidean distance method.

[0098] Example 2

[0099] Reference Figures 3-4This is another embodiment of the present invention. Different from the first embodiment, this embodiment provides a verification test of a distribution network fault detection accuracy calculation method based on distribution network fault simulation. In order to verify and illustrate the technical effects adopted in this method, this embodiment adopts traditional technical solutions and the method of the present invention for testing, and verifies the real effect of this method by means of scientific argumentation.

[0100] Figures 3-4 Some simulation results are given: Figure 3 This is the voltage and current waveform of a single-phase single-cycle early fault, during which the arc burns once and then extinguishes; Figure 4 These are the voltage and current waveforms of a single-phase multi-cycle early fault, during which the arc burns once and then extinguishes. All waveforms are collected at the busbar measurement point. It can be seen that the simulation results are close to the actual fault waveforms, and a series of practical factors such as the number of fault phases, fault duration, and the number of arc burning and extinguishing times are taken into account, providing a sufficient data basis for the experimental verification stage.

[0101] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A distribution network fault detection accuracy calculation method based on distribution network fault simulation, characterized in that: include: Use simulation data to build a distribution network fault simulation model; According to the distribution network fault simulation model, the fault is identified in real time using a twin timing graph network; Compare the judgment result with the input simulation fault data, output the comparison success rate, and complete the calculation of distribution network fault detection accuracy; The simulation data includes: Fault starting angle: the value range is 0 to 360; Fault distance: The ratio of the fault location to the length of the area where it is located ranges from 0 to 1; Fault arc parameters; Load parameters: The range of active power and reactive power change multiples is 0.8 to 1.2; Line parameters: The range of the resistance, inductance and capacitance change multiples per unit length is 0.8 to 1.2; The real-time fault identification based on the twin timing diagram network includes: Convert distribution network simulation operation data into waveform data; Converting the waveform data into a timing diagram network format; Use the defined temporal graph network convolution for feature extraction and mapping; Compare the similarity of two groups of feature vectors in the feature space to determine whether the two groups of input waveforms belong to the same category, and perform fault identification based on the judgment results; Define a single cycle graph convolutional network: With f in Represents the input feature with the number of channels c, after the convolution kernel with size K×K, the single-channel output feature f at position x out It can be expressed as: Among them, the sampling function P is used to obtain the eigenvalues ​​of the h×w range around the position x, and the weight W is used as the inner product with the c-dimensional vector obtained by sampling to obtain the output result. At the same time, the weight W is independent of the position x; Define a graph convolutional network with multiple cycles: Among them, Γ is the temporal convolution kernel scale; The twin temporal graph network is used to directly predict the input category. For the unknown category input w x , calculate its eigenvector f x The average distance to each class, grouping it into the class with the closest distance: Among them, n i represents the number of known samples in category i, k is w x The category prediction results.

2. The method for calculating the distribution network fault detection accuracy based on distribution network fault simulation according to claim 1, characterized in that: The fault arc parameters include: The Kizilcay arc model is adopted, namely: Where g(t) is the arc conductance, i f (t) is the arc current, u f (t) is the arc voltage, τ is the arc time constant, r o is the arc characteristic resistance, u o is the arc characteristic voltage, the above parameters have the following value ranges: τ = 0.2 ~ 0.4ms, u o =300~4000V, r o =0.01~0.015Ω.

3. The method for calculating the distribution network fault detection accuracy based on distribution network fault simulation according to any one of claims 1 or 2, characterized in that: The error function of the twin timing graph network is: Where N is the number of samples, d = |f1-f2| is the Euclidean distance between two feature vectors, the indicator variable y(f1,f2) is equal to 1 when the feature vectors f1 and f2 belong to the same class, otherwise it is equal to 0, and m is the distance threshold between samples of different classes, that is, when the feature vector distance between two samples of different classes exceeds m, the loss is 0, otherwise the loss is (md) 2 .

4. The method for calculating the distribution network fault detection accuracy based on distribution network fault simulation according to claim 1, characterized in that: Comparison between the judgment result and the input simulation fault data includes: Define that each type of simulation fault data has its unique code; Converting the discrimination result into a corresponding fault code, and matching the fault code with a simulation input fault code; If the match is successful, the fault detection is accurate.

Images