Outlet matrix detection method based on relay protection test system

Abstract

The embodiment of the invention discloses an outlet matrix detection method based on a relay protection test system, and belongs to the technical field of circuit detection, and the method comprises the steps: installing a detection sensor at a relay protection outlet pressing plate, collecting signal data, carrying out the data preprocessing of the signal data, and generating a normal sample data set and a fault sample data set; analyzing the normal sample data set and the fault sample data set to obtain the cause of each fault, performing network model recognition training on the convolutional neural network by using the analyzed two-dimensional features to obtain a fault classification result, performing calculation and judgment on to-be-detected target signal data collected by a sensor, and obtaining a fault classification result; according to the embodiment of the invention, a plurality of outlet pressure plates can be quickly and effectively measured, a large amount of manpower and time are saved, and meanwhile, the detection efficiency and accuracy are improved.

Description

Technical Field

[0001] The present application belongs to the technical field of circuit detection, and in particular, relates to an output matrix detection method based on a relay protection test system. Background Art

[0002] At present, the tripping matrix logics corresponding to different time limits of different types of protection for main transformers are different. In the commissioning of new main transformer protection and the regular inspection of old main transformer protection devices, it is very important to verify the correctness of the action and tripping logic of the protection device. Most transformer products of the same type have similar core functions, such as measuring the export information of the tripping matrix, export verification, data storage, report generation, etc., but they differ in some specific functions or details. For example, some products support touch screen operation, some processors are different, some use single-chip microcomputers, and some use PLCs as control processors. The tripping logic after each type of protection action with different time limits is different, which makes the measurement of the main transformer protection tripping matrix more complicated. If you want to detect the export matrices of these different types of transformers, you need to conduct multiple experiments, which is difficult to obtain accurate test results and will consume a lot of manpower and time. Summary of the invention

[0003] In order to solve the above problems and technical defects, the embodiment of the present application adopts the following technical solution, which is based on the output matrix detection method of the relay protection test system, including the following steps:

[0004] Step 1, a detection sensor is arranged at the relay protection outlet pressure plate, and the normal signal data of the relay protection outlet pressure plate in a normal operating state, the fault signal data in a fault operating state and the target signal data to be detected are collected by the detection sensor;

[0005] Step 2: preprocess the normal signal data, the fault signal data and the target signal data to be detected to generate a normal sample data set and a fault sample data set;

[0006] Step 3: Analyze the normal sample data set and the fault sample data set, map the one-dimensional signal to two-dimensional, obtain the pulse spectrum joint features that fully reflect the non-stationary signal, and obtain the cause of each fault;

[0007] Step 4: Perform model recognition training on the convolutional neural network recognition model based on the pulse spectrum joint features to obtain the fault classification results;

[0008] Step 5: Calculate the target signal data to obtain the target signal feature subset, use the trained convolutional neural network recognition model to judge the target signal feature subset, obtain the current fault classification result and fault cause of the outlet pressure plate, and complete the fault detection of the relay protection outlet pressure plate.

[0009] Preferably, the data preprocessing is to perform linear transformation on the original signal data of the normal signal data and the fault signal data based on maximum-minimum standardization, and then perform normalization calculation processing on the transformed data.

[0010] Furthermore, the linear transformation formula is as follows:

[0011]

[0012] Wherein, minA and maxA are respectively an original value x of the original signal data A mapped to the value of the interval [0, 1] through maximum normalization and minimum normalization.

[0013] Furthermore, the normalized calculation formula is as follows:

[0014]

[0015] Among them, z is the normalized data, is the mean of the original value x, and δ is the variance of the original value x.

[0016] Preferably, the analysis adopts a wavelet transform analysis method, and the pulse spectrum joint characteristics are obtained by analyzing the wavelet transform analysis method, and the formula is as follows:

[0017]

[0018] Among them, a is the scale, which corresponds to the pulse frequency, b is the translation, which corresponds to the time of the pulse signal, and t is the time value. is a fixed parameter related to the pulse spectrum, and R is the time domain sequence of the pulse signal.

[0019] Preferably, the calculation of the target signal data includes first using a PCA feature algorithm to calculate the normalized target signal data to be detected to obtain target features; and then using an LPP algorithm to extract discriminative features for dimensionality reduction to obtain a target signal feature subset.

[0020] Furthermore, the PCA feature algorithm calculation process is as follows:

[0021] The characteristic matrix of the target signal data is mean-centered to determine the main direction w with the largest signal projection variance. The formula is as follows:

[0022]

[0023] ||w|| 2 =1

[0024] Among them, X is the characteristic matrix X of the target signal data m×n, m is the number of signals, n is the number of features, and T is the scale parameter;

[0025] Extract the first k main directions to form the load matrix W = (w1, ...w k ), the result after feature extraction of the target signal data to be detected is X new =XW.

[0026] Furthermore, the process of using the LPP algorithm to extract discriminative features for dimensionality reduction to obtain a target signal feature subset is as follows:

[0027] First, establish neighborhood information on the target feature point, derive linear transformation, and retain the original local information in the subspace after dimensionality reduction;

[0028] The feature matrix is ​​X m*n , m is the number of signals, n is the number of features, first at the signal point x i Select its neighboring point x j , establish the weighted adjacency matrix W = (w ij ) m*m ;

[0029]

[0030] where σ 2 is the ratio parameter. After obtaining the connection between the sample points in the original space, the distance between the points is obtained through the weight. The larger the weight, the closer the distance, and the smaller the weight, the farther the distance.

[0031] The area point features of the target signal point are selected as the target signal feature subset.

[0032] Furthermore, the x i and x j Whether they are mutually domain points is determined using the k-domain method. If x i and x j When one of them is one of the k points with the shortest distance to the other, they are judged to be the boundary points of each other.

[0033] Furthermore, the original local information is retained in the subspace after dimensionality reduction, and a linear isometric transformation is performed from the original high-dimensional signal space to the low-dimensional feature subspace. The calculation formula is as follows:

[0034]

[0035]

[0036] L=DW

[0037] W=(w ij ) n*n

[0038] XL T u=λxDX T u

[0039] Y=u T X

[0040] Among them, Y is the feature subset matrix after linear isometry transformation, s is the time domain skewness of the target signal, t is the time value, D is the diagonal matrix, T is the scale parameter, I is the target signal pulse index, n is the number of features, u is the feature weight vector, W is the weighted adjacency matrix, X is the feature matrix of the target signal data, and L is the Laplace matrix.

[0041] Compared with the prior art, the beneficial effects of the embodiments of the present application are:

[0042] This application obtains the cause of each fault and the result of fault classification by analyzing and calculating from historical signal data, and then uses the PCA feature algorithm and LPP algorithm to calculate and judge the target signal feature subset to obtain the current fault classification result and cause of the outlet pressure plate. It can quickly and effectively measure multiple outlet pressure plates, saving a lot of manpower and time, while also improving detection efficiency and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In the attached picture:

[0044] Figure 1 Schematic diagram of the method steps of an embodiment of the present application. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Generally, the components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various different configurations.

[0046] Example

[0047] like Figure 1 As shown, the output matrix detection method based on the relay protection test system includes the following steps:

[0048] A detection sensor is arranged at the relay protection outlet pressure plate, and the normal signal data of the relay protection outlet pressure plate in a normal operating state, the fault signal data in a fault operating state and the target signal data to be detected are collected by the detection sensor;

[0049] Performing data preprocessing on normal signal data, fault signal data and target signal data to be detected to generate a normal sample data set and a fault sample data set;

[0050] The data preprocessing is to perform linear transformation on the original signal data of normal signal data and fault signal data based on maximum standardization and minimum standardization, and then perform normalization calculation processing on the transformed data;

[0051] The linear transformation formula is as follows:

[0052]

[0053] Among them, minA and maxA are respectively an original value x of the original signal data A mapped to the value of the interval [0,1] through maximum normalization and minimum normalization.

[0054] The normalization calculation formula is as follows:

[0055]

[0056] Among them, z is the normalized data, is the mean of the original value x, and δ is the variance of the original value x.

[0057] The normal sample data set and the fault sample data set are analyzed, and the one-dimensional signal is mapped to two dimensions to obtain the pulse spectrum joint features that fully reflect the non-stationary signal, and obtain the cause of each fault;

[0058] The analysis adopts the wavelet transform analysis method, and the pulse spectrum joint characteristics are obtained by the wavelet transform analysis method. The formula is as follows:

[0059]

[0060] Among them, a is the scale, which corresponds to the pulse frequency, b is the translation, which corresponds to the time of the pulse signal, and t is the time value. is a fixed parameter related to the pulse spectrum, and R is the time domain sequence of the pulse signal.

[0061] Use the analyzed two-dimensional features to train the convolutional neural network model recognition and obtain the fault classification results;

[0062] The target signal data to be detected collected by the sensor is calculated to obtain the target signal feature subset, and the trained convolutional neural network recognition model is used to judge the target signal feature subset to obtain the current fault classification result and cause of the outlet pressure plate, thereby completing the fault detection of the relay protection outlet pressure plate.

[0063] The calculation of the target signal data includes first using the PCA feature algorithm to calculate the normalized target signal data to be detected to obtain the target features; then using the LPP algorithm to extract the discriminative features for dimensionality reduction to obtain the target signal feature subset.

[0064] The PCA feature algorithm calculation process is as follows:

[0065] The characteristic matrix of the target signal data is mean-centered to determine the main direction w with the largest signal projection variance. The formula is as follows:

[0066]

[0067] ||w|| 2 =1

[0068] Among them, X is the characteristic matrix X of the target signal data m×n , m is the number of signals, n is the number of features, and T is the scale parameter;

[0069] Extract the first k main directions to form the load matrix W = (w1, ...w k ), the result after feature extraction of the target signal data to be detected is X new =XW.

[0070] The process of using the LPP algorithm to extract discriminative features for dimensionality reduction and obtain the target signal feature subset is as follows:

[0071] First, establish neighborhood information on the target feature point, derive linear transformation, and retain the original local information in the subspace after dimensionality reduction;

[0072] The feature matrix is ​​X m*n , m is the number of signals, n is the number of features, first at the signal point x i Select its neighboring point x j , establish the weighted adjacency matrix W = (w ij ) m*m ;

[0073]

[0074] where σ 2 is the ratio parameter. After obtaining the connection between the sample points in the original space, the distance between the points is obtained through the weight. The larger the weight, the closer the distance, and the smaller the weight, the farther the distance.

[0075] The area point features of the target signal point are selected as the target signal feature subset.

[0076] x i and x j Whether they are mutually domain points is determined using the k-domain method. If x iand x j When one of them is one of the k points with the shortest distance to the other, they are judged to be the boundary points of each other.

[0077] In the subspace after dimensionality reduction, the original local information is retained. When the original high-dimensional signal space is transformed into the low-dimensional feature subspace, a linear isometric transformation is performed. The calculation formula is as follows:

[0078]

[0079]

[0080] L=DW

[0081] W=(w ij ) n*n

[0082] XL T u=λxDX T u

[0083] Y=u T X

[0084] Among them, Y is the feature subset matrix after linear isometry transformation, s is the time domain skewness of the target signal, t is the time value, D is the diagonal matrix, T is the scale parameter, I is the target signal pulse index, n is the number of features, u is the feature weight vector, W is the weighted adjacency matrix, X is the feature matrix of the target signal data, and L is the Laplace matrix.

[0085] The above-mentioned embodiments only express the preferred implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for ordinary technicians in this field, several modifications, improvements and substitutions can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application.

Claims

1. An outlet matrix detection method based on a relay protection test system is characterized in that: The following steps are involved: Step 1, a detection sensor is set at the relay protection outlet pressure plate, and the normal signal data of the relay protection outlet pressure plate in a normal operating state, the fault signal data in a fault operating state, and the target signal data to be detected are collected by the detection sensor; Step 2: preprocess the normal signal data, the fault signal data and the target signal data to be detected to generate a normal sample data set and a fault sample data set; Step 3: Analyze the normal sample data set and the fault sample data set, map the one-dimensional signal to two-dimensional, obtain the pulse spectrum joint features that fully reflect the non-stationary signal, and obtain the cause of each fault; Step 4: Perform model recognition training on the convolutional neural network recognition model based on the pulse spectrum joint features to obtain the fault classification results; Step 5: Calculate the target signal data to obtain the target signal feature subset, use the trained convolutional neural network recognition model to judge the target signal feature subset, obtain the current fault classification result and fault cause of the outlet pressure plate, and complete the fault detection of the relay protection outlet pressure plate.

2. The output matrix detection method based on the relay protection test system according to claim 1 is characterized in that: The data preprocessing is to perform linear transformation on the original signal data of the normal signal data and the fault signal data based on maximum standardization and minimum standardization, and then perform normalization calculation processing on the transformed data.

3. The output matrix detection method based on the relay protection test system according to claim 2 is characterized in that: The linear transformation formula is as follows: Among them, minA and maxA are respectively an original value x of the original signal data A mapped to the value of the interval [0,1] through maximum normalization and minimum normalization.

4. The output matrix detection method based on the relay protection test system according to claim 3 is characterized in that: The normalized calculation formula is as follows: Among them, z is the normalized data, is the mean of the original value x, and δ is the variance of the original value x.

5. The output matrix detection method based on the relay protection test system according to claim 1 is characterized in that: The analysis adopts the wavelet transform analysis method, and the pulse spectrum joint characteristics are obtained by the wavelet transform analysis method. The formula is as follows: Among them, a is the scale, which corresponds to the pulse frequency, b is the translation, which corresponds to the time of the pulse signal, and t is the time value. is a fixed parameter related to the pulse spectrum, and R is the time domain sequence of the pulse signal.

6. The output matrix detection method based on the relay protection test system according to claim 1 is characterized in that: The calculating of the target signal data includes first using a PCA feature algorithm to calculate the normalized target signal data to be detected to obtain target features; Then the LPP algorithm is used to extract discriminative features for dimensionality reduction to obtain the target signal feature subset.

7. The output matrix detection method based on the relay protection test system according to claim 6 is characterized in that: The PCA feature algorithm calculation process is as follows: The characteristic matrix of the target signal data is mean-centered to determine the main direction w with the largest signal projection variance. The formula is as follows: ||w|| 2 ＝1 Among them, X is the characteristic matrix X of the target signal data m×n , m is the number of signals, n is the number of features, and T is the scale parameter; Extract the first k main directions to form the load matrix W = (w1, ...w k ), the result after feature extraction of the target signal data to be detected is X new =XW.

8. The output matrix detection method based on the relay protection test system according to claim 7 is characterized in that: The process of using the LPP algorithm to extract discriminative features for dimensionality reduction to obtain a target signal feature subset is as follows: First, establish neighborhood information on the target feature point, derive linear transformation, and retain the original local information in the subspace after dimensionality reduction; The feature matrix is ​​X m*n , m is the number of signals, n is the number of features, first at the signal point x i Select its neighboring point x j , establish the weighted adjacency matrix W = (w ij ) m*m ; where σ 2 is the ratio parameter. After obtaining the connection between the sample points in the original space, the distance between the points is obtained through the weight. The larger the weight, the closer the distance, and the smaller the weight, the farther the distance. The area point features of the target signal point are selected as the target signal feature subset.

9. The output matrix detection method based on the relay protection test system according to claim 8 is characterized in that: The x i and x j Whether they are mutually domain points is determined using the k-domain method. If x i and x j When one of them is one of the k points with the shortest distance to the other, they are judged to be the boundary points of each other.

10. The output matrix detection method based on the relay protection test system according to claim 9 is characterized in that: The original local information is retained in the subspace after dimensionality reduction, and a linear isometric transformation is performed from the original high-dimensional signal space to the low-dimensional feature subspace. The calculation formula is as follows: L=DW In=(in ij ) n*n XLX T u=λXDX T u Y=u T X Among them, Y is the feature subset matrix after linear isometry transformation, s is the time domain skewness of the target signal, t is the time value, D is the diagonal matrix, T is the scale parameter, I is the target signal pulse index, n is the number of features, u is the feature weight vector, W is the weighted adjacency matrix, X is the feature matrix of the target signal data, and L is the Laplace matrix.

Images