Series arc fault on-line detection method and apparatus for three-phase inverter lines

Abstract

This invention provides an online detection method for series arc faults in three-phase frequency converter lines. The method includes setting a sampling frequency and, based on this sampling frequency, acquiring the loop current signal of any phase of the inverter's front-end line during operation. The acquired loop current signal is normalized and, after frequency sampling, mapped point-by-point equidistantly to an input matrix. The discrete points in the mapped input matrix are then connected to obtain the final input matrix. The final input matrix is ​​then imported into a pre-trained series arc fault identification model to identify whether a series arc fault exists in the three-phase frequency converter line. This invention also provides an online detection device for series arc faults in three-phase frequency converter lines. Implementing this invention accurately detects series arc faults and issues fault warning signals, improving the operational reliability of three-phase frequency converter lines.

Description

Technical Field

[0001] This invention relates to the field of online detection technology for series arc faults, and in particular to a method and apparatus for online detection of series arc faults in three-phase frequency converter lines. Background Technology

[0002] To achieve energy conservation and emission reduction, and to improve the speed regulation performance of three-phase motors, three-phase frequency converters are widely used in various industrial fields such as mining, machinery, and metallurgy. During the operation of three-phase frequency converters and motor power supply lines, series arcing faults frequently occur due to reasons such as loose terminals, broken cable cores, and insulation damage, which can lead to electrical fires in severe cases. Therefore, it is necessary to detect series arcing faults in three-phase frequency converter lines.

[0003] Currently, the mainstream methods for identifying series arc faults at home and abroad are mainly as follows: (1) Detecting series arc faults by using signals such as arc sound, arc light, and electromagnetic radiation noise generated by the arc; (2) Performing time-domain, frequency-domain, or time-frequency-domain analysis on the voltage and / or loop current signals of the load line to find the fault characteristics of the line voltage and / or loop current signals under the series arc fault state, and using nonlinear classifiers such as support vector machines and neural networks to identify series arc faults; (3) Using end-to-end deep learning algorithms (such as convolutional neural networks, recurrent neural networks, etc.) to input the original or simply pre-processed line voltage and / or loop current signals into the identification model to identify series arc faults.

[0004] However, all three methods for identifying series arc faults have shortcomings. Specifically: In method (1), series arc faults are highly random, requiring sensors to be installed at fixed detection points, making it impossible to detect the entire load line's status; in method (2), fault features of series arc faults need to be manually extracted and classifiers manually selected, resulting in strong subjectivity. Furthermore, the fault features of series arc faults in line voltage and / or loop current signals vary significantly under different circuit topologies and load types, making this method less universal; in method (3), while deep learning algorithms can eliminate subjectivity in manually selecting fault features and improve the universality of the identification model, they rarely address hardware deployment and performance improvement on hardware platforms. Most identification models only implement performance testing on computers and cannot run online in embedded devices, failing to meet the actual needs of industrial and mining enterprises.

[0005] Therefore, there is an urgent need for a new method for identifying series arc faults, which can realize online detection and early warning functions for series arc faults in industrial three-phase frequency converter circuits in embedded devices. Summary of the Invention

[0006] The technical problem to be solved by the embodiments of the present invention is to provide an online detection method and device for series arc faults in three-phase frequency converter lines, which can detect series arc faults in three-phase frequency converter lines in a timely and accurate manner and issue fault warning signals, thereby improving the operational reliability of three-phase frequency converter lines.

[0007] To address the aforementioned technical problems, this invention provides an online detection method for series arc faults in three-phase frequency converter lines, the method comprising the following steps:

[0008] Set the sampling frequency, and based on the sampling frequency, obtain the loop current signal of any one phase of the inverter front end during the operation of the three-phase inverter line;

[0009] The acquired loop current signal is normalized and then mapped to the input matrix point by point through frequency sampling. The discrete points in the mapped input matrix are then connected to obtain the final input matrix.

[0010] The final input matrix is ​​then imported into the trained series arc fault identification model to identify whether a series arc fault exists in the three-phase frequency converter line.

[0011] The specific steps of normalizing the acquired loop current signal, sampling it by frequency, mapping it point-by-point to the input matrix, and connecting the discrete points in the mapped input matrix to obtain the final input matrix include:

[0012] Through formula Normalization is performed; where vector V represents a data sample, V(i) represents the i-th element of vector V, and V... max V represents the maximum value of vector V. min V'(i) represents the minimum value of vector V, and V'(i) represents the normalized value of the i-th element of vector V, where i represents a sampling point.

[0013] Frequency sampling is performed on the normalized loop current signal; wherein the sampling point n generated by the frequency sampling is less than or equal to the sampling point N generated corresponding to the sampling frequency;

[0014] Initialize an n×n identity matrix, and establish a horizontal axis x and a vertical axis y with the element in the nth row and 1st column of this matrix as the origin. For each sampling point generated by the frequency sampling, use the formula... Calculate its index on the x-axis and y-axis of the identity matrix in turn, and then set the element at that index position in the identity matrix from 1 to 0 to obtain the input matrix for the completed mapping; where i = 0, 1, 2, ..., n-1;

[0015] For the discrete points (x') in the input matrix that have completed the mapping i ,y' i ), and sequentially put the x'th i and x' i+1 Column, y' i to y' i+1 Set all matrix elements between rows to 0 to obtain the final input matrix.

[0016] The series arc fault identification model is constructed based on a lightweight convolutional neural network and incorporates a channel attention mechanism.

[0017] The series arc fault identification model includes a feature extraction module and a feature calculation module; wherein, the feature extraction module consists of 1 standard convolutional layer, 2 depthwise separable convolutional layers, 3 SE blocks and 3 max pooling layers, and each depthwise separable convolutional layer consists of a depthwise convolutional layer and a point convolutional layer; the feature calculation module consists of 1 unfolded layer and 2 fully connected layers.

[0018] The method further includes:

[0019] The number of times a series arc fault occurs in the three-phase frequency converter line is accumulated. When the total number of accumulated faults exceeds the prediction threshold, the location and time information of the series arc fault are determined and then pushed to the user.

[0020] The loop current signal is obtained by measuring a Hall sensor.

[0021] This invention also provides an online detection device for series arc faults in a three-phase frequency converter line, comprising:

[0022] The signal acquisition unit is used to set the sampling frequency and, based on the sampling frequency, acquire the loop current signal of any one phase of the inverter front end during the operation of the three-phase inverter line.

[0023] The signal processing unit is used to normalize the acquired loop current signal, and after frequency sampling, it maps it point by point to the input matrix at equal intervals. The discrete points in the mapped input matrix are then connected to obtain the final input matrix.

[0024] The online fault detection unit is used to import the final input matrix into the trained series arc fault identification model to identify whether a series arc fault exists in the three-phase frequency converter line.

[0025] The signal processing unit includes:

[0026] The data normalization module is used to process data using formulas. Normalization is performed; where vector V represents a data sample, V(i) represents the i-th element of vector V, and V... max V represents the maximum value of vector V. min V'(i) represents the minimum value of vector V, and V'(i) represents the normalized value of the i-th element of vector V, where i represents a sampling point.

[0027] The data frequency sampling module is used to perform frequency sampling on the normalized loop current signal; wherein the sampling point n generated by the frequency sampling is less than or equal to the sampling point N generated corresponding to the sampling frequency;

[0028] The matrix mapping module is used to initialize an n×n identity matrix, and establish a horizontal axis x and a vertical axis y with the element of the nth row and 1st column of this matrix as the origin. For each sampling point generated by the frequency sampling, the matrix is ​​mapped using the formula... Calculate its index on the x-axis and y-axis of the identity matrix in turn, and then set the element at that index position in the identity matrix from 1 to 0 to obtain the input matrix for the completed mapping; where i = 0, 1, 2, ..., n-1;

[0029] The matrix discrete connection module is used to perform a mapping on discrete points (x') in the input matrix. i ,y' i ), and sequentially put the x'th i and x' i+1 Column, y' i to y' i+1 Set all matrix elements between rows to 0 to obtain the final input matrix.

[0030] The series arc fault identification model is constructed based on a lightweight convolutional neural network and incorporates a channel attention mechanism.

[0031] The series arc fault identification model includes a feature extraction module and a feature calculation module; wherein, the feature extraction module consists of 1 standard convolutional layer, 2 depthwise separable convolutional layers, 3 SE blocks and 3 max pooling layers, and each depthwise separable convolutional layer consists of a depthwise convolutional layer and a point convolutional layer; the feature calculation module consists of 1 unfolded layer and 2 fully connected layers.

[0032] This also includes: a fault push unit;

[0033] The fault push unit is used to accumulate the number of times a series arc fault exists in the three-phase frequency converter line, and when the total number of accumulations exceeds the prediction threshold, it determines the location and time information of the series arc fault and pushes it to the user.

[0034] The loop current signal is obtained by measuring a Hall sensor.

[0035] Implementing the embodiments of the present invention has the following beneficial effects:

[0036] 1. This invention eliminates the subjectivity of human signal analysis and fault feature extraction. It only requires real-time acquisition of the loop current signal of any one phase at the front end of the three-phase frequency converter to accurately detect series arc faults in the three-phase frequency converter line and issue a fault warning signal, thereby improving the operational reliability of the three-phase frequency converter line. It requires fewer current sensors and is easy to implement in engineering.

[0037] 2. The point-by-point equidistant mapping method proposed in this invention can not only enable the signal acquisition unit to adapt to the signal sampling frequency, but also has the advantages of high real-time performance and the ability to intuitively retain sample features;

[0038] 3. This invention introduces the channel attention mechanism into the series arc fault identification model, which improves the reliability of the identification model. It uses depthwise separable convolution to perform lightweight design of the identification model, which simplifies the computational complexity of the model, reduces the computation time of the model, facilitates the embedded deployment of the model and enables it to run efficiently on embedded devices, and ensures the real-time performance of series arc fault identification. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, obtaining other drawings based on these drawings without creative effort still falls within the scope of the present invention.

[0040] Figure 1 A flowchart illustrating an online detection method for series arc faults in a three-phase frequency converter circuit, provided by an embodiment of the present invention;

[0041] Figure 2 This is a schematic diagram of the structure of a series arc fault identification model in an online detection method for series arc faults in a three-phase frequency converter line provided in an embodiment of the present invention;

[0042] Figure 3The logic block diagram for constructing the input matrix in an online detection method for series arc faults in a three-phase frequency converter line provided in an embodiment of the present invention;

[0043] Figure 4 This is a schematic diagram of the structure of an online detection device for series arc faults in a three-phase frequency converter circuit, provided in an embodiment of the present invention. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings.

[0045] like Figure 1 As shown in the figure, an online detection method for series arc faults in a three-phase frequency converter line is proposed in an embodiment of the present invention. The method includes the following steps:

[0046] Step S1: Set the sampling frequency, and based on the sampling frequency, obtain the loop current signal of any one phase of the inverter front end during the operation of the three-phase inverter line.

[0047] Step S2: Normalize the acquired loop current signal, and after frequency sampling, map it point by point to the input matrix at equal intervals. Connect the discrete points in the mapped input matrix to obtain the final input matrix.

[0048] Step S3: Import the final input matrix into the trained series arc fault identification model to identify whether a series arc fault exists in the three-phase frequency converter line.

[0049] The specific process is as follows: Before step S1, a series arc fault identification model is first constructed. This model is based on a lightweight convolutional neural network and incorporates a channel attention mechanism. The model includes a feature extraction module and a feature calculation module. The feature extraction module consists of one standard convolutional layer, two depthwise separable convolutional layers, three SE blocks (Squeeze and Excitation blocks), and three max-pooling layers. Each depthwise separable convolutional layer is composed of depthwise convolutional layers and pointwise convolutional layers. The feature calculation module consists of one unfolded layer and two fully connected layers. It should be noted that the series arc fault identification model outputs a 1×2 vector. If the index of the maximum value in the vector is 0, it indicates that the data sample input to the model is a fault class sample; conversely, if the index of the maximum value in the vector is 1, it indicates that the data sample input to the model is a normal class sample.

[0050] like Figure 2As shown, the feature extraction module consists of one standard convolutional layer C1, two depthwise separable convolutional layers (two depthwise convolutional layers D1-D2 and two pointwise convolutional layers P1-P2), three SE blocks S1-S3, and three max pooling layers M1-M3. The feature calculation module consists of one unfolded layer Z1 and two fully connected layers F1-F2. The connection order between layers is C1-S1-M1-D1-P1-S2-M2-D2-P2-S3-M3-Z1-F1-F2.

[0051] In one example, taking the deep convolutional layer D2 as an example, the shape of the input feature is 36×36×32. D2 contains 32 convolutional kernels of size 3×3×1, with a stride of 1. The shape of the output feature obtained after processing by this layer becomes [(36-3+1) / 1]×[(36-3+1) / 1]×32, that is, 34×34×32.

[0052] In another example, taking point convolutional layer P2 as an example, the shape of the input feature is 34×34×32. P2 contains 64 convolutional kernels of size 1×1×32, with a stride of 1. The shape of the output feature obtained after processing by this layer becomes [(34-1+1) / 1]×[(34-1+1) / 1]×64, that is, 34×34×64.

[0053] In another example, taking SE block S3 as an example, the input feature shape is 34×34×64, and the output feature shape after processing by this block is 34×34×64. In this case, the calculation formulas for the Squeeze and Excitation operations are as follows:

[0054]

[0055] F ex (z,W)=σ(W2RELU(W1z))

[0056] In the formula, u c Given a 34×34×1 feature map of the input SE block, where H and W are the height and width of the feature map, and z represents the area after passing through F... sq (u c The real numbers obtained by the operation are W1 and W2, which represent the weight parameters generated and updated by the Excitation operation, σ represents the Sigmoid activation function, and RELU represents the ReLU activation function.

[0057] In another example, taking the max pooling layer M3 as an example, the shape of the input feature is 34×34×64, and the stride of M3 is set to 2. The shape of the output feature obtained after processing by this layer becomes (34 / 2)×(34 / 2)×64, that is, 17×17×64. The above parameters need to be adjusted according to the effect of the recognition model.

[0058] In another example, taking the fully connected layer F2 as an example, the shape of the input features is 1×128×, and the shape of the output features obtained after processing by this layer becomes 1×2. The output features of the fully connected layer F2 contain two values, representing the prediction probabilities of the recognition model that the data sample belongs to the normal class and the fault class, respectively.

[0059] It should be noted that the above parameters need to be adjusted according to the effect of the recognition model.

[0060] Secondly, obtain samples, including training samples and test samples.

[0061] (21) The sampling frequency is set to 25kHz. Under the condition of 380V 50Hz mains power, one current cycle contains 500 sampling points.

[0062] (22) Based on the sampling frequency, during the operation of the three-phase inverter line, the arc voltage signal and the loop current signal of any phase line at the front end of the inverter are collected by Hall sensors. It should be noted that, based on the arc voltage signal, normal data samples and fault data samples of the loop current signal are distinguished, and labels are added to the data samples of different categories to obtain different datasets.

[0063] (23) The loop current signal is divided into data samples with one current cycle length as the interval, and labels are attached to the data samples of different states to obtain the loop current sample dataset A when the three-phase frequency converter line is running normally, and the loop current sample dataset B when a series arc fault occurs in the three-phase frequency converter line. The sample datasets A and B are further combined and randomly shuffled to obtain the series arc fault dataset C.

[0064] (24) Divide the dataset C into training and testing sets for the model according to a certain ratio. For example, divide the dataset C into training and testing sets for the model according to a ratio of 80% and 20%.

[0065] Next, the training set data samples and the test set data samples are preprocessed into the input data required by the series arc fault identification model.

[0066] (31) Normalization process:

[0067] Through formula The training set data samples and test set data samples containing N=500 sampling points are normalized; where vector V represents a data sample, V(i) represents the i-th element of vector V, and V max V represents the maximum value of vector V. min V'(i) represents the minimum value of vector V, and V'(i) represents the normalized value of the i-th element of vector V, where i represents a sampling point.

[0068] (32) Frequency sampling:

[0069] After normalization, both the training and test data samples contain 500 sampling points. The training and test data samples are then downsampled to n=150 sampling points using frequency sampling.

[0070] (33) Mapping matrix:

[0071] First, initialize a 150×150 identity matrix. Using the element in the 150th row and 1st column of this matrix as the origin, establish the horizontal x-axis and vertical y-axis. For each sampling point in the frequency-sampled training and test sets, calculate its index on the x-axis and y-axis of the identity matrix according to the following formulas. Then, set the element at that index position in the identity matrix from 1 to 0, finally obtaining the input matrix with complete mapping.

[0072] x′ i =ii=0,1,2,...,149.;y i '=(y i -0)×149i=0,1,2,...,149.

[0073] (34) Connecting discrete points:

[0074] For the discrete points (x') in the input matrix that have completed the mapping i ,y' i ), i = 0, 1, 2, ... 149, sequentially assign the x'th... i and x' i+1 Column, y' i to y' i+1 All matrix elements between rows are set from 1 to 0 to obtain the input matrices corresponding to the final training set and test set, respectively.

[0075] Finally, based on the input matrices corresponding to the final training set and test set, the series arc fault identification model is trained and tested to obtain the trained series arc fault identification model.

[0076] At this point, the number of training iterations for the series arc fault identification model is set to 30. The optimizer uses the Adam optimization algorithm and utilizes the class cross-entropy function to measure the loss between the predicted and actual values ​​of the series arc fault identification model. The calculation formula is as follows:

[0077]

[0078] In the formula, y represents the true value, a represents the predicted value of the recognition model, x is the data sample, and n represents the number of data samples.

[0079] During each training iteration, the Adam optimization algorithm is used to update the weight parameters of the series arc fault identification model, and the loss function value and accuracy of the model on the training and test sets are examined. The series arc fault identification model is trained according to the number of training iterations. When the number of iterations reaches 30, if the loss function value of the model approaches 0 and becomes relatively stable, the training iteration process is terminated, and the trained series arc fault identification model is obtained. Otherwise, the number of training iterations is appropriately extended.

[0080] It should be noted that the loss value of the trained series arc fault identification model on the training and test sets is less than 0.001, and the accuracy is higher than 99.90%, indicating that the trained series arc fault identification model has high reliability and can meet the actual detection needs.

[0081] In step S1, based on the sampling frequency (e.g., 25kHz) mentioned above, during the operation of the three-phase inverter line, a Hall-type current transformer is used to collect the loop current signal of any one phase of the inverter front-end line.

[0082] In step S2, the final input matrix of the loop current signal is constructed, and the specific steps are as follows:

[0083] First, through the formula Normalization is performed; where vector V represents a data sample, V(i) represents the i-th element of vector V, and V... max V represents the maximum value of vector V. min V'(i) represents the minimum value of vector V, and V'(i) represents the normalized value of the i-th element of vector V, where i represents a sampling point.

[0084] Secondly, frequency sampling is performed on the normalized loop current signal; wherein, the sampling point n (e.g., 150) generated by frequency sampling is less than or equal to the sampling point N (e.g., 500) generated by the sampling frequency.

[0085] Then, initialize an n×n identity matrix, and establish the horizontal axis x and vertical axis y with the element of the nth row and 1st column of this matrix as the origin. For each sampling point generated by the frequency sampling, use the formula... Calculate its index on the x-axis and y-axis of the identity matrix in turn, and then set the element at that index position in the identity matrix from 1 to 0 to obtain the input matrix for the completed mapping; where i = 0, 1, 2, ..., n-1;

[0086] Finally, for the discrete points (x') in the input matrix that have completed the mappingi ,y' i ), and sequentially put the x'th i and x' i+1 Column, y' i to y' i+1 Set all matrix elements between rows to 0 to obtain the final input matrix.

[0087] In step S3, the final input matrix obtained in step S2 is imported into the trained series arc fault identification model to identify whether a series arc fault exists in the three-phase frequency converter line.

[0088] If the data sample is predicted to be a normal sample, the next online detection begins. If the data sample is predicted to be a fault sample, the status flag is incremented by 1, and the next online detection begins. When the status flag is accumulated to 10 within 0.5 seconds, the status flag is set to 0, and the location and time information of the series arc fault are stored and synchronously sent to the user. Therefore, this method further includes: accumulating the number of times a series arc fault exists in the three-phase inverter line, and when the total number of accumulations exceeds the prediction threshold, determining the location and time information of the series arc fault, and further pushing it to the user.

[0089] like Figure 4 As shown in the figure, an online detection device for series arc faults in a three-phase frequency converter line is provided in an embodiment of the present invention, comprising:

[0090] The signal acquisition unit 110 is used to set the sampling frequency and, based on the sampling frequency, acquire the loop current signal of any one phase of the inverter front end during the operation of the three-phase inverter line.

[0091] The signal processing unit 120 is used to normalize the acquired loop current signal, and after frequency sampling, it maps it to the input matrix point by point at equal intervals. The discrete points in the mapped input matrix are then connected to obtain the final input matrix.

[0092] The online fault detection unit 130 is used to import the final input matrix into the trained series arc fault identification model to identify whether a series arc fault exists in the three-phase frequency converter line.

[0093] The signal processing unit 120 includes:

[0094] The data normalization module is used to process data using formulas. Normalization is performed; where vector V represents a data sample, V(i) represents the i-th element of vector V, and V... max V represents the maximum value of vector V. minV'(i) represents the minimum value of vector V, and V'(i) represents the normalized value of the i-th element of vector V, where i represents a sampling point.

[0095] The data frequency sampling module is used to perform frequency sampling on the normalized loop current signal; wherein the sampling point n generated by the frequency sampling is less than or equal to the sampling point N generated corresponding to the sampling frequency;

[0096] The matrix mapping module is used to initialize an n×n identity matrix, and establish a horizontal axis x and a vertical axis y with the element of the nth row and 1st column of this matrix as the origin. For each sampling point generated by the frequency sampling, the matrix is ​​mapped using the formula... Calculate its index on the x-axis and y-axis of the identity matrix in turn, and then set the element at that index position in the identity matrix from 1 to 0 to obtain the input matrix for the completed mapping; where i = 0, 1, 2, ..., n-1;

[0097] The matrix discrete connection module is used to perform a mapping on discrete points (x') in the input matrix. i ,y' i ), and sequentially put the x'th i and x' i+1 Column, y' i to y' i+1 Set all matrix elements between rows to 0 to obtain the final input matrix.

[0098] The series arc fault identification model is constructed based on a lightweight convolutional neural network and incorporates a channel attention mechanism.

[0099] The series arc fault identification model includes a feature extraction module and a feature calculation module; wherein, the feature extraction module consists of 1 standard convolutional layer, 2 depthwise separable convolutional layers, 3 SE blocks and 3 max pooling layers, and each depthwise separable convolutional layer consists of a depthwise convolutional layer and a point convolutional layer; the feature calculation module consists of 1 unfolded layer and 2 fully connected layers.

[0100] This also includes: a fault push unit;

[0101] The fault push unit is used to accumulate the number of times a series arc fault exists in the three-phase frequency converter line, and when the total number of accumulations exceeds the prediction threshold, it determines the location and time information of the series arc fault and pushes it to the user.

[0102] The loop current signal is obtained by measuring a Hall sensor.

[0103] Implementing the embodiments of the present invention has the following beneficial effects:

[0104] 1. This invention eliminates the subjectivity of human signal analysis and fault feature extraction. It only requires real-time acquisition of the loop current signal of any one phase at the front end of the three-phase frequency converter to accurately detect series arc faults in the three-phase frequency converter line and issue a fault warning signal, thereby improving the operational reliability of the three-phase frequency converter line. It requires fewer current sensors and is easy to implement in engineering.

[0105] 2. The point-by-point equidistant mapping method proposed in this invention can not only enable the signal acquisition unit to adapt to the signal sampling frequency, but also has the advantages of high real-time performance and the ability to intuitively retain sample features;

[0106] 3. This invention introduces the channel attention mechanism into the series arc fault identification model, which improves the reliability of the identification model. It uses depthwise separable convolution to perform lightweight design of the identification model, which simplifies the computational complexity of the model, reduces the computation time of the model, facilitates the embedded deployment of the model and enables it to run efficiently on embedded devices, and ensures the real-time performance of series arc fault identification.

[0107] It is worth noting that in the above-described device embodiments, the various units are divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy differentiation and are not intended to limit the scope of protection of the present invention.

[0108] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as ROM / RAM, disk, optical disk, etc.

[0109] The above description is merely a preferred embodiment of the present invention and should not be construed as limiting the scope of the invention. Therefore, any equivalent variations made in accordance with the claims of the present invention are still within the scope of the present invention.

Claims

1. A method for online detection of series arc faults in a three-phase frequency converter circuit, characterized in that, The method includes the following steps: Set the sampling frequency, and based on the sampling frequency, obtain the loop current signal of any one phase of the inverter front end during the operation of the three-phase inverter line; The acquired loop current signal is normalized and then mapped to the input matrix point by point through frequency sampling. The discrete points in the mapped input matrix are then connected to obtain the final input matrix. The final input matrix is ​​then imported into the trained series arc fault identification model to identify whether a series arc fault exists or not in the three-phase frequency converter line. The specific steps for normalizing the acquired loop current signal, sampling it by frequency, mapping it point-by-point to the input matrix, and connecting the discrete points in the mapped input matrix to obtain the final input matrix include: Through formula Normalization is performed; where the vector V To represent a data sample, V(i) Representing vectors V The i One element, V max Representing vectors V The maximum value, V min Representing vectors V The minimum value, V'(i) Representing vectors V The i The value of each element after normalization. i Represents a sampling point; Frequency sampling is performed on the normalized loop current signal; wherein, the sampling points generated by the frequency sampling n The number of sampling points generated at a sampling frequency less than or equal to the stated sampling frequency N ; Initialize 1 n × n The identity matrix, and with the first [value] of the matrix. n Using the elements of row 1 and column 1 as the origin, establish a horizontal axis x and a vertical axis y. For each sampling point generated by the frequency sampling, use the formula... Calculate its index on the x-axis and y-axis of the identity matrix sequentially, and then set the element at that index position in the identity matrix from 1 to 0 to obtain the input matrix after mapping; where, ; For discrete points in the input matrix that have completed the mapping (x ’ i , y ’ i ) , in turn x ’ i and x ’ i+1 Column, number y ’ i to y ’ i+1 Set all matrix elements between rows to 0 to obtain the final input matrix.

2. The online detection method for series arc faults in a three-phase frequency converter circuit as described in claim 1, characterized in that, The series arc fault identification model is built based on a lightweight convolutional neural network and incorporates a channel attention mechanism; wherein, The series arc fault identification model includes a feature extraction module and a feature calculation module; wherein, the feature extraction module consists of 1 standard convolutional layer, 2 depthwise separable convolutional layers, 3 SE blocks and 3 max pooling layers, and each depthwise separable convolutional layer consists of a depthwise convolutional layer and a point convolutional layer; the feature calculation module consists of 1 unfolded layer and 2 fully connected layers.

3. The online detection method for series arc faults in a three-phase frequency converter circuit as described in claim 1, characterized in that, The method further includes: The number of times a series arc fault occurs in the three-phase frequency converter line is accumulated. When the total number of accumulated faults exceeds the prediction threshold, the location and time information of the series arc fault are determined and then pushed to the user.

4. The online detection method for series arc faults in a three-phase frequency converter circuit as described in claim 1, characterized in that, The loop current signal is obtained by measuring a Hall sensor.

5. An online detection device for series arc faults in a three-phase frequency converter circuit, characterized in that, include; The signal acquisition unit is used to set the sampling frequency and, based on the sampling frequency, acquire the loop current signal of any one phase of the inverter front end during the operation of the three-phase inverter line. The signal processing unit is used to normalize the acquired loop current signal, and after frequency sampling, it maps it point by point to the input matrix at equal intervals. The discrete points in the mapped input matrix are then connected to obtain the final input matrix. The online fault detection unit is used to import the final input matrix into the trained series arc fault identification model to identify whether a series arc fault exists or not in the three-phase frequency converter line. The signal processing unit includes: The data normalization module is used to process data using formulas. Normalization is performed; where the vector V To represent a data sample, V(i) Representing vectors V The i One element, V max Representing vectors V The maximum value, V min Representing vectors V The minimum value, V'(i) Representing vectors V The i The value of each element after normalization. i Represents a sampling point; The data frequency sampling module is used to perform frequency sampling on the normalized loop current signal; wherein, the sampling points generated by the frequency sampling... n The number of sampling points generated at a sampling frequency less than or equal to the stated sampling frequency N ; The matrix mapping module is used to initialize one n × n The identity matrix, and with the first [value] of the matrix. n Using the elements of row 1 and column 1 as the origin, establish a horizontal axis x and a vertical axis y. For each sampling point generated by the frequency sampling, use the formula... Calculate its index on the x-axis and y-axis of the identity matrix sequentially, and then set the element at that index position in the identity matrix from 1 to 0 to obtain the input matrix after mapping; where, ; The matrix discrete connection module is used to perform a mapping on discrete points in the input matrix. (x ’ i , y ’ i ) , in turn x ’ i and x ’ i+1 Column, number y ’ i to y ’ i+1 Set all matrix elements between rows to 0 to obtain the final input matrix.

6. The online detection device for series arc faults in a three-phase frequency converter circuit as described in claim 5, characterized in that, The series arc fault identification model is built based on a lightweight convolutional neural network and incorporates a channel attention mechanism; wherein, The series arc fault identification model includes a feature extraction module and a feature calculation module; wherein, the feature extraction module consists of 1 standard convolutional layer, 2 depthwise separable convolutional layers, 3 SE blocks and 3 max pooling layers, and each depthwise separable convolutional layer consists of a depthwise convolutional layer and a point convolutional layer; the feature calculation module consists of 1 unfolded layer and 2 fully connected layers.

7. The online detection device for series arc faults in a three-phase frequency converter circuit as described in claim 5, characterized in that, Also includes: Fault push unit; The fault push unit is used to accumulate the number of times a series arc fault exists in the three-phase frequency converter line, and when the total number of accumulations exceeds the prediction threshold, it determines the location and time information of the series arc fault and pushes it to the user.

8. The online detection device for series arc faults in a three-phase frequency converter circuit as described in claim 5, characterized in that, The loop current signal is obtained by measuring a Hall sensor.

Images