Detection model training method and system, line detection method, equipment and medium

Abstract

The invention provides a detection model training method and system, a line detection method, equipment and a medium. The detection model training method comprises the following steps: acquiring frequency analysis data of a transformer; processing the frequency analysis data by using sliding correlation to obtain a characteristic heat map of the frequency analysis data; training a convolution kernel of the initial convolutional neural network model by using an algorithm to obtain a target convolution step length of the initial convolutional neural network model; and training the initial convolutional neural network by using the feature heat map and the target convolutional step length to obtain a line fault detection model. The detection model training method can obtain a more accurate line fault detection model while simplifying training data.

Description

Technical Field

[0001] This application belongs to the technical field of line fault analysis, and relates to a method for training a detection model, in particular to a method and system for training a detection model, a line detection method, a device and a medium. Background Art

[0002] In recent years, transformer winding deformation faults account for 30% of the total number of its faults. Frequency Response Analysis (FRA) has become the mainstream detection method due to its high sensitivity and repeatability. However, there are still significant limitations in the existing technology: Firstly, there is a lack of a standardized method for interpreting FRA results, mainly relying on manual visual inspection or mathematical calculation. The former is easily interfered by subjective factors, and the latter is difficult to accurately identify the type and degree of faults. Secondly, there are defects in efficiency and practicality in the existing improvement methods. For example, the patented technology based on the centroid analysis of three-dimensional frequency response curves needs to calculate a large number of centroid coordinates and establish a three-dimensional distribution model, which is a complex process and has a high calculation cost; while the method combining FRA and empirical mode decomposition introduces energy change information, but its diagnostic effect overly relies on the frequency band intervals divided manually, which is prone to result in deviations.

[0003] In summary, although artificial intelligence shows potential in the field of fault detection, the traditional methods have insufficient generalization ability for complex line faults and are difficult to adapt to the diverse fault scenarios in the actual operation of transformers. Summary of the Invention

[0004] In view of the above-mentioned disadvantages of the prior art, the purpose of this application is to provide a method and system for training a detection model, a line detection method, a device and a medium, which are used to solve the problem of the lack of a line fault detection model with high accuracy in the existing technology.

[0005] In the first aspect, this application provides a method for training a detection model. The method for training a detection model includes: obtaining frequency analysis data of a transformer; processing the frequency analysis data by using sliding correlation to obtain a feature heat map of the frequency analysis data; training the convolution kernel of an initial convolutional neural network model by using an algorithm to obtain a target convolution step length of the initial convolutional neural network model; and training the initial convolutional neural network by using the feature heat map and the target convolution step length to obtain a line fault detection model.

[0006] In this application, sliding correlation processing is used to expand the frequency analysis data to obtain a feature heat map. The initial convolutional neural network is preprocessed to obtain the target convolutional stride, and the initial convolutional neural network is trained using the feature heat map and the target convolutional compensation to obtain a line fault detection model. This method of training the detection model breaks through the limitations of manual experience and mathematical simplification through automated feature learning and multi-source information fusion, significantly improving the detection accuracy and efficiency, and providing a more reliable and general diagnostic solution for transformer winding deformation faults.

[0007] In one implementation of the first aspect, obtaining the frequency analysis data of the transformer includes: simulating and experimentally processing the transformer to obtain the frequency analysis data during various faults of the transformer, and the frequency analysis data includes simulation data and experimental data.

[0008] In one implementation of the first aspect, using sliding correlation to process the frequency analysis data to obtain the feature heat map of the frequency analysis data includes: using the dilation algorithm to perform dilation processing on the frequency analysis data to obtain the dilated frequency analysis data; using sliding correlation to perform correlation calculation on the dilated frequency analysis data to obtain the correlation coefficient; generating the feature heat map of the frequency analysis data according to the correlation coefficients of the entire frequency band.

[0009] In one implementation of the first aspect, using an algorithm to train the convolutional kernel of the initial convolutional neural network model to obtain the target convolutional stride of the initial convolutional neural network model includes: using the Sobol sequence to optimize the initial population of the pelican optimization algorithm to obtain the improved pelican optimization algorithm, and obtaining the population size and the maximum number of iterations; obtaining the target optimization path according to the maximum number of iterations and the generated random number; using the target optimization path to calculate the target convolutional kernel size to obtain the target convolutional stride of the initial convolutional neural network model.

[0010] In one implementation of the first aspect, the method for training the detection model further includes: based on the optimization algorithm, updating the target position according to the maximum number of iterations and the generated random number to obtain the target optimization path.

[0011] In one implementation of the first aspect, obtaining the line fault detection model further includes: using a loss function to obtain the loss value of the initial convolutional neural network model; determining whether the loss value meets a preset threshold, if it meets, obtaining the line fault detection model, if it does not meet, updating the weight parameters of the initial convolutional neural network model according to the error level of the loss value.

[0012] Second aspect, the present application provides a detection model training system. The detection model training system includes: a data acquisition module for acquiring frequency analysis data of a transformer; a data processing module for processing the frequency analysis data by using sliding correlation to obtain a feature heatmap of the frequency analysis data; a model optimization module for training a convolution kernel of an initial convolutional neural network model by using an algorithm to obtain a target convolution stride of the initial convolutional neural network model; and a model training module for training the initial convolutional neural network by using the feature heatmap and the target convolution stride to obtain a line fault detection model.

[0013] Third aspect, the present application provides a line detection method. The line detection method includes: acquiring to-be-detected frequency analysis data; processing the to-be-detected frequency analysis data by using sliding correlation to obtain a feature heatmap of the to-be-detected frequency analysis data; detecting and processing the feature heatmap by using a line fault detection model to obtain a line fault detection result; and the line fault detection model is trained by using the detection model training system according to any one of the first aspect.

[0014] Fourth aspect, the present application provides an electronic device. The electronic device includes: a memory for storing a computer program; and a processor for executing the computer program stored in the memory, so that the electronic device executes the detection model training method according to any one of the first aspect and / or the line detection method according to the third aspect.

[0015] Fifth aspect, the present application provides a computer-readable storage medium. A computer program is stored on the computer-readable storage medium, and when the program is executed by a processor, it implements the detection model training method according to any one of the first aspect and / or the line detection method according to the third aspect. Description of the Drawings

[0016] Figure 1A It shows a schematic diagram of an application scenario of the detection model training method described in the present application.

[0017] Figure 1B It shows a schematic diagram of the structure of the end-cloud interaction scenario in these implementation manners.

[0018] Figure 2 It shows a schematic flowchart of the detection model training method described in an embodiment of the present application.

[0019] Figure 3 It shows a schematic flowchart of the detection model training method described in an embodiment of the present application.

[0020] Figure 4 It shows a schematic flowchart of the detection model training method described in an embodiment of the present application.

[0021] Figure 5 It shows a schematic diagram of the detection model training method described in the embodiments of the present application.

[0022] Figure 6 It shows a schematic structural diagram of the detection model training system described in the embodiments of the present application.

[0023] Figure 7 It shows a schematic flow diagram of the line detection method described in the embodiments of the present application.

[0024] Figure 8 It shows a schematic structural diagram of the electronic device described in the embodiments of the present application.

[0025] Description of component labels

[0026] 1 Line detection device

[0027] 11 Transformer circuit board

[0028] 12 Local processor

[0029] 13 Display terminal

[0030] 2 Terminal-cloud interaction system

[0031] 20 Terminal

[0032] 21 Cloud server

[0033] 100 Detection model training system

[0034] 110 Data acquisition module

[0035] 120 Data processing module

[0036] 130 Model optimization module

[0037] 140 Model training module

[0038] 800 Electronic device

[0039] 810 Memory

[0040] 820 Processor

[0041] 830 Display

[0042] Steps S11 to S14

[0043] Steps S121 to S123

[0044] Steps S131 to S133

[0045] Steps S21 to S23 DETAILED DESCRIPTION

[0046] The following describes the embodiments of the present application through specific examples, and those skilled in the art can easily understand other advantages and effects of the present application from the contents disclosed in this specification. The present application can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present application. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0047] It should be noted that in the embodiments of the present application, the words "optionally" or "for example" represent examples, illustrations or descriptions. Any embodiment or design described as "optionally" or "for example" in the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "optionally" or "for example" is intended to present related concepts in a specific way.

[0048] In the embodiments of the present application, "at least one" refers to one or more, and "plurality" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone, where A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can represent: a, b, c, ab, ac, bc or abc, where a, b, c can be single or multiple.

[0049] It should be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present application, and thus the drawings only show components related to the present application rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed at will, and the component layout may also be more complicated.

[0050] According to the statistics of each power grid, among the transformer fault types, winding deformation faults account for 30% of the total number of transformer faults. The Frequency Response Analysis (FRA) method has the advantages of good repeatability and high sensitivity in the actual process and is widely used in the test for detecting whether the transformer winding is deformed. So far, there is still no standard and reliable method to interpret FRA. The analysis of FRA mainly relies on visual inspection or mathematical calculation. The diagnostic results of visual inspection are easily affected by the subjectivity of personnel, and it is not easy to identify the winding fault type and degree through simple mathematical calculation. One implementation method is to calculate the centroid coordinates represented by frequency, amplitude, and phase, visualize the obtained coordinates, obtain the centroid distribution of each three-dimensional frequency response curve, establish a three-dimensional coordinate system, and the centroid distributions of different fault three-dimensional frequency response curves are around it. The fault type of the winding is determined according to the different centroid distribution intervals. This method requires a large amount of calculation and the diagnostic process is cumbersome. In recent years, using artificial intelligence for transformer winding fault detection has become a development trend. Another implementation method is to combine the Frequency Response Analysis (FRA) and the Empirical Mode Decomposition (EMD) method for the diagnosis of winding deformation faults. This method focuses on using the energy change information in each interval during winding deformation faults, and the division of the FRA interval has a great influence on the diagnostic results.

[0051] At least for the above problems, the embodiment of the present application provides a detection model training method. The detection model training method includes: obtaining frequency analysis data of a transformer; processing the frequency analysis data by using sliding correlation to obtain a feature heat map of the frequency analysis data; training the convolution kernel of an initial convolutional neural network model by using an algorithm to obtain a target convolution step length of the initial convolutional neural network model; and training the initial convolutional neural network by using the feature heat map and the target convolution step length to obtain a line fault detection model.

[0052] In the embodiment of the present application, the frequency analysis data is expanded by using sliding correlation to obtain a feature heat map. The initial convolutional neural network is preprocessed to obtain a target convolution step length, and the initial convolutional neural network is trained by using the feature heat map and the target convolution compensation to obtain a line fault detection model. This detection model training method breaks through the limitations of artificial experience and mathematical simplification through automated feature learning and multi-source information fusion, significantly improves the detection accuracy and efficiency, and provides a more reliable and universal diagnostic solution for transformer winding deformation faults.

[0053] Figure 1A Shown is a schematic diagram of an application scenario of the detection model training method described in the present application. The line detection device 1 can be used to implement the detection model training method provided by the embodiment of the present application, but the application scenario of the detection model training method provided by the embodiment of the present application is not limited toFigure 1A The line detection device 1 shown in the figure. As Figure 1A shown, the line detection device 1 includes a transformer circuit board 11, a local processor 12, and a display terminal 13. The detection model training method provided by the embodiments of the present application can be applied to the local processor 12.

[0054] Among them, Figure 1A the local processor 12 in can be a single local processor or a local processor cluster or a cloud computing center composed of multiple local processors, etc., and are not specifically limited here. Although Figure 1A only one transformer circuit board 11, one local processor 12, and one display terminal 13 are shown in, it should be understood that Figure 1A the examples in are only for understanding the solution, and the specific numbers of the local processor 12 and the display terminal 13 should be flexibly determined according to the actual situation.

[0055] In some other implementation manners, the line detection device 1 may also not include the display terminal 13, but only include the local processor 12 with a display function and the transformer circuit board 11. The detection model training method provided by the embodiments of the present application can be applied to the local processor 12. The local processor 12 with a display function may include a tablet computer, a laptop computer, a palm computer, a mobile phone, a personal computer, a voice interaction device, etc., and are not limited here.

[0056] In still some other implementation manners, the detection model training method described in the present application can be applied to the end-cloud interaction scenario. Figure 1B Shown is a schematic structural diagram of the end-cloud interaction scenario in these implementation manners. As Figure 1B shown, the end-cloud interaction system 2 includes a terminal 20 and a cloud server 21, and communication can be performed between the terminal 20 and the cloud server 21, and the communication method is not limited to wired or wireless methods.

[0057] Among them, the terminal 20 can be mobile or fixed. For example, the terminal 20 can be a wireless terminal or a wired terminal. The wireless terminal can refer to a device with wireless transceiver functions and can be deployed indoors, outdoors, and in industrial workshops. The terminal 20 can be a mobile phone, a tablet computer, a laptop computer, etc., and are not limited here. The cloud server 21 can include one or more servers, or include one or more processing nodes, or include one or more virtual machines running on the server. The cloud server 21 can also be referred to as a server cluster, a management platform, a data processing center, etc., and are not limited in the embodiments of the present application.

[0058] Next, the technical solutions in the embodiments of the present application will be described in detail with reference to the accompanying drawings in the embodiments of the present application.

[0059] The following embodiments of this application provide a detection model training method, which can be implemented, for example, by Figure 1A the local processor 12 shown in Figure 1B or the cloud server 21 shown in Figure 2 FIG. shows a schematic flowchart of the detection model training method described in the embodiments of this application. As Figure 2 shown, the detection model training method includes steps S11 to S14.

[0060] Step S11, obtaining frequency analysis data of the transformer.

[0061] Step S12, using sliding correlation to process the frequency analysis data to obtain a feature heat map of the frequency analysis data. The feature heat map is a heat map that uses colors to distinguish correlation coefficients.

[0062] Step S13, using an algorithm to train the convolutional kernels of the initial convolutional neural network model to obtain the target convolutional step size of the initial convolutional neural network model.

[0063] Step S14, using the feature heat map and the target convolutional step size to train the initial convolutional neural network to obtain a line fault detection model. The initial population of the POA population algorithm is optimized by combining the Sobol sequence, and the convolutional kernels of the initial convolutional neural network CNN (Convolutional Neural Network) model are optimized using IPOA to construct a line fault detection model of IPOA-CNN. The line fault detection model can be used for diagnosing transformer winding deformation faults.

[0064] In some possible implementation manners, in the scenario of transformer winding deformation faults, the transformer winding deformation faults include radial deformation RD, disk space variation DSV, and short circuit SC. The fault type of transformer winding deformation can be obtained based on the frequency analysis data. Obtain the frequency analysis data of the transformer, use sliding correlation to process the frequency analysis data, dynamically capture the local correlation between the data points of adjacent frequency analysis data, and obtain a feature heat map of the frequency analysis data. Use an optimization algorithm to train the convolutional kernels of the initial convolutional neural network CNN model to obtain the target convolutional step size of the CNN model. Use the feature heat map and the target convolutional step size to train the CNN model to obtain a line fault detection model.

[0065] In the embodiments of the present application, sliding correlation processing is used to expand the frequency analysis data to obtain a feature heat map. The initial convolutional neural network is preprocessed to obtain a target convolutional stride, and the initial convolutional neural network is trained using the feature heat map and the target convolutional compensation to obtain a line fault detection model. This detection model training method breaks through the limitations of manual experience and mathematical simplification through automated feature learning and multi-source information fusion, significantly improving the detection accuracy and efficiency, and providing a more reliable and universal diagnostic solution for transformer winding deformation faults.

[0066] In one embodiment of the present application, obtaining the frequency analysis data of the transformer includes: simulating and experimentally processing the transformer to obtain the frequency analysis data when the transformer has various faults, and the frequency analysis data includes simulation data and experimental data.

[0067] In some possible implementation manners, a simulation and experimental data set is established for model training. A transformer is simulated and experiments are carried out on an actual transformer. Assume that the simulated transformer is T1 and the experimental transformer is T2. Transformer T1 is an 11.55 / 0.412 KV, 10 kVA transformer, the high-voltage winding consists of 6 discs, each disc is a 1134-turn coil, the low-voltage winding consists of 140 continuous layers, and the parameters of transformer T1 are shown in Table 1.

[0068] Table 1: Electrical parameters of T1

[0069]

[0070] Among them, C g is the winding-to-ground capacitance, C H-L is the capacitor between the high-voltage winding and the low-voltage winding, K H-L is the coupling coefficient between the inductances of the high-voltage winding and the low-voltage winding, K ij is the coupling coefficient of the unilateral winding inductance, R S 、L S and C S are the resistance, inductance and capacitance of the winding respectively, and G is the resistance between the winding and the ground and the resistance between the high-voltage winding and the low-voltage winding.

[0071] Simulate transformer T1, input the scanned voltage information at one end of the high-voltage side of the equivalent circuit, and the voltage at the other end of the high-voltage side is used as the output signal. The simulated faults include disk space variation DSV, radial deformation RD and inter-disk short circuit SC. The DSV fault is simulated using C H-L 、M H-L 、M H and M L simulation, M H-L is the mutual inductance between the high-voltage winding and the low-voltage winding, M H is the self-inductance of the high-voltage winding, ML is the self - inductance of the low - voltage winding. The RD fault is simulated using C H-L 、C g 、M H-L and M H The fault degree is set in the range of 1% - 40%. The SC fault is simulated by connecting two adjacent disks with wires.

[0072] In collecting the frequency - analysis data of the actual transformer, a 10kV / 0.4kV, 400kVA model transformer (T2) is adopted. The detailed parameters of T2 are shown in Table 2.

[0073] Table 2: Parameters of T2

[0074]

[0075] Experiments are carried out on transformer T2. The DSV fault is simulated by connecting capacitors in parallel with some disks, and the fault degree changes with the change of the parallel - capacitor value, including 50pF, 67pF, 100pF, 200pF, 400pF, 600pF and 800pF. The RD fault is simulated by replacing normal disks with deformed disks. The SC fault is simulated by connecting adjacent disks with wires. The simulation data sets and experimental data sets composed of simulation data and experimental data are shown in Table 3.

[0076] Table 3: Simulation data sets and experimental data sets

[0077]

[0078] Figure 3 It is shown as the flow - schematic diagram of the detection - model training method described in the embodiment of the present application. As Figure 3 shown, the step S12 includes steps S121 to S123.

[0079] Step S121, using the dilation algorithm to perform dilation processing on the frequency - analysis data to obtain the dilated frequency - analysis data.

[0080] Step S122, using sliding - correlation processing to perform correlation calculation on the dilated frequency - analysis data to obtain the correlation coefficient.

[0081] Step S123, generating the characteristic heat - map of the frequency - analysis data according to the correlation coefficient of the entire frequency band.

[0082] In some possible implementation manners, preprocessing and visualization processing are performed on the frequency analysis data. There is a problem of data imbalance of different fault types during the winding deformation fault simulation. Taking the insufficient SC data as an example, the SMOTE algorithm is used to expand the SC data by 3 times, and the distribution of the expanded frequency analysis data is the same as that before expansion. The sliding correlation processing is used to perform correlation calculation on the expanded frequency analysis data to obtain the correlation coefficient. The correlation coefficient (Indicator-correlation Coefficient, abbreviated as cc) is the statistical index correlation coefficient, which is used to measure the similarity of two curves. The calculation formula of the correlation coefficient is:

[0083]

[0084] where X i and Y i are two different frequency analysis data sequences respectively, l is the length of the frequency analysis data for the sliding correlation processing, s is the sliding step, i is the starting point of the frequency analysis data, and the initial value is 1.

[0085] After completing one correlation calculation, i is successively added with s step lengths, and the loop ends until l = 1000 - l. The obtained sequence consists of the correlation coefficients of the normal winding frequency analysis data and the fault winding frequency analysis data in different frequency bands. For example, the length of the frequency analysis data is 1000. Starting from the starting position 0Hz of the frequency analysis data, the frequency analysis data with length l is selected for correlation calculation, and after translating it by s units, the frequency analysis data with length l is selected for correlation calculation, and the loop is repeated until the end. The length of the processed frequency analysis data sequence is:

[0086]

[0087] Table 4: Evaluation indexes when l and s are different

[0088]

[0089] where h is correlated with l and s. When s is fixed and l is relatively small, the amount of data for one correlation calculation will decrease, the statistical data of the correlation coefficient will be less, and the correlation coefficient will be closer to 1. When s is fixed and l is relatively large, the amount of data for correlation coefficient calculation increases, h decreases, and the fault features that can be extracted decrease. When l remains unchanged and s increases, h decreases, and the fault features that can be extracted decrease. As can be seen from Table 4, when s = 1 and l = 100, the accuracy rate is the highest and the loss is the smallest.

[0090] In the embodiments of the present application, a method for calculating the correlation coefficient of a sliding window is used to generate a feature heat map by dynamically capturing the local correlation between adjacent frequency analysis data points. It can comprehensively reflect the global impact of different fault types on the frequency analysis data curve and avoid the loss of feature information. Experiments show that the sliding correlation processing increases the model accuracy by 12% compared with the traditional method, reaching 98%. To address the problem of unbalanced transformer winding fault data, the oversampling technique SMOTE is introduced to generate minority class samples by interpolation, expanding the data by 3 times. After SMOTE processing, the recognition accuracy of the model for insufficient faults reaches 100%, which is significantly improved compared with the expanded data, effectively solving the overfitting problem caused by insufficient data.

[0091] Figure 4 It is shown as the schematic flow chart of the detection model training method described in the embodiments of the present application. As Figure 4 shown, the step S13 includes the following steps S131 to S133.

[0092] Step S131, using the Sobol sequence to optimize the initial population of the pelican optimization algorithm to generate an improved pelican optimization algorithm, and obtaining the population size and the maximum number of iterations.

[0093] Step S132, obtaining the target optimization path according to the maximum number of iterations and the generated random number.

[0094] Step S133, using the target optimization path to calculate the target convolution kernel size to obtain the target convolution stride of the initial convolutional neural network model.

[0095] In some possible implementation manners, the iterative Pareto optimization algorithm (IPOA) is used to train the convolution kernel of the initial convolutional neural network to obtain the target convolution stride. The Sobol sequence is used to optimize the initial population of the pelican optimization algorithm to generate an improved pelican optimization algorithm, and the population size and the maximum number of iterations are obtained. In the exploration stage of the IPOA algorithm, it is determined that the prey position moves to this determined area. The calculation formula is:

[0096]

[0097] where is the updated j -th dimension position of the i -th pelican in the exploration stage, rand is a random number between [0, 1], I is a random integer of 1 or 2, P j is the j -th dimension position of the prey, F P is the objective function value of the prey, F i is the fitness function value, and P1 is the initial solution generated by the Sobol sequence.

[0098] In the IPOA algorithm, if the objective function value of this part is improved, the new position of the pelican is accepted. The calculation formula is:

[0099]

[0100] Among them, X i is the new position of the i-th pelican, is the unique part of the i-th pelican, is the objective function value of the new position of the i-th pelican updated based on the first stage, F i is the fitness function value.

[0101] The mathematical modeling formula in the development stage of the IPOA algorithm is:

[0102]

[0103] Among them, is the j-th position of the i-th pelican during the second stage update, rand is a random number between [0, 1], R is a random integer of 0 or 2, t is the number of current iterations, and T is the maximum number of iterations. The objective function value of the updated position will be updated at the end of the development stage. The calculation formula is:

[0104]

[0105] Among them, X i is the new position of the i-th pelican, is the j-th position of the i-th pelican during the second stage update, is the objective function value of the new position of the i-th pelican updated based on the first stage, F i is the fitness function value.

[0106] Obtain the target optimization path according to the maximum number of iterations and the generated random number. Calculate the target convolution kernel size using the target optimization path to obtain the target convolution stride of the initial convolutional neural network model.

[0107] In an embodiment of the present application, the detection model training method further includes: based on an optimization algorithm, update the target position according to the maximum number of iterations and the generated random number to obtain the target optimization path.

[0108] In some possible implementation manners, the optimization algorithm is the IPOA optimization algorithm. The distribution of the initial solutions of the POA swarm intelligence algorithm in the space will significantly affect the convergence speed and optimization accuracy of the algorithm. A uniform distribution of the initial solutions helps to improve the performance of the algorithm. The standard POA algorithm uses random numbers to initialize the population, and the distribution of the initial population is random. In the present invention, the Sobol sequence is used to initialize the population to obtain the IPOA. Under the same population size, maximum number of iterations, and number of runs, the fitness value of the IPOA is the smallest and the error is the smallest. Compared with other algorithms, the fitness of the IPOA is optimal.

[0109] In the embodiments of the present application, the pelican optimization algorithm IPOA is used to dynamically optimize the convolution stride, and the Sobol sequence is used to initialize the population. The IPOA iteratively searches in the exploration and development stages to obtain the target convolution compensation, significantly reducing the number of parameters and calculation time of the convolution operation. The CNN model optimized by the IPOA has a lower loss value than traditional algorithms such as POA (Particle Swarm Optimization) and PSO (Particle Swarm Optimization) under the same data set, and the diagnostic accuracy is significantly better than that of ordinary CNN and BP (Backpropagation Neural Network) networks.

[0110] In an embodiment of the present application, obtaining the line fault detection model further includes: obtaining the loss value of the initial convolutional neural network model by using a loss function; determining whether the loss value meets a preset threshold. If it meets, the line fault detection model is obtained. If it does not meet, the weight parameters of the initial convolutional neural network model are updated according to the error level of the loss value.

[0111] In some possible implementation manners, Figure 5 is shown as a schematic diagram of the detection model training method described in the embodiments of the present application. As Figure 5As shown, obtain the frequency analysis data of the transformer. After dilating and expanding the insufficient data of the frequency analysis data, use sliding correlation to process the dilated frequency analysis data, and convert the obtained correlation coefficient into a characteristic heat map distinguished by colors. Use the algorithm to train with the initial neural network to obtain the target convolution kernel size of the CNN model. Use the Sobol sequence to initialize the population to optimize the POA algorithm, determine the iteration number parameter, update the parameters according to the randomly generated position of the prey, and continuously iterate until the optimal target optimization path is obtained. Use the target optimization path to determine the target convolution step size of the initial convolutional neural network to obtain the optimized IPOA algorithm. After processing through the convolutional layer, pooling layer, and fully connected layer, calculate the output loss function, and determine whether the loss value meets the preset threshold. If it meets, obtain the line fault detection model; if not, update the weight parameters of the initial convolutional neural network model according to the error level of the loss value.

[0112] Figure 6 It is shown as the structural schematic diagram of the detection model training system described in the embodiment of the present application. As Figure 6 shown, the detection model training system 100 includes a data acquisition module 110, a data processing module 120, a model optimization module 130, and a model training module 140.

[0113] The data acquisition module 110 is used to obtain the frequency analysis data of the transformer.

[0114] The data processing module 120 is used to process the frequency analysis data by using sliding correlation to obtain the characteristic heat map of the frequency analysis data.

[0115] The model optimization module 130 is used to train the convolution kernel of the initial convolutional neural network model by using an algorithm to obtain the target convolution step size of the initial convolutional neural network model.

[0116] The model training module 140 is used to train the initial convolutional neural network by using the characteristic heat map and the target convolution step size to obtain a line fault detection model.

[0117] In some possible implementation manners, in a transformer winding deformation fault scenario, the transformer winding deformation fault includes radial deformation RD, disk space variation DSV, and short circuit SC. The fault type of the transformer winding deformation can be obtained according to the frequency analysis data. Obtain the frequency analysis data of the transformer, and use sliding correlation to process the frequency analysis data to dynamically capture the local correlation between the data points of adjacent frequency analysis data, so as to obtain the characteristic heat map of the frequency analysis data. Use an optimization algorithm to train the convolution kernel of the initial convolutional neural network model CNN model to obtain the target convolution step size of the CNN model. Use the characteristic heat map and the target convolution step size to train the CNN model to obtain a line fault detection model.

[0118] In the embodiments of the present application, sliding correlation processing is used to expand the frequency analysis data to obtain a characteristic heat map. The initial convolutional neural network is preprocessed to obtain the target convolution step size, and the initial convolutional neural network is trained using the characteristic heat map and the target convolution compensation to obtain a line fault detection model. This method for training the detection model breaks through the limitations of manual experience and mathematical simplification through automated feature learning and multi-source information fusion, significantly improving the detection accuracy and efficiency, and providing a more reliable and general diagnostic solution for transformer winding deformation faults.

[0119] Figure 7 It shows a schematic flowchart of the line detection method described in the embodiments of the present application. As Figure 7 shown, the line detection method includes the following steps S21 to S23.

[0120] Step S21, obtain the frequency analysis data to be measured.

[0121] Step S22, use sliding correlation to process the frequency analysis data to be measured to obtain the characteristic heat map of the frequency analysis data to be measured.

[0122] Step S23, use the line fault detection model to perform detection processing on the characteristic heat map to obtain a line fault detection result, and the line fault detection model is trained using the detection model training system described in any embodiment of the present application.

[0123] In the embodiments of the present application, a sliding correlation processing is combined with an IPOA-optimized CNN model to construct an end-to-end line fault detection model. The model realizes accurate classification of fault types and effective distinction of fault degrees through heat map input, IPOA dynamic optimization of convolution step size, and feature dimensionality reduction of the full pooling layer. Compared with other models, it has stronger generalization ability, and the comprehensive accuracy rate is increased to 91%.

[0124] In several embodiments provided by the present application, it should be understood that the disclosed system, device or method can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of modules / units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple modules or units can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection to each other can be through some interfaces, and the indirect coupling or communication connection of devices or modules or units can be in electrical, mechanical or other forms.

[0125] The modules / units described as separate components may or may not be physically separated, and the components displayed as modules / units may or may not be physical modules, that is, they can be located in one place, or distributed to multiple network units. Some or all of the modules / units can be selected according to actual needs to achieve the purpose of the embodiments of the present application. For example, in each embodiment of the present application, the functional modules / units can be integrated in a processing module, or each module / unit can exist physically alone, or two or more modules / units can be integrated in one module / unit.

[0126] Those of ordinary skill in the art should further realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described according to functions in the above description. Whether these functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present application.

[0127] The embodiments of the present application also provide an electronic device. Figure 8 The structure diagram of the electronic device 800 described in the embodiments of the present application is shown. As Figure 8 shown, in this embodiment, the electronic device 800 includes a memory 810 and a processor 820.

[0128] The memory 810 is used to store computer programs; preferably, the memory 810 includes various media that can store program codes, such as ROM, RAM, magnetic disks, USB flash drives, memory cards, or optical discs.

[0129] Specifically, the memory 810 may include a computer system readable medium in the form of volatile memory, such as random access memory (RAM) and / or cache memory. The electronic device 800 may further include other removable / non-removable, volatile / non-volatile computer system storage media. The memory 810 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of the embodiments of the present application. It can be understood that the memory 810 may be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. Among them, the non-volatile memory may be read-only memory (ROM, Read Only Memory), programmable read-only memory (PROM, Programmable Read-Only Memory), which is used as an external cache. By way of example but not limitation, many forms of RAM are available, such as static random access memory (SRAM, Static Random Access Memory), synchronous static random access memory (SSRAM, Synchronous Static Random Access Memory). The memory described in the embodiments of the present invention is intended to include but not limited to these and any other suitable categories of memory.

[0130] The processor 820 is connected to the memory 810 and is configured to execute the computer program stored in the memory 810, so that the electronic device 800 executes the detection model training method described in any embodiment of the present application and / or the line detection method described in the embodiments of the present application.

[0131] Optionally, the processor 820 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0132] Optionally, the electronic device 800 in this embodiment may further include a display 830. The display 830 is communicatively connected to the memory 810 and the processor 820, and is configured to display a graphical user interface (GUI) interaction interface related to the detection model training method described in the embodiments of the present application and / or the circuit detection method described in other embodiments of the present application.

[0133] An embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the detection model training method described in any embodiment of the present application and / or the circuit detection method described in other embodiments of the present application.

[0134] The terms "component", "module", "system", etc. used in this specification are used to represent computer-related entities, hardware, firmware, combinations of hardware and software, software, or software in execution. For example, a component may be, but is not limited to, a process running on a processor, a processor, an object, an executable file, an execution thread, a program, and / or a computer. By way of illustration, an application running on a computing device and the computing device can both be components. One or more components may reside in a process and / or an execution thread, and a component may be located on one computer and / or distributed between two or more computers. In addition, these components may execute from various computer-readable media storing various data structures. A component may, for example, communicate through local and / or remote processes according to a signal having one or more data packets (such as data from two components interacting with another component in a local system, a distributed system, and / or a network, such as the Internet interacting with other systems through a signal).

[0135] The descriptions of the processes or structures corresponding to the above respective drawings each have their own focuses. For parts not detailed in a certain process or structure, reference may be made to the relevant descriptions of other processes or structures.

[0136] The above embodiments are only illustrative of the principles and effects of the present application, and are not used to limit the present application. Any person familiar with this technology can modify or change the above embodiments without departing from the spirit and scope of the present application. Therefore, all equivalent modifications or changes completed by those with ordinary knowledge in the technical field without departing from the spirit and technical idea disclosed by the present application should still be covered by the claims of the present application.

Claims

1. A method for training a detection model, characterized in that Including: Obtain the frequency analysis data of the transformer; Use sliding correlation to process the frequency analysis data to obtain the characteristic heat map of the frequency analysis data; Use an algorithm to train the convolution kernels of the initial convolutional neural network model to obtain the target convolution step size of the initial convolutional neural network model; Use the characteristic heat map and the target convolution step size to train the initial convolutional neural network to obtain a line fault detection model.

2. The detection model training method according to claim 1, wherein Obtaining the frequency analysis data of the transformer includes: Perform simulation and experimental processing on the transformer to obtain the frequency analysis data during various faults of the transformer, and the frequency analysis data includes simulation data and experimental data.

3. The detection model training method according to claim 1, wherein Using sliding correlation to process the frequency analysis data to obtain the characteristic heat map of the frequency analysis data includes: Use the dilation algorithm to perform dilation processing on the frequency analysis data to obtain the dilated frequency analysis data; Use sliding correlation to perform correlation calculations on the dilated frequency analysis data to obtain the correlation coefficient; Generate the characteristic heat map of the frequency analysis data according to the correlation coefficients of the entire frequency band.

4. The detection model training method according to claim 1, characterized in that Using an algorithm to train the convolution kernels of the initial convolutional neural network model to obtain the target convolution step size of the initial convolutional neural network model includes: Adopt the Sobol sequence to optimize the initial population of the pelican optimization algorithm to obtain an improved pelican optimization algorithm, and obtain the population size and the maximum number of iterations; Obtain the target optimization path according to the maximum number of iterations and the generated random number; Use the target optimization path to calculate the target convolution kernel size to obtain the target convolution step size of the initial convolutional neural network model.

5. The detection model training method according to claim 4, wherein Also included: Based on the optimization algorithm, update the target position according to the maximum number of iterations and the generated random number to obtain the target optimization path.

6. The detection model training method according to claim 1, characterized in that Obtaining the line fault detection model also includes: Use the loss function to obtain the loss value of the initial convolutional neural network model; Judge whether the loss value meets the preset threshold. If it meets, obtain the line fault detection model. If it does not meet, update the weight parameters of the initial convolutional neural network model according to the error level of the loss value.

7. A detection model training system, characterized in that, Including: A data acquisition module for obtaining the frequency analysis data of the transformer; A data processing module for using sliding correlation to process the frequency analysis data to obtain the characteristic heat map of the frequency analysis data; A model optimization module for using an algorithm to train the convolution kernels of the initial convolutional neural network model to obtain the target convolution step size of the initial convolutional neural network model; A model training module for using the characteristic heat map and the target convolution step size to train the initial convolutional neural network to obtain a line fault detection model.

8. A line detection method, characterized in that, Including: Obtain the frequency analysis data to be measured; Use sliding correlation to process the frequency analysis data to be measured to obtain the characteristic heat map of the frequency analysis data to be measured; Use the line fault detection model to perform detection processing on the characteristic heat map to obtain a line fault detection result; the line fault detection model is trained by the detection model training system according to any one of claims 1 to 6.

9. An electronic device, characterized in that, Including: A memory for storing a computer program; A processor configured to execute the computer program stored in the memory, so that the electronic device executes the detection model training method according to any one of claims 1 to 6 and / or the circuit detection method according to claim 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the detection model training method according to any one of claims 1 to 6 and / or the circuit detection method according to claim 8.

Images