Transformer winding fault detection method, device, electronic equipment and storage medium

Abstract

The embodiments of the present invention disclose a method, device, electronic device and storage medium for detecting transformer winding faults. The method includes: obtaining current signals at the low-voltage end of a target transformer under a plurality of preset loads respectively, and extracting instantaneous parameters of the current signals; performing multi-feature extraction and feature fusion based on the current signals and the instantaneous parameters of the current signals to obtain a fusion feature map of the current signals of the target transformer; splitting the fusion feature map, calculating the feature differences between the splitting results to obtain feature difference data; and determining whether the winding of the target transformer has a fault based on the magnitude of the feature difference data. Considering that the deformation of the transformer winding will have an instantaneous impact on the current signal during closing, the fault is detected by using the feature differences of the instantaneous parameters caused by the deformation of the transformer winding, the judgment of the early deformation fault of the winding is realized, the pre-maintenance of the coil deformation can be effectively and timely realized, and the stability of the transformer operation can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of transformers, and in particular, to a method, device, electronic device, and storage medium for detecting transformer winding faults. Background Art

[0002] A transformer is a key device in the power system. Among them, winding faults are one of the common fault types, which directly affect the safe and stable operation of the transformer. During the operation of the transformer, a large amount of heat is generated, and the winding temperature fluctuates with the change of the load. Metal wires (such as copper or aluminum) expand at high temperatures and contract when cooled. Therefore, long-term thermal cycling will cause the expansion and contraction of the winding material, resulting in winding deformation and relaxation.

[0003] When the transformer is switched on, inrush current phenomenon will occur. At this time, the transformer core has not been magnetized yet, and a certain amount of time and energy are required for the magnetization process. Therefore, a large current will be generated instantaneously, usually 5 to 10 times the normal working current. This process will generate a strong electromagnetic force, which acts on the relaxed winding, causing the winding to generate mechanical vibration or displacement, and even local deformation. If the local deformation is serious, it will affect the stability of the subsequent operation of the transformer, and even may cause faults or accidents.

[0004] In the prior art, for the fault detection of transformers, the feature differences between the target data points and the domain data can be analyzed to determine whether there is a fault at that moment, but the feature differences of the operation data caused by the deformation of the transformer winding are not revealed, so it is impossible to effectively promote the early maintenance of the coil deformation. In addition, the fault detection can also be performed by capturing the vibration signal caused by the winding deformation. However, for safety reasons, external devices are usually not allowed to access the inside of the transformer. Although vibration measurement can be performed outside the transformer, its accuracy often cannot meet the expectations.

[0005] Therefore, the existing transformer fault detection technologies are difficult to realize the judgment of early transformer winding deformation faults, which greatly affects the stability of transformer operation. Summary of the Invention

[0006] Based on this, it is necessary to propose a method, device, electronic device, and storage medium for detecting transformer winding faults to solve the problem of difficult to accurately judge early transformer winding deformation faults.

[0007] To achieve the above object, the first aspect of the present application provides a method for detecting transformer winding faults, the method includes:

[0008] Obtain the current signals at the low-voltage end of the target transformer under a plurality of preset loads respectively, and extract the instantaneous parameters of the current signals;

[0009] Perform multi - feature extraction and feature fusion based on the current signal and the instantaneous parameters of the current signal to obtain a fused feature map of the target transformer current signal;

[0010] Split the fused feature map, calculate the feature differences between the split results, and obtain feature difference data;

[0011] Determine whether the windings of the target transformer are faulty based on the magnitude of the feature difference data.

[0012] Further, the performing multi - feature extraction and feature fusion based on the current signal and the instantaneous parameters of the current signal to obtain a fused feature map of the target transformer current signal specifically includes:

[0013] Sort the current signal and the instantaneous parameters of the current signal in the channel direction to form a first feature map, and the first feature map includes at least the instantaneous features of the instantaneous parameters of the current signal;

[0014] Use a feature extraction model to extract features from the first feature map to obtain a second feature map and a third feature map. Among them, the second feature map includes instantaneous features with a change degree higher than a change threshold, and the third feature map includes instantaneous features with a change degree lower than the change threshold;

[0015] Use a feature alignment model to align and splice the second feature map and the third feature map to obtain the fused feature map.

[0016] Further, the feature extraction model includes a first feature network and a second feature network. The first feature network includes a number of cascaded convolutional blocks, and the second feature network includes a number of cascaded deep learning models;

[0017] Then the using a feature extraction model to extract features from the first feature map to obtain a second feature map and a third feature map specifically includes:

[0018] Use the first feature network to extract features from the first feature map and output the second feature map;

[0019] Use the second feature network to extract features from the first feature map and output the third feature map.

[0020] Further, the feature alignment model includes a connected multi - layer perceptron, a loss function block, and an image processing block;

[0021] Then the using a feature alignment model to align and splice the second feature map and the third feature map to obtain a fused feature map includes:

[0022] Perform image processing on the second feature map and the third feature map using the multi-layer perceptron to obtain a first feature map to be processed and a second feature map to be processed with the same image size, where the image size is determined by the preset image height, image width, and number of channels of the fused feature map;

[0023] Use the loss function configured by the loss function block to align the distributions of the first feature map to be processed and the second feature map to be processed, obtaining a first target feature map and a second target feature map;

[0024] Use the image processing block to splice the first target feature map and the second target feature map according to the number of channels to obtain a fused feature map.

[0025] Furthermore, the loss function is determined by the following formula:

[0026]

[0027] In the formula, is the loss function, is the number of channels, is the preset temperature parameter, is the image height, is the image width, is the value of the c th feature point in the i th channel of the first feature map to be processed, is the calculated probability distribution corresponding to the value of the c th feature point in the i th channel of the first feature map to be processed, represents the value of the c th feature point in the i th channel of the second feature map to be processed, represents the calculated probability distribution corresponding to the value of the c th feature point in the i th channel of the second feature map to be processed.

[0028] Furthermore, the splitting of the fused feature map, calculating the feature differences between the splitting results to obtain feature difference data specifically includes:

[0029] Split the fused feature map into a fused feature sequence in the image width direction, where the fused feature sequence contains several sub-features;

[0030] Calculate the mean square error between two adjacent sub-features in the fused feature sequence to obtain feature difference data.

[0031] Further, the extraction of the instantaneous parameters of the current signal specifically includes:

[0032] Performing a Hilbert transform on the current signal to obtain a target current signal;

[0033] Calculating the instantaneous amplitude and instantaneous phase of the current signal according to the current signal and the target current signal;

[0034] Calculating the instantaneous frequency of the current signal according to the instantaneous phase;

[0035] The instantaneous parameters include the instantaneous amplitude, the instantaneous phase, and the instantaneous frequency;

[0036] Among them, the instantaneous amplitude is calculated by the following formula:

[0037]

[0038] In the formula, is the instantaneous amplitude, is the target current signal, is the current signal;

[0039] The instantaneous phase is calculated by the following formula:

[0040]

[0041] In the formula, is the instantaneous phase;

[0042] The instantaneous frequency is calculated by the following formula:

[0043]

[0044] Among them, is the instantaneous frequency.

[0045] To achieve the above object, a second aspect of the present application provides a transformer winding fault detection device, and the device includes: a signal acquisition module, a feature extraction module, and an abnormality determination module;

[0046] The signal acquisition module is used to acquire current signals at the low-voltage end of a target transformer under a plurality of preset loads respectively, and extract the instantaneous parameters of the current signals;

[0047] The feature extraction module is used to perform multi-feature extraction and feature fusion according to the current signal and the instantaneous parameters of the current signal to obtain a fusion feature map of the current signal of the target transformer;

[0048] The abnormal judgment module is used to split the fused feature map, calculate the feature differences between the split results, and obtain feature difference data;

[0049] Based on the magnitude of the feature difference data, it is determined whether the winding of the target transformer has a fault.

[0050] To achieve the above object, a third aspect of the present application provides a computer-readable storage medium storing a computer program, which when executed by a processor causes the processor to execute the steps of the method described in the first aspect.

[0051] To achieve the above object, a fourth aspect of the present application provides a computer device including a memory and a processor, where the memory stores a computer program, and when the computer program is executed by the processor, it causes the processor to execute the steps of the method described in the first aspect.

[0052] Adopting the embodiments of the present invention has the following beneficial effects:

[0053] The embodiments of the present invention propose a method for detecting transformer winding faults. The method includes: obtaining current signals at the low-voltage end of a target transformer under a plurality of preset loads respectively, and extracting instantaneous parameters of the current signals; performing multi-feature extraction and feature fusion based on the current signals and the instantaneous parameters of the current signals to obtain a fused feature map of the target transformer current signal; splitting the fused feature map, calculating the feature differences between the split results, and obtaining feature difference data; based on the magnitude of the feature difference data, it is determined whether the winding of the target transformer has a fault. The present invention takes into account that the deformation of the transformer winding will have an instantaneous impact on the current signal during closing, and thus uses the feature differences of the instantaneous parameters caused by the deformation of the transformer winding for fault detection, realizes the judgment of early deformation faults of the winding, can effectively and timely achieve the early maintenance of the coil deformation, and can effectively improve the stability of the transformer operation. Description of the Drawings

[0054] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0055] Among them:

[0056] Figure 1 Flow chart of the method for detecting transformer winding faults in the embodiments of the present invention;

[0057] Figure 2Schematic diagram of the feature extraction model and the feature alignment model in the embodiments of the present invention;

[0058] Figure 3 Schematic diagram of the convolutional block in the embodiments of the present invention;

[0059] Figure 4 Detailed structure diagram of the feature alignment model in the embodiments of the present invention;

[0060] Figure 5 Structural block diagram of the transformer winding fault detection device in the embodiments of the present invention;

[0061] Figure 6 Internal structure diagram of the computer device in the embodiments of the present invention. Detailed implementation manners

[0062] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without making creative efforts shall fall within the protection scope of the present invention.

[0063] To detect early faults in transformer windings, an embodiment of the present invention proposes a transformer winding fault detection method, which can be referred to Figure 1 , Figure 1 Flow schematic diagram of the transformer winding fault detection method in the embodiments of the present invention. The method includes Step 110 - Step 140.

[0064] Step 110: Obtain current signals at the low-voltage end of the target transformer under a plurality of preset loads, and extract instantaneous parameters of the current signals.

[0065] In the embodiments of the present invention, the target transformer is any one of the transformers to be detected. The target transformer is equipped with a current sensor, such as a Hall effect current sensor, a current transformer and other current sensors, and the current sensor is used to collect the magnitudes of the current signals at the low-voltage end of the target transformer under multiple loads.

[0066] The transformer winding fault detection method of this embodiment can be applied to electronic devices such as processors. Exemplarily, the processor has a plurality of current acquisition interfaces and can simultaneously collect current signals under multiple different loads sent by the current sensor. For example, connect the current acquisition interface to a load with a step span to obtain current signals with step differences. The current signal can be expressed as , is the Nth current signal, and N is the total number of current signals.

[0067] By collecting the current signal at the low-voltage end of the target transformer, it is possible to detect whether the transformer winding is faulty, without the need to externally connect other detection devices, improving portability and at the same time improving the accuracy of detection.

[0068] If the winding of the target transformer is deformed, inrush current phenomenon will occur when the target transformer is switched on. At this time, the characteristics of the current signal change instantaneously. Therefore, by detecting the instantaneous parameters of the current signal and analyzing the changes in the instantaneous parameters, it is possible to judge the early deformation fault of the transformer winding.

[0069] Step 120: Perform multi-feature extraction and feature fusion based on the current signal and the instantaneous parameters of the current signal to obtain the fusion feature map of the current signal of the target transformer.

[0070] In the embodiment of the present invention, feature extraction is performed on the current signal and the instantaneous parameters of the current signal to determine the instantaneous features of the current signal. The instantaneous features at least include features such as non-linear features and non-symmetry features, and the obtained instantaneous features are fused to obtain a fusion feature map for fault analysis based on the fusion feature map. Considering that the deformation of the transformer winding will have a non-linear and non-symmetric impact on the current signal during switching on, so features such as non-linear features and non-symmetry features of the current signal are extracted to utilize the feature changes caused by the deformation of the transformer winding for fault detection, realize the accurate judgment of the early deformation fault of the winding, and can effectively and timely achieve the early maintenance of the coil deformation, then the stability of the transformer operation can be effectively improved.

[0071] Step 130: Split the fusion feature map, calculate the feature differences between the split results, and obtain the feature difference data.

[0072] In the embodiment of the present invention, the fusion feature map is split to obtain multiple elements, that is, the split results. Then, the feature differences between each split result are calculated to obtain the feature difference data between each element. For fault detection based on the feature differences caused by the deformation of the transformer winding, and to realize the accurate judgment of the early deformation fault of the winding.

[0073] Step 140: Determine whether the winding of the target transformer is faulty based on the magnitude of the feature difference data.

[0074] Due to the influence brought by the deformation of the transformer winding, it only appears in the switching-on situation in the early stage. Most of the data in other situations represents the normal operation state. Therefore, it is possible to determine whether the transformer winding is faulty by the magnitude of the difference between the features.

[0075] Specifically, the feature difference data can be compared with a preset abnormal threshold. When it is detected that any feature difference data is greater than the abnormal threshold, it is considered that the difference between the splitting results is relatively obvious, that is, the difference between the normal state and the abnormal state is large, that is, it is determined that the winding of the target transformer has a fault, and a winding fault result is generated accordingly to achieve a fault reminder.

[0076] The present invention takes into account that the deformation of the transformer winding will have a non-linear and asymmetric impact on the current signal when closing the switch. Therefore, the feature difference of the instantaneous parameters caused by the deformation of the transformer winding is used for fault detection, and the judgment of the early deformation fault of the winding is realized, which can effectively and timely achieve the early maintenance of the coil deformation, and then can effectively improve the stability of the transformer operation.

[0077] In an embodiment of the present invention, in step 110, extracting the instantaneous parameters of the current signal specifically includes: performing a Hilbert transform on the current signal to obtain a target current signal; calculating the instantaneous amplitude and instantaneous phase of the current signal according to the current signal and the target current signal; calculating the instantaneous frequency of the current signal according to the instantaneous phase; the instantaneous parameters include the instantaneous amplitude, the instantaneous phase and the instantaneous frequency.

[0078] The Hilbert transform is a linear transform of a real-valued signal, which describes the complex envelope of a signal modulated by a real-valued carrier. Calculating the instantaneous amplitude, instantaneous frequency and instantaneous phase of the current signal based on the target current signal obtained by the Hilbert transform can maintain the original characteristics of the signal, facilitate calculation, improve the efficiency of signal processing, and thus improve the efficiency of transformer fault detection, and achieve timely detection of transformer winding faults.

[0079] Specifically, the target current signal can be calculated by the following formula:

[0080]

[0081] Where represents the target current signal, is the signal length of the current signal, represents the integration variable, represents the time variable, represents the current signal.

[0082] In an embodiment of the present invention, the instantaneous amplitude represents the amplitude size of the current signal at a certain moment, and the instantaneous amplitude can be obtained by calculating the modulus of the target current signal.

[0083] Optionally, the instantaneous amplitude can be calculated by the following formula:

[0084]

[0085] In the formula, is the instantaneous amplitude, is the target current signal, is the current signal.

[0086] In the embodiment of the present invention, the instantaneous phase represents the phase value of the current signal at a certain moment, and the instantaneous phase can be obtained by calculating the phase of the target current signal.

[0087] Optionally, the instantaneous phase is calculated by the following formula:

[0088]

[0089] In the formula, is the instantaneous phase.

[0090] In the embodiment of the present invention, the instantaneous frequency represents the frequency magnitude of the current signal at a certain moment, and the instantaneous frequency can be obtained by differentiating the instantaneous phase.

[0091] Optionally, the instantaneous frequency is calculated by the following formula:

[0092]

[0093] Wherein, is the instantaneous frequency.

[0094] In an embodiment of the present invention, step 120, multi-feature extraction and feature fusion are performed according to the current signal and the instantaneous parameters of the current signal to obtain a fusion feature map of the target transformer current signal, specifically including:

[0095] Step 210, sort the current signal and the instantaneous parameters of the current signal in the channel direction to form a first feature map, and the first feature map at least includes the instantaneous features of the instantaneous parameters of the current signal.

[0096] Specifically, the current signal, the instantaneous amplitude, the instantaneous frequency, and the instantaneous phase are sorted in the channel direction to form a first feature map, and the size of the first feature map is determined by the signal length and the total amount of the current signal. For example, the current signals and their corresponding instantaneous amplitudes, instantaneous frequencies, and instantaneous phases are arranged in the channel direction, and it is assumed that the signal length of each current signal is, and the size of the first feature map formed by all current signals is where, is the total amount of the current signal.

[0097] The first feature map at least includes the instantaneous features of the instantaneous parameters of the current signal, and the instantaneous features at least include non-linear features and non-symmetry features.

[0098] Step 220: Use the feature extraction model to extract features from the first feature map, obtaining a second feature map and a third feature map. The second feature map contains instantaneous features with a change degree higher than the change threshold, and the third feature map contains instantaneous features with a change degree lower than the change threshold.

[0099] In the embodiments of the present invention, the feature extraction model is used to extract features from the first feature map to extract the instantaneous features of instantaneous parameters, such as the first non-linear feature related to the instantaneous amplitude, the second non-linear feature related to the instantaneous frequency, and the asymmetric feature related to the instantaneous phase. Among them, the first non-linear feature and the second non-linear feature are used to indicate the features that enable (such as the instantaneous amplitude, instantaneous frequency, and instantaneous phase) to meet the preset non-linear system conditions, and the asymmetric feature is used to indicate the feature that enables the instantaneous phase to meet the preset asymmetry condition.

[0100] The non-linear feature may refer to the characteristics that conform to the non-linear system in the current signal, reflecting the non-linearity of the current signal. Exemplarily, the first non-linear feature of the instantaneous amplitude may be kurtosis, which reflects the steepness of the amplitude distribution. A high kurtosis value may indicate the existence of abnormal shocks caused by faults; the second non-linear feature of the instantaneous frequency may be the fluctuation of the frequency and the frequency modulation index, and the frequency modulation index reflects the degree of frequency fluctuation. For example, winding faults may cause frequency modulation; the non-linear feature of the instantaneous phase may be phase jitter and phase synchronization index. The phase jitter represents the non-stationarity of the phase change, and the phase synchronization index may be the reduction of the phase synchronization between different sensor signals, reflecting the non-linear coupling failure.

[0101] The asymmetric feature is the feature that enables the current feature data (such as instantaneous amplitude, instantaneous frequency, and instantaneous phase) to meet the asymmetry condition, that is, the asymmetric feature may refer to the characteristics of the asymmetry of the current signal in dimensions such as time and amplitude. Exemplarily, the asymmetric feature of the instantaneous phase may represent the asymmetry in the rising and falling processes of the phase, such as the difference in the phase change rate in different directions (i.e., the skewness of the phase change rate); the asymmetric feature of the instantaneous amplitude may represent the asymmetry of the amplitude distribution (such as skewness) and the difference in the average amplitude of the positive and negative half-cycles (such as the amplitude ratio of the positive and negative half-cycles); the asymmetric feature of the instantaneous frequency may represent the asymmetry of the frequency change in different time periods.

[0102] For the instantaneous amplitude and the instantaneous frequency, obvious winding deformation causes an increase in the non-linear feature. For the instantaneous phase, obvious winding deformation causes an increase in the asymmetric feature. Therefore, in the embodiments of the present invention, by extracting features from the instantaneous amplitude, instantaneous frequency, and instantaneous phase, the first non-linear feature, the second non-linear feature, and the asymmetric feature are obtained, so as to form a fusion feature that can reflect the winding deformation, and then a fusion feature map is generated.

[0103] After extracting the instantaneous features, compare the degree of change of the instantaneous features with the change threshold to generate a second feature map and a third feature map. Among them, the second feature map contains the instantaneous features whose degree of change is higher than the change threshold, and the third feature map contains the instantaneous features whose degree of change is lower than the change threshold. That is to say, the second feature map can contain the features in the first feature map that change significantly (i.e., the degree of change and the change rate are relatively high). For example, the non-linear features and asymmetric features with a relatively high degree of change; the third feature map can contain the features in the first feature map that change insignificantly (i.e., the degree of change and the change rate are relatively low). For example, the non-linear features and asymmetric features with a relatively low degree of change.

[0104] The structure of the feature extraction model can be referred to Figure 2 , Figure 2 is a schematic structural diagram of the feature extraction model and the feature alignment model in the embodiment of the present invention. In the embodiment of the present invention, the feature extraction model 21 includes a first feature network 211 and a second feature network 212. The first feature network 211 includes a plurality of cascaded convolutional blocks, and the second feature network 212 includes a plurality of cascaded deep learning models.

[0105] Then, in Step 220, use the feature extraction model to extract features from the first feature map to obtain a second feature map and a third feature map, specifically including: use the first feature network to extract features from the first feature map and output the second feature map; use the second feature network to extract features from the first feature map and output the third feature map.

[0106] In an embodiment of the present invention, the first feature network 211 includes a plurality of cascaded convolutional blocks, such as convolutional block 1, convolutional block 2, and convolutional block 3. The second feature network 212 includes a plurality of cascaded deep learning models, such as deep learning model 1, deep learning model 2, and deep learning model 3, as Figure 2 shown in the feature extraction model, the first feature network is provided with 3 convolutional blocks, and the second feature network is provided with 3 deep learning models.

[0107] Specifically, the structure of each convolutional block can be referred to Figure 3 , Figure 3The following is a schematic structural diagram of a convolutional block according to an embodiment of the present invention. The convolutional block 300 includes a first convolutional layer 301, an activation function 302, and a second convolutional layer 303 that are connected in sequence. Each convolutional block has two layers of convolution. Therefore, the first feature network 211 composed of convolutional blocks belongs to a shallow convolutional network, which is responsible for extracting significantly changing non-linear features and asymmetric features to form a second feature map. For features with insignificant changes, the convolutional neural network will regard them as noise, resulting in losses. In this embodiment, the second feature network 212 composed of multiple deep learning models is used to extract non-obvious features to form a third feature map. Optionally, the deep learning model can be a transformer block with a self-attention mechanism.

[0108] Therefore, in the embodiment of the present invention, the first feature network 211 composed of multiple convolutional blocks is configured to extract significantly changing non-linear features and asymmetric features, and the second feature network 212 composed of multiple deep learning models is configured to extract non-significantly changing non-linear features and asymmetric features, thereby obtaining a second feature map and a third feature map. Feature fusion is performed based on the second feature map and the third feature map, taking into account the correlation between features, capable of comprehensively capturing features in instantaneous parameters, and improving the accuracy of fault detection.

[0109] Step230: Use the feature alignment model to align and splice the second feature map and the third feature map to obtain a fused feature map.

[0110] In the embodiment of the present invention, the second feature map and the third feature map are input into the feature alignment model, and after feature alignment and splicing, a fused feature map is obtained. At this time, the fused feature map contains first non-linear features, second non-linear features, and asymmetric features, and the fused feature is composed of the first non-linear features, the second non-linear features, and the asymmetric features.

[0111] The structure of the feature alignment model can refer to Figure 4 , Figure 4 which is a detailed structure diagram of the feature alignment model according to an embodiment of the present invention. The feature alignment model 22 includes a multi-layer perceptron 221, a loss function block 222, and an image processing block 223 that are connected.

[0112] Then Step230: Use the feature alignment model to align and splice the second feature map and the third feature map to obtain a fused feature map, includes:

[0113] Step231: Use the multi-layer perceptron to perform image processing on the second feature map and the third feature map to obtain a first feature map to be processed and a second feature map to be processed with the same image size, where the image size is determined by the preset image height, image width, and number of channels of the fused feature map.

[0114] In an embodiment of the present invention, the multi-layer perceptron 221 is a feedforward neural network composed of an input layer, one or more hidden layers, and an output layer. The second feature map and the third feature map are unified in scale through the multi-layer perceptron 221 to obtain a first to-be-processed feature map and a second to-be-processed feature map with the same image size. Wherein, the image size is , represents the number of channels, represents the image height of the fused feature map, represents the image width of the fused feature map.

[0115] Step232: Use the loss function configured by the loss function block to align the distributions of the first to-be-processed feature map and the second to-be-processed feature map to obtain a first target feature map and a second target feature map.

[0116] In an embodiment of the present invention, the loss function is determined by the following formula:

[0117]

[0118] In the formula, is the loss function, is the number of channels, is a preset temperature parameter, is the image height, is the image width, is the value of the c th feature point in the i th channel of the first to-be-processed feature map, is the calculated probability distribution corresponding to the value of the c th feature point in the i th channel of the first to-be-processed feature map, represents the value of the c th feature point in the i th channel of the second to-be-processed feature map, represents the calculated probability distribution corresponding to the value of the c th feature point in the i th channel of the second to-be-processed feature map.

[0119] Step233: Use the image processing block to splice the first target feature map and the second target feature map according to the number of channels to obtain a fused feature map.

[0120] Fuse the first target feature map with obvious changes and the second feature map with less obvious changes to obtain a fused feature map containing rich change features of the current signal for fault analysis.

[0121] In the embodiment of the present invention, the second feature map and the third feature map are feature-aligned and then spliced through a multi-layer perceptron, a loss function block, and an image processing block to obtain a fused feature map, realizing distribution alignment, improving the accuracy of the fused feature map, and at the same time improving the subsequent calculation efficiency.

[0122] In an embodiment of the present invention, in step 130, the fused feature map is split, and the feature differences between the split results are calculated to obtain feature difference data, which specifically includes: splitting the fused feature map into a fused feature sequence in the image width direction, where the fused feature sequence contains several sub-features; calculating the mean square error between two adjacent sub-features in the fused feature sequence to obtain the feature difference data.

[0123] Specifically, in the image width direction , the fused features in the fused feature map are split into a fused feature sequence , and multiple elements in the fused feature sequence are sub-features. Subsequently, the mean square error between two adjacent sub-features in the fused feature sequence is calculated, and all the mean square errors constitute the feature difference data. In order to compare each mean square error in the feature difference data with a preset anomaly threshold, if any mean square error in the detected feature difference data is greater than the anomaly threshold, the sub-feature corresponding to the mean square error greater than the anomaly threshold is used as a fault point. When there is at least one fault point, a winding fault result is generated to achieve a fault reminder.

[0124] The embodiment of the present invention takes into account that the deformation of the transformer winding will have a non-linear and asymmetric impact on the current signal during closing, and uses the error between the instantaneous features generated by the current signal to judge the early deformation fault of the winding, effectively and timely realizing the early maintenance of the coil deformation. At the same time, by calculating the mean square error, the detection accuracy is improved.

[0125] In an embodiment of the present invention, a transformer winding fault detection device is also proposed, which can be referred to Figure 5 , Figure 5 is the structural block diagram of the transformer winding fault detection device according to the embodiment of the present invention. The device includes: a signal acquisition module 501, a feature extraction module 502, and an anomaly judgment module 503.

[0126] The signal acquisition module 501 is used to acquire the current signals at the low-voltage end of the target transformer under a preset plurality of loads and extract the instantaneous parameters of the current signals.

[0127] The feature extraction module 502 is used to perform multi-feature extraction and feature fusion according to the current signal and the instantaneous parameters of the current signal to obtain the fused feature map of the current signal of the target transformer.

[0128] Anomaly judgment module 503 is used to split the fused feature map, calculate the feature differences between the split results, and obtain feature difference data; and determine whether there is a fault in the winding of the target transformer based on the magnitude of the feature difference data.

[0129] The feature extraction module 502 includes:

[0130] A feature sorting unit is used to sort the current signal and the instantaneous parameters of the current signal in the channel direction to form a first feature map, and the first feature map includes at least the instantaneous features of the instantaneous parameters of the current signal.

[0131] A feature extraction unit is used to extract features from the first feature map by using a feature extraction model to obtain a second feature map and a third feature map, where the second feature map includes instantaneous features with a change degree higher than a change threshold, and the third feature map includes instantaneous features with a change degree lower than the change threshold.

[0132] A feature alignment unit is used to align and splice the second feature map and the third feature map by using a feature alignment model to obtain a fused feature map.

[0133] The feature extraction model includes a first feature network and a second feature network. The first feature network includes a number of cascaded convolutional blocks, and the second feature network includes a number of cascaded deep learning models; the feature extraction unit is further used to extract features from the first feature map by using the first feature network and output the second feature map; and extract features from the first feature map by using the second feature network and output the third feature map.

[0134] The feature alignment model includes a connected multi-layer perceptron, a loss function block, and an image processing block; the feature alignment unit is further used to perform image processing on the second feature map and the third feature map by using the multi-layer perceptron to obtain a first to-be-processed feature map and a second to-be-processed feature map with the same image size, where the image size is determined by the preset image height, image width, and number of channels of the fused feature map; use the loss function configured by the loss function block to perform distribution alignment on the first to-be-processed feature map and the second to-be-processed feature map to obtain a first target feature map and a second target feature map; and use the image processing block to splice the first target feature map and the second target feature map according to the number of channels to obtain a fused feature map.

[0135] Wherein, the loss function is determined by the following formula:

[0136]

[0137] In the formula, is the loss function, is the number of channels, is the preset temperature parameter, is the image height, is the image width, in the first feature map to be processed at the c value of the i-th feature point in the channel; c in the i channel of the first feature map to be processed, the calculated probability distribution corresponding to the value of the represents the value of the j-th feature point in the second feature map to be processed at the c channel; i j-th represents the calculated probability distribution corresponding to the value of the j-th feature point in the second feature map to be processed at the c channel. i

[0138] The transformer winding fault detection device proposed in the embodiment of the present invention takes into account that the deformation of the transformer winding will have a non-linear and asymmetric impact on the current signal during closing. Therefore, the characteristic differences of the instantaneous parameters caused by the deformation of the transformer winding are used for fault detection, realizing the judgment of the early deformation fault of the winding, and being able to effectively and timely achieve the early maintenance of the coil deformation, thus effectively improving the stability of the transformer operation.

[0139] The anomaly judgment module 503 is further configured to split the fused feature map into a fused feature sequence in the image width direction, where the fused feature sequence includes a plurality of sub-features; calculate the mean square error between two adjacent sub-features in the fused feature sequence to obtain feature difference data.

[0140] The signal acquisition module 501 is further configured to perform Hilbert transform on the current signal to obtain a target current signal;

[0141] Calculate the instantaneous amplitude and instantaneous phase of the current signal according to the current signal and the target current signal;

[0142] Calculate the instantaneous frequency of the current signal according to the instantaneous phase;

[0143] The instantaneous parameters include instantaneous amplitude, instantaneous phase and instantaneous frequency;

[0144] Among them, the instantaneous amplitude is calculated by the following formula:

[0145]

[0146] In the formula, is the instantaneous amplitude, is the target current signal, is the current signal;

[0147] The instantaneous phase is calculated by the following formula:

[0148] ​

[0149] In the formula, is the instantaneous phase;

[0150] The instantaneous frequency is calculated by the following formula:

[0151]

[0152] where is the instantaneous frequency.

[0153] Figure 6 FIG. shows the internal structure diagram of a computer device in an embodiment of the present invention. The computer device may specifically be a terminal or a system. As Figure 6 shown, the computer device includes a processor, a memory, and a network interface connected through a system bus. Among them, the memory includes a non-volatile storage medium and an internal memory. The non-volatile storage medium of the computer device stores an operating system and may also store a computer program. When the computer program is executed by the processor, the processor can implement each step in the above method embodiment. The internal memory may also store a computer program. When the computer program is executed by the processor, the processor can execute each step in the above method embodiment. Those skilled in the art can understand that Figure 6 the structure shown in

[0154] is only a block diagram of some structures related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0155] In an embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the processor executes each step in the above method embodiment.

[0156] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium used in the various embodiments provided in the present application can include non-volatile and / or volatile memories. Non-volatile memories can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memories can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0157] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0158] The above-described embodiments merely represent several implementation manners of the present application. The description is relatively specific and detailed, but it should not be construed as a limitation on the patent scope of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the appended claims.

Claims

1. A method for detecting transformer winding faults, characterized in that, The method includes: Obtaining current signals at the low-voltage end of the target transformer under a plurality of preset loads respectively, and extracting instantaneous parameters of the current signals; Performing multi-feature extraction and feature fusion according to the current signals and the instantaneous parameters of the current signals to obtain a fused feature map of the current signals of the target transformer; Splitting the fused feature map, calculating the feature differences between the splitting results to obtain feature difference data; Determining whether a winding of the target transformer fails based on the magnitude of the feature difference data; Wherein, the performing multi-feature extraction and feature fusion according to the current signals and the instantaneous parameters of the current signals to obtain a fused feature map of the current signals of the target transformer specifically includes: Sorting the current signals and the instantaneous parameters of the current signals in the channel direction to form a first feature map, and the first feature map includes at least the instantaneous features of the instantaneous parameters of the current signals; Using a feature extraction model to perform feature extraction on the first feature map to obtain a second feature map and a third feature map, wherein the second feature map includes instantaneous features with a change degree higher than a change threshold, and the third feature map includes instantaneous features with a change degree lower than the change threshold; Using a feature alignment model to align and splice the second feature map and the third feature map to obtain the fused feature map; Wherein, the feature extraction model includes a first feature network and a second feature network, the first feature network includes a plurality of cascaded convolutional blocks, and the second feature network includes a plurality of cascaded deep learning models; then the using the feature extraction model to perform feature extraction on the first feature map to obtain a second feature map and a third feature map specifically includes: Using the first feature network to perform feature extraction on the first feature map and outputting the second feature map; Using the second feature network to perform feature extraction on the first feature map and outputting the third feature map; Wherein, the splitting the fused feature map, calculating the feature differences between the splitting results to obtain feature difference data specifically includes: Splitting the fused feature map into a fused feature sequence in the image width direction, and the fused feature sequence includes a plurality of sub-features; Calculating the mean square error between two adjacent sub-features in the fused feature sequence to obtain feature difference data.

2. The method according to claim 1, wherein The feature alignment model includes a connected multi-layer perceptron, a loss function block, and an image processing block; Then the using the feature alignment model to align and splice the second feature map and the third feature map to obtain a fused feature map includes: Using the multi-layer perceptron to perform image processing on the second feature map and the third feature map to obtain a first to-be-processed feature map and a second to-be-processed feature map with the same image size, wherein the image size is determined by the preset image height, image width, and number of channels of the fused feature map; Using the loss function configured by the loss function block to perform distribution alignment on the first to-be-processed feature map and the second to-be-processed feature map to obtain a first target feature map and a second target feature map; Using the image processing block, the first target feature map and the second target feature map are stitched according to the number of channels to obtain a fused feature map.

3. The method according to claim 2, wherein The loss function is determined by the following formula: In the formula, is the loss function, is the number of channels, is the preset temperature parameter, is the image height, is the image width, is the value of the c -th feature point in the -th channel of the first feature map to be processed, c is the calculation probability distribution corresponding to the value of the i -th feature point in the -th channel of the first feature map to be processed, c represents the value of the i -th feature point in the -th channel of the second feature map to be processed, c is the calculation probability distribution corresponding to the value of the i -th feature point in the -th channel of the second feature map to be processed.

4. The method according to claim 1, wherein The extraction of the instantaneous parameters of the current signal specifically includes: Performing a Hilbert transform on the current signal to obtain a target current signal; Calculating the instantaneous amplitude and instantaneous phase of the current signal according to the current signal and the target current signal; Calculating the instantaneous frequency of the current signal according to the instantaneous phase; The instantaneous parameters include the instantaneous amplitude, the instantaneous phase, and the instantaneous frequency; Among them, the instantaneous amplitude is calculated by the following formula: Wherein, is the instantaneous amplitude, is the target current signal, is the current signal; The instantaneous phase is calculated by the following formula: In the formula, is the instantaneous phase; The instantaneous frequency is calculated by the following formula: Among them, is the instantaneous frequency.

5. A transformer winding fault detection device, characterized in that, The device includes: a signal acquisition module, a feature extraction module, and an anomaly judgment module; The signal acquisition module is configured to acquire current signals at the low-voltage end of a target transformer under a plurality of preset loads respectively, and extract the instantaneous parameters of the current signals; The feature extraction module is configured to perform multi-feature extraction and feature fusion according to the current signal and the instantaneous parameters of the current signal to obtain a fused feature map of the current signal of the target transformer; The anomaly judgment module is configured to split the fused feature map, calculate the feature differences between the split results, and obtain feature difference data; Based on the magnitude of the feature difference data, it is determined whether the winding of the target transformer fails; The feature extraction module is further configured to sort the current signal and the instantaneous parameters of the current signal in the channel direction to form a first feature map, and the first feature map at least includes the instantaneous features of the instantaneous parameters of the current signal; Using a feature extraction model to perform feature extraction on the first feature map to obtain a second feature map and a third feature map, wherein the second feature map includes instantaneous features with a change degree higher than a change threshold, and the third feature map includes instantaneous features with a change degree lower than the change threshold; Using a feature alignment model to align and stitch the second feature map and the third feature map to obtain the fused feature map; Among them, the feature extraction model includes a first feature network and a second feature network, the first feature network includes a plurality of cascaded and sorted convolutional blocks, and the second feature network includes a plurality of cascaded and sorted deep learning models; The first feature network is configured to perform feature extraction on the first feature map and output a second feature map; The second feature network is configured to perform feature extraction on the first feature map and output a third feature map; The anomaly judgment module is further configured to split the fused feature map into a fused feature sequence in the image width direction, and the fused feature sequence includes a plurality of sub-features; Calculating the mean square error between two adjacent sub-features in the fused feature sequence to obtain feature difference data.

6. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, the processor is caused to execute the steps of the method according to any one of claims 1 to 4.

7. A computer device, comprising a memory and a processor, characterized in that, The memory stores a computer program, and when the computer program is executed by the processor, the processor is caused to execute the steps of the method according to any one of claims 1 to 4.

Images