Intelligent gas relay for transformers and method

Abstract

The application relates to the field of intelligent detection, and specifically discloses an intelligent gas relay for a transformer and a method, which monitors and collects pressure values, temperature values and oil level values in real time through a sensor group, and performs time sequence correlation and collaborative analysis on the pressure values, the temperature values and the oil level values by using a data processing and analysis algorithm based on a deep learning neural network, so as to automatically judge whether the transformer has a fault according to the real-time running state of various parameters in the transformer, and generate a warning signal to prompt when the transformer has a fault. In this way, the fault problem can be found in time and warned, so that corresponding measures can be taken, the timeliness of fault diagnosis is improved, the false alarm problem caused by gas accumulation in the traditional scheme can be avoided, the accuracy of fault detection is improved, and the safety and stability of equipment are ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent detection, and more specifically, to an intelligent gas relay for a transformer and a method. BACKGROUND

[0002] A transformer is an electrical device used to change the voltage of alternating current, which converts the input voltage into output voltage through the principle of electromagnetic induction. A gas relay is a relay that uses gas (usually nitrogen or hydrogen sulfide gas) as an insulating medium and arc extinguishing medium, which can monitor the gas concentration inside the transformer and provide additional safety protection and control functions in the transformer. Specifically, when internal faults or overloads occur in the transformer, the gas relay can detect these abnormal conditions and issue a warning signal to prevent the accident from further expanding.

[0003] During normal operation, a small amount of gas will inevitably be generated in the transformer, which will gradually accumulate inside the transformer. However, the traditional transformer fault detection method usually triggers the protection function of the gas relay according to the degree of gas accumulation. In this way, even a small amount of gas accumulation generated during normal operation may be misjudged as a fault, resulting in false alarms and maintenance costs for the transformer. In addition, the traditional method cannot comprehensively consider the real-time running state of the transformer, thereby affecting the timeliness of the operation and maintenance of the equipment.

[0004] Therefore, an optimized intelligent gas relay for a transformer is desired. SUMMARY

[0005] To solve the above technical problems, the present application is proposed. The embodiments of the present application provide an intelligent gas relay for a transformer and a method, which monitors and collects pressure values, temperature values and oil level values in real time through a sensor group, and uses a data processing and analysis algorithm based on a deep learning neural network to perform time sequence correlation analysis of the pressure values, the temperature values and the oil level values. In this way, according to the real-time running state of each parameter in the transformer, it can automatically determine whether the transformer has a fault, and generate a warning signal when the transformer has a fault. In this way, the fault problem can be discovered and warned in time, so that appropriate measures can be taken, improving the timeliness of fault diagnosis, and avoiding false alarms caused by gas accumulation in traditional solutions, improving the accuracy of fault detection, thereby ensuring the safety and stability of the equipment.

[0006] According to one aspect of the present application, an intelligent gas relay for a transformer is provided, comprising:

[0007] The gas chamber, the oil chamber and the controller, wherein the gas chamber and the oil chamber are connected through a diaphragm, the gas chamber is provided with a pressure sensor and a temperature sensor, the oil chamber is provided with an oil level sensor and an oil quality sensor, the controller is electrically connected with the pressure sensor, the temperature sensor, the oil level sensor and the oil quality sensor, the controller is also electrically connected with a display screen and an alarm device, the controller monitors the running state of the transformer in real time according to the signals of the pressure sensor, the temperature sensor, the oil level sensor and the oil quality sensor ， and displays the monitoring result on the display screen, when the monitoring result exceeds the preset range, the controller sends an alarm signal through the alarm device

[0008] The controller, characterized in that, comprises:

[0009] The transformer running parameter acquisition module is used to acquire the time series of the pressure value, the temperature value and the oil level value collected by the sensor group;

[0010] The time series data regularization module is used to perform data regularization on the time series of the pressure value, the temperature value and the oil level value according to the time dimension to obtain a pressure time series input vector, a temperature time series input vector and an oil level time series input vector;

[0011] The feature correction module is used to perform gamma correction on the pressure time series input vector, the temperature time series input vector and the oil level time series input vector to obtain a corrected pressure time series input vector, a corrected temperature time series input vector and a corrected oil level time series input vector;

[0012] The pressure-temperature correlation feature module is used to calculate the sample covariance correlation matrix between the corrected pressure time series input vector and the corrected temperature time series input vector to obtain a pressure-temperature time series correlation matrix;

[0013] The pressure-temperature time series correlation feature extraction module is used to perform pressure-temperature time series pattern correlation feature extraction on the pressure-temperature time series correlation matrix to obtain a pressure-temperature time series pattern correlation feature map;

[0014] The vector-image conversion module is used to convert the corrected oil level time series input vector through the vector-image conversion module to obtain an oil level time series image;

[0015] The oil level time series feature extraction module is used to perform oil level time series feature extraction on the oil level time series image to obtain an oil level time series pattern feature map;

[0016] an attention fusion module configured to fuse the pressure-temperature time-series correlation feature map and the oil level time-series pattern feature map through a global mean pooling based attention fusion network to obtain a transformer state representation feature map as a transformer state representation feature;

[0017] a transformer fault early warning module configured to determine whether a transformer has a fault based on the transformer state representation feature, and determine whether to generate an early warning signal prompt.

[0018] According to another aspect of the present application, an intelligent gas relay method for a transformer is provided, which comprises:

[0019] obtaining time series of pressure values, temperature values and oil level values collected by a sensor group;

[0020] performing data regularization on the time series of pressure values, temperature values and oil level values according to a time dimension to obtain a pressure time-series input vector, a temperature time-series input vector and an oil level time-series input vector;

[0021] performing gamma correction on the pressure time-series input vector, the temperature time-series input vector and the oil level time-series input vector to obtain a corrected pressure time-series input vector, a corrected temperature time-series input vector and a corrected oil level time-series input vector;

[0022] calculating a sample covariance correlation matrix between the corrected pressure time-series input vector and the corrected temperature time-series input vector to obtain a pressure-temperature time-series correlation matrix;

[0023] performing pressure-temperature time-series pattern correlation feature extraction on the pressure-temperature time-series correlation matrix to obtain a pressure-temperature time-series pattern correlation feature map;

[0024] converting the corrected oil level time-series input vector through a vector-image conversion module to obtain an oil level time-series image;

[0025] performing oil level time-series feature extraction on the oil level time-series image to obtain an oil level time-series pattern feature map;

[0026] fusing the pressure-temperature time-series pattern correlation feature map and the oil level time-series pattern feature map through a global mean pooling based attention fusion network to obtain a transformer state representation feature map as a transformer state representation feature;

[0027] determining whether a transformer has a fault based on the transformer state representation feature, and determining whether to generate an early warning signal prompt.

[0028] Compared with the prior art, the intelligent gas relay for a transformer and the method provided by the application can realize real-time monitoring and collection of pressure values, temperature values and oil level values through a sensor group, and perform time sequence correlation and collaborative analysis on the pressure values, the temperature values and the oil level values by using a data processing and analysis algorithm based on a deep learning neural network, so as to automatically judge whether the transformer has a fault according to real-time running states of various parameters in the transformer, and generate a warning signal to prompt when the transformer has a fault. In this way, the fault problem can be found in time and a warning can be given, so that corresponding countermeasures can be taken, the timeliness of fault diagnosis is improved, the false alarm problem caused by gas accumulation in the traditional scheme can be avoided, the accuracy of fault detection is improved, and the safety and stability of the equipment are ensured. BRIEF DESCRIPTION OF DRAWINGS

[0029] The above and other objects, features and advantages of the present application will become more apparent from the following detailed description of embodiments of the present application taken in conjunction with the accompanying drawings. The drawings provided in the Drawings serve to provide a further understanding of the embodiments of the present application, constitute a part of the specification and are included to explain the present application, and do not limit the present application. In the drawings, the same reference numerals generally refer to the same components or steps.

[0030] Figure 1 A block diagram of the intelligent gas relay for a transformer according to the embodiments of the present application.

[0031] Figure 2 An architectural schematic diagram of the intelligent gas relay for a transformer according to the embodiments of the present application.

[0032] Figure 3 A block diagram of an attention fusion module in the controller of the intelligent gas relay for a transformer according to the embodiments of the present application.

[0033] Figure 4 A block diagram of a training module in the intelligent gas relay for a transformer according to the embodiments of the present application.

[0034] Figure 5 A flowchart of the intelligent gas relay method for a transformer according to the embodiments of the present application. DETAILED DESCRIPTION

[0035] Embodiments of the present disclosure will be described in greater detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms, and should not be interpreted as being limited to the embodiments set forth herein, but rather, these embodiments are provided to make the present disclosure more thorough and complete. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes, and are not intended to limit the scope of protection of the present disclosure.

[0036] It should be understood that the various steps recited in the method embodiments of the present disclosure can be performed in different orders and / or in parallel. In addition, the method embodiments can include additional steps and / or omit performing the steps shown. The scope of the present disclosure is not limited in this regard.

[0037] In the description of embodiments of the present disclosure, the term "comprising" and its conjugations should be understood to encompass the meanings of "including" and "consisting of" and its conjugations. The term "based on" should be understood as "based at least in part on". The term "one embodiment" or "an embodiment" should be understood as "at least one embodiment". The terms "first", "second", etc. can refer to different or the same objects. Other explicit and implicit definitions can also be included below.

[0038] It should be noted that the modification of "one" or "multiple" in the present disclosure is illustrative and not limiting, and those skilled in the art should understand that unless otherwise explicitly indicated in the context, it should be understood as "one or more".

[0039] A transformer is an electrical device used to change the voltage of alternating current, which converts the input voltage into the output voltage through the principle of electromagnetic induction. A gas relay is a relay that uses gas (usually nitrogen or hydrogen sulfide gas) as an insulating medium and arc extinguishing medium, which can monitor the gas concentration inside the transformer and provide additional safety protection and control functions in the transformer. Specifically, when the transformer has internal faults or overloads, the gas relay can detect these abnormal conditions and issue a warning signal to prevent the accident from further expanding.

[0040] During normal operation, a small amount of gas will inevitably be generated in the transformer, which will gradually accumulate inside the transformer. However, the traditional transformer fault detection method usually triggers the protection function of the gas relay according to the degree of gas accumulation. In this way, even a small amount of gas accumulation generated during normal operation may be misjudged as a fault, resulting in false alarms and maintenance costs for the transformer. In addition, the traditional method cannot comprehensively consider the real-time running state of the transformer, thereby affecting the timeliness of the operation and maintenance of the equipment.

[0041] Based on this, in the technical scheme of the present application, an intelligent gas relay for a transformer is proposed, which comprises a gas chamber, an oil chamber and a controller, wherein the gas chamber and the oil chamber are connected through a diaphragm, a pressure sensor and a temperature sensor are arranged in the gas chamber, an oil level sensor and an oil quality sensor are arranged in the oil chamber, the controller is electrically connected with the pressure sensor, the temperature sensor, the oil level sensor and the oil quality sensor, the controller is also electrically connected with a display screen and an alarm device, the controller monitors the running state of the transformer in real time according to the signals of the pressure sensor, the temperature sensor, the oil level sensor and the oil quality sensor ， and displays the monitoring results on the display screen, when the monitoring results exceed the preset range, the controller sends an alarm signal through the alarm device.

[0042] Correspondingly, in the above controller, the technical concept of the present application is to monitor and collect the pressure value, temperature value and oil level value in real time through the sensor group, and to perform time sequence correlation and collaborative analysis on the pressure value, temperature value and oil level value by using a data processing and analysis algorithm based on deep learning neural network, so as to automatically judge whether the transformer has a fault according to the real-time running state of each parameter in the transformer, and to generate a warning signal when the transformer has a fault. In this way, the fault problem can be found in time and a warning can be given, so that corresponding measures can be taken, the timeliness of fault diagnosis is improved, and the accuracy of fault detection is improved, thereby ensuring the safety and stability of the equipment.

[0043] Figure 1 The block diagram of the controller in the intelligent gas relay for a transformer according to the embodiment of the present application. Figure 2 The architecture schematic diagram of the controller in the intelligent gas relay for a transformer according to the embodiment of the present application. As Figure 1 and Figure 2As shown, the intelligent gas relay 100 for transformer according to the embodiment of the present application comprises: a transformer operating parameter acquisition module 110, configured to acquire time series of pressure values, temperature values and oil level values collected by a sensor group; a time series data regularization module 120, configured to perform data regularization on the time series of the pressure values, the temperature values and the oil level values according to time dimensions to obtain a pressure time series input vector, a temperature time series input vector and an oil level time series input vector; a feature correction module 130, configured to perform gamma correction on the pressure time series input vector, the temperature time series input vector and the oil level time series input vector to obtain a corrected pressure time series input vector, a corrected temperature time series input vector and a corrected oil level time series input vector; a pressure-temperature correlation feature module 140, configured to calculate a sample covariance correlation matrix between the corrected pressure time series input vector and the corrected temperature time series input vector to obtain a pressure-temperature time series correlation matrix; a pressure-temperature time series correlation feature extraction module 150, configured to perform pressure-temperature time series mode correlation feature extraction on the pressure-temperature time series correlation matrix to obtain a pressure-temperature time series mode correlation feature map; a vector-image conversion module 160, configured to convert the corrected oil level time series input vector through a vector-image conversion module to obtain an oil level time series image; an oil level time series feature extraction module 170, configured to perform oil level time series feature extraction on the oil level time series image to obtain an oil level time series mode feature map; an attention fusion module 180, configured to fuse the pressure-temperature time series mode correlation feature map and the oil level time series mode feature map through an attention fusion network based on global mean pooling to obtain a transformer state representation feature map as a transformer state representation feature; and a transformer fault early warning module 190, configured to determine whether a transformer has a fault and determine whether to generate a warning signal based on the transformer state representation feature.

[0044] In the embodiment of the present application, the transformer operating parameter acquisition module 110 is configured to acquire time series of pressure values, temperature values and oil level values collected by a sensor group. It can be understood that, considering that the pressure values, the temperature values and the oil level values are important parameters for normal operation of a transformer, changes thereof can reflect the running state inside the transformer, so as to make a judgment on a transformer fault. Specifically, monitoring of the pressure values and the temperature values can help ensure that the transformer operates within a safe range, avoiding dangerous situations caused by overheating or overloading. Monitoring of the oil level values can ensure normal working state of insulating oil inside the transformer, maintaining stability of the equipment. Based on this, in the technical solution of the present application, time series of pressure values, temperature values and oil level values collected by a sensor group are acquired, and time series cooperative correlation analysis is performed on the pressure values, the temperature values and the oil level values, so that abnormal changes of the transformer can be found in time, potential faults can be predicted, and thus equipment safety can be improved and service life of the equipment can be prolonged.

[0045] In the embodiment of the present application, the time series data normalization module 120 is configured to normalize the time series of the pressure value, the temperature value and the oil level value according to the time dimension to obtain a pressure time series input vector, a temperature time series input vector and an oil level time series input vector. Correspondingly, it is considered that the pressure value, the temperature value and the oil level value all contain the correlation and trend change in the time dimension, that is, the pressure value, the temperature value and the oil level value all have time series characteristic information in the time dimension. Therefore, in order to retain the time series characteristic information of the pressure value, the temperature value and the oil level value, so as to subsequently analyze the dynamic change rule and fluctuation trend of the time series characteristics more accurately, in the technical solution of the present application, the time series of the pressure value, the temperature value and the oil level value are normalized according to the time dimension to obtain a pressure time series input vector, a temperature time series input vector and an oil level time series input vector, which provides help for subsequent analysis of the time series dynamic change and mutual collaborative influence among the pressure value, the temperature value and the oil level value.

[0046] In the embodiment of the present application, the feature correction module 130 is configured to perform gamma correction on the pressure time series input vector, the temperature time series input vector and the oil level time series input vector to obtain a corrected pressure time series input vector, a corrected temperature time series input vector and a corrected oil level time series input vector. It can be understood that, considering that the sensor may be affected by various factors such as saturation effect, temperature influence, electromagnetic interference and the like under different working conditions, these factors make the relationship between the output signal and the input signal of the sensor not a simple linear relationship, but a complex nonlinear characteristic. Therefore, in the technical solution of the present application, in order to convert the nonlinear response of the sensor into a nearly linear response, so as to make it closer to the true value, and improve the accuracy and reliability of data measurement, in the technical solution of the present application, the pressure time series input vector, the temperature time series input vector and the oil level time series input vector are subjected to gamma correction to obtain a corrected pressure time series input vector, a corrected temperature time series input vector and a corrected oil level time series input vector. It can be understood that gamma correction is a kind of nonlinear operation, through which the nonlinear effect of the sensor can be eliminated, and the data collected by different sensors can be standardized, so that the data has a unified dimension and range, facilitating subsequent data processing and analysis, so as to be better applied to the processing and analysis process of pressure, temperature and oil level time series data.

[0047] Specifically, in the embodiment of the present application, the feature correction module is configured to perform gamma correction on the pressure time series input vector according to the following correction formula to obtain the corrected pressure time series input vector; wherein the correction formula is:

[0048]

[0049] wherein, v i is the eigenvalue of the i-th position of the pressure time-series input vector, V i is the eigenvalue of the i-th position of the corrected pressure time-series input vector, A, B, C and D are corresponding adjustment parameters. Similarly, the encoding process of the temperature time-series input vector and the oil level time-series input vector is consistent with the encoding process of the pressure time-series input vector.

[0050] In the embodiments of the present application, the pressure-temperature correlation feature module 140 is configured to calculate a sample covariance correlation matrix between the corrected pressure time-series input vector and the corrected temperature time-series input vector to obtain a pressure-temperature time-series correlation matrix. Accordingly, considering that there is a certain correlation between the pressure and the temperature of the transformer, that is, the pressure and the temperature have a certain mutual influence relationship in time series as time goes on. Therefore, in the technical solution of the present application, the sample covariance correlation matrix between the corrected pressure time-series input vector and the corrected temperature time-series input vector is calculated to obtain the pressure-temperature time-series correlation matrix. It is worth mentioning that the sample covariance is a statistical quantity for measuring the correlation between two variables. By calculating the sample covariance between the corrected pressure time-series input vector and the corrected temperature time-series input vector, the strength and direction of the correlation between them can be measured, which can more accurately reveal the time-series correlation relationship between the pressure and the temperature, and thus more accurately detect the fault of the transformer.

[0051] Specifically, in the embodiments of the present application, the pressure-temperature correlation feature module is configured to calculate the sample covariance correlation matrix between the corrected pressure time-series input vector and the corrected temperature time-series input vector to obtain the pressure-temperature time-series correlation matrix by using the following sample covariance formula:

[0052] M = V T XX T V

[0053] wherein, V is the corrected pressure time-series input vector, X is the corrected temperature time-series input vector, and M is the pressure-temperature time-series correlation matrix.

[0054] In the embodiments of the present application, the pressure-temperature time sequence correlation feature extraction module 150 is configured to perform pressure-temperature time sequence pattern correlation feature extraction on the pressure-temperature time sequence correlation matrix to obtain a pressure-temperature time sequence pattern correlation feature map. Specifically, in the embodiments of the present application, the pressure-temperature time sequence correlation feature extraction module is configured to: pass the pressure-temperature time sequence correlation matrix through a pressure-temperature time sequence pattern correlation feature extractor based on a convolutional neural network model to obtain the pressure-temperature time sequence pattern correlation feature map. It should be understood that the pressure-temperature time sequence correlation matrix contains key time sequence correlation patterns and features, and the convolutional neural network has good performance in processing implicit correlation information between time sequence data and can effectively capture implicit time sequence correlation features in time sequence data. Therefore, in the technical solution of the present application, the pressure-temperature time sequence correlation matrix is passed through a pressure-temperature time sequence pattern correlation feature extractor based on a convolutional neural network model to capture time sequence correlation feature information between the pressure and the temperature, thereby obtaining a pressure-temperature time sequence pattern correlation feature map with more characteristic expressions.

[0055] In the embodiments of the present application, the vector-image conversion module 160 is configured to pass the corrected oil level time sequence input vector through a vector-image conversion module to obtain an oil level time sequence image. Accordingly, it should be understood that a time sequence input vector represents simple time sequence numerical change information, while an image can provide more rich feature information than a vector, such as the trend of oil level change, the frequency of oil level fluctuation, the rate of oil level change, etc. Based on this, in order to better mine potential time sequence dynamic information in oil level data for subsequent accurate analysis, in the technical solution of the present application, the corrected oil level time sequence input vector is passed through a vector-image conversion module to obtain an oil level time sequence image. It should be understood that by converting the corrected oil level time sequence input vector into an image, the trend of change and fluctuation of the oil level in the time sequence feature can be more intuitively understood, for example, the change of the pixel value in the oil level time sequence image with time can reflect the trend of the oil level change, such as rising, falling or stabilizing, or the frequency of the pixel value fluctuation in the oil level time sequence image can reflect the frequency of the oil level fluctuation, such as high-frequency fluctuation or low-frequency fluctuation, and these time sequence feature change information provides more possibilities for subsequent transformer fault judgment.

[0056] In the embodiment of the present application, the oil level time sequence feature extraction module 170 is configured to perform oil level time sequence feature extraction on the oil level time sequence image to obtain an oil level time sequence pattern feature map. Specifically, in the embodiment of the present application, the oil level time sequence feature extraction module is configured to: pass the oil level time sequence image through an oil level time sequence feature extractor based on a convolutional neural network model to obtain the oil level time sequence pattern feature map. It should be understood that the oil level time sequence image contains implicit feature information about the oil level time sequence, and the convolutional neural network has good feature extraction capability in processing image implicit features. Therefore, in the technical solution of the present application, the oil level time sequence image is passed through the oil level time sequence feature extractor based on the convolutional neural network model to extract the implicit features about the oil level time sequence, so as to obtain the oil level time sequence pattern feature map.

[0057] In the embodiment of the present application, the attention fusion module 180 is configured to pass the pressure-temperature time sequence pattern association feature map and the oil level time sequence pattern feature map through a global mean pooling based attention fusion network to obtain a transformer state representation feature map as the transformer state representation feature. Figure 3 The block diagram of the attention fusion module in the controller of the intelligent gas relay for the transformer according to the embodiment of the present application is shown in FIG. 8. Specifically, in the embodiment of the present application, as shown in FIG. 8, the attention fusion module 180 is configured to pass the pressure-temperature time sequence pattern association feature map and the oil level time sequence pattern feature map through a global mean pooling based attention fusion network to obtain a transformer state representation feature map as the transformer state representation feature. Figure 3As shown, the attention fusion module 180 includes: a global mean pooling unit 181 configured to perform global mean pooling on each feature matrix of the pressure-temperature time sequence mode associated feature map along the channel dimension to obtain a pressure-temperature time sequence mode associated feature vector; a fully connected encoding unit 182 configured to perform fully connected encoding on the pressure-temperature time sequence mode associated feature vector using a fully connected layer to obtain a fully connected encoding pressure-temperature time sequence mode associated feature vector; a weighted fusion unit 183 configured to perform weighted fusion of the fully connected encoding pressure-temperature time sequence mode associated feature vector and the oil level time sequence mode feature map along the channel dimension to obtain a weighted fusion oil level time sequence mode feature map; and a fusion unit 184 configured to fuse the weighted fusion oil level time sequence mode feature map and the pressure-temperature time sequence mode associated feature map to obtain the transformer state representation feature map. Accordingly, in order to comprehensively integrate the time sequence cooperative correlation relationship between various operating states of the transformer, the state of the transformer is accurately represented. Specifically, the pressure-temperature mode can reflect the thermodynamic state of the equipment, and the oil level mode can reflect the mechanical state of the equipment. By comprehensively considering the two modes, a more in-depth understanding of the overall state of the equipment can be obtained. Therefore, in the technical solution of the present application, the pressure-temperature time sequence mode associated feature map and the oil level time sequence mode feature map are further fused through the attention fusion network based on global mean pooling to obtain the transformer state representation feature map. It is worth mentioning that the attention fusion network based on global mean pooling performs global average pooling on the high-level feature map along the channel dimension, then multiplies the pooled features as attention information with the low-level feature map, and finally adds the generated low-level feature map and high-level feature map for feature fusion to obtain a feature map with richer semantic information. Specifically, the global mean pooling operation can aggregate all information in the pressure-temperature time sequence mode associated feature map into a single pooling vector, thereby extracting the global pressure-temperature time sequence correlation mode. Then, based on the pooling vector, the correlation and importance between the oil level time sequence mode feature maps are weighted and fused, and finally the weighted fusion oil level time sequence mode feature map and the pressure-temperature time sequence mode associated feature map are fused to obtain a more comprehensive transformer state representation.

[0058] In the embodiment of the present application, the transformer fault early warning module 190 is configured to determine whether the transformer has a fault based on the transformer state representation feature, and determine whether to generate a warning signal prompt. Specifically, in the embodiment of the present application, the transformer fault early warning module includes a transformer fault judgment unit configured to pass the transformer state representation feature map through a transformer fault detector based on a classifier to obtain a detection result, wherein the detection result is used to indicate whether the transformer has a fault; and a warning signal unit configured to generate a warning signal prompt in response to the detection result indicating that the transformer has a fault. That is, the transformer state representation feature obtained by fusing the pressure-temperature time series pattern correlation feature map and the oil level time series pattern feature map through attention is used for classification processing, so as to automatically determine whether the transformer has a fault according to the real-time running state of each parameter in the transformer, and generate a warning signal prompt when the transformer has a fault. In this way, the fault problem can be found in time and warned, so that corresponding measures can be taken, the timeliness of fault diagnosis is improved, and the accuracy of fault detection is improved, thereby ensuring the safety and stability of the equipment.

[0059] It is worth mentioning that, before applying the deep neural network model for inference, the deep neural network model needs to be trained to enable the deep neural network to achieve a specific function.

[0060] Specifically, in the technical solution of the present application, the intelligent gas relay for transformers further includes a training module for training the pressure-temperature time series pattern correlation feature extractor based on the convolutional neural network model, the vector-image conversion module, the oil level time series feature extractor of the convolutional neural network model, the attention fusion network based on global mean pooling, and the transformer fault detector based on the classifier.

[0061] Figure 4 The block diagram of the training module in the intelligent gas relay for transformers according to the embodiment of the present application is shown in FIG. 8. Specifically, in the embodiment of the present application, as shown in FIG. 8, the training module includes a pressure-temperature time series pattern correlation feature extractor based on a convolutional neural network model, a vector-image conversion module, an oil level time series feature extractor of a convolutional neural network model, an attention fusion network based on global mean pooling, and a transformer fault detector based on a classifier. Figure 4As shown, the training module 200 comprises: a training data acquisition unit 210 configured to acquire training data, the training data comprising time series of training pressure values, training temperature values, and training oil level values collected by a sensor group, and a true value of whether the transformer has a fault; a training data regularization unit 220 configured to perform data regularization on the time series of training pressure values, training temperature values, and training oil level values according to a time dimension to obtain a training pressure time sequence input vector, a training temperature time sequence input vector, and a training oil level time sequence input vector; a training data gamma correction unit 230 configured to perform gamma correction on the training pressure time sequence input vector, the training temperature time sequence input vector, and the training oil level time sequence input vector to obtain a corrected training pressure time sequence input vector, a corrected training temperature time sequence input vector, and a corrected training oil level time sequence input vector; a training time sequence correlation unit 240 configured to calculate a sample covariance correlation matrix between the corrected training pressure time sequence input vector and the corrected training temperature time sequence input vector to obtain a training pressure-temperature time sequence correlation matrix; a training time sequence pattern correlation feature unit 250 configured to pass the training pressure-temperature time sequence correlation matrix through the pressure-temperature time sequence pattern correlation feature extractor based on the convolutional neural network model to obtain a training pressure-temperature time sequence pattern correlation feature map; a training vector-image conversion unit 260 configured to pass the corrected training oil level time sequence input vector through the vector-image conversion module to obtain a training oil level time sequence image; a training oil level time sequence feature extraction unit 270 configured to pass the training oil level time sequence image through the oil level time sequence feature extractor based on the convolutional neural network model to obtain a training oil level time sequence pattern feature map; a training feature fusion unit 280 configured to pass the training pressure-temperature time sequence pattern correlation feature map and the training oil level time sequence pattern feature map through the attention fusion network based on global mean pooling to obtain a training transformer state representation feature map; a training feature optimization unit 290 configured to optimize the training transformer state representation feature map based on a feature matrix distribution of the training transformer state representation feature map along a channel dimension to obtain an optimized training transformer state representation feature map; a classification loss calculation unit 300 configured to pass the optimized training transformer state representation feature map through the transformer fault detector based on the classifier to obtain a classification loss function value; and a training unit 310 configured to train the pressure-temperature time sequence pattern correlation feature extractor based on the convolutional neural network model, the vector-image conversion module, the oil level time sequence feature extractor of the convolutional neural network model, the attention fusion network based on global mean pooling, and the transformer fault detector based on the classifier based on the classification loss function value and through back propagation of gradient descent.

[0062] Specifically, the training feature optimization unit 290 is configured to optimize the training transformer state representation feature map along the feature matrix distribution of the channel dimension to obtain an optimized training transformer state representation feature map. It should be understood that in the above technical solution, each feature matrix of the training pressure-temperature time series mode correlation feature map represents a local high-order time series correlation feature of the full time series covariance correlation of the training pressure value and the temperature value, and each feature matrix of the training oil level time series mode feature map represents a local time domain intra-local time domain inter local time series correlation feature of the training oil level value in the local time domain determined by the vector-image conversion from the global time domain. Therefore, considering the significant difference in the time series feature distribution of each feature matrix of the training pressure-temperature time series mode correlation feature map and the training oil level time series mode feature map in the distribution dimension and the distribution order, after the training pressure-temperature time series mode correlation feature map and the training oil level time series mode feature map are fused through the attention fusion network based on global mean pooling, since the attention fusion based on global mean pooling is based on the overall feature space distribution of each feature matrix to perform overall distribution weighting on the feature matrix, the training transformer state representation feature map has more significant feature distribution difference between each feature matrix, thereby affecting the classification regression effect of each feature matrix of the training transformer state representation feature map based on the overall numerical distribution of the feature value, and affecting the accuracy of the classification result obtained by the transformer fault detector based on the classifier.

[0063] Specifically, in the embodiment of the present application, the training feature optimization unit comprises: a probability coefficient calculation subunit configured to perform scenario-based class probability logical inference on each feature matrix of the training transformer state representation feature map along the channel dimension to obtain a probability vector; and a weighted optimization subunit configured to perform weighted optimization on each feature matrix of the training transformer state representation feature map along the channel dimension by taking the probability coefficient value at each position in the probability vector as a weighting coefficient to obtain the optimized training transformer state representation feature map.

[0064] More specifically, in the embodiment of the present application, the probability coefficient calculation subunit is configured to perform scenario-based class probability logical inference on each feature matrix of the training transformer state representation feature map along the channel dimension to obtain the probability vector by using the following probability calculation formula:

[0065]

[0066] wherein, m i,j is the feature value of the (i, j) position of each feature matrix of the training transformer state representation feature map, φ(m i,j ) represents a probabilistic function of the feature value, i.e., a probabilistic function that maps the feature value m i,j to the interval [0, 1], S is the scale, i.e., the width multiplied by the height, of each feature matrix of the training transformer state representation feature map, p is the class probability value obtained by the classifier for the training transformer state representation feature map, and a is a weight hyperparameter, and ω is the probability coefficient value of each position in the probability vector.

[0067] In particular, that is, for each feature matrix of the training transformer state representation feature map, the corresponding feature scene is associated with the scene saturation class probability reasoning logic by the probability distribution foreground constraint and the relative probability mapping response hypothesis, so as to endow the feature set of each feature matrix of the training transformer state representation feature map with scene concept ontology cognition, that is, to internally align the overall distribution with the scene-based class probability logic reasoning in the classification process, so as to improve the understanding ability of the feature matrix scene distribution of the training transformer state representation feature map for class cognition. In this way, by using the probability coefficient value of each position in the probability vector as a weighting coefficient to weight and optimize the corresponding feature matrix of the transformer state representation feature map, the accuracy of the classification result obtained by the transformer fault detector based on the classifier can be improved. In this way, the fault problem can be discovered and warned in time, so that corresponding measures can be taken, the timeliness of fault diagnosis is improved, and the accuracy of fault detection is improved, thereby ensuring the safety and stability of the equipment.

[0068] In summary, the intelligent gas relay 100 for the transformer based on the embodiments of the present application is illustrated, which monitors and collects the pressure value, temperature value and oil level value in real time through the sensor group, and uses the data processing and analysis algorithm based on the deep learning neural network to perform time sequence correlation analysis on the pressure value, the temperature value and the oil level value. In this way, according to the real-time running state of each parameter in the transformer, it is automatically judged whether the transformer has a fault, and a warning signal is generated when the transformer has a fault. In this way, the fault problem can be discovered and warned in time, so that corresponding measures can be taken, the timeliness of fault diagnosis is improved, and the accuracy of fault detection is improved, thereby ensuring the safety and stability of the equipment.

[0069] As described above, the intelligent gas relay 100 for transformers according to embodiments of this application can be implemented in various wireless terminals, such as servers with online monitoring algorithms for gas from oil-immersed transformers. In one possible implementation, the intelligent gas relay 100 for transformers according to embodiments of this application can be integrated into a wireless terminal as a software module and / or a hardware module. For example, the intelligent gas relay 100 for transformers can be a software module in the operating system of the wireless terminal, or it can be an application developed for the wireless terminal; of course, the intelligent gas relay 100 for transformers can also be one of many hardware modules of the wireless terminal.

[0070] Alternatively, in another example, the transformer smart gas relay 100 and the wireless terminal can also be separate devices, and the transformer smart gas relay 100 can be connected to the wireless terminal via wired and / or wireless networks, and transmit interactive information in accordance with an agreed data format.

[0071] Figure 5 This is a flowchart of a smart gas relay method for transformers according to an embodiment of this application. Figure 5 As shown, the intelligent gas relay method for transformers according to an embodiment of this application includes: S110, acquiring time series of pressure values, temperature values, and oil level values ​​collected by a sensor group; S120, performing data normalization on the time series of pressure values, temperature values, and oil level values ​​according to the time dimension to obtain pressure time series input vector, temperature time series input vector, and oil level time series input vector; S130, performing gamma correction on the pressure time series input vector, the temperature time series input vector, and the oil level time series input vector to obtain corrected pressure time series input vector, corrected temperature time series input vector, and corrected oil level time series input vector; S140, calculating the sample covariance correlation matrix between the corrected pressure time series input vector and the corrected temperature time series input vector to obtain the pressure-temperature... S150, extract pressure-temperature time-series correlation features from the pressure-temperature time-series correlation matrix to obtain a pressure-temperature time-series correlation feature map; S160, convert the corrected oil level time-series input vector into an oil level time-series image through a vector-image conversion module; S170, extract oil level time-series features from the oil level time-series image to obtain an oil level time-series pattern feature map; S180, pass the pressure-temperature time-series correlation feature map and the oil level time-series pattern feature map through a global mean pooling-based attention fusion network to obtain a transformer state characterization feature map as the transformer state characterization feature; and S190, based on the transformer state characterization feature, determine whether the transformer has a fault and whether to generate a warning signal.

[0072] Here, those skilled in the art can understand that the specific operations of each step in the above intelligent gas relay method for transformers have been described in detail above with reference to the description of the intelligent gas relay for transformers Figures 1 to 4 , and thus repetitive descriptions thereof will be omitted.

[0073] The implementations of the disclosure have been described above with the foregoing specification, which are exemplary, rather than exhaustive, of the implementations. And also not limited to the disclosed implementations, many modifications and changes will be obvious to those of ordinary skill in the art without departing from the scope and spirit of the described implementations. The choice of words used in this document is intended to best explain the principles of the implementations, practical applications, or improvements to the technology in the market, or to enable other ordinary skilled persons in the art to understand the various implementation methods disclosed herein.

Claims

1. A smart gas relay for transformers, comprising: The transformer comprises a gas chamber, an oil chamber, and a controller. The gas chamber and the oil chamber are connected by a diaphragm. The gas chamber contains a pressure sensor and a temperature sensor, and the oil chamber contains an oil level sensor and an oil quality sensor. The controller is electrically connected to the pressure sensor, the temperature sensor, the oil level sensor, and the oil quality sensor. The controller is also electrically connected to a display screen and an alarm device. The controller is used to monitor the operating status of the transformer in real time based on the signals from the pressure sensor, the temperature sensor, the oil level sensor, and the oil quality sensor, and displays the monitoring results on the display screen. When the monitoring results exceed a preset range, the controller issues an alarm signal through the alarm device. The controller is characterized in that it comprises: The transformer operating parameter acquisition module is used to acquire the time series of pressure, temperature and oil level values ​​collected by the sensor group; The time series data normalization module is used to normalize the time series of the pressure value, temperature value and oil level value according to the time dimension to obtain the pressure time series input vector, temperature time series input vector and oil level time series input vector. The feature correction module is used to perform gamma correction on the pressure time-series input vector, the temperature time-series input vector and the oil level time-series input vector to obtain the corrected pressure time-series input vector, the corrected temperature time-series input vector and the corrected oil level time-series input vector. The pressure-temperature correlation feature module is used to calculate the sample covariance correlation matrix between the corrected pressure time-series input vector and the corrected temperature time-series input vector to obtain the pressure-temperature time-series correlation matrix; The pressure-temperature time series correlation feature extraction module is used to extract pressure-temperature time series pattern correlation features from the pressure-temperature time series correlation matrix to obtain a pressure-temperature time series pattern correlation feature map. A vector-to-image conversion module is used to convert the corrected oil level timing input vector into an oil level timing image. The oil level time-series feature extraction module is used to extract oil level time-series features from the oil level time-series image to obtain an oil level time-series pattern feature map. The attention fusion module is used to pass the pressure-temperature time series pattern association feature map and the oil level time series pattern feature map through an attention fusion network based on global mean pooling to obtain a transformer state characterization feature map as a transformer state characterization feature. The transformer fault early warning module is used to determine whether a transformer has a fault based on the transformer's state characterization characteristics, and to determine whether to generate an early warning signal.

2. The intelligent gas relay for transformers according to claim 1, characterized in that, The feature correction module is used to: perform gamma correction on the pressure time series input vector using the following correction formula to obtain the corrected pressure time series input vector; The correction formula is as follows: Among them, v i V is the feature value at the i-th position of the pressure time-series input vector. i Let be the feature value at the i-th position of the corrected pressure timing input vector, and let A, B, C, and D be the corresponding adjustment parameters.

3. The intelligent gas relay for transformers according to claim 2, characterized in that, The pressure-temperature correlation feature module is used to: calculate the sample covariance correlation matrix between the corrected pressure time-series input vector and the corrected temperature time-series input vector using the following sample covariance formula to obtain the pressure-temperature time-series correlation matrix; The formula for the sample covariance is as follows: M=V T XX T V Wherein, V is the corrected pressure time series input vector, X is the corrected temperature time series input vector, and M is the pressure-temperature time series correlation matrix.

4. The intelligent gas relay for transformers according to claim 3, characterized in that, The pressure-temperature time series correlation feature extraction module is used to: extract the pressure-temperature time series correlation matrix through a pressure-temperature time series pattern correlation feature extractor based on a convolutional neural network model to obtain the pressure-temperature time series pattern correlation feature map.

5. The intelligent gas relay for transformers according to claim 4, characterized in that, The oil level time-series feature extraction module is used to: pass the oil level time-series image through an oil level time-series feature extractor based on a convolutional neural network model to obtain the oil level time-series pattern feature map.

6. The intelligent gas relay for transformers according to claim 5, characterized in that, The attention fusion module includes: A global mean pooling unit is used to perform global mean pooling on each feature matrix along the channel dimension of the pressure-temperature time series pattern association feature map to obtain the pressure-temperature time series pattern association feature vector. A fully connected encoding unit is used to perform fully connected encoding on the pressure-temperature time series pattern association feature vector using a fully connected layer to obtain a fully connected encoded pressure-temperature time series pattern association feature vector. The weighted fusion unit is used to perform weighted fusion along the channel dimension of the fully connected encoded pressure-temperature time series pattern associated feature vector and the oil level time series pattern feature map to obtain the weighted fused oil level time series pattern feature map; The fusion unit is used to fuse the weighted fused oil level timing pattern feature map and the pressure-temperature timing pattern association feature map to obtain the transformer state characterization feature map.

7. The intelligent gas relay for transformers according to claim 6, characterized in that, The transformer fault early warning module includes: A transformer fault judgment unit is used to pass the transformer state characterization feature map through a classifier-based transformer fault detector to obtain a detection result, which is used to indicate whether the transformer has a fault. The warning signal unit is used to generate a warning signal in response to the detection result indicating that there is a fault in the transformer.

8. The intelligent gas relay for transformers according to claim 7, characterized in that, It also includes a training module for training: the pressure-temperature time-series pattern correlation feature extractor based on the convolutional neural network model, the vector-image conversion module, the oil level time-series feature extractor based on the convolutional neural network model, the attention fusion network based on global mean pooling, and the transformer fault detector based on the classifier; The training module includes: The training data acquisition unit is used to acquire training data, which includes time series of training pressure values, training temperature values, and training oil level values ​​collected by the sensor group, as well as the actual value of whether the transformer has a fault. The training data normalization unit is used to normalize the time series of the training pressure value, training temperature value and training oil level value according to the time dimension to obtain the training pressure time series input vector, the training temperature time series input vector and the training oil level time series input vector. The training data gamma correction unit is used to perform gamma correction on the training pressure time-series input vector, the training temperature time-series input vector and the training oil level time-series input vector to obtain the corrected training pressure time-series input vector, the corrected training temperature time-series input vector and the corrected training oil level time-series input vector. The training time series correlation unit is used to calculate the sample covariance correlation matrix between the corrected training pressure time series input vector and the corrected training temperature time series input vector to obtain the training pressure-temperature time series correlation matrix; The training time-series pattern association feature unit is used to pass the training pressure-temperature time-series association matrix through the pressure-temperature time-series pattern association feature extractor based on the convolutional neural network model to obtain the training pressure-temperature time-series pattern association feature map. The training vector-image conversion unit is used to pass the corrected training oil level time series input vector through the vector-image conversion module to obtain a training oil level time series image. The training oil level time series feature extraction unit is used to pass the training oil level time series image through the oil level time series feature extractor based on the convolutional neural network model to obtain the training oil level time series pattern feature map; The training feature fusion unit is used to pass the training pressure-temperature time series pattern association feature map and the training oil level time series pattern feature map through the attention fusion network based on global mean pooling to obtain the training transformer state characterization feature map. The training feature optimization unit is used to optimize the training transformer state representation feature map based on the feature matrix distribution along the channel dimension of the training transformer state representation feature map to obtain an optimized training transformer state representation feature map. The classification loss calculation unit is used to pass the optimized training transformer state representation feature map through the classifier-based transformer fault detector to obtain the classification loss function value; The training unit is used to train the pressure-temperature time-series pattern correlation feature extractor based on the convolutional neural network model, the vector-image conversion module, the oil level time-series feature extractor based on the convolutional neural network model, the attention fusion network based on global mean pooling, and the transformer fault detector based on the classifier, based on the classification loss function value and through backpropagation of gradient descent.

9. The intelligent gas relay for transformers according to claim 8, characterized in that, The training feature optimization unit includes: The probability coefficient calculation subunit is used to perform scenario-based probabilistic logic reasoning on each feature matrix along the channel dimension of the training transformer state representation feature map to obtain a probability vector. The weighted optimization subunit is used to perform weighted optimization on each feature matrix along the channel dimension of the training transformer state representation feature map using the probability coefficient values ​​at each position in the probability vector as weighting coefficients to obtain the optimized training transformer state representation feature map.

10. A smart gas relay method for transformers, characterized in that, include: Acquire time series of pressure, temperature, and oil level values ​​collected by the sensor array; The time series of pressure, temperature and oil level values ​​are normalized according to the time dimension to obtain pressure time series input vector, temperature time series input vector and oil level time series input vector. Gamma correction is performed on the pressure timing input vector, the temperature timing input vector, and the oil level timing input vector to obtain the corrected pressure timing input vector, the corrected temperature timing input vector, and the corrected oil level timing input vector. Calculate the sample covariance correlation matrix between the corrected pressure time-series input vector and the corrected temperature time-series input vector to obtain the pressure-temperature time-series correlation matrix; The pressure-temperature time series correlation matrix is ​​subjected to pressure-temperature time series pattern correlation feature extraction to obtain a pressure-temperature time series pattern correlation feature map; The corrected oil level timing input vector is passed through a vector-image conversion module to obtain an oil level timing image; Oil level time-series images are subjected to oil level time-series feature extraction to obtain oil level time-series pattern feature maps; The pressure-temperature time series pattern feature map and the oil level time series pattern feature map are used by an attention fusion network based on global mean pooling to obtain a transformer state characterization feature map as the transformer state characterization feature. Based on the transformer status characteristics, it is determined whether the transformer has a fault and whether a warning signal should be generated.

Images