Transformer fault diagnosis method and device based on space-time sequence classification algorithm

Abstract

The embodiment of the invention discloses a transformer fault diagnosis method and device based on a space-time sequence classification algorithm, and the method comprises the steps: obtaining a first magnetic flux leakage signal related to the internal time and space of a transformer through a magnetic flux sensor, and carrying out the classification of a transformer fault through a preset space-time sequence classification algorithm model; whether the transformer has a small-turn short-circuit fault is determined, and when the small-turn short-circuit fault occurs, an alarm message is sent to a worker side, so that the worker can timely discover fault hidden dangers, and accident early warning caused by transformer short circuit is provided for a power grid operation company and accidents are prevented. The method comprises the following steps: receiving a first magnetic flux leakage signal sent by a magnetic flux sensor; inputting the first leakage flux signal into a preset space-time sequence classification algorithm model to output a transformer tag; and if the transformer tag is a fault tag, sending an alarm message to a worker side.

Description

Technical Field

[0001] This application relates to the technical field of transformers, and particularly to a transformer fault diagnosis method and device based on a spatio-temporal sequence classification algorithm. Background Art

[0002] Before some major safety accidents occur in a transformer, there will be some precursor signs. Conducting predictive maintenance for faults can avoid the occurrence of accidents and disasters, and avoid economic and even personal safety hazards and losses to power grid companies and users.

[0003] When a small-turn short-circuit fault occurs in a transformer, it will not immediately cause the transformer to stop working, but will continue to work in a state similar to being ill. At this time, the transformer has already had a certain adverse impact on the power system. For example, voltage fluctuations and a decline in power quality. A small-turn short-circuit will cause abnormal local voltage fluctuations, affecting the power supply quality and making the electrical equipment connected to the transformer unable to work properly.

[0004] Although the working parameters of the transformer in this state do not change significantly, if it is allowed to continue to develop, more serious consequences will occur. Specifically, it includes: further damage to the transformer winding, an increase in the short-circuit current due to the short-circuit, resulting in winding deformation and a sharp increase in copper loss, further exacerbating the short-circuit fault, an increase in temperature, the winding insulation will accelerate aging and damage due to high temperature, and ultimately lead to the burning of the winding; the overvoltage generated by the short-circuit may cause the main insulation or longitudinal insulation of the transformer to break down, ultimately leading to more serious fault accidents such as phase-to-phase short-circuit and winding-to-ground short-circuit; the impact force generated by the short-circuit may also damage other components of the transformer, such as the damage of components such as leads and tap changers. The short-circuit fault will also trigger a chain reaction in the power system, resulting in system oscillations, voltage collapse, and even large-scale power outages. Especially when a short-circuit occurs in a transformer near a hub substation or an important transmission line, the impact range is wider.

[0005] It can be seen that if signs of a transformer fault, that is, a small-turn short-circuit, can be diagnosed at the initial stage of the fault occurrence, major property and personal safety losses can be avoided. Therefore, how to identify transformer faults has become a technical problem that needs to be solved urgently by those skilled in the art. Summary of the Invention

[0006] This application provides a transformer fault diagnosis method and device based on a spatio-temporal sequence classification algorithm. The method obtains a first leakage magnetic flux signal regarding the time and space inside the transformer through a magnetic flux sensor, uses a preset spatio-temporal sequence classification algorithm model to determine whether there is a small-turn short-circuit fault in the transformer, and sends an alarm message to the staff terminal when a small-turn short-circuit fault occurs, enabling the staff to discover potential fault hazards in a timely manner, providing an accident warning caused by transformer short-circuit for the power grid operation company, and preventing accidents from occurring.

[0007] In a first aspect, an embodiment of the present application provides a transformer fault diagnosis method based on a spatio-temporal sequence classification algorithm. A plurality of magnetic flux sensors are arranged around the transformer. The method includes: Receiving a first leakage magnetic flux signal sent by the magnetic flux sensor; Inputting the first leakage magnetic flux signal into a preset spatio-temporal sequence classification algorithm model to output a transformer label. The preset spatio-temporal sequence classification algorithm model is trained according to the second leakage magnetic flux signals sent by the magnetic flux sensors in the fault states of small turn-to-turn short circuits occurring at different positions of the transformer and in the normal operation state; If the transformer label is a fault label, send an alarm message to the staff terminal.

[0008] In some embodiments, the transformer is a three-phase transformer. The number of magnetic flux sensors is eight, and the magnetic flux sensors acquire leakage magnetic flux signals in three directions of x, y, and z. The preset spatio-temporal sequence classification algorithm model includes a shape transformation module, a first one-dimensional convolution module, a second one-dimensional convolution module, a message passing graph module, and a fully connected layer module. The training steps of the preset spatio-temporal sequence classification algorithm model include: Based on the operating frequency of the three-phase transformer, determine the operating cycle of the three-phase transformer; In the fault states of small turn-to-turn short circuits occurring at different positions of the three-phase transformer and in the normal operation state, respectively, based on the preset number of acquisitions in one operating cycle of the magnetic flux sensors, acquire the second leakage magnetic flux signals sent by the magnetic flux sensors of the three-phase transformer in the preset operating cycle; Based on the preset number of acquisitions and the preset operating cycle, determine the time slot length; Based on the number of magnetic flux sensors and the number of directions included in the second leakage magnetic flux signal, determine the total number of sensing data in the second leakage magnetic flux signals acquired by all magnetic flux sensors in one operating cycle; Based on the total number of sensing data and the time slot length, form data with a first target matrix shape from the second leakage magnetic flux signal, where the first target matrix shape is batch_size, 24, 128; Input the input data with the first target matrix shape into the shape transformation module to output data with a second target matrix shape, where the second target matrix shape is batch_size×24×8, 1, 16; Pass the data with the second target matrix shape through the first one-dimensional convolution module to output first target data; Input the first target data into the second one-dimensional convolution module to output second target data; Input the second target data into the message passing graph module to output third target data; Input the third target data into the fully connected layer module.

[0009] In some embodiments, the first one-dimensional convolutional module includes a first one-dimensional convolutional layer, a first one-dimensional batch normalization layer, a first activation layer, a first max pooling layer, and a dropout layer; the step of passing the data with the second target matrix shape through the first one-dimensional convolutional module to output first target data includes: Input the data with the second target matrix shape into the first one-dimensional convolutional layer to output data with a third target matrix shape, where the third target matrix shape is batch_size×24×8,64,16; Input the data with the third target matrix shape into the first one-dimensional batch normalization layer; Input the data output from the first one-dimensional batch normalization layer into the first activation layer; Input the data output from the first activation layer into the first max pooling layer to output data with a fourth target matrix shape, where the fourth target matrix shape is batch_size×24×8,64,9; Input the data with the fourth target matrix shape into the dropout layer to output first target data.

[0010] In some embodiments, the second one-dimensional convolutional module includes a second one-dimensional convolutional layer, a second one-dimensional batch normalization layer, a second activation layer, a second max pooling layer, a first shape transformation unit, a first linear layer, a third one-dimensional batch normalization layer, and a second shape transformation unit; the step of inputting the first target data into the second one-dimensional convolutional module to output second target data includes: Input the first target data into the second one-dimensional convolutional layer to output data with a fifth target matrix shape, where the fifth target matrix shape is batch_size×24×8,18,7; Input the data with the fifth target matrix shape into the second one-dimensional batch normalization layer; Input the data output from the second one-dimensional batch normalization layer into the second activation layer; Input the data output from the second activation layer into the second max pooling layer to output data with a sixth target matrix shape, where the sixth target matrix shape is batch_size×24×8,18,4; Input the data with the shape of the sixth target matrix into the first shape transformation unit to output data with the shape of the seventh target matrix, where the shape of the seventh target matrix is batch_size×24,576; Input the data with the shape of the seventh target matrix into the first linear layer to output data with the shape of the eighth target matrix, where the shape of the eighth target matrix is batch_size×24,32; Input the data with the shape of the eighth target matrix into the third one-dimensional batch normalization layer; Input the data output by the third one-dimensional batch normalization layer into the second shape transformation unit to output the second target data, where the matrix shape of the second target data is batch_size,24,32.

[0011] In some embodiments, the message passing graph module includes a relational matrix unit, a second linear layer, a fourth one-dimensional batch normalization layer, and a third activation layer; the step of inputting the second target data into the message passing graph module to output the third target data includes: Input the second target data into the relational matrix unit to generate a relational matrix; Perform a matrix multiplication operation on the relational matrix and the second target data to obtain data with the shape of the tenth target matrix, where the shape of the tenth target matrix is batch_size,24,32; Input the data with the shape of the tenth target matrix into the second linear layer to obtain data with the shape of the eleventh target matrix, where the shape of the eleventh target matrix is batch_size,24,16; Input the data with the shape of the eleventh target matrix into the fourth one-dimensional batch normalization layer; Input the data output by the fourth one-dimensional batch normalization layer into the third activation layer to obtain the third target data.

[0012] In some embodiments, the relational matrix unit includes a fourth activation layer and a softmax layer; the step of inputting the second target data into the relational matrix unit to generate a relational matrix includes: Transpose the second target data, perform a matrix multiplication on the transposed second target data and the second target data to obtain data with the shape of the twelfth target matrix, where the shape of the twelfth target matrix is batch_size×24×24; Perform a dot product operation on the data with the shape of the twelfth target matrix and the attenuation matrix, and then input the result into the fourth activation layer; Input the data output by the fourth activation layer into the softmax layer to generate a relational matrix.

[0013] In some embodiments, the fully connected layer module includes a third shape change unit, a third linear layer, a fourth linear layer, a fifth linear layer, a fifth activation layer, a sixth activation layer, and a seventh activation layer; the step of inputting the third target data into the fully connected layer module includes: Input the third target data into the third shape change unit to output data with a thirteenth target matrix shape, where the thirteenth target matrix shape is batch_size, 384; Input the data with the thirteenth target matrix shape into the fifth activation layer through the third linear layer to obtain data with a fourteenth target matrix shape, where the fourteenth target matrix shape is batch_size, 128; Input the data with the fourteenth target matrix shape into the sixth activation layer through the fourth linear layer to obtain data with a fifteenth target matrix shape, where the fifteenth target matrix shape is batch_size, 64; Input the data with the fifteenth target matrix shape into the seventh activation layer through the fifth linear layer to obtain the third target data, where the shape of the third target data is batch_size, 28.

[0014] In some embodiments, the fault labels include high-voltage U-phase high short circuit, high-voltage U-phase middle short circuit, high-voltage U-phase low short circuit, high-voltage V-phase high short circuit, high-voltage V-phase middle short circuit, high-voltage V-phase low short circuit, high-voltage W-phase high short circuit, high-voltage W-phase middle short circuit, high-voltage W-phase low short circuit, medium-voltage U-phase high short circuit, medium-voltage U-phase middle short circuit, medium-voltage U-phase low short circuit, medium-voltage V-phase high short circuit, medium-voltage V-phase middle short circuit, medium-voltage V-phase low short circuit, medium-voltage W-phase high short circuit, medium-voltage W-phase middle short circuit, medium-voltage W-phase low short circuit, low-voltage U-phase high short circuit, low-voltage U-phase middle short circuit, low-voltage U-phase low short circuit, low-voltage V-phase high short circuit, low-voltage V-phase middle short circuit, low-voltage V-phase low short circuit, low-voltage W-phase high short circuit, low-voltage W-phase middle short circuit, or low-voltage W-phase low short circuit.

[0015] In some embodiments, the four magnetic flux sensors are evenly distributed on one side of the three-phase transformer and are centrosymmetric with the remaining four magnetic flux sensors about the coil center point of the three-phase transformer.

[0016] In a second aspect, an embodiment of the present application provides a transformer fault diagnosis device based on a spatio-temporal sequence classification algorithm, with a plurality of magnetic flux sensors arranged around the transformer; the device includes: A receiving unit for receiving the first leakage magnetic flux signal sent by the magnetic flux sensor; An input unit for inputting the first leakage magnetic flux signal into a preset spatio-temporal sequence classification algorithm model to output a transformer label; wherein the preset spatio-temporal sequence classification algorithm model is trained according to the second leakage magnetic flux signals sent by the magnetic flux sensors in the fault states of small turn-to-turn short circuits occurring at different positions of the transformer and in the normal operating state. A sending unit for sending an alarm message to the staff terminal if the transformer label is a fault label.

[0017] In the above embodiment, a transformer fault diagnosis method and device based on a spatio-temporal sequence classification algorithm are provided. The method obtains a first leakage magnetic flux signal regarding the time and space inside the transformer through a magnetic flux sensor, determines whether there is a small turn-to-turn short circuit fault in the transformer by using a preset spatio-temporal sequence classification algorithm model, and sends an alarm message to the staff terminal when a small turn-to-turn short circuit fault occurs, enabling the staff to discover potential fault hazards in a timely manner, providing accident warnings caused by transformer short circuits for the power grid operation company, and preventing accidents from occurring. The method includes: receiving the first leakage magnetic flux signal sent by the magnetic flux sensor; inputting the first leakage magnetic flux signal into a preset spatio-temporal sequence classification algorithm model to output a transformer label; wherein the preset spatio-temporal sequence classification algorithm model is trained according to the second leakage magnetic flux signals sent by the magnetic flux sensors in the fault states of small turn-to-turn short circuits occurring at different positions of the transformer and in the normal operating state; and sending an alarm message to the staff terminal if the transformer label is a fault label. Description of the Drawings

[0018] Figure 1 Exemplarily shows a flowchart of a transformer fault diagnosis method based on a spatio-temporal sequence classification algorithm according to some embodiments; Figure 2 Exemplarily shows a schematic diagram of the distribution of magnetic flux sensors according to some embodiments; Figure 3 Exemplarily shows a flowchart of another transformer fault diagnosis method based on a spatio-temporal sequence classification algorithm according to some embodiments; Figure 4 Exemplarily shows a schematic structural diagram of a transformer fault diagnosis device based on a spatio-temporal sequence classification algorithm according to some embodiments. Detailed Embodiments

[0019] To make the objectives and implementation manners of the present application clearer, the following will clearly and completely describe the exemplary implementation manners of the present application with reference to the drawings in the exemplary embodiments of the present application. Obviously, the described exemplary embodiments are only a part of the embodiments of the present application, rather than all of the embodiments.

[0020] It should be noted that the brief description of the terms in this application is only for facilitating the understanding of the following described embodiments, rather than intending to limit the embodiments of this application. Unless otherwise specified, these terms should be understood in their ordinary and common meanings.

[0021] In this application, the terms "first", "second", "third", etc. in the specification, claims and the above-mentioned drawings are used to distinguish similar or like objects or entities, and do not necessarily mean to limit a specific order or sequence, unless otherwise noted. It should be understood that such terms can be interchanged under appropriate circumstances.

[0022] The terms "comprising" and "having" and any variations thereof are intended to cover but not exclusively include. For example, a product or device comprising a series of components does not necessarily have to be limited to all the components clearly listed, but may include other components not clearly listed or inherent to these products or devices.

[0023] Before some major safety accidents occur in a transformer, there will be some precursor signs. Conducting fault predictive maintenance can avoid the occurrence of accidents and disasters, and avoid the economic and even personal safety hazards and losses of power grid companies and users.

[0024] When a small-turn short-circuit fault occurs in a transformer, it will not immediately cause the transformer to stop working, but will continue to work in a state similar to an abnormal state. At this time, the transformer has already had a certain adverse impact on the power system. For example: voltage fluctuations and the decline of power quality. A small-turn short-circuit will cause abnormal local voltage fluctuations, affect the power supply quality, and make the electrical equipment connected to this transformer unable to work properly.

[0025] Although the working parameters of the transformer in this state do not change significantly, if it is allowed to continue to develop, more serious consequences will occur. Specifically, it includes: further damage to the transformer winding, an increase in the short-circuit current and the short-circuit current rising, causing the winding to deform and the copper loss to increase sharply, further exacerbating the short-circuit fault, the temperature rising, the winding insulation will age and be damaged due to high temperature, and finally the winding will be burned out; the overvoltage generated by the short-circuit may cause the main insulation or longitudinal insulation of the transformer to break down, and finally lead to more serious fault accidents such as phase-to-phase short-circuit and winding-to-ground short-circuit; the impact force generated by the short-circuit may also damage other components of the transformer, such as the damage of components such as leads and tap changers. The short-circuit fault will also trigger a chain reaction in the power system, resulting in system oscillation, voltage collapse, and even large-scale power outages. Especially when a short-circuit occurs in a transformer near a hub substation or an important transmission line, its influence range is wider.

[0026] It can be seen that if signs can be diagnosed at the initial stage of transformer faults, namely small-turn short circuits, major property and personal safety losses can be avoided. Therefore, how to identify transformer faults has become a technical problem that needs to be solved urgently by those skilled in the art.

[0027] To solve the above technical problem, the embodiment of the present application provides a transformer fault diagnosis method based on a spatio-temporal sequence classification algorithm. The method obtains a first leakage magnetic flux signal about the time and space inside the transformer through a magnetic flux sensor, and uses a preset spatio-temporal sequence classification algorithm model to determine whether there is a small-turn short circuit fault in the transformer. When a small-turn short circuit fault occurs, an alarm message is sent to the staff terminal, so that the staff can timely discover potential faults, provide accident warnings caused by transformer short circuits for the power grid operation company, and prevent accidents from occurring. The method includes: receiving the first leakage magnetic flux signal sent by the magnetic flux sensor; inputting the first leakage magnetic flux signal into a preset spatio-temporal sequence classification algorithm model to output a transformer label; wherein the preset spatio-temporal sequence classification algorithm model is trained according to the fault states of small-turn short circuits occurring at different positions of the transformer and the second leakage magnetic flux signals sent by the magnetic flux sensor under normal operating conditions; if the transformer label is a fault label, an alarm message is sent to the staff terminal.

[0028] Figure 1 Exemplarily shows a flowchart of a transformer fault diagnosis method based on a spatio-temporal sequence classification algorithm provided according to some embodiments. The method includes S100 - S300.

[0029] S100, receiving the first leakage magnetic flux signal sent by the magnetic flux sensor.

[0030] In the embodiment of the present application, the first leakage magnetic flux signal is the leakage magnetic flux signal collected by the magnetic flux sensor when the transformer is working.

[0031] In the embodiment of the present application, multiple magnetic flux sensors are arranged around the transformer. In one example, eight magnetic flux sensors can be set. In some embodiments, the transformer is a three-phase transformer. Four of the eight magnetic flux sensors are evenly distributed on one side of the three-phase transformer and are centrosymmetric with the remaining four magnetic flux sensors about the coil center point of the three-phase transformer. As Figure 2 shown, Figure 2 shows the positions of the eight magnetic flux sensors.

[0032] In the embodiment of the present application, the transformer fault diagnosis method based on the spatio-temporal sequence classification algorithm can be built into the transformer terminal unit.

[0033] S200. Input the first leakage magnetic flux signal into a preset spatio-temporal sequence classification algorithm model to output a transformer label, where the preset spatio-temporal sequence classification algorithm model is trained based on the second leakage magnetic flux signals sent by the magnetic flux sensors in the fault states of small turn-to-turn short circuits at different positions of the transformer and in the normal operation state.

[0034] In some embodiments, the step of inputting the first leakage magnetic flux signal into a preset spatio-temporal sequence classification algorithm model to output a transformer label includes: forming target data in the shape of a first target matrix based on the total number of sensing data and the time slot length of the first leakage magnetic flux signal, and inputting the target data into the preset spatio-temporal sequence classification algorithm model to output a transformer label. The determination methods of the total number of sensing data and the time slot length are described in detail below and will not be elaborated here. In this embodiment, the purpose of forming the target data of the first leakage magnetic flux signal in the shape of a first target matrix is to convert the first leakage magnetic flux signal into the same format as the sample data used for training the preset spatio-temporal sequence classification algorithm model.

[0035] In some embodiments, the transformer is a three-phase transformer; the number of magnetic flux sensors is eight, and the magnetic flux sensors acquire leakage magnetic flux signals in three directions of x, y, and z; the preset spatio-temporal sequence classification algorithm model includes a shape transformation module, a first one-dimensional convolution module, a second one-dimensional convolution module, a message passing graph module, and a fully connected layer module. Figure 3 Exemplarily shows a flowchart of another transformer fault diagnosis method based on a spatio-temporal sequence classification algorithm provided according to some embodiments. The training steps of the preset spatio-temporal sequence classification algorithm model include S400 - S1300: S400. Determine the operating period of the three-phase transformer based on the operating frequency of the three-phase transformer.

[0036] In one example, the operating frequency of the three-phase transformer is 50 Hz, and the operating period of the three-phase transformer is 1 / 50 second.

[0037] S500. In the fault states of small turn-to-turn short circuits at different positions of the three-phase transformer and in the normal operation state, respectively, based on the preset number of acquisitions of the magnetic flux sensors in one operating period, acquire the second leakage magnetic flux signals sent by the magnetic flux sensors within the preset operating period.

[0038] In this embodiment, in order to obtain a preset spatio-temporal sequence classification algorithm model that can accurately predict the transformer label, the samples used for training the preset spatio-temporal sequence classification algorithm model include both the second leakage magnetic flux signals in the fault states of small turn-to-turn short circuits at different positions of the three-phase transformer and the second leakage magnetic flux signals in the normal operation state.

[0039] In this embodiment, the second leakage flux signal can be actually collected or obtained by simulation software.

[0040] In this embodiment, the transformer labels include a fault label and a normal operation label. The fault label includes high-voltage U-phase short circuit at a high position, high-voltage U-phase short circuit at a middle position, high-voltage U-phase short circuit at a low position, high-voltage V-phase short circuit at a high position, high-voltage V-phase short circuit at a middle position, high-voltage V-phase short circuit at a low position, high-voltage W-phase short circuit at a high position, high-voltage W-phase short circuit at a middle position, high-voltage W-phase short circuit at a low position, medium-voltage U-phase short circuit at a high position, medium-voltage U-phase short circuit at a middle position, medium-voltage U-phase short circuit at a low position, medium-voltage V-phase short circuit at a high position, medium-voltage V-phase short circuit at a middle position, medium-voltage V-phase short circuit at a low position, medium-voltage W-phase short circuit at a high position, medium-voltage W-phase short circuit at a middle position, medium-voltage W-phase short circuit at a low position, low-voltage U-phase short circuit at a high position, low-voltage U-phase short circuit at a middle position, low-voltage U-phase short circuit at a low position, low-voltage V-phase short circuit at a high position, low-voltage V-phase short circuit at a middle position, low-voltage V-phase short circuit at a low position, low-voltage W-phase short circuit at a high position, low-voltage W-phase short circuit at a middle position, or low-voltage W-phase short circuit at a low position.

[0041] In one example, the preset number of acquisitions of the magnetic flux sensor in one working cycle is 16 times. The preset working cycle is eight working cycles. At this time, in the fault states and normal operation states of small-turn short circuits occurring at different positions of the three-phase transformer, based on the preset number of acquisitions of the magnetic flux sensor in one working cycle, that is, 16 times, the second leakage flux signals sent by the magnetic flux sensor within eight working cycles are obtained.

[0042] S600. Determine the time slot length based on the preset number of acquisitions and the preset working cycle.

[0043] In this embodiment, the preset number of acquisitions can be 16 times, and the preset acquisition period is 8. At this time, the time slot length is 16×8, that is, 128.

[0044] S700. Determine the total number of sensing data in the second leakage flux signals collected by all magnetic flux sensors in one working cycle based on the number of magnetic flux sensors and the number of directions included in the second leakage flux signal.

[0045] In this embodiment, determining the total number of sensing data in the second leakage flux signals collected by all magnetic flux sensors in one working cycle based on the number of magnetic flux sensors and the number of directions included in the second leakage flux signal may include: multiplying the number of magnetic flux sensors by the number of directions included in the second leakage flux signal to obtain the total number of sensing data in the second leakage flux signals collected by all magnetic flux sensors in one working cycle. In this embodiment, the second leakage flux signal includes sub-signals in three directions, x, y, and z. Therefore, the number of directions included in the second leakage flux signal is three. The number of magnetic flux sensors is eight. At this time, the total number of sensing data is 24.

[0046] S800. Based on the total number of the sensing data and the time slot length, form data with a first target matrix shape from the second leakage flux signal, where the first target matrix shape is batch_size, 24, 128.

[0047] In the embodiments of the present application, a large number of second leakage flux signals can be obtained to form data with multiple first target matrix shapes. The model is trained using the data with multiple first target matrix shapes. The present application can also obtain other second leakage flux signals as a validation data set to verify the accuracy of the model using the validation data set. In this embodiment, the ratio of the number of second leakage flux signals in the training data set to the number of second leakage flux signals in the validation data set can be 4:1. It can be understood that the more the number of second leakage flux signals, the more comprehensive the coverage of defects and normal operating conditions, and the better the training effect.

[0048] S900. Input the input data with the first target matrix shape into the shape transformation module to output data with a second target matrix shape, where the second target matrix shape is batch_size×24×8, 1, 16.

[0049] In this embodiment, the main function of the shape transformation module is to segment the data with the first target matrix shape according to the periodic characteristics of each magnetic flux sensor and reorganize them into a matrix with a new shape in sequence, so as to facilitate further feature extraction in the follow-up. In this embodiment, the first target matrix shape is batch_size, 24, 128, where 24 indicates that there are 8 magnetic flux sensors in total, and the second leakage flux signals collected by each magnetic flux sensor include data in 3 directions. Multiply the two to obtain the total number of sensing data, that is, 24. Each magnetic flux sensor collects data 16 times within one working cycle of each magnetic flux transformer, and the algorithm inputs data of 8 three-phase transformer working cycles each time. Therefore, for 1 direction of each magnetic flux sensor, the algorithm needs to input 128 time slot data. The shape transformation module further divides the 128 time slot data into a shape of 8×16 according to the working characteristics of inputting 8 working cycles each time and 16 time slot data within each working cycle, and then integrates the previous data dimensions. Therefore, after the input data with the shape of batch_size, 24, 128 passes through the shape transformation module, the matrix shape becomes batch_size×24×8, 1, 16. The extra dimension with a shape of 1 in the middle here is for storing the features to be extracted subsequently.

[0050] S1000. Pass the data with the second target matrix shape through the first one-dimensional convolution module to output the first target data.

[0051] In some embodiments, the first one-dimensional convolutional module includes a first one-dimensional convolutional layer, a first one-dimensional batch normalization layer, a first activation layer, a first max pooling layer, and a dropout layer; the step of passing the data of the second target matrix shape through the first one-dimensional convolutional module to output first target data includes: Input the data of the second target matrix shape into the first one-dimensional convolutional layer to output data of a third target matrix shape, where the third target matrix shape is batch_size×24×8,64,16.

[0052] In this embodiment, the main function of the first one-dimensional convolutional layer is to extract more data features in the time dimension. After the input data passes through this layer, the matrix shape changes from batch_size×24×8,1,16 to batch_size×24×8,64,16. Here, 64 indicates that 64 data features are extracted from every 16 time slot data.

[0053] Input the data of the third target matrix shape into the first one-dimensional batch normalization layer.

[0054] In this embodiment, the main function of the first one-dimensional batch normalization layer is to standardize the input distribution, which can reduce overfitting of the neural network and improve the generalization ability of the algorithm.

[0055] Input the data output from the first one-dimensional batch normalization layer into the first activation layer.

[0056] In this embodiment, the first activation layer can solve the problem of vanishing gradients during model training and enhance the model's learning ability in terms of non-linearity. The first activation layer does not change the data dimension and shape.

[0057] Input the data output from the first activation layer into the first max pooling layer to output data of a fourth target matrix shape, where the fourth target matrix shape is batch_size×24×8,64,9; In this embodiment, the function of the first max pooling layer is to reduce the data dimension, reduce the amount of calculation, and can eliminate a part of the noise. After passing through the first max pooling layer, the data shape becomes batch_size×24×8,64,9.

[0058] Input the data of the fourth target matrix shape into the dropout layer to output the first target data.

[0059] In this embodiment, the dropout layer can improve the generalization ability of the neural network and reduce the risk of overfitting. The dropout layer also does not change the data dimension shape.

[0060] S1100. Input the first target data into the second one-dimensional convolutional module to output second target data; In this embodiment, the main function of the second one-dimensional convolutional module is to compress and refine data features and prepare for the next step of data entering the message passing graph module.

[0061] The second one-dimensional convolutional module includes a second one-dimensional convolutional layer, a second one-dimensional batch normalization layer, a second activation layer, a second max pooling layer, a first shape transformation unit, a first linear layer, a third one-dimensional batch normalization layer, and a second shape transformation unit. The step of inputting the first target data into the second one-dimensional convolutional module to output second target data includes: Input the first target data into the second one-dimensional convolutional layer to output data with a fifth target matrix shape, where the fifth target matrix shape is batch_size×24×8,18,7; In this embodiment, the second one-dimensional convolutional layer compresses and refines data features. After the data passes through this second one-dimensional convolutional layer, the shape becomes batch_size×24×8,18,7.

[0062] Input the data with the fifth target matrix shape into the second one-dimensional batch normalization layer; In this embodiment, the function of the second one-dimensional batch normalization layer is to standardize the data distribution and reduce overfitting.

[0063] Input the data output by the second one-dimensional batch normalization layer into the second activation layer; In this embodiment, the function of the second activation layer is still to reduce the risk of gradient vanishing and enhance the model's non-linear learning ability. The data dimension shape does not change after passing through the second one-dimensional batch normalization layer and the second activation layer.

[0064] Input the data output by the second activation layer into the second max pooling layer to output data with a sixth target matrix shape, where the sixth target matrix shape is batch_size×24×8,18,4; In this embodiment, the main function of the second max pooling layer is data dimensionality reduction and further extraction of main information. After the data passes through the second max pooling layer, the dimension shape becomes batch_size×24×8,18,4.

[0065] Input the data with the sixth target matrix shape into the first shape transformation unit to output data with a seventh target matrix shape, where the seventh target matrix shape is batch_size×24,576; In this embodiment, the first shape transformation unit mainly flattens the other data in the direction of each magnetic flux sensor in the batch_size. After the data undergoes this shape transformation, the dimension shape becomes batch_size×24,576.

[0066] Input the data with the shape of the seventh target matrix into the first linear layer to output data with the shape of the eighth target matrix, where the shape of the eighth target matrix is batch_size×24,32; In this embodiment, the role of the first linear layer is to further reduce the data dimension and refine the data features. After passing through this first linear layer, the data dimension shape becomes batch_size×24,32.

[0067] Input the data with the shape of the eighth target matrix into the third one-dimensional batch normalization layer; In this embodiment, the role of the third one-dimensional batch normalization layer is still to standardize the data distribution and reduce overfitting, and this layer does not change the data dimension shape.

[0068] Input the data output by the third one-dimensional batch normalization layer into the second shape transformation unit to output the second target data, where the matrix shape of the second target data is batch_size,24,32; In this embodiment, the last second shape transformation unit changes the data dimension shape to batch_size,24,32 to prepare for the subsequent model calculation.

[0069] S1200. Input the second target data into the message passing graph module to output the third target data.

[0070] In some embodiments, the message passing graph module includes a relationship matrix unit, a second linear layer, a fourth one-dimensional batch normalization layer, and a third activation layer; the step of inputting the second target data into the message passing graph module to output the third target data includes: Input the second target data into the relationship matrix unit to generate a relationship matrix.

[0071] In some embodiments, the relationship matrix unit includes a fourth activation layer and a softmax layer; the step of inputting the second target data into the relationship matrix unit to generate a relationship matrix includes: Transpose the second target data, perform matrix multiplication on the transposed second target data and the second target data to obtain data with the shape of the twelfth target matrix, where the shape of the twelfth target matrix is batch_size×24×24; After performing a dot product operation on the data in the shape of the twelfth target matrix and the attenuation matrix, input it into the fourth activation layer; Input the data output by the fourth activation layer into the softmax layer to generate a relationship matrix.

[0072] In this embodiment, first, the output result of the second one-dimensional convolution module is transposed and then multiplied by itself to obtain a matrix with a dimension shape of batch_size×24×24. This operation calculates the characteristic relationship matrix of each magnetic flux sensor with other magnetic flux sensors in each time slot in a multiplicative manner, that is, the data in the shape of the twelfth target matrix. This matrix then performs a dot product operation with an attenuation matrix to weaken the relationship with magnetic flux sensors that are relatively far away from their respective magnetic flux sensors. The attenuation matrix is a square matrix with the same number of rows and columns as the characteristic relationship matrix, with 1 on the diagonal and gradually decreasing to 0 on both sides, and the dimension shape is batch_size, 24, 24. Subsequently, the result passes through the fourth activation layer. This fourth activation layer uses leaky_relu, and leaky_relu can also retain a very small value when the value is negative to avoid the situation of neuron death. Finally, through the Softmax layer, the data in the direction of magnetic flux sensors with relatively close spatio-temporal relationships is further strengthened and the data in the direction of magnetic flux sensors with relatively far spatio-temporal relationships is weakened.

[0073] Perform a matrix multiplication operation on the relationship matrix and the second target data to obtain data in the shape of the tenth target matrix, and the shape of the tenth target matrix is batch_size, 24, 32; In this embodiment, after the relationship matrix is calculated, it performs a matrix multiplication operation with the output result of the second one-dimensional convolution module. At this time, the data dimension shape is batch_size, 24, 32.

[0074] Input the data in the shape of the tenth target matrix into the second linear layer to obtain data in the shape of the eleventh target matrix, and the shape of the eleventh target matrix is batch_size, 24, 16; In this embodiment, the role of the second linear layer is to reduce the dimension and compress the data features. At this time, the data dimension shape is batch_size, 24, 16.

[0075] Input the data in the shape of the eleventh target matrix into the fourth one-dimensional batch normalization layer; Input the data output by the fourth one-dimensional batch normalization layer into the third activation layer to obtain the third target data.

[0076] In this embodiment, the data in the shape of the eleventh target matrix is input into the third activation layer through the fourth one-dimensional batch normalization layer. The functions of these two layers are the same as those before, which are to standardize the data distribution and improve the non-linear learning ability and the generalization ability of the model respectively.

[0077] S1300. Input the third target data into the fully connected layer module.

[0078] The fully connected layer module includes a third shape change unit, a third linear layer, a fourth linear layer, a fifth linear layer, a fifth activation layer, a sixth activation layer, and a seventh activation layer; the step of inputting the third target data into the fully connected layer module includes: Input the third target data into the third shape change unit to output data with a thirteenth target matrix shape, where the thirteenth target matrix shape is batch_size, 384; Input the data with the thirteenth target matrix shape into the fifth activation layer through the third linear layer to obtain data with a fourteenth target matrix shape, where the fourteenth target matrix shape is batch_size, 128; Input the data with the fourteenth target matrix shape into the sixth activation layer through the fourth linear layer to obtain data with a fifteenth target matrix shape, where the fifteenth target matrix shape is batch_size, 64; Input the data with the fifteenth target matrix shape into the seventh activation layer through the fifth linear layer to obtain the third target data, where the shape of the third target data is batch_size, 28.

[0079] In this embodiment, the functions of the linear layer and the activation layer are to compress and extract data features and improve the generalization ability of the model respectively. After the data passes through the combination of 3 linear layers and activation layers, the dimension shapes change to batch_size, 128, batch_size, 64, and batch_size, 28 in sequence.

[0080] In this embodiment, the accuracy rate of the trained model is basically above 99.5%.

[0081] S300. If the transformer label is a fault label, send an alarm message to the staff terminal.

[0082] In the embodiment of the present application, the transformer label includes a fault label and a normal operation label. When the transformer label output by the model is a fault label, it indicates that there is a short circuit with a small number of turns in the transformer. At this time, send an alarm message to the staff terminal. The staff can view the alarm message through the staff terminal and determine the specific location of the short circuit with a small number of turns in the transformer according to the specific content of the fault label. Exemplarily, when the fault label is a high - altitude short circuit in phase U, it indicates that the high - altitude position of the high - voltage phase U coil is short - circuited. The staff can repair the high - altitude position of the high - voltage phase U coil.

[0083] The method in the embodiments of the present application has the following advantages: 1. Improve the safety of the transformer. The method in the embodiments of the present application can detect abnormalities in a timely manner when a small-turn short-circuit fault occurs in the transformer. After the grid staff obtains the abnormality and makes timely handling, it can effectively avoid the further expansion of the fault and ensure the safety of personnel's lives and the property of the substation. 2. Reduce the losses caused by the fault. The method in the embodiments of the present application can alarm at the first time when a fault occurs and inform the fault location, providing accurate fault location information for the maintenance personnel, improving the fault handling efficiency, reducing the power outage time caused by fault troubleshooting. It reduces the maintenance cost, also reduces the power outage time of users caused by the fault, and reduces the losses brought by the fault to the power grid.

[0084] In the above embodiments, a method and device for diagnosing early turn-to-turn short-circuit faults of a transformer based on magnetic flux leakage analysis are provided. The method diagnoses faults by monitoring the magnetic field changes during the operation of the transformer, without the need to disassemble or damage the transformer, and has the advantage of non-invasiveness; the method can also realize real-time detection of the magnetic field changes during the operation of the transformer and quickly diagnose faults, and has the advantage of strong real-time performance; combining signal processing technology and machine learning algorithms for fault diagnosis has high accuracy; in addition, the method is applicable to various types and sizes of transformers for turn-to-turn short-circuit fault triage and has wide applicability.

[0085] The embodiments of the present application further provide a transformer fault diagnosis device based on a spatio-temporal sequence classification algorithm, and a plurality of magnetic flux sensors are arranged around the transformer. Figure 4 Exemplarily shows a schematic structural diagram of a transformer fault diagnosis device based on a spatio-temporal sequence classification algorithm provided according to some embodiments. The device includes: a receiving unit 401, an input unit 402, and a sending unit 403.

[0086] The receiving unit 401 is configured to receive a first magnetic flux leakage signal sent by the magnetic flux sensor; The input unit 402 is configured to input the first magnetic flux leakage signal into a preset spatio-temporal sequence classification algorithm model to output a transformer label; wherein the preset spatio-temporal sequence classification algorithm model is trained according to the second magnetic flux leakage signals sent by the magnetic flux sensors in the fault states and normal operation states of small-turn short circuits occurring at different positions of the transformer; The sending unit 403 is configured to send an alarm message to the staff terminal if the transformer label is a fault label.

[0087] The embodiments of the present application further provide an electronic device, including a memory and a processor, where the memory stores a computer program, and the processor implements the steps of the transformer fault diagnosis method based on the spatio-temporal sequence classification algorithm when executing the computer program.

[0088] An embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the transformer fault diagnosis method based on the spatio-temporal sequence classification algorithm are implemented.

[0089] Through the description of the above embodiments, those skilled in the art can clearly understand that the present application can be implemented by means of software plus a necessary general hardware platform, or can be implemented by hardware.

[0090] Those skilled in the art can understand that the drawings are only schematic diagrams of a preferred embodiment scenario, and the units or processes in the drawings are not necessarily essential for implementing the present application. Those skilled in the art can understand that the units in the device in the embodiment scenario can be distributed in the device in the embodiment scenario according to the description of the embodiment scenario, or can be correspondingly changed and located in one or more devices different from this embodiment scenario. The units in the above embodiment scenario can be combined into one unit, or can be further split into multiple sub-units.

[0091] The above serial numbers of the present application are only for description and do not represent the advantages or disadvantages of the embodiment scenarios. The above-disclosed are only several specific embodiment scenarios of the present application. However, the present application is not limited thereto, and any changes that can be thought of by those skilled in the art should fall within the protection scope of the present application.

Claims

1. A transformer fault diagnosis method based on a spatio-temporal sequence classification algorithm, characterized in that, A plurality of magnetic flux sensors are arranged around the transformer; the method includes: Receiving a first leakage magnetic flux signal sent by the magnetic flux sensors; Inputting the first leakage magnetic flux signal into a preset spatio-temporal sequence classification algorithm model to output a transformer label; wherein the preset spatio-temporal sequence classification algorithm model is trained according to the second leakage magnetic flux signals sent by the magnetic flux sensors in the fault state of small turn-to-turn short circuits occurring at different positions of the transformer and in the normal operation state; If the transformer label is a fault label, send an alarm message to the staff terminal.

2. The method according to claim 1, wherein The transformer is a three-phase transformer; the number of the magnetic flux sensors is eight, and the magnetic flux sensors acquire leakage magnetic flux signals in three directions of x, y, and z; the preset spatio-temporal sequence classification algorithm model includes a shape transformation module, a first one-dimensional convolution module, a second one-dimensional convolution module, a message passing graph module, and a fully connected layer module; The training steps of the preset spatio-temporal sequence classification algorithm model include: Based on the operating frequency of the three-phase transformer, determine the operating cycle of the three-phase transformer; In the fault state of small turn-to-turn short circuits occurring at different positions of the three-phase transformer and in the normal operation state, respectively based on the preset number of acquisitions of the magnetic flux sensors in one operating cycle, acquire the second leakage magnetic flux signals sent by the magnetic flux sensors of the three-phase transformer within the preset operating cycle; Based on the preset number of acquisitions and the preset operating cycle, determine the time slot length; Based on the number of the magnetic flux sensors and the number of directions included in the second leakage magnetic flux signal, determine the total number of sensing data in the second leakage magnetic flux signals acquired by all the magnetic flux sensors in one operating cycle; Based on the total number of sensing data and the time slot length, form data with a first target matrix shape from the second leakage magnetic flux signal, wherein the first target matrix shape is batch_size, 24, 128; Input the input data with the first target matrix shape into the shape transformation module to output data with a second target matrix shape, wherein the second target matrix shape is batch_size×24×8, 1, 16; Pass the data with the second target matrix shape through the first one-dimensional convolution module to output first target data; Input the first target data into the second one-dimensional convolution module to output second target data; Input the second target data into the message passing graph module to output third target data; Input the third target data into the fully connected layer module.

3. The method according to claim 2, wherein The first one-dimensional convolution module includes a first one-dimensional convolution layer, a first one-dimensional batch normalization layer, a first activation layer, a first max pooling layer, and a dropout layer; The step of passing the data with the second target matrix shape through the first one-dimensional convolution module to output first target data includes: Input the data with the second target matrix shape into the first one-dimensional convolution layer to output data with a third target matrix shape, and the third target matrix shape is batch_size×24×8, 64, 16; Input the data with the shape of the third target matrix into the first one-dimensional batch normalization layer; Input the data output from the first one-dimensional batch normalization layer into the first activation layer; Input the data output from the first activation layer into the first max pooling layer to output data with the shape of the fourth target matrix, where the shape of the fourth target matrix is batch_size×24×8,64,9; Input the data with the shape of the fourth target matrix into the dropout layer to output the first target data.

4. The method according to claim 3, wherein The second one-dimensional convolutional module includes a second one-dimensional convolutional layer, a second one-dimensional batch normalization layer, a second activation layer, a second max pooling layer, a first shape transformation unit, a first linear layer, a third one-dimensional batch normalization layer, and a second shape transformation unit; The step of inputting the first target data into the second one-dimensional convolutional module to output the second target data includes: Input the first target data into the second one-dimensional convolutional layer to output data with the shape of the fifth target matrix, where the shape of the fifth target matrix is batch_size×24×8,18,7; Input the data with the shape of the fifth target matrix into the second one-dimensional batch normalization layer; Input the data output from the second one-dimensional batch normalization layer into the second activation layer; Input the data output from the second activation layer into the second max pooling layer to output data with the shape of the sixth target matrix, where the shape of the sixth target matrix is batch_size×24×8,18,4; Input the data with the shape of the sixth target matrix into the first shape transformation unit to output data with the shape of the seventh target matrix, where the shape of the seventh target matrix is batch_size×24,576; Input the data with the shape of the seventh target matrix into the first linear layer to output data with the shape of the eighth target matrix, where the shape of the eighth target matrix is batch_size×24,32; Input the data with the shape of the eighth target matrix into the third one-dimensional batch normalization layer; Input the data output from the third one-dimensional batch normalization layer into the second shape transformation unit to output the second target data, where the matrix shape of the second target data is batch_size,24,32.

5. The method according to claim 4, wherein The message passing graph module includes a relational matrix unit, a second linear layer, a fourth one-dimensional batch normalization layer, and a third activation layer; The step of inputting the second target data into the message passing graph module to output the third target data includes: Input the second target data into the relational matrix unit to generate a relational matrix; Perform a matrix multiplication operation on the relational matrix and the second target data to obtain data with the shape of the tenth target matrix, where the shape of the tenth target matrix is batch_size,24,32; Input the data with the shape of the tenth target matrix into the second linear layer to obtain data with the shape of the eleventh target matrix, where the shape of the eleventh target matrix is batch_size,24,16; Input the data with the shape of the eleventh target matrix into the fourth one-dimensional batch normalization layer; Input the data output by the fourth one-dimensional batch normalization layer into the third activation layer to obtain the third target data.

6. The method according to claim 5, wherein The relationship matrix unit includes a fourth activation layer and a softmax layer; the step of inputting the second target data into the relationship matrix unit to generate a relationship matrix includes: Transpose the second target data, perform matrix multiplication on the transposed second target data and the second target data to obtain data with the shape of the twelfth target matrix, and the shape of the twelfth target matrix is batch_size×24×24; After performing a dot product operation on the data with the shape of the twelfth target matrix and the attenuation matrix, input it into the fourth activation layer; Input the data output by the fourth activation layer into the softmax layer to generate a relationship matrix.

7. The method according to claim 6, wherein The fully connected layer module includes a third shape change unit, a third linear layer, a fourth linear layer, a fifth linear layer, a fifth activation layer, a sixth activation layer, and a seventh activation layer; The step of inputting the third target data into the fully connected layer module includes: Input the third target data into the third shape change unit to output data with the shape of the thirteenth target matrix, and the shape of the thirteenth target matrix is batch_size,384; Input the data with the shape of the thirteenth target matrix into the fifth activation layer through the third linear layer to obtain data with the shape of the fourteenth target matrix, and the shape of the fourteenth target matrix is batch_size,128; Input the data with the shape of the fourteenth target matrix into the sixth activation layer through the fourth linear layer to obtain data with the shape of the fifteenth target matrix, and the shape of the fifteenth target matrix is batch_size,64; Input the data with the shape of the fifteenth target matrix into the seventh activation layer through the fifth linear layer to obtain the third target data, and the shape of the third target data is batch_size,28.

8. The method according to claim 7, characterized in that, The fault labels include high-voltage phase U high short circuit, high-voltage phase U middle short circuit, high-voltage phase U low short circuit, high-voltage phase V high short circuit, high-voltage phase V middle short circuit, high-voltage phase V low short circuit, high-voltage phase W high short circuit, high-voltage phase W middle short circuit, high-voltage phase W low short circuit, medium-voltage phase U high short circuit, medium-voltage phase U middle short circuit, medium-voltage phase U low short circuit, medium-voltage phase V high short circuit, medium-voltage phase V middle short circuit, medium-voltage phase V low short circuit, medium-voltage phase W high short circuit, medium-voltage phase W middle short circuit, medium-voltage phase W low short circuit, low-voltage phase U high short circuit, low-voltage phase U middle short circuit, low-voltage phase U low short circuit, low-voltage phase V high short circuit, low-voltage phase V middle short circuit, low-voltage phase V low short circuit, low-voltage phase W high short circuit, low-voltage phase W middle short circuit, or low-voltage phase W low short circuit.

9. The method according to claim 8, wherein The four magnetic flux sensors are evenly distributed on one side of the three-phase transformer and are centrosymmetric with the remaining four magnetic flux sensors about the coil center point of the three-phase transformer.

10. A transformer fault diagnosis device based on a spatio-temporal sequence classification algorithm, characterized in that, A plurality of magnetic flux sensors are arranged around the transformer; the device includes: a receiving unit, configured to receive a first leakage magnetic flux signal sent by the magnetic flux sensors; an input unit, configured to input the first leakage magnetic flux signal into a preset spatio-temporal sequence classification algorithm model to output a transformer label; wherein the preset spatio-temporal sequence classification algorithm model is trained according to second leakage magnetic flux signals sent by the magnetic flux sensors in a fault state of small turn-to-turn short circuits occurring at different positions of the transformer and in a normal operation state; a sending unit, configured to send an alarm message to the staff terminal if the transformer label is a fault label.

Images