Transformer winding insulation monitoring system

Abstract

The invention belongs to the technical field of winding insulation monitoring, and discloses a transformer winding insulation monitoring system which comprises the following steps: acquiring transformer winding data and an aging degree label; performing image enhancement processing, temperature correction, hot spot detection and temperature marking on the transformer winding data to generate a hot spot temperature distribution diagram; constructing a winding insulation aging prediction model for predicting the insulation aging degree of the winding based on the hot-spot temperature distribution diagram and the aging degree label; real-time transformer winding data are collected, prediction is carried out through the winding insulation aging prediction model based on the real-time transformer winding data, the winding insulation aging degree is obtained, and early warning judgment is carried out according to the winding insulation aging degree; the initiative of transformer maintenance is greatly improved, a user is helped to take necessary measures before a fault occurs, the risks of power failure accidents and equipment damage are reduced, and the service life of equipment is prolonged.

Description

Technical Field

[0001] The present invention relates to the technical field of winding insulation monitoring, and more specifically, to a transformer winding insulation monitoring system. Background Art

[0002] The patent application publication number CN112710929A discloses a power electronic transformer winding interlayer The invention discloses an insulation monitoring method and system. Before using a power electronic transformer, a swept common-mode voltage excitation is applied, and the ratio of the common-mode component amplitude is used as a healthy state transfer function. The curve of the healthy state transfer function changing with frequency is obtained, and the effective frequency band for monitoring interlayer insulation is clearly defined. After use, when it is necessary to monitor interlayer insulation, the harmonic component of the output voltage of the power electronic transformer is adjusted to cover the effective frequency band, the voltage at the winding entry point and the neutral point is measured online, the common-mode component of the two is calculated, and the online transfer function changing with frequency curve is obtained. The online transfer function changing with frequency curve is compared with the healthy state transfer function changing with frequency curve to determine the insulation state of the power electronic transformer winding interlayer. By adopting a power electronic transformer with a multi-layer winding structure, the online monitoring of the insulation of the power electronic transformer winding interlayer is realized by controlling the converter switch modulation strategy.

[0003] Under long-term operation, the insulation materials of transformer windings are prone to thermal aging in high-temperature environments, resulting in a decrease in insulation performance. The chemical structure of materials such as insulating paper and oil may change, reducing their dielectric strength. Under the existing technology, the aging problem of transformer winding insulation is solved by regular maintenance methods, which cannot be detected in real time, and the reliability and stability of the power system are low.

[0004] In view of this, the present invention proposes a transformer winding insulation monitoring system to solve the above problem. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned purpose, the present invention provides the following technical solutions: A transformer winding insulation monitoring system, comprising: a data acquisition module: collecting transformer winding data and aging degree labels; Data processing module: Perform image enhancement processing on transformer winding data to generate enhanced winding image data, perform temperature conversion processing based on the enhanced winding image data, perform temperature correction in combination with ambient temperature, and generate actual temperature data; perform hotspot detection and temperature marking based on actual temperature data to generate hotspot temperature distribution map; Model building module: building a winding insulation aging prediction model for predicting the degree of winding insulation aging based on the hot spot temperature distribution map and aging degree labels; Real-time detection module: collects real-time transformer winding data, predicts the winding insulation aging prediction model based on the real-time transformer winding data, obtains the degree of winding insulation aging, and makes early warning judgments based on the degree of winding insulation aging.

[0006] Furthermore, the transformer winding data includes: winding thermal image and ambient temperature; aging degree labels include: normal, slightly aged, moderately aged and severely aged.

[0007] Furthermore, the method of performing image enhancement processing on the transformer winding data includes: The upper left corner of the winding thermal image is used as the coordinate origin, and the number of pixels of the length and width of the winding thermal image is used as the coordinate axis scale to construct a pixel coordinate system. Based on the pixel coordinate system, a preset length of , width is The winding thermal image is smoothed based on the pixel neighborhood and the smoothing formula is: ;in, Represents the coordinates in the pixel coordinate system The smoothed pixel value of the pixel at Represents the horizontal axis value of the pixel coordinate system, Represents the vertical axis value of the pixel coordinate system, represents the length of the pixel neighborhood, represents the width of the pixel neighborhood, Represents the coordinates in the pixel coordinate system The pixel value of the pixel at Represents the standard deviation of the pixel values ​​of the winding thermal image. represents pi; The winding thermal image is regionally enhanced based on the smoothed pixel value. The calculation formula for regional enhancement is: ;in, Represents the coordinates in the pixel coordinate system The enhanced pixel value of the pixel at Represents the coordinates in the pixel coordinate system The second-order derivative of the pixel point in the horizontal direction of the pixel coordinate system, Represents the coordinates in the pixel coordinate system The second-order derivative of the pixel point in the vertical direction of the pixel coordinate system.

[0008] Furthermore, the method of generating actual temperature data includes: Temperature conversion based on enhanced pixel value, formula for temperature conversion ;in, Represents the coordinates in the pixel coordinate system The pixel temperature of the pixel at and represents the calibration constant, represents the material emissivity; The pixel temperature is corrected based on the ambient temperature. The temperature correction formula is: ;in, Represents the coordinates in the pixel coordinate system The actual temperature of the pixel at Represents the ambient temperature.

[0009] Furthermore, the method of generating the hotspot temperature distribution map includes: The temperature threshold is preset, and the hotspot detection is performed on the actual temperature based on the temperature threshold. The formula for hotspot detection is: ;in, Represents the hotspot labeling matrix, 1 represents a hotspot, 0 represents a non-hotspot, represents the temperature threshold; Based on the hot spot marking matrix, the hot spot temperature of the winding thermal image is marked to obtain the hot spot temperature distribution map. The formula for hot spot temperature marking is: ;in, Represents the coordinates in the pixel coordinate system The temperature value of the pixel.

[0010] Furthermore, the winding insulation aging prediction model is constructed in the following manner: The winding insulation aging prediction model is defined, and the defined winding insulation aging prediction model is trained and optimized through the optimization algorithm to obtain the best winding insulation aging prediction model.

[0011] Furthermore, the method of defining the model includes: The winding insulation aging prediction model is a neural network model. The winding insulation aging prediction model takes the hot spot temperature distribution map as input, the winding insulation aging degree as output, and the aging degree label as the prediction target for model training. The architecture of the winding insulation aging prediction model includes: convolution layer, pooling layer and fully connected layer; the formula for defining the convolution operation of the winding insulation aging prediction model is: ;in, Representative The convolution kernel is located at the hot spot temperature distribution map. The convolution value at Represents the index of the long side pixel of the hotspot temperature distribution map, Represents the index of the wide edge pixel of the hotspot temperature distribution map, represents the index of the convolution kernel, Representative The weight matrix of the convolution kernel, and represents the index of the element position in the weight matrix, The representative hot spot temperature distribution diagram is at the location The pixel value at Representative The bias term of the convolution kernel, represents the activation function; The loss function of the winding insulation aging prediction model is defined as: ;in, represents the loss function value, Represents the total number of aging degree label categories, Represents the aging degree label category index, Represents the true distribution The probability of aging degree labels, Represents the first The probability of an aging degree label.

[0012] Furthermore, the step of performing training and tuning includes: Step 1: Initialize E ants; initialize the parameter space of the winding insulation aging prediction model, which includes the weight matrix of the convolution kernel and the bias term of the convolution kernel, and assign each ant to a different position in the parameter space; initialize pheromone: assign an initial pheromone concentration to each ant; Step 2: Each ant moves in the parameter space and selects the direction of the position to be updated next based on the path probability. The calculation formula of the path probability is: ;in, Represents the ant from the position Select Path to Location The probability of Representative position To location The pheromone concentration, Pheromone weight coefficient, and represents the position index, and , Representative position The inverse of the loss function value of the winding insulation aging prediction model in the corresponding parameter space, The index representing the selected position; Step 3: After each ant selects a path, it updates its position and calculates the loss function value corresponding to each ant through the loss function of the winding insulation aging prediction model according to the parameters at the selected position; Step 4: Update the pheromone concentration of the found path according to the loss function value corresponding to each ant. The formula for updating the pheromone concentration is: ;in, Representative position To location The updated value of the pheromone concentration, represents the pheromone volatility coefficient, represents the index of the ant, Representative The loss function value of an ant, represents the total number of ants, represents a constant; Step 5: When the maximum number of iterations is reached or the loss function of the winding insulation aging prediction model converges, the parameters corresponding to the positions where the largest number of ants gather are the parameter space that minimizes the loss function value of the winding insulation aging prediction model. When the maximum number of iterations is not reached and the loss function of the winding insulation aging prediction model has not converged, repeat steps 2, 3, and 4.

[0013] Furthermore, the method of performing early warning judgment according to the degree of winding insulation aging includes: Collect real-time transformer winding data, including real-time winding thermal images and real-time ambient temperature. Perform corresponding processing on the real-time transformer winding data to obtain hotspot temperature distribution map. Use the hotspot temperature distribution map as the input of the winding insulation aging prediction model to predict the degree of winding insulation aging. Activate an alarm when the degree of winding insulation aging is moderate or severe.

[0014] The technical effects and advantages of a transformer winding insulation monitoring system of the present invention are as follows: By performing hotspot detection and temperature marking on transformer winding data, the present invention can better show the local temperature changes of the winding, effectively identify abnormal hotspot areas, improve the accuracy of hotspot detection, and ensure the reliability and accuracy of detection; through the neural network model, the accurate prediction of the aging degree of winding insulation is achieved, and the adaptability and prediction accuracy of the system in different environments are improved; the model parameters are optimized through the ant colony algorithm, the intelligent search mechanism of the biological population is simulated, the convergence speed and optimization effect of the model are improved, and thus the prediction accuracy of the model is effectively improved; by accurately predicting the aging degree of the transformer winding, the initiative of transformer maintenance is greatly improved, helping users to take necessary measures before a fault occurs, reducing the risk of power outages and equipment damage, and extending the service life of the equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1A schematic diagram of a transformer winding insulation monitoring system of the present invention; Figure 2 It is a schematic diagram of a transformer winding insulation monitoring method of the present invention. DETAILED DESCRIPTION

[0016] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0017] Example 1 See also Figure 1 As shown, a transformer winding insulation monitoring system described in this embodiment includes: a data acquisition module: collecting transformer winding data and aging degree labels; Data processing module: Perform image enhancement processing on transformer winding data to generate enhanced winding image data, perform temperature conversion processing based on the enhanced winding image data, perform temperature correction in combination with ambient temperature, and generate actual temperature data; perform hotspot detection and temperature marking based on actual temperature data to generate hotspot temperature distribution map; Model building module: building a winding insulation aging prediction model for predicting the degree of winding insulation aging based on the hot spot temperature distribution map and aging degree labels; Real-time detection module: collects real-time transformer winding data, predicts the winding insulation aging prediction model based on the real-time transformer winding data, obtains the degree of winding insulation aging, and makes early warning judgments based on the degree of winding insulation aging; Each module is connected by wired and / or wireless means to achieve data transmission between modules; Transformer winding data includes: winding thermal images and ambient temperature; aging degree labels are obtained through manual judgment by experts, and aging degree labels include: normal, slight aging, moderate aging and severe aging; experts label aging degree labels to make the data source more accurate and targeted; The winding thermal image is obtained by scanning the transformer winding insulation with a high-precision infrared thermal imager, and the ambient temperature is obtained with a temperature measuring instrument; The upper left corner of the winding thermal image is used as the coordinate origin, and the number of pixels of the length and width of the winding thermal image is used as the coordinate axis scale to construct a pixel coordinate system. Based on the pixel coordinate system, a preset length of , width is The winding thermal image is smoothed based on the pixel neighborhood and the smoothing formula is: ;in, Represents the coordinates in the pixel coordinate system The smoothed pixel value of the pixel at Represents the horizontal axis value of the pixel coordinate system, Represents the vertical axis value of the pixel coordinate system, represents the length of the pixel neighborhood, represents the width of the pixel neighborhood, Represents the coordinates in the pixel coordinate system The pixel value of the pixel at Represents the standard deviation of the pixel values ​​of the winding thermal image. represents pi; The winding thermal image is regionally enhanced based on the smoothed pixel value. The calculation formula for regional enhancement is: ;in, Represents the coordinates in the pixel coordinate system The enhanced pixel value of the pixel at Represents the coordinates in the pixel coordinate system The second-order derivative of the pixel point in the horizontal direction of the pixel coordinate system, Represents the coordinates in the pixel coordinate system The second-order derivative of the pixel point in the vertical direction of the pixel coordinate system; Temperature conversion based on enhanced pixel value, formula for temperature conversion ;in, Represents the coordinates in the pixel coordinate system The pixel temperature of the pixel at and represents the calibration constant (obtained from the production parameters of the high-precision infrared thermal imager), Represents the material emissivity (obtained from the material parameters of the transformer winding). The emissivity of the surface of an object varies from material to material. The higher the emissivity, the more energy the surface of the object radiates. The pixel temperature is corrected based on the ambient temperature. The temperature correction formula is: ;in, Represents the coordinates in the pixel coordinate system The actual temperature of the pixel at Represents the ambient temperature; The temperature threshold is preset, and the hotspot detection is performed on the actual temperature based on the temperature threshold. The formula for hotspot detection is: ;in, represents the hotspot labeling matrix, 1 represents a hotspot (i.e., an abnormal point), and 0 represents a non-hotspot (i.e., a normal point). represents the temperature threshold; Based on the hot spot marking matrix, the hot spot temperature of the winding thermal image is marked to obtain the hot spot temperature distribution map. The formula for hot spot temperature marking is: ;in, Represents the coordinates in the pixel coordinate system The temperature value of the pixel at The temperature information in the winding thermal image is effectively processed through pixel coordinate system, smoothing and area enhancement operations; in particular, the smoothing and enhancement of temperature data can better show the local temperature changes of the winding and improve the accuracy of hot spot detection; the formula for detecting the actual temperature based on the temperature threshold can effectively identify abnormal hot spot areas and ensure the reliability and accuracy of detection; A winding insulation aging prediction model is constructed based on the hot spot temperature distribution map and the aging degree label. The construction method of the winding insulation aging prediction model includes: The winding insulation aging prediction model is a neural network model. The winding insulation aging prediction model takes the hot spot temperature distribution map as input, the winding insulation aging degree as output, and the aging degree label as the prediction target for model training. The architecture of the winding insulation aging prediction model includes: convolution layer, pooling layer and fully connected layer; the formula for defining the convolution operation of the winding insulation aging prediction model is: ;in, Representative The convolution kernel is located at the hot spot temperature distribution map. The convolution value at Represents the index of the long side pixel of the hotspot temperature distribution map, Represents the index of the wide edge pixel of the hotspot temperature distribution map, represents the index of the convolution kernel, Representative The weight matrix of the convolution kernel, and represents the index of the element position in the weight matrix, The representative hot spot temperature distribution diagram is at the location The pixel value at Representative The bias term of the convolution kernel enables the model to learn more complex features. Represents the activation function, which is used to improve the performance of the model; Through the neural network model, the convolution layer, pooling layer and fully connected layer are used to learn and extract the features of the hot spot temperature distribution map, realizing the accurate prediction of the degree of winding insulation aging. This prediction method based on the actual temperature distribution improves the adaptability and prediction accuracy of the system in different environments. The loss function of the winding insulation aging prediction model is defined as: ;in, represents the loss function value, Represents the total number of aging degree label categories, Represents the aging degree label category index, Represents the true distribution The probability of aging degree labels, Represents the first The probability of aging degree labels; The steps of obtaining the optimal winding insulation aging prediction model by minimizing the value of the loss function of the winding insulation aging prediction model include: Step 1: Initialize E ants; initialize the parameter space of the winding insulation aging prediction model, which includes the weight matrix of the convolution kernel and the bias term of the convolution kernel, and assign each ant to a different position in the parameter space; initialize pheromone: assign an initial pheromone concentration to each ant; Step 2: Each ant moves in the parameter space and selects the direction of the position to be updated next based on the path probability. The calculation formula of the path probability is: ;in, Represents the ant from the position Select Path to Location The probability of Representative position To location Pheromones are chemical markers left by ants on their paths. The higher the concentration, the more attractive the path. Pheromone weight coefficient, which controls the influence of pheromone on path selection. and represents the position index, and , Representative position The inverse of the loss function value of the winding insulation aging prediction model in the corresponding parameter space, Represents the index of the selected position (position indices of all locations where selections were made); Step 3: After each ant selects a path, it updates its position and calculates the loss function value corresponding to each ant through the loss function of the winding insulation aging prediction model according to the parameters at the selected position. That is, each ant updates its current position selection according to the loss function value at each step; Step 4: Update the pheromone concentration of the found path according to the loss function value corresponding to each ant. The formula for updating the pheromone concentration is: ;in, Representative position To location The updated value of the pheromone concentration, Represents the pheromone volatility coefficient, which is used to simulate the natural volatility of pheromones. represents the index of the ant, Representative The loss function value of an ant, represents the total number of ants, represents a constant, used to adjust the pheromone increment; Step 5: When the maximum number of iterations is reached or the loss function of the winding insulation aging prediction model converges, the parameter corresponding to the position where the largest number of ants gather is the parameter space that minimizes the loss function value of the winding insulation aging prediction model. When the maximum number of iterations is not reached and the loss function of the winding insulation aging prediction model has not converged, repeat steps 2, 3 and 4; The parameters of the winding insulation aging prediction model are optimized through the ant colony algorithm. After multiple rounds of iterative optimization of the loss function, the effectiveness of parameter tuning is ensured. The ants' path selection and pheromone concentration update process simulate the intelligent search mechanism of the biological community, which improves the convergence speed and optimization effect of the model, thereby effectively improving the prediction accuracy of the model. Collect real-time transformer winding data, including real-time winding thermal image and real-time ambient temperature, process the real-time transformer winding data accordingly, and obtain the hot spot temperature distribution map. The hot spot temperature distribution map is used as the input of the winding insulation aging prediction model to predict the degree of winding insulation aging. When the degree of winding insulation aging is moderate or severe, an alarm is triggered. This embodiment performs hotspot detection and temperature marking on transformer winding data to better demonstrate local temperature changes in the winding, effectively identify abnormal hotspot areas, improve the accuracy of hotspot detection, and ensure the reliability and accuracy of detection; through the neural network model, accurate prediction of the degree of winding insulation aging is achieved, and the adaptability and prediction accuracy of the system in different environments are improved; the model parameters are optimized through the ant colony algorithm, the intelligent search mechanism of the biological population is simulated, the convergence speed and optimization effect of the model are improved, and thus the prediction accuracy of the model is effectively improved; by accurately predicting the aging degree of the transformer winding, the initiative of transformer maintenance is greatly improved, helping users to take necessary measures before a fault occurs, reducing the risk of power outages and equipment damage, and extending the service life of the equipment.

[0018] Example 2 See also Figure 2 As shown, the part not described in detail in this embodiment is described in Example 1, and a transformer winding insulation monitoring method is provided, including: S1. Collect transformer winding data and aging degree labels; S2. Perform image enhancement processing on the transformer winding data to generate enhanced winding image data, perform temperature conversion processing based on the enhanced winding image data, perform temperature correction in combination with the ambient temperature, and generate actual temperature data; perform hotspot detection and temperature marking based on the actual temperature data to generate a hotspot temperature distribution map; S3. Constructing a winding insulation aging prediction model for predicting the degree of winding insulation aging based on the hot spot temperature distribution map and the aging degree label; S4. Collect real-time transformer winding data, make predictions based on the real-time transformer winding data through the winding insulation aging prediction model, obtain the degree of winding insulation aging, and make early warning judgments based on the degree of winding insulation aging.

[0019] Example 3 This embodiment discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the operation mode of the transformer winding insulation monitoring method provided above is implemented.

[0020] Since the electronic device introduced in this embodiment is an electronic device used to implement a transformer winding insulation monitoring method in the embodiment of the present application, based on the transformer winding insulation monitoring method introduced in the embodiment of the present application, a person skilled in the art can understand the specific implementation of the electronic device of the present embodiment and its various variations, so how the electronic device implements the method in the embodiment of the present application is not described in detail here. As long as a person skilled in the art implements the electronic device used in the transformer winding insulation monitoring method in the embodiment of the present application, it belongs to the scope of protection of the present application.

[0021] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and thresholds in the formula are set by technicians in this field according to actual conditions.

[0022] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technical users in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A transformer winding insulation monitoring system, characterized in that: include: Data acquisition module: collects transformer winding data and aging degree labels; Data processing module: Perform image enhancement processing on transformer winding data to generate enhanced winding image data, perform temperature conversion processing based on the enhanced winding image data, perform temperature correction in combination with ambient temperature, and generate actual temperature data; Perform hot spot detection and temperature marking based on actual temperature data to generate hot spot temperature distribution map; Model building module: building a winding insulation aging prediction model for predicting the degree of winding insulation aging based on the hot spot temperature distribution map and aging degree labels; Real-time detection module: collects real-time transformer winding data, predicts the winding insulation aging prediction model based on the real-time transformer winding data, obtains the degree of winding insulation aging, and makes early warning judgments based on the degree of winding insulation aging.

2. A transformer winding insulation monitoring system according to claim 1, characterized in that: The transformer winding data includes: winding thermal image and ambient temperature; aging degree labels include: normal, slightly aged, moderately aged and severely aged.

3. A transformer winding insulation monitoring system according to claim 2, characterized in that: The method of performing image enhancement processing on transformer winding data includes: The upper left corner of the winding thermal image is used as the coordinate origin, and the number of pixels of the length and width of the winding thermal image is used as the coordinate axis scale to construct a pixel coordinate system. Based on the pixel coordinate system, a preset length of , width is The winding thermal image is smoothed based on the pixel neighborhood and the smoothing formula is: ;in, Represents the coordinates in the pixel coordinate system The smoothed pixel value of the pixel at Represents the horizontal axis value of the pixel coordinate system, Represents the vertical axis value of the pixel coordinate system, represents the length of the pixel neighborhood, represents the width of the pixel neighborhood, Represents the coordinates in the pixel coordinate system The pixel value of the pixel at Represents the standard deviation of the pixel values ​​of the winding thermal image. represents pi; The winding thermal image is regionally enhanced based on the smoothed pixel value. The calculation formula for regional enhancement is: ;in, Represents the coordinates in the pixel coordinate system The enhanced pixel value of the pixel at Represents the coordinates in the pixel coordinate system The second-order derivative of the pixel point in the horizontal direction of the pixel coordinate system, Represents the coordinates in the pixel coordinate system The second-order derivative of the pixel point in the vertical direction of the pixel coordinate system.

4. A transformer winding insulation monitoring system according to claim 3, characterized in that: The method of generating actual temperature data includes: Temperature conversion based on enhanced pixel value, formula for temperature conversion ;in, Represents the coordinates in the pixel coordinate system The pixel temperature of the pixel at and represents the calibration constant, represents the material emissivity; The pixel temperature is corrected based on the ambient temperature. The temperature correction formula is: ;in, Represents the coordinates in the pixel coordinate system The actual temperature of the pixel at Represents the ambient temperature.

5. A transformer winding insulation monitoring system according to claim 4, characterized in that: The method of generating the hotspot temperature distribution map includes: The temperature threshold is preset, and the hotspot detection is performed on the actual temperature based on the temperature threshold. The formula for hotspot detection is: ;in, Represents the hotspot labeling matrix, 1 represents a hotspot, 0 represents a non-hotspot, represents the temperature threshold; Based on the hot spot marking matrix, the hot spot temperature of the winding thermal image is marked to obtain the hot spot temperature distribution map. The formula for hot spot temperature marking is: ;in, Represents the coordinates in the pixel coordinate system The temperature value of the pixel.

6. A transformer winding insulation monitoring system according to claim 5, characterized in that: The method for constructing the winding insulation aging prediction model includes: The winding insulation aging prediction model is defined, and the defined winding insulation aging prediction model is trained and optimized through the optimization algorithm to obtain the best winding insulation aging prediction model.

7. A transformer winding insulation monitoring system according to claim 6, characterized in that: The method of defining the model includes: The winding insulation aging prediction model is a neural network model. The winding insulation aging prediction model takes the hot spot temperature distribution map as input, the winding insulation aging degree as output, and the aging degree label as the prediction target for model training. The architecture of the winding insulation aging prediction model includes: convolution layer, pooling layer and fully connected layer; the formula for defining the convolution operation of the winding insulation aging prediction model is: ;in, Representative The convolution kernel is located at the hot spot temperature distribution map. The convolution value at Represents the index of the long side pixel of the hotspot temperature distribution map, Represents the index of the wide edge pixel of the hotspot temperature distribution map, represents the index of the convolution kernel, Representative The weight matrix of the convolution kernel, and represents the index of the element position in the weight matrix, The representative hot spot temperature distribution diagram is at the location The pixel value at Representative The bias term of the convolution kernel, represents the activation function; The loss function of the winding insulation aging prediction model is defined as: ;in, represents the loss function value, Represents the total number of aging degree label categories, Represents the aging degree label category index, Represents the true distribution The probability of aging degree labels, Represents the first The probability of an aging degree label.

8. A transformer winding insulation monitoring system according to claim 7, characterized in that: The steps of performing training and tuning include: Step 1: Initialize E ants; initialize the parameter space of the winding insulation aging prediction model, which includes the weight matrix of the convolution kernel and the bias term of the convolution kernel, and assign each ant to a different position in the parameter space; initialize pheromone: assign an initial pheromone concentration to each ant; Step 2: Each ant moves in the parameter space and selects the direction of the position to be updated next based on the path probability. The calculation formula of the path probability is: ;in, Represents the ant from the position Select Path to Location The probability of Representative position To location The pheromone concentration, Pheromone weight coefficient, and represents the position index, and , Representative position The inverse of the loss function value of the winding insulation aging prediction model in the corresponding parameter space, Represents the index of the selected position; Step 3: After each ant selects a path, it updates its position and calculates the loss function value corresponding to each ant through the loss function of the winding insulation aging prediction model according to the parameters at the selected position; Step 4: Update the pheromone concentration of the found path according to the loss function value corresponding to each ant. The formula for updating the pheromone concentration is: ;in, Representative position To location The updated value of the pheromone concentration, represents the pheromone volatility coefficient, represents the index of the ant, Representative The loss function value of an ant is represents the total number of ants, represents a constant; Step 5: When the maximum number of iterations is reached or the loss function of the winding insulation aging prediction model converges, the parameters corresponding to the positions where the largest number of ants gather are the parameter space that minimizes the loss function value of the winding insulation aging prediction model. When the maximum number of iterations is not reached and the loss function of the winding insulation aging prediction model has not converged, repeat steps 2, 3, and 4.

9. A transformer winding insulation monitoring system according to claim 8, characterized in that: The method of performing early warning judgment according to the degree of winding insulation aging includes: Collect real-time transformer winding data, including real-time winding thermal images and real-time ambient temperature. Perform corresponding processing on the real-time transformer winding data to obtain hotspot temperature distribution map. Use the hotspot temperature distribution map as the input of the winding insulation aging prediction model to predict the degree of winding insulation aging. Activate an alarm when the degree of winding insulation aging is moderate or severe.

Images