A Transformer Hotspot Information Measurement Method and System Based on Machine Learning Algorithms

Abstract

This invention relates to the field of transformer technology, and specifically designs a method and system for measuring transformer hotspot information based on machine learning algorithms. The method includes the following steps: establishing a transformer geometric model and setting several ultrasonic sensors on the geometric model; performing sound field simulation based on the geometric location and temperature values ​​of several hotspot faults; obtaining acoustic signals of several hotspot faults and a sound field simulation dataset using ultrasonic detection; extracting feature parameters from the acoustic signals of several hotspot faults and parameterizing them to establish hotspot types; training a machine learning algorithm using the hotspot types to obtain a trained machine learning algorithm, and establishing a hotspot fault type library based on the training results; classifying the feature parameters of the hotspots of the transformer under test using the trained machine learning algorithm, and obtaining transformer hotspot information based on the classification results.

Description

Technical Field

[0001] This invention relates to the field of transformer technology, specifically to a method and system for measuring transformer hotspot information based on machine learning algorithms. Background Technology

[0002] As the most crucial piece of electrical equipment in a power system, the power transformer plays a vital role in power conversion and transmission. Internal heating during transformer operation severely impacts the insulation level and electrical performance of the equipment. Hot spot temperatures in the transformer windings are a significant factor leading to insulation failure and transformer malfunctions. Therefore, automatic detection of hot spot temperatures in the internal windings of power transformers is of paramount importance.

[0003] Among existing direct methods for detecting the internal temperature of transformers, oil surface thermometer measurement is easily affected by the installation location and can only directly measure the temperature of the top layer of oil. Infrared thermometry can only measure the surface temperature of the transformer tank and the outer surface of the transformer, and is easily affected by the ambient temperature, limiting its measurement range. Resistance temperature detectors (RTDs) and thermocouples suffer from problems such as the susceptibility of metal wires to corrosion and sensor probe encapsulation, making it difficult to meet future development needs. Fiber optic temperature sensors are relatively suitable for measuring the temperature at the target location inside the transformer, but the placement of the sensor within the windings makes it difficult to choose between the measurement location and the insulation level of the equipment, limiting its applicability to newly manufactured transformers. Furthermore, indirect temperature measurement methods such as national standard recommended algorithms, thermal simulation methods, and thermoelectric analogy methods analyze internal heat transfer from a mathematical perspective, but all have certain performance drawbacks, only able to determine the approximate internal temperature of the transformer, making it difficult to determine the temperature and actual location of hot spots.

[0004] Ultrasonic temperature measurement technology, as a non-contact temperature measurement method, uses the relationship between the acoustic characteristics of ultrasonic waves propagating in a medium and the temperature of the medium to measure temperature. It has advantages such as wide temperature measurement range, fast response speed, high sensitivity, and safe use. However, there are many types of transformers and their internal environment is complex, which inevitably increases the difficulty for workers in the actual measurement process. Machine learning algorithms, as a product of artificial intelligence research, meet people's needs for mining and applying massive amounts of data. Summary of the Invention

[0005] To address the aforementioned problems, the present invention provides, in its first aspect, a method for measuring transformer hotspot information based on machine learning algorithms, the steps of which are as follows:

[0006] A geometric model of the transformer was established, and several ultrasonic sensors were set on the geometric model. Sound field simulation was performed based on the geometric location and temperature value of several hot spot faults. The sound wave signals of several hot spot faults and the sound field simulation dataset were obtained by ultrasonic detection.

[0007] Feature parameters are extracted from acoustic signals of several hotspot faults and then parameterized to establish hotspot types.

[0008] The machine learning algorithm is trained using hotspot types to obtain the trained machine learning algorithm, and a hotspot fault type library is built based on the training results.

[0009] The trained machine learning algorithm is used to classify the feature parameters of the hot spots of the transformer under test, and the transformer hot spot information is obtained based on the classification results.

[0010] The method simulates the propagation of ultrasonic waves in a transformer geometric model using simulation techniques. It traverses the possible locations of hot spot faults inside the transformer, acquires the received acoustic signals under different hot spot fault conditions, extracts the waveform and feature parameters of the acoustic waves under the corresponding hot spot fault, trains a machine learning algorithm based on a large amount of simulation data, and establishes a hot spot fault type library. This method is then applied to the actual temperature measurement process of transformers to obtain real-time information on actual transformer hot spot faults.

[0011] In some implementations of the first aspect, the sound field simulation first performs steady-state simulation of the hotspot temperature field, and then performs transient simulation of the ultrasonic sound field.

[0012] Steady-state simulation of hot spot temperature field: Select a point as heat source in the geometric model of transformer, set the heat source power and radius to perform simulation, and obtain the steady-state temperature field inside the transformer.

[0013] The transient simulation of the ultrasonic sound field is performed by using a one-transmitter-multiple-receiver ultrasonic detection method to traverse all fixed positions of ultrasonic sensors on the geometric model of the transformer to simulate the propagation of the sound field transiently.

[0014] In some implementations of the first aspect, the transformer geometric model includes transformer oil, core, windings, and outer wall structure, and the geometric model has a plurality of ultrasonic sensors installed on the outer wall.

[0015] In some implementations of the first aspect, the sound field simulation is performed based on the geometric location and temperature value of several hot spot faults. Then, background noise and ultra-high frequency harmonics are filtered out by a bandpass filter model to obtain the sound wave signal of several hot spot faults.

[0016] In some implementations of the first aspect, the establishment of hotspot types specifically refers to:

[0017] The characteristic parameters of the acoustic signals of hot spot faults are classified and organized according to geometric location regions and temperature value ranges.

[0018] The geometric position region is divided based on the iron core and winding inside the geometric model. A coordinate system is established in the horizontal direction with the winding axis as the origin. The horizontal position is divided into y1 regions, and the winding axis is divided into y2 regions, for a total of y3 = y1 × y2 regions.

[0019] Temperature ranges are divided into z1℃ intervals, from 40℃ to 140℃, for a total of z2 = (140-40) / z1 intervals; the location regions and temperature ranges are combined to form y3×z2 hotspot types.

[0020] Preferably, the machine learning algorithm is a neural network classifier.

[0021] In some implementations of the first aspect, the construction of an ultrasonic inversion model of transformer hotspots based on a hotspot fault type library and an acoustic field simulation dataset includes:

[0022] For a single hotspot type, select m1 points on the edges of a hexahedron around the center of the region as the geometric positions of the reference hotspots, and select m2 temperature values ​​within the temperature range of the hotspot type as the temperatures of the reference hotspots, thus obtaining m3 = m1 × m2 different reference hotspots;

[0023] By further refining the location regions and temperature ranges of the m3 reference hot spots, a dataset of acoustic field simulation waveforms corresponding to the m3 reference hot spots is obtained. The correlation degree and functional relationship of different characteristic parameters of hot spots within a single hot spot type are fitted, and an ultrasonic inversion model of transformer hot spots is constructed.

[0024] The second aspect provides a transformer hotspot information measurement system based on machine learning algorithms, including:

[0025] The acoustic field simulation module is used to establish a geometric model of the transformer and set up several ultrasonic sensors on the geometric model. Based on the geometric location and temperature value of several hot spot faults, acoustic field simulation is performed. The acoustic wave signals of several hot spot faults and the acoustic field simulation dataset are obtained by ultrasonic detection.

[0026] The feature parameter extraction module extracts feature parameters from the acoustic signals of several hotspot faults and performs parameterization to establish hotspot types.

[0027] The processing module is used to train the machine learning algorithm using hotspot types, obtain the trained machine learning algorithm, and establish a hotspot fault type library based on the training results; it is also used to classify the feature parameters of the hotspots of the transformer under test using the trained machine learning algorithm, and obtain transformer hotspot information based on the classification results.

[0028] The storage module is used to store the type library of transformer hot spot faults and the ultrasonic inversion model.

[0029] The third aspect provides a transformer hotspot temperature device based on a machine learning algorithm, including a processor and a memory, wherein the processor executes program data stored in the memory to implement the transformer hotspot information measurement method based on the machine learning algorithm as described above.

[0030] The fourth aspect provides a computer-readable storage medium for storing control program data, wherein the control program data, when executed by a processor, implements the transformer hotspot information measurement method based on machine learning algorithms as described above.

[0031] The beneficial effects of this invention are as follows:

[0032] (1) This method solves the limitations of existing temperature measurement methods for power transformers, such as being susceptible to the measurement environment, having a short service life, and being unable to accurately obtain the temperature of hot spots inside the transformer;

[0033] (2) The propagation of ultrasonic waves is simulated and deduced in the geometric model of the transformer by simulation. The possible locations of hot spot faults inside the transformer are traced, and the acoustic wave signals received under different hot spot fault conditions are obtained. The waveform and feature parameters of the acoustic waves under the corresponding hot spot fault are extracted. The classification neural network is trained based on a large amount of simulation data, and a hot spot fault type library is established. The trained classification neural network is then applied to the actual transformer temperature measurement process to obtain actual transformer hot spot fault information. Maintenance personnel can verify the hot spot fault type library and the measured information.

[0034] (3) While retaining the original advantages of ultrasonic temperature measurement technology, it greatly avoids the damage and wear and tear of experimental equipment, saves a lot of manpower and material costs, and improves the accuracy and precision of measurement results. Attached Figure Description

[0035] Figure 1 A flowchart illustrating the method for measuring transformer hotspot information;

[0036] Figure 2 This is a schematic diagram showing some of the benchmark hotspot locations;

[0037] Figure 3 This is a schematic diagram of the transformer hotspot information measurement system. Detailed Implementation

[0038] Exemplary embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings.

[0039] Example

[0040] See Figure 1 This invention provides a method for measuring transformer hotspot information based on machine learning algorithms, and the specific steps are as follows:

[0041] Step 1: Establish a geometric model of the transformer and set up several ultrasonic sensors on the geometric model. Perform sound field simulation based on the geometric location and temperature value of several hot spot faults. Use ultrasonic detection to obtain the sound wave signals of several hot spot faults and the sound field simulation dataset.

[0042] Based on the propagation law of ultrasonic waves inside the transformer, a geometric model of the transformer is established, including transformer oil, core, windings, and outer wall structure. Several ultrasonic sensors are set on the outer wall of the geometric model.

[0043] As a specific implementation method, x1 ultrasonic sensors are provided on the outer wall of the geometric model to transmit and receive sound waves. The ultrasonic sensors are used to enable one ultrasonic sensor to receive the transmitted signals of other ultrasonic sensors.

[0044] Meanwhile, considering the safe and stable operation of the transformer, the parameters of the emitted sound source were set based on the transformer's own vibration frequency and the ultrasonic cavitation effect of the transformer oil, taking into account the actual internal structural characteristics of the transformer and the quality differences of the collected ultrasonic signals. The functional relationship between sound velocity and medium temperature in the actual transformer oil and outer wall material was obtained through experiments, and this functional relationship was transformed and substituted into the geometric model.

[0045] In acoustics, because the expansion and compression of the volume element of a medium are very fast, the sound propagation process in an ideal medium does not have enough time for heat exchange. Therefore, the sound wave propagation process can be assumed to be an adiabatic process, and the sound field simulation of the transformer hot spot can be performed.

[0046] Specifically, a single hotspot fault within the geometric model is selected for acoustic field simulation. First, a steady-state simulation of the hotspot's temperature field is performed by setting the heat source power and radius to obtain the steady-state temperature field inside the transformer. Then, given the known temperature field distribution of the transformer, a transient ultrasonic acoustic field simulation is conducted. A multi-transmitter ultrasonic detection method is used to traverse all fixed locations on the transformer's geometric model to simulate the transient propagation of the acoustic field. That is, a single ultrasonic sensor is selected each time to apply an excitation signal of a specific waveform, and an appropriate computation time step is chosen to simulate the propagation of the sound wave. Other ultrasonic sensors are used at fixed locations to receive and collect the sound wave signals, establishing a dataset of x2 = x1*(x1-1) acoustic field simulation waveforms corresponding to a specific hotspot.

[0047] Step 2: By changing the geometric location and temperature value of a single hot spot fault, sound field simulation is performed to obtain several hot spot fault sound wave signals and sound field simulation datasets. Based on the time domain and frequency domain variation parameterization, the characteristic parameters of the hot spot fault sound wave signals are extracted.

[0048] Inside the transformer geometric model, the geometric location and temperature value of the hot spot fault were changed, and a sound field simulation experiment was conducted to obtain x2 sound wave waveform signals under the corresponding hot spot. Background noise and ultra-high frequency harmonics were filtered out by bandpass filtering.

[0049] In the time domain, the filtered acoustic waveform is segmented by identifying abrupt changes through singular value decomposition, and then compared and studied according to the waveform amplitude and its corresponding time.

[0050] In the frequency domain, the acoustic waveform in each band after time-domain segmentation undergoes preprocessing processes such as pre-emphasis and windowing, and an appropriate frequency reduction coefficient is selected to optimize the acoustic signal.

[0051] By aggregating the acoustic signals collected by ultrasonic sensors at different locations, the variations in acoustic signals in the time and frequency domains caused by changes in the geometric location and temperature of hotspots are parameterized. The mRMR algorithm is then used to extract feature parameters that are highly correlated with the hotspot type from parameters such as peak time, time domain indices (average amplitude, root mean square value, peak value, standard deviation, skewness coefficient, kurtosis coefficient, peak factor, waveform factor, etc.), frequency domain indices (centroid frequency, root mean square frequency, frequency standard deviation, etc.), and sample entropy.

[0052] In practical applications, the acoustic wave signals from the sound field simulation can be experimentally verified. By randomly selecting the acoustic field simulation results of several transformer hotspots, hotspots with the same parameters and sensor arrangement strategy are set at the same location on the actual transformer. The received acoustic wave waveforms are processed using the same bandpass filtering, and the characteristic parameters and waveforms obtained in the simulation are verified.

[0053] Step 3: Extract feature parameters from the acoustic signals of several hotspot faults and perform parameterization to establish hotspot types;

[0054] The characteristic parameters of the hotspot acoustic wave signal are classified and organized according to geometric location region and temperature value range;

[0055] The geometric position region is divided based on the iron core and winding inside the geometric model. A coordinate system is established in the horizontal direction with the winding axis as the origin. The horizontal position is divided into y1 regions, and the winding axis is divided into y2 regions, for a total of y3 = y1 × y2 regions.

[0056] Temperature ranges are divided into z1℃ intervals, from 40℃ to 140℃, for a total of z2 = (140-40) / z1 intervals; the location regions and temperature ranges are combined to form y3×z2 hotspot types.

[0057] Step 4: Train the machine learning algorithm using hotspot types to obtain the trained machine learning algorithm, and build a hotspot fault type library based on the training results;

[0058] We organized the data from all sound field simulations within different hotspot types and trained a model based on machine learning algorithms to analyze the acoustic characteristic parameters corresponding to the changes in different temperature ranges and distribution locations of the hotspots.

[0059] The machine learning algorithm is a BP neural network classifier. The classification neural network is trained by hotspot types. By labeling different hotspot types, autonomous machine learning is performed to obtain the trained classification neural network. The classification training results are stored to establish a type library of transformer hotspot faults. The data types of the hotspot fault type library include time-domain indicators, frequency-domain indicators, and sample entropy of acoustic signals.

[0060] Step 5: Use the trained machine learning algorithm to classify the feature parameters of the hot spots in the transformer under test, and obtain the transformer hot spot information based on the classification results;

[0061] When there are unknown hot spots inside the actual transformer, the same sensor arrangement strategy as the transformer geometric model is adopted on the outer wall of the transformer. A specific ultrasonic signal is transmitted into the transformer and the acoustic signal is collected by a one-to-many method.

[0062] Furthermore, an ultrasonic inversion model of transformer hotspots was constructed based on a hotspot fault type library and an acoustic field simulation dataset.

[0063] like Figure 2 As shown, for a single hotspot type in the hotspot fault type library, m1 points on the edges of a hexahedron around the center of the region are selected as the geometric positions of the reference hotspots. Within the temperature range of the hotspot type, m2 temperature values ​​are selected as the temperatures of the reference hotspots, resulting in m3 = m1 × m2 different reference hotspots.

[0064] By further refining the location regions and temperature ranges of the m3 reference hot spots, a dataset of acoustic field simulation waveforms corresponding to the m3 reference hot spots is obtained. The correlation degree and functional relationship of different characteristic parameters of hot spots within a single hot spot type are fitted, and an ultrasonic inversion model of transformer hot spots is constructed.

[0065] The collected acoustic signals of the actual hot spots to be measured are processed to extract the corresponding feature parameters. The trained classification neural network is used to classify the feature parameters to obtain the hot spot fault information of the actual transformer under test. At the same time, maintenance personnel can search the hot spot fault type database and reasonably infer the hot spot fault information based on the ultrasonic inversion model. The results are then compared with the classification neural network results to ensure the accuracy of the data. This helps maintenance personnel to find the hot spot location inside the actual transformer, making it easier to manually disassemble the transformer at that location for maintenance and improve the efficiency of transformer hot spot fault inspection.

[0066] Furthermore, this invention provides a transformer hotspot information measurement system based on machine learning algorithms, such as... Figure 3 As shown, the system includes:

[0067] The acoustic field simulation module is used to establish a geometric model of the transformer and set up several ultrasonic sensors on the geometric model. Based on the geometric location and temperature value of several hot spot faults, acoustic field simulation is performed. The acoustic wave signals of several hot spot faults and the acoustic field simulation dataset are obtained by ultrasonic detection.

[0068] Furthermore, the sound field simulation model also includes a hot spot temperature field steady-state simulation model and an ultrasonic sound field transient simulation model. The hot spot temperature field steady-state simulation model is used to select a point as a heat source in the transformer geometric model, set the heat source power and radius for simulation, and obtain the steady-state temperature field inside the transformer. The ultrasonic sound field transient simulation uses a one-transmitter-multiple-receiver ultrasonic detection method to traverse all fixed positions of ultrasonic sensors on the transformer geometric model to perform transient sound field simulation propagation.

[0069] Furthermore, the system also includes a bandpass filter module, which is used to acquire the acoustic waveform signal of the hotspot and filter out background noise and ultra-high frequency harmonics through bandpass filtering;

[0070] The feature parameter extraction module extracts feature parameters from the acoustic signals of several hotspot faults and performs parameterization to establish hotspot types.

[0071] The processing module is used to train the machine learning algorithm using hotspot types, obtain the trained machine learning algorithm, and establish a hotspot fault type library based on the training results; it is also used to classify the feature parameters of the hotspots of the transformer under test using the trained machine learning algorithm, and obtain transformer hotspot information based on the classification results.

[0072] The storage module is used to store a library of transformer hot spot fault types and ultrasonic inversion models.

[0073] Furthermore, the present invention also provides a transformer hotspot temperature device based on machine learning algorithms, including a processor and a memory, wherein the processor executes program data stored in the memory to implement a transformer hotspot information measurement method based on machine learning algorithms.

[0074] Finally, the present invention also provides a computer-readable storage medium for storing control program data, wherein the control program data, when executed by a processor, implements a transformer hotspot information measurement method based on a machine learning algorithm.

[0075] It should be noted that the exemplary embodiments mentioned in this invention describe methods or systems based on a series of steps or apparatus. However, this invention is not limited to the order of the steps described above; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

Claims

1. A method for measuring transformer hotspot information based on machine learning algorithms, characterized in that, Includes the following steps: A geometric model of the transformer was established, and several ultrasonic sensors were set on the geometric model. Sound field simulation was performed based on the geometric location and temperature value of several hot spot faults. The sound wave signals of several hot spot faults and the sound field simulation dataset were obtained by ultrasonic detection. By changing the geometric location and temperature value of a single hot spot fault, acoustic field simulation is performed to obtain acoustic signals of several hot spot faults and acoustic field simulation datasets. Based on the time domain and frequency domain variation parameterization, the characteristic parameters of the acoustic signals of the hot spot faults are extracted. Feature parameters are extracted from acoustic signals of several hotspot faults and then parameterized to establish hotspot types, specifically: The characteristic parameters of the acoustic signals of hot spot faults are classified and organized according to geometric location regions and temperature value ranges. The geometric location region is defined by the iron core and windings inside the geometric model. A coordinate system is established horizontally with the winding axis as the origin, and the region is divided horizontally. y One region, divided along the winding axis. y 2 areas, totaling y 3= y 1× y 2 areas; Temperature range division based on z 1℃ is one interval, from 40℃ to 140℃, totaling z 2 = (140 - 40) / z One interval; formed by combining a location area with a temperature value interval. y 3× z Two types of hot topics; The machine learning algorithm is trained using hotspot types to obtain the trained machine learning algorithm, and a hotspot fault type library is built based on the training results. The trained machine learning algorithm is used to classify the feature parameters of the hot spots of the transformer under test, and the transformer hot spot information is obtained based on the classification results.

2. The method for measuring transformer hotspot information based on machine learning algorithms according to claim 1, characterized in that, The sound field simulation first performs a steady-state simulation of the hotspot temperature field, and then performs a transient simulation of the ultrasonic sound field. Steady-state simulation of hot spot temperature field: Select a point as heat source in the geometric model of transformer, set the heat source power and radius to perform simulation, and obtain the steady-state temperature field inside the transformer. The transient simulation of the ultrasonic sound field is performed by using a one-transmitter-multiple-receiver ultrasonic detection method to traverse all fixed positions of ultrasonic sensors on the geometric model of the transformer to simulate the propagation of the sound field transiently.

3. The method for measuring transformer hotspot information based on machine learning algorithms according to claim 1, characterized in that, The transformer geometric model includes transformer oil, core, windings, and outer wall structure. Several ultrasonic sensors are installed on the outer wall of the geometric model.

4. The method for measuring transformer hotspot information based on machine learning algorithms according to claim 1, characterized in that, The sound field simulation is performed based on the geometric location and temperature value of several hot spot faults. Then, the background noise and ultra-high frequency harmonics are filtered out by a bandpass filter model to obtain the sound wave signals of several hot spot faults.

5. The method for measuring transformer hotspot information based on machine learning algorithms according to claim 1, characterized in that, The machine learning algorithm is a neural network classifier.

6. The method for measuring transformer hotspot information based on machine learning algorithms according to claim 1, characterized in that, It also includes constructing an ultrasonic inversion model of transformer hotspots based on a hotspot fault type library and acoustic field simulation dataset, including: For a single hotspot type, select the edges of a hexahedron around the center of the region. m One point is selected as the geometric location of the baseline hotspot, within the temperature range of the hotspot type. m The two temperature values ​​are used as the reference hot spot temperature to obtain m 3= m 1× m Two different benchmark hotspots; By m The location and temperature range of the hotspot type are refined using three benchmark hotspots to obtain... m The dataset of acoustic field simulation waveforms corresponding to three benchmark hotspots was used to fit the correlation degree and functional relationship of different characteristic parameters of hotspots within a single hotspot type, and to construct an ultrasonic inversion model of transformer hotspots.

7. A transformer hotspot information measurement system based on machine learning algorithms, used to implement the transformer hotspot information measurement method based on machine learning algorithms as described in claim 1, characterized in that, include: The acoustic field simulation module is used to establish a geometric model of the transformer and set up several ultrasonic sensors on the geometric model. Based on the geometric location and temperature value of several hot spot faults, acoustic field simulation is performed. The acoustic wave signals of several hot spot faults and the acoustic field simulation dataset are obtained by ultrasonic detection. The feature parameter extraction module extracts feature parameters from the acoustic signals of several hotspot faults and performs parameterization to establish hotspot types. The processing module is used to train machine learning algorithms using hotspot types, obtain trained machine learning algorithms, and build a hotspot fault type library based on the training results. The processing module is also used to classify the feature parameters of the hot spot of the transformer under test using a trained machine learning algorithm, and obtain the transformer hot spot information based on the classification results; The storage module is used to store the type library of transformer hot spot faults and the ultrasonic inversion model.

8. A transformer hotspot temperature device based on machine learning algorithms, characterized in that, The device includes a processor and a memory, wherein the processor executes program data stored in the memory to implement the transformer hotspot information measurement method based on machine learning algorithm as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, Used to store control program data, wherein the control program data, when executed by a processor, implements the transformer hotspot information measurement method based on machine learning algorithm as described in any one of claims 1-6.

Images