Transformer state sensing method based on multi-sensor information fusion
Through the multi-sensor information fusion method, the sound signals of the transformer and the operating status of the cooling fan are collected and processed, and the softmax classifier is used to judge the operating status of the transformer, which solves the problem that the state of different types of transformers cannot be effectively distinguished in the prior art, and achieves more accurate and efficient state perception.
Patent Information
- Application Number
- CN202411771292.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-04
- Publication Date
- 2025-03-14
AI Technical Summary
The existing transformer state perception methods cannot effectively distinguish the sound signals of different transformers in the same state, resulting in the inability to accurately sense the operating status of the transformer.
The multi-sensor information fusion method is adopted, by setting a sound sensor and a temperature sensor, the sound signals of the transformer in different states and the operating state of the cooling fan are collected, noise reduction processing and feature extraction are performed, and different sound signals are used to distinguish between different sound signals and determine the operating state of the transformer.
It realizes accurate perception of different transformers in the same state, improves the accuracy and efficiency of transformer status judgment, and can detect abnormalities in a timely manner and take preventive measures to avoid failures.
Smart Images
Figure CN119618308A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of transformers, and in particular to a transformer state perception method based on multi-sensor information fusion. Background Art
[0002] Transformers are the core equipment of the power system, responsible for the transmission and distribution of electric energy, and play a key role in the stability and reliability of power supply. Long-term operation and external factors may cause abnormal operation of transformers, which may threaten the safe operation of the power grid and even cause power outages or more serious consequences. Through state recognition technology, the operating status of the transformer can be monitored in real time and abnormalities can be discovered in time, which helps to take preventive measures to avoid failures, thereby improving the reliability and stability of the power system.
[0003] In the prior art, the transformer state is sensed by sensors. However, the existing transformers are of different models, such as dry and wet transformers, oil transformers, etc., and the sounds emitted by different transformers are different, so it is impossible to effectively sense the transformers in the same state through sensors. Therefore, a transformer state sensing method based on multi-sensor information fusion is provided to solve the problems raised in the above background technology. Summary of the invention
[0004] The purpose of the present invention is to provide a transformer state perception method based on multi-sensor information fusion to solve the problems raised in the above background technology.
[0005] To achieve the above object, the present invention provides the following technical solutions:
[0006] A transformer state perception method based on multi-sensor information fusion includes the following steps:
[0007] Step 1: setting a plurality of sound sensors and temperature sensors, wherein the sound sensor collects sound signals of the transformer in different states, and the temperature sensor is used to determine whether the cooling fan of the transformer is in operation;
[0008] Step 2: Perform noise reduction processing on the sound signal collected by the sound sensor;
[0009] Step 3: extracting the sound signal after noise reduction processing to characterize the operating status of the transformer;
[0010] Step 4: distinguishing the transformer sound signal and the cooling fan sound signal by using a softmax classifier;
[0011] Step 5: Determine and output the current status of the transformer and the cooling fan by distinguishing the results.
[0012] As a further solution of the present invention: wherein, the sound sensor performs a convolution layer operation on the collected sound signal, and the convolution layer can perform a convolution operation on the input data to generate a local connection of the feature vector, and the input data and the convolution kernel in the convolution layer are connected in a local area.
[0013] As a further solution of the present invention: wherein, a plurality of the sound sensors form a sensor acquisition array, and the sensor acquisition array collects sound signals of the transformer in a normal state, sound signals of a winding damaged state, sound signals of a core fault state, as well as sound signals of the cooling fan in a normal state and state signals of the cooling fan in a damaged state.
[0014] As a further solution of the present invention: wherein, the number of cooling fans turned on and the speed data of the cooling fans are obtained and input into a softmax classifier to obtain a cooling influence factor on the transformer, and the temperature drop value of the transformer is collected and corrected according to the cooling influence factor.
[0015] As a further solution of the present invention: wherein, the transformer temperature reduction numerical calculation formula is:
[0016]
[0017] Among them, G1 is the cooling influence factor. When the value is 0.3, it means air cooling, when it is 0.5, it means forced oil cooling, and when it is 1, it means natural oil cooling. D is the temperature reduction value.
[0018] As a further solution of the present invention: wherein, the sound signal is feature extracted to obtain MFCC feature parameters and then GFCC feature parameters, and the PSO-SVM algorithm is used to obtain a prediction model through the MFCC feature parameters, the GFCC feature parameters, and the sound signal characteristic values of the cooling fan to determine the operating status of the transformer.
[0019] As a further solution of the present invention: wherein the operating state of the transformer includes a normal operating state, an abnormal operating state and a fault operating state.
[0020] As a further solution of the present invention: wherein, the loudness threshold of the transformer is constructed by the rotation speed of the cooling fan, the output power of the cooling fan, and the weight of the transformer, thereby further reducing the technical problem of inaccurate operation status judgment caused by the influence of the sound signal of the transformer caused by the operation of the cooling fan, and ensuring the efficiency and accuracy of the transformer status judgment
[0021] As a further solution of the present invention: wherein, the transformer sounds in different operating states are collected according to the sensor acquisition array, and the collected signals are preliminarily denoised using the spatial filtering function of beamforming to obtain 150 groups of data, each group of data is collected for 10 seconds, and each group of data is sliced, each slice interval is 2.5 seconds, to obtain 600 groups of original data.
[0022] As a further solution of the present invention: wherein, the original data is subjected to LCD noise reduction processing, that is, the data is first decomposed into 8 intrinsic scale components and one residual component, the first 6 components with higher energy are screened out from ISC for reconstruction, and the background noise with lower energy is eliminated to obtain the denoised transformer acoustic data.
[0023] As a further solution of the present invention: wherein, feature extraction is performed on the 600 groups of original data after noise reduction, the frequency complexity, odd and even harmonics and main frequency proportion of the 600 groups of data are calculated, and 300 groups of feature values are randomly selected as training sets, and the remaining 300 groups are used as test sets, and cross-validation is performed 10 times to obtain a transformer state recognition model; finally, through a new test, the data is substituted into the data to obtain the transformer state.
[0024] Compared with the prior art, the present invention has the following beneficial effects:
[0025] By setting up multiple sound sensors and temperature sensors, the working status of the transformer can be sensed, and then different sound signals can be distinguished through the softmax classifier, and the operating status of different transformers can be determined through the prediction model.
[0026] The sound signal is processed by the sensor acquisition array, and the background sound of the transformer is extracted, so that the effective acoustic data of the transformer can be effectively obtained, so that the operating status of the transformer can be judged more accurately. BRIEF DESCRIPTION OF THE DRAWINGS
[0027] Figure 1 It is a schematic diagram of the flow of the sound sensor and the temperature sensor in the present invention;
[0028] Figure 2 It is a schematic diagram of the sound sensor process in the present invention;
[0029] Figure 3 It is a schematic diagram of the sound sensor acquisition array process in the present invention;
[0030] Figure 4 This is a schematic diagram of the effective sound data flow in the present invention;
[0031] Figure 5 It is a schematic diagram of LCD noise reduction process in the present invention; DETAILED DESCRIPTION
[0032] See also Figures 1 to 5 :A transformer state perception method based on multi-sensor information fusion, comprising the following steps:
[0033] Step 1: Set up multiple sound sensors and temperature sensors, where the sound sensor collects sound signals of the transformer in different states, and the temperature sensor determines whether the cooling fan of the transformer is in operation;
[0034] Step 2: Perform noise reduction processing on the sound signal collected by the sound sensor;
[0035] Step 3: extracting the sound signal after noise reduction processing to characterize the operating status of the transformer;
[0036] Step 4: Use the softmax classifier to distinguish the transformer sound signal and the cooling fan sound signal;
[0037] Step 5: Determine and output the current status of the transformer and cooling fan by differentiating the results.
[0038] Among them, the working status of the transformer can be sensed by setting up multiple sound sensors and temperature sensors, and then different sound signals can be distinguished by the softmax classifier, and the operating status of different transformers can be judged by the prediction model.
[0039] Preferably, the sound sensor performs a convolution layer operation on the collected sound signal. The convolution layer can perform a convolution operation on the input data to generate a local connection of the feature vector. The input data and the convolution kernel in the convolution layer are connected in a local area.
[0040] Among them, after the data collected by the sound sensor has been processed by different numbers of convolution layers, the feature expressions gradually separate as the number of convolution layers increases, indicating that the model is learning the features to distinguish different states.
[0041] Furthermore, it can be divided into five layers of convolution; after the first layer of convolution, the features of different states are almost indistinguishable, indicating that the initial feature extraction has not fully separated different categories. After the second and third layers of convolution, the features begin to show a distinguishing trend, but have not yet been completely separated. After the fourth and fifth layers of convolution, the feature expressions can be clearly separated, indicating that the model simulates the characteristics of different states of the transformer.
[0042] Preferably, a plurality of sound sensors form a sensor acquisition array, which collects sound signals of the transformer in a normal state, sound signals of a winding damaged state, sound signals of a core fault state, as well as sound signals of a cooling fan in a normal state and state signals of a cooling fan damaged.
[0043] Among them, the sensor acquisition array is used to collect the sound signals of the transformer in normal state, the sound signals of the winding damage state, the sound signals of the core fault state and the sound signals of the cooling fan in normal state, and then the spatial filtering function of beamforming is used to reduce the noise of the collected signals to obtain effective sound signal data.
[0044] Preferably, the number of cooling fans turned on and the speed data of the cooling fans are obtained and input into a softmax classifier to obtain a cooling influence factor on the transformer, and the temperature drop value of the transformer is collected and corrected according to the cooling influence factor.
[0045] Among them, by using different input data and prediction models according to the different settings of the cooling fan, the operating status of the transformer is judged, thereby further avoiding the influence of the sound signal caused by the opening of the cooling fan of the transformer, improving the accuracy of the operating status judgment, and by further combining the cooling method of the transformer, the real-time operating load of the transformer, and the weight of the transformer, the operating status of the transformer is judged from multiple angles, and at the same time, the technical problem of inaccurate operating status judgment results caused by the single use of sound characteristic signals is avoided.
[0046] Preferably, the transformer temperature reduction numerical calculation formula is:
[0047]
[0048] Among them, G1 is the cooling influence factor. When the value is 0.3, it means air cooling, when it is 0.5, it means forced oil cooling, and when it is 1, it means natural oil cooling. D is the temperature reduction value.
[0049] Among them, the transformer temperature reduction coefficient is different according to the different cooling methods of the transformer and the different numbers of cooling fans. In the above formula, G1 is the cooling influencing factor, which is specifically manifested as the external temperature of the transformer or the speed of the cooling fan, which will affect the cooling of the transformer. When the value of G1 is different, it will affect the subsequent transformer cooling value D.
[0050] Preferably, feature extraction is performed on the sound signal to obtain MFCC feature parameters and then GFCC feature parameters, and a prediction model is obtained by using the MFCC feature parameters, GFCC feature parameters, and the sound signal feature values of the cooling fan using a PSO-SVM algorithm to determine the operating state of the transformer.
[0051] Among them, the loudness threshold of the transformer is constructed by the speed of the cooling fan, the output power of the cooling fan, and the weight of the transformer, thereby further reducing the technical problem of inaccurate operation status judgment caused by the influence of the cooling fan operation on the sound signal of the transformer, and ensuring the efficiency and accuracy of the transformer status judgment.
[0052] Preferably, the operating state of the transformer includes a normal operating state, an abnormal operating state and a fault operating state.
[0053] Among them, the sound signal is denoised by the sensor acquisition array, and the background sound of the transformer is extracted, so that the effective acoustic data of the transformer can be effectively obtained, so that the operating status of the transformer can be judged more accurately and whether the operating status of the transformer is normal.
[0054] Preferably, the transformer sounds in different operating states are collected according to the sensor collection array, and the collected signals are preliminarily denoised using the spatial filtering function of beamforming to obtain 150 groups of data, with each group of data collecting for 10 seconds. Each group of data is sliced, with each slice interval being 2.5 seconds, to obtain 600 groups of original data.
[0055] Among them, the noise interference of the terminal's power equipment on the detection object is used to obtain a pure and unadulterated sound signal; then the noise reduction data is extracted through the voiceprint feature analysis module to output the frequency complexity, odd-even harmonic ratio and main frequency proportion.
[0056] Preferably, the original data is subjected to LCD noise reduction processing, that is, the data is first decomposed into 8 intrinsic scale components and one residual component, the first 6 components with higher energy are screened out from ISC for reconstruction, and the background noise with lower energy is eliminated to obtain the denoised transformer acoustic data.
[0057] Among them, the characteristic values are substituted into the neural network model for training through the voiceprint feature analysis module, and the transformer fault diagnosis model is constructed to finally identify the operating status of the transformer. It can accurately determine whether the transformer is faulty without affecting the operation of the transformer itself, thereby improving the operation and maintenance management capabilities of the substation.
[0058] Preferably, feature extraction is performed on 600 sets of denoised original data, the frequency complexity, odd and even harmonics and main frequency proportion of the 600 sets of data are calculated, and 300 sets of feature values are randomly selected as training sets, and the remaining 300 sets are used as test sets. Cross-validation is performed 10 times to obtain a transformer state recognition model; finally, a new test is performed and the data is substituted to obtain the transformer state.
[0059] Furthermore, the sound signal is subjected to noise reduction processing by the sensor acquisition array, and the background sound of the transformer is extracted, so that the effective acoustic data of the transformer can be effectively obtained, thereby enabling a more accurate judgment of the operating status of the transformer.
[0060] Working principle: Step 1: First, the sound of the transformer in different operating states is collected according to the sensor acquisition array, and the spatial filtering function of beamforming is used to perform preliminary noise reduction on the collected signal to obtain 150 sets of data. The collection time of each set of data is 10s. Each set of data is sliced, and the interval between each slice is 2.5s, so 600 sets of original data are obtained;
[0061] Step 2: Perform LCD noise reduction on the original data, that is, first decompose the data into 8 intrinsic scale components and one residual component, select the first 6 components with higher energy from ISC for reconstruction, and remove the background noise with lower energy to obtain the denoised transformer acoustic data, wherein the noise interference of the terminal power equipment on the detection object is used to obtain a pure and unadulterated sound signal; then extract the features of the noise reduction data through the voiceprint feature analysis module, output the frequency complexity, odd and even harmonic ratio and main frequency proportion, and finally substitute the feature value into the neural network model for training through the voiceprint feature analysis module, build a transformer fault diagnosis model, and finally identify the transformer operation status, which can accurately judge whether the transformer is faulty without affecting the operation of the transformer itself, thereby improving the operation and maintenance management capabilities of the substation;
[0062] Step 3: Extract features from 600 sets of denoised original data, calculate the frequency complexity, odd and even harmonics and main frequency ratio of the 600 sets of data, and randomly select 300 sets of feature values as training sets, and the remaining 300 sets as test sets, cross-validate 10 times, and obtain the transformer state recognition model; finally, through a new test, substitute the data to obtain the transformer state, wherein the sound signal is denoised through the set sensor acquisition array, and the background sound of the transformer is proposed, so as to effectively obtain the effective acoustic data of the transformer, so as to more accurately judge the operating state of the transformer.
[0063] What is described above is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.
Claims
1. A transformer state perception method based on multi-sensor information fusion, characterized in that: The following steps are involved: Step 1: setting a plurality of sound sensors and temperature sensors, wherein the sound sensor collects sound signals of the transformer in different states, and the temperature sensor is used to determine whether the cooling fan of the transformer is in operation; Step 2: Perform noise reduction processing on the sound signal collected by the sound sensor; Step 3: extracting the sound signal after noise reduction processing to characterize the operating status of the transformer; Step 4: distinguishing the transformer sound signal and the cooling fan sound signal by using a softmax classifier; Step 5: Determine and output the current status of the transformer and the cooling fan by distinguishing the results.
2. A transformer state perception method based on multi-sensor information fusion according to claim 1, characterized in that: The sound sensor performs a convolution layer operation on the collected sound signal. The convolution layer can perform a convolution operation on the input data to generate a local connection of the feature vector. The input data and the convolution kernel in the convolution layer are both locally connected.
3. The transformer state perception method based on multi-sensor information fusion according to claim 1 is characterized in that: A plurality of the sound sensors form a sensor acquisition array, which collects sound signals of the transformer in a normal state, sound signals of a winding damaged state, sound signals of a core fault state, sound signals of the cooling fan in a normal state, and state signals of the cooling fan in a damaged state.
4. The transformer state perception method based on multi-sensor information fusion according to claim 1 is characterized in that: The number of cooling fans turned on and the speed data of the cooling fans are obtained, and the temperature sensor measures the front and rear temperatures of the transformer, and inputs them into a softmax classifier to obtain the cooling influence factor of the transformer, and collects and corrects the temperature drop value of the transformer according to the cooling influence factor.
5. A transformer state perception method based on multi-sensor information fusion according to claim 4, characterized in that: The transformer temperature reduction numerical calculation formula is: Among them, G1 is the cooling influence factor. When the value is 0.3, it means air cooling, when it is 0.5, it means forced oil cooling, and when it is 1, it means natural oil cooling. D is the temperature reduction value.
6. The transformer state perception method based on multi-sensor information fusion according to claim 1 is characterized in that: The sound signal is feature extracted to obtain MFCC feature parameters and then GFCC feature parameters, and a prediction model is obtained by using the MFCC feature parameters, the GFCC feature parameters, and the sound signal feature values of the cooling fan using a PSO-SVM algorithm to determine the operating state of the transformer.
7. The transformer state perception method based on multi-sensor information fusion according to claim 3 is characterized in that: The operating state of the transformer includes a normal operating state, an abnormal operating state and a fault operating state.
8. The transformer state perception method based on multi-sensor information fusion according to claim 7 is characterized in that: The transformer sounds in different operating states are collected according to the sensor collection array, and the collected signals are preliminarily denoised using the spatial filtering function of beamforming to obtain 150 sets of data. The collection time of each set of data is 10 seconds. Each set of data is sliced, and the interval between each slice is 2.5 seconds, so as to obtain 600 sets of original data.
9. The transformer state perception method based on multi-sensor information fusion according to claim 8 is characterized in that: The original data is subjected to LCD noise reduction processing, that is, the data is first decomposed into 8 intrinsic scale components and one residual component, the first 6 components with higher energy are screened out from ISC for reconstruction, and the background noise with lower energy is eliminated to obtain the denoised transformer acoustic data.
10. A transformer state perception method based on multi-sensor information fusion according to claim 9, characterized in that: Feature extraction is performed on the 600 sets of original data after denoising, and the frequency complexity, odd and even harmonics and main frequency proportion of the 600 sets of data are calculated. 300 sets of feature values are randomly selected as training sets, and the remaining 300 sets are used as test sets. Cross-validation is performed 10 times to obtain the transformer state recognition model. Finally, a new test is performed and the data is substituted to obtain the transformer state.