Method, medium and system for evaluating water content in transformer insulation paper

Abstract

The invention provides a method, medium and system for evaluating the water content in transformer insulation paper, and belongs to the technical field of transformer insulation paper.The method comprises the steps that a plurality of collection points are evenly distributed on the transformer insulation paper, and a reflection signal is obtained on each point through an ultrasonic transmitting and receiving device; the collected reflection signals are preprocessed, and then the reflection signals and the emission signals are aligned through cross-correlation analysis to obtain emission-reflection signal pairs under all frequencies. The amplitude similarity, the phase similarity and the frequency spectrum similarity of each emission-reflection signal pair are calculated, and the amplitude similarity, the phase similarity and the frequency spectrum similarity are spliced into a similarity matrix according to frequency; and inputting the similarity matrix into a pre-trained water content evaluation model to obtain the water content of the insulation paper at each acquisition point. And calculating the water content distribution of the whole insulation paper by adopting an interpolation method based on the acquisition point coordinates and the water content data. The water content of the insulation paper is evaluated by utilizing the change of the ultrasonic reflection characteristic, and online continuous efficient detection can be quickly realized without dismounting the transformer.

Description

Technical Field

[0001] The present invention belongs to the technical field of transformer insulation paper, and in particular, relates to a method, a medium and a system for evaluating the moisture content in transformer insulation paper. Background Art

[0002] Transformers are indispensable key equipment in power systems, and the quality of their insulation performance is directly related to the safe operation of the equipment. As the main insulating material inside the transformer, the moisture content of insulation paper is one of the important indicators for judging insulation performance. Excessive moisture content will reduce insulation strength, lead to insulation breakdown, and cause serious power accidents. Therefore, accurate evaluation of the moisture content of transformer insulation paper has important engineering significance.

[0003] Existing methods for detecting moisture content of insulating paper mainly include:

[0004] 1. Laboratory sampling and testing method. This method is to take samples of transformer insulation paper back to the laboratory and measure the moisture content by drying and weighing. This method is highly accurate, but it requires stopping the machine for sampling, which is complicated to operate and cannot be monitored continuously.

[0005] 2. Insulating oil dissolved moisture detection method. This method indirectly estimates the moisture content of insulating paper by measuring the moisture content dissolved in the insulating oil. This method is simple to operate, but it can only reflect the moisture in the insulating oil, and cannot directly reflect the moisture content of the insulating paper itself.

[0006] 3. Infrared spectroscopy. This method uses the absorption characteristics of insulating paper in a specific infrared band to detect moisture content. This method does not require sampling and can achieve online monitoring, but it is greatly affected by factors such as ambient temperature and humidity, and its accuracy needs to be improved.

[0007] Although the above existing detection methods have their own advantages and disadvantages, generally speaking, they are unable to achieve online, continuous and efficient detection of the moisture content in transformer insulation paper. Summary of the invention

[0008] In view of this, the present invention provides a method, medium and system for evaluating the moisture content in transformer insulation paper, which can solve the technical problem that the prior art cannot achieve online, continuous and efficient detection of the moisture content in transformer insulation paper.

[0009] The present invention is achieved in that:

[0010] A first aspect of the present invention provides a method for evaluating the moisture content in transformer insulating paper, wherein the method is used to collect a reflection signal of the insulating paper in the transformer using an ultrasonic transceiver, and use the reflection signal to evaluate the moisture content of the insulating paper, and specifically comprises the following steps:

[0011] S10, determining multiple collection points in a uniformly distributed manner on the insulation paper of the transformer;

[0012] S20, for each collection point, after adjusting the extension line of the transmitting end of the ultrasonic transmitter and the extension line of the receiving end of the ultrasonic receiver to intersect at the collection point, control the ultrasonic transmitter to continuously emit ultrasonic waves of an equidistant frequency sequence, recorded as a transmission signal group, and simultaneously receive the reflection signal group collected by the ultrasonic receiver;

[0013] S30, preprocessing the reflection signal group to obtain a preprocessed reflection signal group;

[0014] S40, aligning the transmission signal group and the pre-processed reflection signal group by a cross-correlation method to obtain transmission-reflection signal pairs of different frequencies;

[0015] S50, calculating a fusion similarity array of a transmission signal and a reflection signal in a transmission-reflection signal pair of each frequency, including amplitude similarity, phase similarity, and spectrum similarity;

[0016] S60, concatenating the fusion similarity arrays corresponding to each frequency into a similarity matrix in order of frequency from low to high;

[0017] S70, inputting the similarity matrix into a pre-trained transformer insulation paper moisture content assessment model to obtain the moisture content of the insulation paper, which is recorded as the moisture content of the set point;

[0018] S80. Based on the coordinates of each acquisition point and its moisture content, the moisture content of the entire insulation paper is calculated using an interpolation method and output.

[0019] The ultrasonic transceiver device includes an ultrasonic transmitter and an ultrasonic receiver. The ultrasonic transmitter is used to transmit ultrasonic waves of different frequencies, and the ultrasonic receiver is used to receive ultrasonic waves reflected by insulating paper.

[0020] Wherein, the arithmetic difference frequency sequence is more than 100 arithmetic difference frequencies.

[0021] Furthermore, the frequency range of the ultrasonic waves of different frequencies is between 100 kHz and 1 MHz.

[0022] Wherein, the step of preprocessing the reflection signal group includes filtering, normalization and time domain windowing.

[0023] Filtering: Use a bandpass filter to filter the reflected signal to remove high-frequency noise and low-frequency drift and retain valid frequency band information. The cutoff frequency of the filter can be set between 50kHz and 2MHz.

[0024] Normalization: Normalize the reflected signal so that the amplitude of all signals is between 0 and 1, eliminating the impact of amplitude differences.

[0025] Time domain windowing: Using a Hamming window to window the reflected signal can suppress the peaks at both ends of the time domain and reduce frequency domain leakage.

[0026] The amplitude similarity is calculated using the Euclidean distance, the phase similarity is calculated using the standard deviation of the phase difference, and the spectrum similarity is calculated using the Euclidean distance of the spectrum power difference.

[0027] Wherein, the moisture content evaluation model in the transformer insulation paper adopts a multi-layer perceptron or a convolutional neural network.

[0028] Wherein, the interpolation method used to calculate the moisture content of the entire insulating paper is a bilinear interpolation method.

[0029] Wherein, the step S10 specifically includes: dividing the transformer insulation paper into a number of small squares, each small square having a size of 5 cm x 5 cm, and then determining a collection point at the center of each small square, and determining multiple collection points in this uniformly distributed manner, and the coordinates of the collection points are represented by (x, y), wherein x represents the horizontal coordinate and y represents the vertical coordinate, so that the entire insulation paper can be fully covered, providing comprehensive data support for subsequent moisture content analysis.

[0030] Wherein, the step S20 specifically includes: firstly, adjusting the positions of the ultrasonic transmitter and the ultrasonic receiver so that their extension lines intersect at the collection point, ensuring that the emitted ultrasonic wave can directly irradiate the surface of the insulating paper at the collection point, and the reflected signal can also be captured by the receiver; secondly, controlling the ultrasonic transmitter to continuously emit a group of ultrasonic pulses with equal frequency, the frequency range is between 100 kHz and 1 MHz, and the frequency interval is 10 kHz. By emitting such a group of ultrasonic pulses with continuously changing frequencies, the reflection characteristics of the insulating paper at different frequencies can be obtained; at the same time, the ultrasonic receiver will capture the echo signals of these ultrasonic pulses after being reflected on the surface of the insulating paper, which are recorded as the reflection signal group.

[0031] The step S30 specifically includes: first, filtering the reflected signal with a bandpass filter to remove high-frequency noise and low-frequency drift, retaining effective frequency band information, and setting the filter cutoff frequency between 50 kHz and 2 MHz; then, normalizing the reflected signal so that the amplitude of all signals is between 0 and 1, eliminating the influence of amplitude differences; finally, windowing the reflected signal with a Hamming window to suppress the peaks at both ends of the time domain and reduce frequency domain leakage. Through the above preprocessing steps, a clearer and cleaner reflected signal group can be obtained, laying the foundation for subsequent correlation analysis.

[0032] Wherein, the step S40 specifically includes: using the cross-correlation method to align the transmission signal group and the pre-processed reflection signal group, by calculating the peak position of the cross-correlation function, the time alignment relationship between the transmission signal and the reflection signal can be determined, for each transmission frequency, a pair of corresponding transmission-reflection signals can be obtained, and these signal pairs constitute transmission-reflection signal pairs at different frequencies. In this way, the reflection characteristics of the insulating paper in each frequency band can be obtained, providing a basis for subsequent moisture content analysis.

[0033] Wherein, the step S50 specifically includes: in order to comprehensively evaluate the similarity between the transmitted signal and the reflected signal, it is necessary to analyze from three perspectives: amplitude similarity, phase similarity and spectrum similarity. The amplitude similarity can be calculated using the Euclidean distance, the phase similarity can be calculated using the standard deviation of the phase difference, and the spectrum similarity can be calculated using the Euclidean distance of the spectrum power difference. Through these three similarity indicators, the similarity between the transmitted signal and the reflected signal can be comprehensively evaluated, providing a basis for the subsequent water content evaluation.

[0034] Among them, the step S60 specifically includes: arranging the fused similarity arrays (including amplitude similarity, phase similarity and spectrum similarity) at various frequencies calculated in step S50 in order from low to high frequency to form a similarity matrix, each row of the matrix corresponds to a similarity index, and each column corresponds to a frequency point. Such a similarity matrix can reflect the overall reflection characteristics of the insulating paper at different frequencies, and provide comprehensive data support for subsequent moisture content evaluation.

[0035] Wherein, the step S70 specifically comprises: firstly, collecting a batch of transformer insulation paper samples with known moisture content, and using the ultrasonic measurement method described in the present invention to obtain the similarity matrix of each sample, the moisture content of these samples can be obtained by laboratory measurement, and used as the label of the training set; then using a deep learning-based regression model, such as a multi-layer perceptron (MLP) or a convolutional neural network (CNN), taking the similarity matrix as input and the moisture content as a label, to perform model training until the prediction effect of the model on the validation set reaches a satisfactory level; after the training is completed, a pre-trained transformer insulation paper moisture content assessment model is obtained, and in practical applications, it is only necessary to input the similarity matrix obtained in step S60 into the model to obtain the moisture content of the insulation paper at the corresponding collection point.

[0036] Wherein, the step S80 specifically includes: using the bilinear interpolation method, according to the known acquisition point coordinates and their moisture content, estimating the moisture content distribution of the entire insulating paper area, the bilinear interpolation formula can realize smooth interpolation on a two-dimensional plane, in this way, a continuous insulating paper moisture content distribution map can be obtained, which not only reflects the moisture content of each acquisition point, but also can estimate the moisture content of the unmeasured area, providing a basis for further analysis and decision-making.

[0037] Optionally, the arithmetical frequency sequence emitted by the ultrasonic transmitter includes more than 100 frequency points with a frequency interval of 10 kHz, so that the reflection characteristics of the insulating paper in a wider frequency band can be obtained, providing richer data support for moisture content assessment.

[0038] Preferably, the filter cutoff frequency of the preprocessed reflection signal group is between 50 kHz and 2 MHz. This frequency band can cover the main frequency band of the reflection characteristics of the insulating paper, remove high-frequency noise and low-frequency drift, and retain effective frequency band information.

[0039] Preferably, the size of the similarity matrix is ​​3 rows and N columns, wherein the 3 rows correspond to amplitude similarity, phase similarity and spectrum similarity, respectively, and the N columns correspond to different frequency points. Such a matrix structure can fully reflect the reflection characteristics of the insulating paper in each frequency band, and provide a sufficient basis for the subsequent moisture content evaluation.

[0040] Optionally, the pre-trained transformer insulation paper moisture content assessment model adopts a multi-layer perceptron (MLP) structure, which can effectively learn the mapping relationship between the input similarity matrix and the moisture content. In practical applications, the moisture content of the insulation paper can be obtained by simply inputting the similarity matrix into the model.

[0041] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the above-mentioned method for evaluating moisture content in transformer insulation paper.

[0042] A third aspect of the present invention provides a system for evaluating the moisture content in transformer insulating paper, which comprises an ultrasonic transceiver and the above-mentioned computer-readable storage medium, wherein the ultrasonic transceiver is arranged in a transformer whose insulating paper moisture content is to be measured, and the ultrasonic transceiver comprises an ultrasonic transmitter and an ultrasonic receiver, wherein the ultrasonic transmitter is used to transmit ultrasonic waves of different frequencies toward the insulating paper of the transformer, and the ultrasonic receiver is used to receive ultrasonic waves reflected by the insulating paper.

[0043] Compared with the prior art, the method, medium and system for evaluating moisture content in transformer insulation paper provided by the present invention have the following beneficial effects:

[0044] 1. Continuous monitoring of the overall moisture content of the insulation paper is achieved. The method of the present invention uses multiple evenly distributed collection points to fully reflect the overall moisture content distribution of the insulation paper, rather than being limited to local or indirect monitoring.

[0045] 2. Greatly improved measurement accuracy and sensitivity. The present invention uses ultrasonic measurement technology, combined with advanced digital signal processing algorithms, to accurately capture the tiny changes in the reflection characteristics of insulating paper, thereby accurately evaluating its moisture content. Compared with existing methods such as infrared spectroscopy, the present method is less affected by environmental factors and has higher measurement reliability.

[0046] 3. Simplified the detection process and reduced the detection cost. The method of the present invention does not require shutdown for sampling, and only requires a small number of ultrasonic sensors to be installed on the transformer to achieve online monitoring. The operation is simple, and the equipment investment and maintenance costs are low, which is conducive to wide application in power systems.

[0047] In summary, the method for evaluating the moisture content in transformer insulation paper proposed in the present invention realizes rapid, accurate and continuous monitoring of the overall moisture content of the insulation paper, and solves the technical problem that the prior art cannot realize online, continuous and efficient detection of the moisture content in transformer insulation paper. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A flow chart of the method provided by the present invention. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0050] like Figure 1 As shown, it is a flow chart of a method for evaluating moisture content in transformer insulation paper provided by the present invention. The method is used to collect the reflected signal of the insulation paper in the transformer by using an ultrasonic transceiver, and evaluate the moisture content of the insulation paper by using the reflected signal, and specifically comprises the following steps:

[0051] S10, determining multiple collection points in a uniformly distributed manner on the insulation paper of the transformer;

[0052] S20, for each collection point, after adjusting the extension line of the transmitting end of the ultrasonic transmitter and the extension line of the receiving end of the ultrasonic receiver to intersect at the collection point, control the ultrasonic transmitter to continuously emit ultrasonic waves of an equidistant frequency sequence, recorded as a transmission signal group, and simultaneously receive the reflection signal group collected by the ultrasonic receiver;

[0053] S30, preprocessing the reflection signal group to obtain a preprocessed reflection signal group;

[0054] S40, aligning the transmission signal group and the pre-processed reflection signal group by using a cross-correlation method to obtain transmission-reflection signal pairs of different frequencies;

[0055] S50, calculating a fusion similarity array of a transmission signal and a reflection signal in a transmission-reflection signal pair of each frequency, including amplitude similarity, phase similarity, and spectrum similarity;

[0056] S60, concatenating the fusion similarity arrays corresponding to each frequency into a similarity matrix in order of frequency from low to high;

[0057] S70, inputting the similarity matrix into a pre-trained transformer insulation paper moisture content assessment model to obtain the moisture content of the insulation paper, which is recorded as the moisture content of the set point;

[0058] S80. Based on the coordinates of each acquisition point and its moisture content, the moisture content of the entire insulation paper is calculated using an interpolation method and output.

[0059] The specific implementation methods of the above steps are described in detail below:

[0060] Step S10: Determine multiple collection points in a uniformly distributed manner on the insulation paper of the transformer

[0061] The specific implementation method is: first, the transformer insulation paper is divided into several small squares, and the size of each small square is determined by the actual situation, generally 5 cm x 5 cm. Then a collection point is determined at the center of each small square, so that multiple evenly distributed collection points are obtained. The coordinates of the collection points can be expressed as (x, y), where x represents the horizontal coordinate and y represents the vertical coordinate. By determining the collection points in this evenly distributed manner, the entire insulation paper can be fully covered, providing comprehensive data support for subsequent moisture content analysis.

[0062] Step S20: For each collection point, after adjusting the extension line of the transmitting end of the ultrasonic transmitter and the extension line of the receiving end of the ultrasonic receiver to intersect at the collection point, control the ultrasonic transmitter to continuously emit ultrasonic waves of an equidistant frequency sequence, recorded as a transmission signal group, and simultaneously receive the reflection signal group collected by the ultrasonic receiver

[0063] The specific implementation method is: first, adjust the positions of the ultrasonic transmitter and the ultrasonic receiver so that their extension lines intersect at the collection point. This ensures that the emitted ultrasonic wave can directly irradiate the surface of the insulating paper at the collection point, and the reflected signal can also be captured by the receiver.

[0064] Secondly, the ultrasonic transmitter is controlled to continuously emit a group of ultrasonic pulses with equal frequency, which is recorded as the transmission signal group. The frequency range of this group of ultrasonic pulses is generally between 100kHz and 1MHz, and the frequency interval is 10kHz. By emitting such a group of ultrasonic pulses with continuously changing frequencies, the reflection characteristics of the insulating paper at different frequencies can be obtained.

[0065] At the same time, the ultrasonic receiver will capture the echo signals after these ultrasonic pulses are reflected on the surface of the insulating paper, which are recorded as the reflection signal group. These reflection signals carry the structural information of the insulating paper and can be used to evaluate its moisture content.

[0066] Step S30: preprocess the reflection signal group to obtain a preprocessed reflection signal group

[0067] The specific implementation method is: for the acquired reflection signal group, the following preprocessing steps need to be performed:

[0068] 1. Filtering: Use a bandpass filter to filter the reflected signal to remove high-frequency noise and low-frequency drift and retain valid frequency band information. The cutoff frequency of the filter can be set between 50kHz and 2MHz.

[0069] 2. Normalization: Normalize the reflected signal so that the amplitude of all signals is between 0 and 1, eliminating the influence of amplitude differences. Normalization can use the maximum and minimum value normalization formula:

[0070] 3. Time domain windowing: Using the Hamming window to process the reflected signal can suppress the peaks at both ends of the time domain and reduce frequency domain leakage. The expression of the Hamming window function is: Where N is the signal length.

[0071] Through the above preprocessing steps, a clearer and cleaner reflection signal group can be obtained, laying the foundation for subsequent related analysis.

[0072] Step S40: Align the transmission signal group and the pre-processed reflection signal group using a cross-correlation method to obtain transmission-reflection signal pairs of different frequencies.

[0073] The specific implementation method is: Cross-correlation is a method commonly used in signal analysis, which can be used to measure the similarity and time delay between two signals. In this step, the transmission signal group and the pre-processed reflection signal group are cross-correlated to find the time alignment relationship between the transmission signal and the corresponding reflection signal.

[0074] The expression of the cross-correlation function is:

[0075] Where x(t) is the transmitted signal, y(t) is the reflected signal, and τ is the time delay. xy The peak position of (τ) can determine the time alignment relationship between the transmitted signal and the reflected signal.

[0076] For each transmission frequency, a pair of corresponding transmission-reflection signals can be obtained, and these signal pairs constitute transmission-reflection signal pairs at different frequencies. In this way, the reflection characteristics of insulating paper in various frequency bands can be obtained, providing a basis for subsequent moisture content analysis.

[0077] Step S50: Calculate the fusion similarity array of the transmitted signal and the reflected signal in each frequency transmitted-reflected signal pair, including amplitude similarity, phase similarity and spectrum similarity.

[0078] The specific implementation method is: in order to comprehensively evaluate the similarity between the transmitted signal and the reflected signal, it is necessary to analyze from multiple angles, including amplitude similarity, phase similarity and spectrum similarity.

[0079] 1. Amplitude similarity: To calculate the amplitude difference between the transmitted signal and the reflected signal, the Euclidean distance formula can be used: Where x is the transmitted signal, y is the reflected signal, and N is the signal length. The amplitude similarity is

[0080] 2. Phase similarity: Calculate the phase difference between the transmitted signal and the reflected signal. The standard deviation of the phase difference can be used: where φ x and φ y are the phases of the transmitted signal and the reflected signal respectively. The phase similarity is

[0081] 3. Spectral similarity: To calculate the spectrum difference between the transmitted signal and the reflected signal, the Euclidean distance of the spectrum power difference can be used: Where P is the power spectrum density and M is the number of spectrum lines. The spectrum similarity is

[0082] Through the above three similarity indicators, the similarity between the transmitted signal and the reflected signal can be comprehensively evaluated, providing a basis for subsequent water content evaluation.

[0083] Step S60: Concatenate the fusion similarity arrays corresponding to each frequency into a similarity matrix in order from low to high frequency.

[0084] The specific implementation method is: the fusion similarity arrays (including amplitude similarity, phase similarity and spectrum similarity) at various frequencies calculated in step S50 are arranged in order from low to high frequency to form a similarity matrix.

[0085] Suppose there are N frequencies, each frequency corresponds to a 3D fusion similarity array, then the size of the similarity matrix is ​​3xN. Each row of the matrix corresponds to a similarity index, and each column corresponds to a frequency point.

[0086] Such a similarity matrix can reflect the overall reflection characteristics of insulating paper at different frequencies and provide comprehensive data support for subsequent moisture content evaluation.

[0087] Step S70: Input the similarity matrix into the pre-trained transformer insulation paper moisture content assessment model to obtain the moisture content of the insulation paper, which is recorded as the moisture content of the set point.

[0088] The specific implementation method is: a moisture content evaluation model for transformer insulation paper needs to be pre-trained, and the model can accept a similarity matrix as input and output the moisture content of the insulation paper.

[0089] Training data set: A batch of transformer insulation paper samples with known moisture content needs to be collected, and the similarity matrix of each sample is obtained using the ultrasonic measurement method described in the present invention. The moisture content of these samples can be obtained by laboratory measurement and used as the label of the training set.

[0090] Model structure: A deep learning-based regression model can be used, such as a multi-layer perceptron (MLP) or a convolutional neural network (CNN). Such models can effectively learn the mapping relationship between the input similarity matrix and the moisture content.

[0091] Training process: First, the training data is normalized, and then the similarity matrix is ​​used as input and the water content is used as the label. The model parameters are iteratively trained using optimization algorithms such as gradient descent until the prediction effect of the model on the validation set reaches a satisfactory level.

[0092] After the training is completed, a pre-trained transformer insulation paper moisture content assessment model is obtained. In practical applications, the similarity matrix obtained in step S60 is simply input into the model to obtain the insulation paper moisture content of the corresponding collection point.

[0093] Step S80: Based on the coordinates of each acquisition point and its moisture content, the moisture content of the entire insulation paper is calculated using the interpolation method and output

[0094] The specific implementation method is: in step S70, the moisture content of the insulation paper at each collection point has been obtained. Next, the moisture content distribution of the entire insulation paper needs to be calculated using the interpolation method.

[0095] Interpolation is a commonly used data fitting method, which can estimate the value in the unknown area based on the known data points. In this step, bilinear interpolation can be used, which can achieve smooth interpolation on a two-dimensional plane.

[0096] The bilinear interpolation formula is:

[0097] Among them, a i is the interpolation coefficient, φ i (x, y) are basis functions, which can be calculated based on the known coordinates of the four sampling points and their corresponding water contents.

[0098] By applying bilinear interpolation to the entire insulation paper area, a continuous moisture content distribution map can be obtained. This distribution map not only reflects the moisture content of each collection point, but also estimates the moisture content of the unmeasured area, providing a basis for further analysis and decision-making. Finally, this insulation paper moisture content distribution map can be output as the final result of the method of the present invention.

[0099] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the above-mentioned method for evaluating moisture content in transformer insulation paper.

[0100] A third aspect of the present invention provides a system for evaluating the moisture content in transformer insulating paper, which comprises an ultrasonic transceiver and the above-mentioned computer-readable storage medium, wherein the ultrasonic transceiver is arranged in a transformer whose insulating paper moisture content is to be measured, and the ultrasonic transceiver comprises an ultrasonic transmitter and an ultrasonic receiver, wherein the ultrasonic transmitter is used to transmit ultrasonic waves of different frequencies toward the insulating paper of the transformer, and the ultrasonic receiver is used to receive ultrasonic waves reflected by the insulating paper.

[0101] Specifically, the principle of the present invention is: by adopting ultrasonic measurement technology, combined with advanced digital signal processing algorithms, an accurate assessment of the overall moisture content of insulating paper is achieved. The technical principle is as follows: the moisture content has a significant effect on the acoustic properties of insulating paper: moisture will change the acoustic impedance of insulating paper, thereby affecting the propagation and reflection of ultrasonic waves in insulating paper. Generally speaking, the higher the moisture content, the lower the acoustic impedance of insulating paper, and the amplitude, phase and spectrum characteristics of the reflected signal of ultrasonic waves will also change accordingly. Specifically, an increase in the moisture content in insulating paper will cause the propagation speed of sound waves inside the paper to decrease, thereby causing the phase of the reflected signal to change; an increase in the moisture content in insulating paper will also cause the acoustic impedance of the paper to change, thereby changing the amplitude of the reflected signal; changes in the moisture content in insulating paper will also affect the sound wave spectrum inside the paper, which is manifested as changes in the spectrum of the reflected signal. The key to this method is to use the changes in these acoustic parameters to establish a prediction model for the moisture content of insulating paper: by emitting ultrasonic waves of different frequencies and receiving reflected signals, frequency-related acoustic parameter changes (amplitude, phase, spectrum) can be obtained. By correlating these acoustic parameters with the actual moisture content of the insulation paper, a prediction model can be established.

[0102] Specifically:

[0103] 1. Collect insulation paper reflection signal

[0104] First, multiple evenly distributed collection points are determined on the transformer insulation paper, and ultrasonic transmitters and receivers are used to measure at each collection point. The ultrasonic transmitter continuously emits a set of ultrasonic pulses with continuously changing frequencies, ranging from 100kHz to 1MHz, with a frequency interval of 10kHz. These ultrasonic pulses can directly irradiate the surface of the insulation paper and are reflected back to the ultrasonic receiver by the insulation paper.

[0105] 2. Extract reflection signal features

[0106] Next, the obtained reflected signal is preprocessed by filtering, normalizing and windowing to remove noise and anomalies and extract effective frequency domain and time domain features. Then, the cross-correlation algorithm is used to time-align the transmitted signal and the reflected signal to obtain the transmitted-reflected signal pairs at different frequencies.

[0107] 3. Calculate the similarity feature matrix

[0108] For each transmission-reflection signal pair, the amplitude similarity, phase similarity and spectrum similarity are calculated to form a 3×N similarity feature matrix, where N is the number of measured frequencies. This similarity feature matrix fully reflects the reflection characteristics of the insulating paper at different frequencies.

[0109] 4. Moisture content assessment based on machine learning

[0110] Finally, the similarity feature matrix is ​​input into the pre-trained deep learning regression model, which can learn the mapping relationship between the input features and the moisture content of the insulation paper. Through the inference of the model, the moisture content of the insulation paper at each collection point can be obtained.

[0111] The reason why the solution of the present invention can effectively solve the problems existing in the prior art is mainly due to the following reasons:

[0112] First of all, the use of ultrasonic measurement technology can comprehensively collect the reflection characteristic information of the insulating paper, rather than being limited to surface or indirect detection, and thus can more accurately reflect the overall moisture content of the insulating paper.

[0113] Secondly, through in-depth digital signal processing of the reflected signal, including time-frequency domain feature extraction and similarity calculation, it is possible to capture the slight changes in the reflection characteristics of the insulating paper, thereby greatly improving the accuracy and sensitivity of moisture content assessment.

[0114] Furthermore, the model training and inference process based on machine learning can effectively learn the complex mapping relationship between input features and moisture content, thereby achieving accurate estimation of the moisture content of insulating paper. This data-driven method has stronger adaptability and generalization ability than empirical formulas or model assumptions.

[0115] In order to better understand and implement the present invention, a specific embodiment 1 of the present invention is provided below. The steps of this embodiment 1 are specifically described as follows: Step S10: Determine multiple collection points in a uniformly distributed manner on the insulation paper of the transformer

[0116] The specific implementation method is: divide the transformer insulation paper into M×N small squares, and the size of each small square is L x ×L y Then determine a collection point at the center of each small square. The coordinates of the collection point can be expressed as (x i ,y j ),in Where i = 1, 2, ..., M; j = 1, 2, ..., N. By determining multiple collection points in this evenly distributed manner, the entire insulating paper can be fully covered, providing comprehensive data support for subsequent moisture content analysis.

[0117] Step S20: For each collection point, after adjusting the extension line of the transmitting end of the ultrasonic transmitter and the extension line of the receiving end of the ultrasonic receiver to intersect at the collection point, control the ultrasonic transmitter to continuously emit ultrasonic waves of an equidistant frequency sequence, recorded as a transmission signal group, and simultaneously receive the reflection signal group collected by the ultrasonic receiver

[0118] The specific implementation method is: first, adjust the positions of the ultrasonic transmitter and the ultrasonic receiver so that their extension lines are at the collection point (x i ,y j ), which ensures that the emitted ultrasonic wave can directly irradiate the surface of the insulating paper at the collection point, and the reflected signal can also be captured by the receiver.

[0119] Secondly, the ultrasonic transmitter is controlled to continuously emit a set of ultrasonic pulses with equal frequency, and the frequency range is f min to f max , the frequency interval is Δf, and a total of N f frequency pulses, this group of ultrasonic pulses is recorded as the transmitting signal group where f n =f min +(n-1)·Δf,n=1,2,…,N f By emitting such a set of ultrasonic pulses with continuously changing frequencies, the reflection characteristics of insulating paper at different frequencies can be obtained.

[0120] At the same time, the ultrasonic receiver will capture the echo signals after these ultrasonic pulses are reflected on the surface of the insulating paper, which are recorded as the reflected signal group. These reflected signals carry structural information of the insulating paper and can be used to assess its moisture content.

[0121] Step S30: preprocess the reflection signal group to obtain a preprocessed reflection signal group

[0122] The specific implementation is: for the acquired reflection signal group The following preprocessing steps are required:

[0123] 1. Filtering: Use a bandpass filter to filter the reflected signal to remove high-frequency noise and low-frequency drift and retain effective frequency band information. The cutoff frequency of the filter can be set to f low and f high The filtered reflection signal group is recorded as

[0124] 2. Normalization: Normalize the reflected signal so that the amplitude of all signals is between 0 and 1, eliminating the influence of amplitude differences. Normalization can use the maximum and minimum value normalization formula:

[0125] 3. Time domain windowing: Using the Hamming window to process the reflected signal can suppress the peaks at both ends of the time domain and reduce frequency domain leakage. The expression of the Hamming window function is: Where N is the signal length.

[0126] Through the above preprocessing steps, a clearer and cleaner preprocessed reflection signal group can be obtained. Lay the foundation for subsequent related analysis.

[0127] Step S40: Align the transmission signal group and the pre-processed reflection signal group using a cross-correlation method to obtain transmission-reflection signal pairs of different frequencies.

[0128] The specific implementation method is: Cross-correlation is a method commonly used in signal analysis, which can be used to measure the similarity and time delay between two signals. and preprocessed reflection signal group The purpose of performing cross-correlation calculations is to find the time alignment relationship between the transmitted signal and the corresponding reflected signal.

[0129] The cross-correlation function is defined as:

[0130] Where τ is the time delay. By calculating The peak position τ n , the time alignment relationship between the nth transmitted signal and the corresponding reflected signal can be determined.

[0131] For each transmission frequency f n , we can get a pair of corresponding transmission-reflection signals These signal pairs constitute the emission-reflection signal pairs at different frequencies. In this way, the reflection characteristics of the insulating paper in each frequency band can be obtained, providing a basis for the subsequent moisture content analysis.

[0132] Step S50: Calculate the fusion similarity array of the transmitted signal and the reflected signal in each frequency transmitted-reflected signal pair, including amplitude similarity, phase similarity and spectrum similarity.

[0133] The specific implementation is: In order to comprehensively evaluate the transmitted signal x n (t) and reflected signal The similarity of the two images needs to be analyzed from multiple perspectives, including amplitude similarity, phase similarity and spectrum similarity.

[0134] 1. Amplitude similarity: To calculate the amplitude difference between the transmitted signal and the reflected signal, the Euclidean distance formula can be used: The amplitude similarity is

[0135] 2. Phase similarity: Calculate the phase difference between the transmitted signal and the reflected signal. The standard deviation of the phase difference can be used: in and are the phases of the transmitted signal and the reflected signal respectively, and the phase similarity is

[0136] 3. Spectral similarity: To calculate the spectrum difference between the transmitted signal and the reflected signal, the Euclidean distance of the spectrum power difference can be used: in and are the power spectral densities of the transmitted signal and the reflected signal respectively, and the spectrum similarity is

[0137] Through the above three similarity indicators, the similarity between the transmitted signal and the reflected signal can be comprehensively evaluated, providing a basis for subsequent water content evaluation.

[0138] Step S60: Concatenate the fusion similarity arrays corresponding to each frequency into a similarity matrix in order from low to high frequency.

[0139] The specific implementation method is: the fusion similarity arrays (including amplitude similarity) at various frequencies calculated in step S50 are Phase Similarity and spectrum similarity ) are arranged in order from low to high frequency to form a similarity matrix S.

[0140] The size of the similarity matrix S is 3×N f , where each row corresponds to a similarity index and each column corresponds to a frequency point, which can be expressed as:

[0141]

[0142] Such a similarity matrix can reflect the overall reflection characteristics of insulating paper at different frequencies and provide comprehensive data support for subsequent moisture content evaluation.

[0143] Step S70: Input the similarity matrix into the pre-trained transformer insulation paper moisture content assessment model to obtain the moisture content of the insulation paper, which is recorded as the moisture content of the set point.

[0144] The specific implementation method is: a moisture content evaluation model for transformer insulation paper needs to be pre-trained, and the model can accept a similarity matrix S as input and output the moisture content of the insulation paper.

[0145] Training data set: A batch of transformer insulation paper samples with known moisture content need to be collected, and the ultrasonic measurement method described in the present invention is used to obtain the similarity matrix S of each sample (i) , where i = 1, 2, ..., N train Indicates the number of the training sample. The moisture content of these samples can be obtained by laboratory measurement and used as the label of the training set.

[0146] Model structure: A deep learning-based regression model can be used, such as a multi-layer perceptron (MLP) or a convolutional neural network (CNN). Such models can effectively learn the mapping relationship between the input similarity matrix S and the moisture content y.

[0147] Training process: First, normalize the training data and preprocess it, then convert the similarity matrix S (i) As input, the water content y (i) As a label, the loss function is used where f θ (·) represents the model output. The model parameters θ are iteratively trained using optimization algorithms such as gradient descent until the prediction effect of the model on the validation set reaches a satisfactory level.

[0148] After the training is completed, a pre-trained transformer insulation paper moisture content assessment model is obtained. In practical applications, the similarity matrix S obtained in step S60 is simply input into the model to obtain the insulation paper moisture content y at the corresponding collection point.

[0149] Step S80: Based on the coordinates of each acquisition point and its moisture content, the moisture content of the entire insulation paper is calculated using the interpolation method and output

[0150] The specific implementation method is: in step S70, each acquisition point (x i ,y j ) of the insulation paper moisture content y i,j Next, the moisture content distribution of the entire insulation paper needs to be calculated using the interpolation method.

[0151] Interpolation is a commonly used data fitting method, which can estimate the value in the unknown area based on the known data points. In this step, bilinear interpolation can be used, which can achieve smooth interpolation on a two-dimensional plane.

[0152] The bilinear interpolation formula is:

[0153]

[0154] Among them, a i is the interpolation coefficient, φ i (x, y) are basis functions, which can be calculated based on the known coordinates of the four sampling points and their corresponding water content y i,j Calculated.

[0155] In order to further better understand and implement the present invention, the following provides Example 2 of a specific application scenario of the present invention: A power operation and maintenance team is responsible for a 500MVA transformer, which is a key device in the power grid, and its operating status directly affects the stability of the entire power grid. In order to grasp the moisture content of the transformer insulation paper in real time, it is decided to use the ultrasonic measurement technology proposed in the present invention to monitor the transformer online.

[0156] First, the maintenance personnel evenly arranged 25 collection points on the surface of the transformer insulation paper. The coordinates of each collection point can be expressed as (xi, yj), where i = 1, 2, ..., 5, j = 1, 2, ..., 5, and the collection point spacing is 50 cm. At each collection point, an ultrasonic transmitter and an ultrasonic receiver are installed, and their extension lines intersect at the collection point.

[0157] These ultrasonic transmitters and receivers form a measurement system for obtaining the reflected signal from the surface of the insulating paper. Specifically, each ultrasonic transmitter continuously emits a set of ultrasonic pulses with a frequency between 100kHz and 1MHz and a frequency interval of 10kHz, for a total of 100 frequency points. These ultrasonic pulses can directly irradiate the surface of the insulating paper and be reflected by the insulating paper back to the corresponding ultrasonic receiver.

[0158] The reflected signal captured by the receiver must first be preprocessed. Specifically:

[0159] 1. Filtering: Use a bandpass filter to filter the reflected signal, and the cutoff frequency of the filter is 50kHz and 2MHz. This can effectively remove high-frequency noise and low-frequency drift, and retain effective frequency band information.

[0160] 2. Normalization: Normalize the amplitude of the reflected signal so that the amplitude of all signals is between 0 and 1, eliminating the influence of amplitude differences. Normalization uses the maximum and minimum value normalization formula:

[0161] 3. Windowing: Using a Hamming window to process the reflected signal can suppress the peaks at both ends of the time domain and reduce leakage in the frequency domain. The expression of the Hamming window function is: Where N is the signal length.

[0162] Through the above preprocessing, a clearer and cleaner reflection signal sequence can be obtained.

[0163] Next, the transmitted signal sequence and the preprocessed reflected signal sequence are cross-correlated to find the time alignment relationship between the transmitted signal and the corresponding reflected signal. The cross-correlation function is defined as: Where x(t) is the transmitted signal, y(t) is the reflected signal, and τ is the time delay. xy The peak position of (τ) can determine the time alignment relationship between the transmitted signal and the reflected signal. In this way, the transmitted-reflected signal pairs at 100 frequency points are obtained.

[0164] Then, the three indicators of amplitude similarity, phase similarity and spectrum similarity of each transmission-reflection signal pair are calculated to form a 3×100 similarity feature matrix.

[0165] The specific calculation method is as follows:

[0166] 1. Amplitude similarity: The amplitude difference between the transmitted signal and the reflected signal is calculated using the Euclidean distance formula: Where x is the transmitted signal, y is the reflected signal, and N is the signal length.

[0167] The amplitude similarity is

[0168] 2. Phase similarity: Calculate the standard deviation of the phase difference between the transmitted signal and the reflected signal: where φ x and φ y are the phases of the transmitted signal and the reflected signal respectively.

[0169] The phase similarity is

[0170] 3. Spectral similarity: Calculate the Euclidean distance of the spectrum difference between the transmitted signal and the reflected signal: Where P is the power spectrum density and M is the number of spectrum lines. The spectrum similarity is

[0171] The obtained similarity feature matrix data is shown in Table 1 below:

[0172] Table 1 Similarity index table

[0173] Similarity index 100kHz 110kHz ... 1MHz Amplitude Similarity 0.92 0.88 ... 0.83 Phase Similarity 0.95 0.91 ... 0.86 Spectral similarity 0.90 0.87 ... 0.81

[0174] Next, this similarity feature matrix is ​​input into a pre-trained deep learning regression model, which can learn the mapping relationship between the input features and the moisture content of the insulation paper. The training data set of the model includes 100 transformer insulation paper samples with known moisture content. The ultrasonic measurement of these samples is carried out using the method of the present invention to obtain the corresponding similarity feature matrix, which is then used as a label for model training with the actual moisture content of the samples.

[0175] Through training, the model has learned a more complex input-output mapping function and can more accurately predict the moisture content of the insulation paper corresponding to a given similarity feature matrix. In practical applications, simply input the similarity feature matrix of the 25 collection points obtained in the step into the model to obtain the moisture content data of the insulation paper at each collection point, as shown in Table 2:

[0176] Table 2 Water content of collection points

[0177] Collection point number Water content (%) (1,1) 0.85 (1,2) 0.92 ... ... (5,5) 0.79

[0178] Finally, the bilinear interpolation method was used to calculate the moisture content distribution of the entire insulation paper area based on the moisture content data of these 25 collection points.

[0179] The team summarized the application of the method of the present invention:

[0180] 1. Continuous monitoring of the overall moisture content of transformer insulation paper is achieved. 25 evenly distributed collection points can fully reflect the moisture content distribution of insulation paper, rather than being limited to local or indirect monitoring.

[0181] 2. The measurement accuracy and sensitivity are greatly improved. This method uses advanced ultrasonic measurement and digital signal processing technology to accurately capture the tiny changes in the reflection characteristics of the insulation paper, thereby accurately evaluating its moisture content, which is more reliable than methods such as infrared spectroscopy.

[0182] 3. The detection process is simple, and the equipment investment and maintenance costs are low. No need to stop the machine for sampling, only a small number of ultrasonic sensors need to be installed on the transformer to achieve online monitoring, which greatly reduces the difficulty and cost of detection.

[0183] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for evaluating moisture content in transformer insulation paper, characterized in that: The method is used to collect the reflection signal of the insulating paper in the transformer by using an ultrasonic transceiver, and evaluate the moisture content of the insulating paper by using the reflection signal, and specifically comprises the following steps: S10, determining multiple collection points in a uniformly distributed manner on the insulation paper of the transformer; S20, for each collection point, after adjusting the extension line of the transmitting end of the ultrasonic transmitter and the extension line of the receiving end of the ultrasonic receiver to intersect at the collection point, control the ultrasonic transmitter to continuously emit ultrasonic waves of an equidistant frequency sequence, recorded as a transmission signal group, and simultaneously receive the reflection signal group collected by the ultrasonic receiver; S30, preprocessing the reflection signal group to obtain a preprocessed reflection signal group; S40, aligning the transmission signal group and the pre-processed reflection signal group by a cross-correlation method to obtain transmission-reflection signal pairs of different frequencies; S50, calculating a fusion similarity array of a transmission signal and a reflection signal in a transmission-reflection signal pair of each frequency, including amplitude similarity, phase similarity, and spectrum similarity; S60, concatenating the fusion similarity arrays corresponding to each frequency into a similarity matrix in order of frequency from low to high; S70, inputting the similarity matrix into a pre-trained transformer insulation paper moisture content assessment model to obtain the moisture content of the insulation paper, which is recorded as the moisture content of the set point; S80. Based on the coordinates of each acquisition point and its moisture content, the moisture content of the entire insulation paper is calculated using an interpolation method and output.

2. The method for evaluating moisture content in transformer insulation paper according to claim 1, characterized in that: The ultrasonic transceiver comprises an ultrasonic transmitter and an ultrasonic receiver. The ultrasonic transmitter is used to transmit ultrasonic waves of different frequencies to the transformer insulation paper, and the ultrasonic receiver is used to receive the ultrasonic waves reflected by the insulation paper.

3. The method for evaluating moisture content in transformer insulation paper according to claim 1, characterized in that: The arithmetical frequency sequence is more than 100 arithmetical frequencies.

4. The method for evaluating moisture content in transformer insulation paper according to claim 2, characterized in that: The frequency range of ultrasonic waves of different frequencies is between 100kHz and 1MHz.

5. The method for evaluating moisture content in transformer insulation paper according to claim 1, characterized in that: The step of preprocessing the reflection signal group includes filtering, normalization and time domain windowing.

6. The method for evaluating moisture content in transformer insulation paper according to claim 1, characterized in that: The amplitude similarity is calculated using the Euclidean distance, the phase similarity is calculated using the standard deviation of the phase difference, and the spectrum similarity is calculated using the Euclidean distance of the spectrum power difference.

7. The method for evaluating moisture content in transformer insulation paper according to claim 1, characterized in that: The moisture content evaluation model in the transformer insulation paper adopts a multi-layer perceptron or a convolutional neural network.

8. The method for evaluating moisture content in transformer insulation paper according to claim 1, characterized in that: The interpolation method used to calculate the moisture content of the entire insulating paper adopts a bilinear interpolation method.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the method for evaluating moisture content in transformer insulation paper according to any one of claims 1 to 8.

10. A system for evaluating moisture content in transformer insulation paper, characterized in that: The invention comprises an ultrasonic transceiver and a computer-readable storage medium as described in claim 9, wherein the ultrasonic transceiver is arranged in a transformer whose insulation paper moisture content is to be measured, and the ultrasonic transceiver comprises an ultrasonic transmitter and an ultrasonic receiver, wherein the ultrasonic transmitter is used to transmit ultrasonic waves of different frequencies toward the insulation paper of the transformer, and the ultrasonic receiver is used to receive ultrasonic waves reflected by the insulation paper.

Images