A high-precision photoacoustic spectroscopy method for monitoring gas in transformer oil

Abstract

This invention discloses a high-precision photoacoustic spectral monitoring method for gases in transformer oil, belonging to the field of transformer oil and gas monitoring technology. The method includes the following steps: constructing a photoacoustic spectral detection system to sample gases in the transformer oil and acquire photoacoustic spectral data of the gases. The photoacoustic spectral detection system includes a near-infrared tunable fiber laser, a photoacoustic cell, a lock-in amplifier, and a data acquisition system. The acquired photoacoustic spectral data is preprocessed, and fault diagnosis-related features are extracted from the preprocessed photoacoustic spectral data. Chemometric methods are used to decouple the photoacoustic spectral data of the mixed gases. This invention, by real-time acquisition and analysis of photoacoustic spectral data of gases in transformer oil, can detect abnormalities inside the transformer early. By combining chemometrics and machine learning algorithms to decouple and separate the spectral data of the mixed gases, it can accurately identify and quantitatively analyze various gas components.

Description

Technical Field

[0001] This invention relates to the field of transformer oil and gas monitoring technology, specifically to a high-precision photoacoustic spectroscopy monitoring method for gases in transformer oil. Background Technology

[0002] As a key piece of equipment in the power system, the stable operation of transformers is crucial for ensuring the safety and reliability of the power grid. Transformers use the principle of electromagnetic induction to transform voltage from one level to another, realizing the transmission and distribution of electrical energy. In power plants and substations, transformers typically need to operate under high loads for extended periods, and their operating status directly affects the stability and economic benefits of the entire power system. Transformer oil, as an important insulating and cooling medium for transformers, dissolves or generates various gases during transformer operation. The types and amounts of these gases are closely related to the transformer's operating status and fault types. By monitoring and analyzing the gases in transformer oil, potential transformer faults can be detected in a timely manner, providing important information for transformer maintenance and repair.

[0003] In existing technologies, when detecting multiple mixed gases, the absorption spectra of different gases overlap, leading to cross-interference, which in turn affects fault diagnosis and monitoring based on photoacoustic spectral data. Therefore, how to decouple and separate the spectral data of mixed gases and mine and analyze the photoacoustic spectral data to improve the accuracy of fault diagnosis and monitoring is the problem we need to solve. To this end, we propose a high-precision photoacoustic spectral monitoring method for gases in transformer oil. Summary of the Invention

[0004] The purpose of this invention is to provide a high-precision photoacoustic spectroscopy monitoring method for gases in transformer oil, so as to solve the problems mentioned in the background art.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0006] A high-precision photoacoustic spectroscopy method for monitoring gas in transformer oil includes the following steps:

[0007] Step 1: Construct a photoacoustic spectroscopy detection system to sample the gas in the transformer oil and obtain the photoacoustic spectral data of the gas. The photoacoustic spectroscopy detection system includes a near-infrared tunable fiber laser, a photoacoustic cell, a lock-in amplifier, and a data acquisition system. The near-infrared tunable fiber laser emits infrared radiation of a specific wavelength. The frequency of the light source is modulated by a modulation disk to ensure that the light source changes periodically at a specific frequency.

[0008] Step 2: Preprocess the acquired photoacoustic spectral data, including filtering, noise reduction, and baseline correction, to improve data quality and extract fault diagnosis-related features from the preprocessed photoacoustic spectral data.

[0009] Step 3: Decouple the photoacoustic spectral data of the mixed gas using chemometric methods, and use machine learning algorithms to separate and identify the photoacoustic spectral characteristics of different gases, removing cross-interference between different gases;

[0010] Step 4: Analyze the decoupled and separated photoacoustic spectral data, extract feature information related to transformer fault diagnosis, and establish a fault diagnosis model to identify different fault types and states;

[0011] Step 5: Integrate the established fault diagnosis model into the real-time monitoring system to perform real-time analysis of the gas in the transformer oil. Combine the constructed fault diagnosis model with historical fault cases to set different fault levels and assess the severity of the fault.

[0012] Step 6: Based on the data analysis and fault assessment results, generate a transformer fault monitoring report and issue an early warning signal to prompt further inspection and maintenance. The report includes information on fault type, fault severity, development trend, and recommended maintenance measures.

[0013] A further improvement to the technical solution of this invention lies in the following: In step 1, the process of constructing the photoacoustic spectroscopy detection system and acquiring gas photoacoustic spectral data is as follows:

[0014] Step 101: Prepare the components of the photoacoustic spectroscopy detection system, including a near-infrared tunable fiber laser, a photoacoustic cell, a lock-in amplifier, and a data acquisition system. Connect and debug the near-infrared tunable fiber laser, modulation disk, photoacoustic cell, lock-in amplifier, and data acquisition system to ensure that the optical paths, circuits, and signal transmission paths between the components are unobstructed.

[0015] Step 102: The gas sample in the transformer oil is degassed by vacuum and placed into the photoacoustic cell. The near-infrared tunable fiber laser is started and the required output wavelength is set. The frequency of the light source is modulated by the modulation disk so that the laser periodically irradiates the gas sample in the photoacoustic cell at a specific frequency.

[0016] Step 103: In the photoacoustic cell, gas molecules absorb laser energy and generate heat energy, which in turn triggers an acoustic signal. The acoustic signal generated in the photoacoustic cell is detected by a lock-in amplifier, and the acoustic signal is phase-locked and amplified. The amplified signal is clearer and facilitates subsequent data acquisition and analysis.

[0017] Step 104: The data acquisition system acquires the photoacoustic signal output by the lock-in amplifier in real time and converts it into a digital signal for storage, ensuring that the data acquisition system is synchronized with the modulation frequency of the laser so as to correctly analyze the spectral data.

[0018] A further improvement to the technical solution of this invention lies in the following: In step 2, the extraction process of fault diagnosis-related features is as follows:

[0019] Step 201: Perform preprocessing operations such as filtering, denoising, and baseline correction on the acquired photoacoustic spectral data;

[0020] Step 202: Use a feature selection algorithm to extract fault diagnosis-related features from the preprocessed photoacoustic spectral data. The features are related to the absorption characteristics, spectral shape, and intensity parameters of gas molecules. Among them, the fault diagnosis-related features include gas absorption peaks, absorption peak intensity and area, absorption peak width and shape, relative intensity ratio of the spectrum, and spectral slope and curvature. By identifying the gas absorption peaks, the type of gas present can be determined. Combined with the intensity and area of ​​the absorption peaks, the approximate gas concentration can be quantitatively determined. The width and shape of the absorption peaks provide clues about the interactions between gas molecules to help identify the type of gas. The intensity ratio between specific absorption peaks of different gases can be used to distinguish and identify gases. The slope and curvature of the spectrum can be used to analyze the trend of gas concentration over time.

[0021] Step 203: Standardize the extracted fault diagnosis-related feature data to ensure the comparability between different features, and integrate the feature data to form a fault feature dataset for subsequent fault diagnosis model training and validation.

[0022] A further improvement to the technical solution of this invention lies in the following: In step 3, the decoupling process of the photoacoustic spectral data and the separation process of the photoacoustic spectral features are as follows:

[0023] Step 301: The preprocessed photoacoustic spectral data is analyzed using multivariate statistical analysis methods in chemometrics to extract the characteristic information of different gas components;

[0024] Step 302: Based on the correlation data of the extracted characteristic information of different gas components, a multiple linear regression model is trained to establish a chemometric model. The photoacoustic spectral data of each component in the mixed gas are decoupled, and the mixed photoacoustic spectrum is decomposed into the spectral contribution of each gas component. According to the decoupling result, the spectral contribution of each gas component in the mixed spectrum is separated to obtain the individual gas component spectral data. Based on the decoupled photoacoustic spectral data and combined with the absorption characteristics of gas molecules, the concentration of each gas component is calculated using the chemometric model.

[0025] Step 303: Extract features related to different gas components from the decoupled photoacoustic spectral data, including the position, intensity, area, width and shape of absorption peaks. Use known gas component spectral data as a training set and use a convolutional neural network to train the gas component recognition model so that it can learn the feature patterns of different gas components.

[0026] Step 304: Input the preprocessed and decoupled mixed gas photoacoustic spectral data into the trained gas component identification model for identification and separation. The gas component identification model determines the components contained in the mixed gas based on the input spectral characteristics and outputs the identification results of each component and the predicted concentration of each gas, thereby reducing cross-interference between different gas components and improving the accuracy of the detection results.

[0027] A further improvement to the technical solution of this invention lies in: the calculation of the concentration of each gas component using a chemometric model, the calculation expression of which is:

[0028]

[0029] Among them, C i Let A be the concentration of the i-th gaseous component, where i is the index of the component. ij Let w be the absorbance value of the i-th gas at wavelength j, where j is the wavelength index of the spectral data. ij b is the weighting factor. ij m is an exponential parameter related to wavelength j and component i, reflecting the trend of absorption intensity as concentration increases, representing the saturation effect of absorption with increasing concentration. i As the baseline value;

[0030] The calculation expression for the identification results of each component and the predicted concentration of each gas is as follows:

[0031]

[0032] in, Let X be the predicted gas component concentration vector, containing the predicted concentration of each gas component, and W be the preprocessed and decoupled spectral data matrix. k Here, K represents the weight of the k-th convolutional kernel, and K is the number of convolutional layers in the gas component identification model. and These represent the Fourier transform and inverse Fourier transform, respectively. * indicates the convolution operation, which is used for feature extraction in convolutional neural networks. h k Let z be the activation function of the k-th layer. k is the complexity coefficient of the k-th layer, used to adjust the predicted value, and B is the baseline vector, used to adjust the final concentration prediction.

[0033] A further improvement to the technical solution of this invention lies in the following: In step 4, the process of establishing the fault diagnosis model is as follows:

[0034] Step 401: Analyze the photoacoustic spectral data after decoupling and separation to identify potential features related to transformer faults;

[0035] Step 402: Extract quantitative and qualitative features related to transformer fault diagnosis from the photoacoustic spectral data of the separated gas components, including the position, intensity, area, width and shape of the gas absorption peaks, as well as the relative intensity ratio, slope and curvature of the spectrum, and analyze the fault type and fault severity to obtain a fault feature dataset.

[0036] Step 403: Select a classification model based on data characteristics and fault type, use a fault feature dataset containing known fault types and states as a training set, train the classification model to establish a fault diagnosis model, and identify different fault types and states.

[0037] Step 404: Input the preprocessed and decoupled mixed gas photoacoustic spectral data into the trained fault diagnosis model. Based on the input spectral feature information, identify the fault type and state of the transformer. Calculate the fault diagnosis coefficient by combining the output results of the fault diagnosis model and the relevant data of the fault feature dataset, and analyze the severity and urgency of the fault.

[0038] A further improvement to the technical solution of the present invention is that the expression for the fault diagnosis coefficient is:

[0039]

[0040] Where FC is the fault diagnosis coefficient, used to assess the severity and urgency of the fault, n is the number of absorption peaks, α is the weight of the g-th absorption peak, assigned according to the importance of the peak, and p g Let s be the position of the g-th absorption peak. g M represents the intensity of the g-th absorption peak. g Let be the area of ​​the g-th absorption peak, t be the number of spectral data points, and r be the area of ​​the g-th absorption peak. f I is the weight of the relative intensity ratio of the f-th data point. f Let f be the intensity of the f-th data point, H be the set containing slope and curvature, and α, β, and γ be coefficients used to adjust the contribution of each part in the FC calculation. The larger the value of FC, the higher the severity and urgency of the fault.

[0041] A further improvement to the technical solution of this invention lies in the following: In step 5, the process of setting different fault levels and assessing the severity of the fault is as follows:

[0042] Step 501: Deploy the fault diagnosis model to the real-time monitoring system. Combine the constructed fault diagnosis model and historical fault cases to set different fault levels, namely Level 1 fault level, Level 2 fault level, Level 3 fault level and Level 4 fault level. The fault level increases from Level 1 to Level 4 as the severity of the fault increases. Combine the fault diagnosis coefficient to match the corresponding evaluation threshold for each fault level.

[0043] Step 502: Real-time acquisition of gas photoacoustic spectral data in transformer oil, and preprocessing of the acquired data. Using decoupling and separation methods, the mixed photoacoustic spectral data is decomposed into the spectral contributions of each gas component.

[0044] Step 503: Extract fault diagnosis-related features from the decoupled spectral data, input the extracted features into the fault diagnosis model, identify the fault type and state, calculate the fault diagnosis coefficient, determine the corresponding fault level based on the changing trend of the fault diagnosis coefficient, and assess the severity of the fault.

[0045] A further improvement of the technical solution of the present invention is that: the multiple fault levels correspond to multiple evaluation thresholds, wherein the evaluation thresholds include an upper limit threshold and a lower limit threshold;

[0046] The multiple fault levels and the multiple evaluation thresholds satisfy the following relationship:

[0047] Level 1 Fault (FC) <FC L This indicates a minimal level of abnormality, and may not require immediate action.

[0048] Level 2 Fault (FC) L ≤FC <FC M The reading is moderately abnormal; further diagnosis or monitoring is recommended.

[0049] Level 3 Fault FC H ≤FC <FC H This indicates a serious problem that may require repair or replacement of parts.

[0050] Level 4 fault grade FC≥FC H This indicates an emergency requiring immediate action to prevent system failure.

[0051] Where FC is the fault diagnosis coefficient, FC L FC represents the lower threshold for level 2 fault and the upper threshold for level 1 fault. M FC represents the lower threshold for level 3 fault and the upper threshold for level 2 fault. H These are the lower threshold values ​​corresponding to Level 4 fault level and the upper threshold values ​​corresponding to Level 3 fault level.

[0052] A further improvement to the technical solution of this invention is that: in step 6, the process of generating a transformer fault monitoring report and issuing an early warning signal is as follows:

[0053] Step 601: Summarize the photoacoustic spectral data collected by the real-time monitoring system and the output results of the fault diagnosis model. Based on the analysis of the fault diagnosis model, determine the specific type of fault. Combine the fault diagnosis coefficient and fault level to assess the current severity of the fault, and then analyze the development speed and potential impact of the fault.

[0054] Step 602: Summarize the results of fault monitoring and assessment, and compile a transformer fault monitoring report, including the report title, basic information, fault overview, fault severity assessment, development trend analysis, and recommended maintenance measures;

[0055] Step 603: Based on the fault level and assessment results, automatically generate an early warning signal and issue an early warning message. The early warning message includes key information such as fault type, fault level, location of occurrence, and suggested initial response measures. Send the early warning signal and message to relevant personnel, including maintenance personnel, technical personnel, and management, to ensure that they can respond quickly and take action.

[0056] Step 604: During the maintenance process, continuously track the maintenance progress and effects to ensure that the maintenance measures are effectively implemented. After the maintenance is completed, evaluate the maintenance effect to verify the effectiveness of the maintenance measures and whether the fault has been completely resolved. Archive the transformer fault monitoring report for future reference and analysis.

[0057] Due to the adoption of the above technical solution, the technical progress achieved by this invention compared to the prior art is as follows:

[0058] 1. This invention provides a high-precision photoacoustic spectral monitoring method for gases in transformer oil. By real-time acquisition and analysis of photoacoustic spectral data of gases in transformer oil, abnormalities inside the transformer can be detected early. By combining chemometrics and machine learning algorithms to decouple and separate the spectral data of mixed gases, various gas components can be accurately identified and quantitatively analyzed. By extracting features such as the position, intensity, area, width, and shape of gas absorption peaks, as well as information such as the relative intensity ratio, slope, and curvature of the spectrum, the health status of the transformer can be comprehensively assessed, and fault diagnosis coefficients can be calculated. This improves the accuracy of fault detection, reduces the possibility of false alarms and missed alarms, and makes fault detection more timely.

[0059] 2. This invention provides a high-precision photoacoustic spectral monitoring method for gas in transformer oil. By constructing and integrating a fault diagnosis model, it can identify different types of faults. By analyzing the decoupled spectral data, it extracts features related to fault diagnosis and inputs them into the trained fault diagnosis model to identify the fault type and state of the transformer. In addition, by setting different fault levels and matching corresponding evaluation thresholds for each fault level, the severity of the fault can be accurately assessed, enabling maintenance personnel to quickly locate the problem and take targeted maintenance measures, reducing unnecessary inspection and maintenance work and improving maintenance efficiency. Attached Figure Description

[0060] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0061] Figure 1 This is a flowchart of the method of the present invention;

[0062] Figure 2 This is a flowchart illustrating the decoupling of photoacoustic spectral data and the separation of photoacoustic spectral features in this invention.

[0063] Figure 3 This is a flowchart illustrating the establishment of the fault diagnosis model of the present invention.

[0064] Figure 4 This is a flowchart for setting different fault levels and assessing the severity of faults in this invention. Detailed Implementation

[0065] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0066] Example 1, such as Figure 1 , Figure 2 As shown, this invention provides a high-precision photoacoustic spectroscopy monitoring method for gases in transformer oil, comprising the following steps:

[0067] Step 1: Construct a photoacoustic spectroscopy detection system to sample the gas in transformer oil and acquire its photoacoustic spectral data. This system includes a near-infrared tunable fiber laser, a photoacoustic cell, a lock-in amplifier, and a data acquisition system. The near-infrared tunable fiber laser emits infrared radiation of a specific wavelength. The frequency of the light source is modulated by a modulation disk to ensure that the light source changes periodically at a specific frequency. Prepare the components of the photoacoustic spectroscopy detection system, including the near-infrared tunable fiber laser, photoacoustic cell, lock-in amplifier, and data acquisition system. Connect and debug the near-infrared tunable fiber laser, modulation disk, photoacoustic cell, lock-in amplifier, and data acquisition system to ensure that each component functions correctly. The optical path, circuit, and signal transmission path between the components are unobstructed. A near-infrared tunable fiber laser serves as the light source, emitting infrared radiation of a specific wavelength. This wavelength is selected based on the absorption characteristics of gases that may be present in transformer oil (such as H2, CO, CO2, CH4, C2H2, C2H4, C2H6, etc.). Tunability allows the laser's output wavelength to be adjusted within a certain range to meet the detection requirements of different gases. The photoacoustic cell serves as the site of interaction between the gas sample and the laser. Within the photoacoustic cell, gas molecules absorb laser energy and convert it into heat energy through a non-radiative relaxation process, thereby generating an acoustic signal. The photoacoustic cell enhances the resonant cavity of the acoustic signal, improving detection sensitivity. The modulation disk is used to frequency modulate the light emitted by the laser, causing the light source to periodically change at a specific frequency. This helps distinguish background noise from photoacoustic signals caused by gas absorption. The frequency of light passing through the optical path is changed by rotation or vibration. The lock-in amplifier is used to detect and amplify photoacoustic signals that are the same as or related to the modulation frequency. The lock-in amplifier can suppress noise unrelated to the modulation frequency, improve the signal-to-noise ratio, and uses feedback control principles to perform phase locking and amplification of the input signal. The data acquisition system is used to acquire the photoacoustic signal output by the lock-in amplifier, convert it into a digital signal for storage and analysis, and sample gas from transformer oil by vacuum degassing and then placing it into the optical path. In the photoacoustic cell, a near-infrared tunable fiber laser is activated and the desired output wavelength is set. The light source is frequency modulated by a modulation disk, so that the laser periodically irradiates the gas sample in the photoacoustic cell at a specific frequency. In the photoacoustic cell, gas molecules absorb laser energy and generate heat energy, which in turn triggers an acoustic signal. The acoustic signal generated in the photoacoustic cell is detected by a lock-in amplifier, and the acoustic signal is phase-locked and amplified. The amplified signal is clearer, which facilitates subsequent data acquisition and analysis. The data acquisition system acquires the photoacoustic signal output by the lock-in amplifier in real time and converts it into a digital signal for storage. This ensures that the data acquisition system is synchronized with the modulation frequency of the laser to accurately analyze the spectral data.

[0068] Step 2 involves preprocessing the acquired photoacoustic spectral data, including filtering, denoising, and baseline correction, to improve data quality. Fault diagnosis-related features are extracted from the preprocessed photoacoustic spectral data. The preprocessing operations, including filtering, denoising, and baseline correction, improve data quality. Filtering eliminates random noise, thermal noise, and impulse interference in the spectral data, improving the signal-to-noise ratio. Methods include moving average filtering, SG smoothing filtering, median filtering, FFT filtering, and wavelet transform filtering. Appropriate filtering methods are selected based on the characteristics of the spectral data and the type of noise. Denoising further reduces noise interference and improves the clarity of the spectral data, primarily using transform domain-based and spatial domain-based methods. A suitable denoising method is selected based on the three-dimensional characteristics of the spectral data (spatial and spectral domains). Baseline correction eliminates baseline drift in the spectral data, making the spectral curve more accurately reflect the true information of the sample. Methods include polynomial fitting, least squares, adaptive filtering, and iterative reweighted least squares. Based on the complexity of baseline drift and the characteristics of spectral data, a suitable baseline correction method is selected. A feature selection algorithm is used to extract fault diagnosis-related features from the preprocessed photoacoustic spectral data. These features are related to the absorption characteristics, spectral shape, and intensity parameters of gas molecules. Fault diagnosis-related features include gas absorption peaks, absorption peak intensity and area, absorption peak width and shape, relative intensity ratio of the spectrum, and spectral slope and curvature. By identifying gas absorption peaks, the type of gas present is determined. Combined with the intensity and area of ​​absorption peaks, the approximate gas concentration is quantitatively determined. The width and shape of absorption peaks provide clues about the interactions between gas molecules to help identify the type of gas. Gases are distinguished and identified by the intensity ratio between specific absorption peaks of different gases. The trend of gas concentration over time is analyzed by the slope and curvature of the spectrum. The extracted fault diagnosis-related feature data is standardized to ensure the comparability between different features. The feature data are then integrated to form a fault feature dataset for subsequent fault diagnosis model training and validation.

[0069] Step 3: Decouple the photoacoustic spectral data of the mixed gas using chemometric methods, and use machine learning algorithms to separate and identify the photoacoustic spectral characteristics of different gases, removing cross-interference between different gases. Analyze the preprocessed photoacoustic spectral data using multivariate statistical analysis methods in chemometrics to extract the characteristic information of different gas components. Based on the correlation data of the extracted characteristic information of different gas components, train a multivariate linear regression model to establish a chemometric model, decouple the photoacoustic spectral data of each component in the mixed gas, decompose the mixed photoacoustic spectrum into the spectral contributions of each gas component, and separate the spectral contributions of each gas component in the mixed spectrum according to the decoupling results, obtaining individual gas component spectral data. Then, based on the decoupled photoacoustic spectral data... By combining the absorption characteristics of gas molecules, the concentration of each gas component is calculated using a chemometric model. Features related to different gas components are extracted from the decoupled photoacoustic spectral data, including the position, intensity, area, width, and shape of absorption peaks. Using known gas component spectral data as a training set, a convolutional neural network is used to train the gas component recognition model, enabling it to learn the characteristic patterns of different gas components. The preprocessed and decoupled mixed gas photoacoustic spectral data is input into the trained gas component recognition model for identification and separation. Based on the input spectral features, the gas component recognition model determines the components contained in the mixed gas and outputs the identification results of each component and the predicted concentration of each gas, reducing cross-interference between different gas components and improving the accuracy of detection results.

[0070] Furthermore, the concentrations of each gaseous component are calculated using a stoichiometric model, and the calculation expression is as follows:

[0071]

[0072] Among them, C i Let A be the concentration of the i-th gaseous component, where i is the index of the component. ij Let w be the absorbance value of the i-th gas at wavelength j, where j is the wavelength index of the spectral data. ij b is the weighting factor. ij m is an exponential parameter related to wavelength j and component i, reflecting the trend of absorption intensity as concentration increases, representing the saturation effect of absorption with increasing concentration. i This serves as a baseline value, used to adjust the model to match actual measured values.

[0073] The output shows the identification results of each component and the predicted concentration of each gas. The calculation expression is as follows:

[0074]

[0075] in, Let X be the predicted gas component concentration vector, which is a vector containing the predicted concentration of each gas component. Let W be the preprocessed and decoupled spectral data matrix. k Here, K represents the weight of the k-th convolutional kernel, and K is the number of convolutional layers in the gas component identification model. and These represent the Fourier transform and the inverse Fourier transform, respectively. The Fourier transform is used to transform data from the spatial domain to the frequency domain, and the inverse Fourier transform is used to transform data from the frequency domain back to the spatial domain. * represents the convolution operation, which is used for feature extraction in convolutional neural networks. h k The activation function for the k-th layer is a nonlinear function, such as the ReLU function, used to introduce nonlinearity. k is the complexity coefficient of the k-th layer, used to adjust the predicted value; B is the baseline value vector, used to adjust the final concentration prediction.

[0076] Step 4: Analyze the decoupled and separated photoacoustic spectral data, extract feature information related to transformer fault diagnosis, and establish a fault diagnosis model to identify different fault types and states;

[0077] Step 5: Integrate the established fault diagnosis model into the real-time monitoring system to perform real-time analysis of the gas in the transformer oil. Combine the constructed fault diagnosis model with historical fault cases to set different fault levels and assess the severity of the fault.

[0078] Step 6: Based on the data analysis and fault assessment results, generate a transformer fault monitoring report and issue an early warning signal to prompt further inspection and maintenance. The report includes information on fault type, fault severity, development trend, and recommended maintenance measures.

[0079] Example 2, as Figure 3 , Figure 4 As shown, based on Embodiment 1, the present invention provides a technical solution: Preferably, in step 4, the process of establishing the fault diagnosis model is as follows:

[0080] Analyzing the decoupled and separated photoacoustic spectral data, potential features related to transformer faults are identified. Quantitative and qualitative features related to transformer fault diagnosis are extracted from the separated gas component photoacoustic spectral data, including the position, intensity, area, width, and shape of gas absorption peaks, as well as the relative intensity ratio, slope, and curvature of the spectrum. Fault types and severity are analyzed to obtain a fault feature dataset. A classification model is selected based on data characteristics and fault types. The fault feature dataset containing known fault types and states is used as the training set to train the classification model and establish a fault diagnosis model to identify different fault types and states. The preprocessed and decoupled mixed gas photoacoustic spectral data is input into the trained fault diagnosis model. Based on the input spectral feature information, the fault type and state of the transformer are identified. The fault diagnosis coefficient is calculated by combining the output results of the fault diagnosis model and the relevant data from the fault feature dataset to analyze the severity and urgency of the fault.

[0081] Furthermore, the expression for the fault diagnosis coefficient is:

[0082]

[0083] Where FC is the fault diagnosis coefficient, used to assess the severity and urgency of the fault, n is the number of absorption peaks, α is the weight of the g-th absorption peak, assigned according to the importance of the peak, and p g Let s be the position of the g-th absorption peak. g M represents the intensity of the g-th absorption peak. g Let be the area of ​​the g-th absorption peak, t be the number of spectral data points, and r be the area of ​​the g-th absorption peak. f I is the weight of the relative intensity ratio of the f-th data point. f Let f be the intensity of the f-th data point, H be the set containing slope and curvature, α, β, and γ be coefficients used to adjust the contribution of each part in the FC calculation, and ln be the natural logarithm function used to handle the product of area and intensity. The larger the value of FC, the higher the severity and urgency of the fault.

[0084] Step 5 involves setting different fault levels and assessing the severity of the fault as follows:

[0085] The fault diagnosis model is deployed to the real-time monitoring system. Based on the constructed fault diagnosis model and historical fault cases, different fault levels are set: Level 1, Level 2, Level 3, and Level 4. The fault level increases progressively from Level 1 to Level 4 as the severity of the fault increases. The fault diagnosis coefficient is used to match the corresponding evaluation threshold for each fault level. The gas photoacoustic spectral data in transformer oil is collected in real time. The collected data is preprocessed. Using decoupling and separation methods, the mixed photoacoustic spectral data is decomposed into the spectral contributions of each gas component. Fault diagnosis-related features are extracted from the decoupled spectral data. The extracted features are input into the fault diagnosis model to identify the fault type and state, calculate the fault diagnosis coefficient, and determine the corresponding fault level and assess the severity of the fault based on the changing trend of the fault diagnosis coefficient.

[0086] Furthermore, multiple fault levels correspond to multiple evaluation thresholds, where the evaluation thresholds include an upper threshold and a lower threshold;

[0087] Multiple fault levels and multiple evaluation thresholds satisfy the following relationship:

[0088] Level 1 Fault (FC) <FC L This indicates a minimal level of abnormality, and may not require immediate action.

[0089] Level 2 Fault (FC) L ≤FC <FC M The reading is moderately abnormal; further diagnosis or monitoring is recommended.

[0090] Level 3 Fault FC M ≤FC <FC H This indicates a serious problem that may require repair or replacement of parts.

[0091] Level 4 fault grade FC≥FC H This indicates an emergency requiring immediate action to prevent system failure.

[0092] Where FC is the fault diagnosis coefficient, FC L FC represents the lower threshold for level 2 fault and the upper threshold for level 1 fault. M FC represents the lower threshold for level 3 fault and the upper threshold for level 2 fault. H These are the lower threshold corresponding to level four fault level and the upper threshold corresponding to level three fault level.

[0093] Step 6, the process of generating a transformer fault monitoring report and issuing an early warning signal, is as follows:

[0094] This paper summarizes the photoacoustic spectral data collected by the real-time monitoring system and the output results of the fault diagnosis model. Based on the analysis of the fault diagnosis model, the specific type of fault is determined. Combining the fault diagnosis coefficient and fault level, the current severity of the fault is assessed, and the development speed and potential impact of the fault are analyzed. The results of fault monitoring and assessment are summarized, and a transformer fault monitoring report is prepared. The report includes a title, basic information, fault overview, fault severity assessment, development trend analysis, and recommended maintenance measures. The report title and basic information include the report title, report date, transformer number, and monitoring point location. The fault overview briefly describes the fault type, occurrence time, and preliminary judgment of the cause. The fault severity assessment details the assessment basis and results of the fault level, explaining the severity of the fault. The development trend analysis, based on the data analysis results, describes the fault's... The system analyzes development trends, including potential expansion and system impact. Based on the fault type and severity, it proposes specific maintenance recommendations, including emergency response measures, maintenance procedures, required materials and tools. Based on the fault level and assessment results, it automatically generates and issues warning signals. These warning messages include key information such as fault type, fault level, location, and suggested initial response measures. The warning signals and messages are sent to relevant personnel, including maintenance staff, technicians, and management, ensuring rapid response and action. During maintenance, the system continuously tracks progress and effectiveness to ensure effective implementation of maintenance measures. After maintenance, the system evaluates the results to verify the effectiveness of maintenance measures and whether the fault has been completely resolved. The transformer fault monitoring report is archived for future review and analysis.

[0095] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A high-precision photoacoustic spectroscopy method for monitoring gas in transformer oil, characterized in that: Includes the following steps: Step 1: Construct a photoacoustic spectroscopy detection system to sample the gas in the transformer oil and obtain the photoacoustic spectral data of the gas. The photoacoustic spectroscopy detection system includes a near-infrared tunable fiber laser, a photoacoustic cell, a lock-in amplifier, and a data acquisition system. Step 2: Preprocess the collected photoacoustic spectral data and extract fault diagnosis-related features from the preprocessed photoacoustic spectral data; Step 3: Decouple the photoacoustic spectral data of the mixed gas using chemometric methods, and use machine learning algorithms to separate and identify the photoacoustic spectral characteristics of different gases, removing cross-interference between different gases. The process of decoupling the photoacoustic spectral data and separating the photoacoustic spectral characteristics is as follows: Step 301: The preprocessed photoacoustic spectral data is analyzed using multivariate statistical analysis methods in chemometrics to extract the characteristic information of different gas components. Step 302: Based on the correlation data of the extracted characteristic information of different gas components, a multiple linear regression model is trained to establish a chemometric model. The photoacoustic spectral data of each component in the mixed gas are decoupled, and the mixed photoacoustic spectrum is decomposed into the spectral contribution of each gas component. According to the decoupling result, the spectral contribution of each gas component in the mixed spectrum is separated to obtain the individual gas component spectral data. Based on the decoupled photoacoustic spectral data and combined with the absorption characteristics of gas molecules, the concentration of each gas component is calculated using the chemometric model. Step 303: Extract features related to different gas components from the decoupled photoacoustic spectral data, including the position, intensity, area, width and shape of absorption peaks. Use known gas component spectral data as a training set and use a convolutional neural network to train the gas component recognition model so that it can learn the feature patterns of different gas components. Step 304: Input the preprocessed and decoupled mixed gas photoacoustic spectral data into the trained gas component identification model for identification and separation. The gas component identification model determines the components contained in the mixed gas based on the input spectral characteristics and outputs the identification results of each component and the predicted concentration of each gas, thereby reducing cross-interference between different gas components. Step 4: Analyze the decoupled and separated photoacoustic spectral data, extract feature information related to transformer fault diagnosis, and establish a fault diagnosis model to identify different fault types and states; Step 5: Integrate the established fault diagnosis model into the real-time monitoring system to perform real-time analysis of the gas in the transformer oil. Combine the constructed fault diagnosis model with historical fault cases to set different fault levels and assess the severity of the fault. Step 6: Based on the data analysis and fault assessment results, generate a transformer fault monitoring report and issue an early warning signal.

2. The high-precision photoacoustic spectroscopy monitoring method for gas in transformer oil according to claim 1, characterized in that: In step 1, the process of constructing the photoacoustic spectroscopy detection system and acquiring gas photoacoustic spectral data is as follows: Step 101: Prepare the components of the photoacoustic spectroscopy detection system, including a near-infrared tunable fiber laser, a photoacoustic cell, a lock-in amplifier, and a data acquisition system. Connect and debug the near-infrared tunable fiber laser, modulation disk, photoacoustic cell, lock-in amplifier, and data acquisition system to ensure that the optical paths, circuits, and signal transmission paths between the components are unobstructed. Step 102: The gas sample in the transformer oil is degassed by vacuum and placed into the photoacoustic cell. The near-infrared tunable fiber laser is started and the required output wavelength is set. The frequency of the light source is modulated by the modulation disk so that the laser periodically irradiates the gas sample in the photoacoustic cell. Step 103: In the photoacoustic cell, gas molecules absorb laser energy and generate heat energy, which in turn triggers an acoustic signal. The acoustic signal generated in the photoacoustic cell is detected by a lock-in amplifier, and the acoustic signal is phase-locked and amplified. Step 104: The data acquisition system acquires the photoacoustic signal output by the lock-in amplifier in real time and converts it into a digital signal for storage.

3. The high-precision photoacoustic spectroscopy monitoring method for gas in transformer oil according to claim 2, characterized in that: In step 2, the process of extracting fault diagnosis-related features is as follows: Step 201: Perform preprocessing operations such as filtering, denoising, and baseline correction on the acquired photoacoustic spectral data; Step 202: Use a feature selection algorithm to extract fault diagnosis-related features from the preprocessed photoacoustic spectral data. The features are related to the absorption characteristics, spectral shape, and intensity parameters of gas molecules. Among them, the fault diagnosis-related features include gas absorption peaks, absorption peak intensity and area, absorption peak width and shape, relative intensity ratio of the spectrum, and spectral slope and curvature. Step 203: Standardize the extracted fault diagnosis-related feature data and integrate the feature data to form a fault feature dataset.

4. The high-precision photoacoustic spectroscopy monitoring method for gas in transformer oil according to claim 3, characterized in that: The concentration of each gas component is calculated using a chemometric model, and the calculation expression is as follows: ； in, For the first The concentration of each gaseous component For component indexing, To be at wavelength First Absorbance of the gas, For wavelength indexing of spectral data, As a weighting factor, To match the wavelength and components The relevant index parameters reflect the trend of absorption intensity as concentration increases, indicating the saturation effect of absorption with increasing concentration. As the baseline value; The calculation expression for the identification results of each component and the predicted concentration of each gas is as follows: ； in, For the predicted gas component concentration vector, This is the preprocessed and decoupled spectral data matrix. For the first The weights of the convolutional kernels, The number of convolutional layers in the gas component identification model. and These represent the Fourier transform and the inverse Fourier transform, respectively. The convolution operation represents the feature extraction operation used in convolutional neural networks. For the first The activation function of the layer, For the first The complexity coefficient of the layer, This serves as a baseline vector, used to adjust the final concentration prediction.

5. The high-precision photoacoustic spectroscopy monitoring method for gas in transformer oil according to claim 4, characterized in that: In step 4, the process of establishing the fault diagnosis model is as follows: Step 401: Analyze the photoacoustic spectral data after decoupling and separation to identify potential features related to transformer faults; Step 402: Extract quantitative and qualitative features related to transformer fault diagnosis from the photoacoustic spectral data of the separated gas components, including the position, intensity, area, width and shape of the gas absorption peaks, as well as the relative intensity ratio, slope and curvature of the spectrum, and analyze the fault type and fault severity to obtain a fault feature dataset. Step 403: Select a classification model based on data characteristics and fault type, use a fault feature dataset containing known fault types and states as a training set, train the classification model to establish a fault diagnosis model, and identify different fault types and states. Step 404: Input the preprocessed and decoupled mixed gas photoacoustic spectral data into the trained fault diagnosis model. Based on the input spectral feature information, identify the fault type and state of the transformer. Calculate the fault diagnosis coefficient by combining the output results of the fault diagnosis model and the relevant data of the fault feature dataset, and analyze the severity and urgency of the fault.

6. The high-precision photoacoustic spectroscopy monitoring method for gas in transformer oil according to claim 5, characterized in that: The expression for the fault diagnosis coefficient is: ； in, This is a fault diagnosis coefficient used to assess the severity and urgency of a fault. The number of absorption peaks, No. The weight of each absorption peak For the first The position of each absorption peak. For the first The intensity of each absorption peak, For the first The area of ​​each absorption peak The number of spectral data points. For the first The weights of the relative intensity ratio of each data point. For the first The intensity of each data point For a set containing slope and curvature, , , The coefficient is used to adjust the various parts in It has a high contribution to the calculation.

7. The high-precision photoacoustic spectroscopy monitoring method for gas in transformer oil according to claim 6, characterized in that: In step 5, the process of setting different fault levels and assessing the severity of the fault is as follows: Step 501: Deploy the fault diagnosis model to the real-time monitoring system. Combine the constructed fault diagnosis model and historical fault cases to set different fault levels, namely Level 1 fault level, Level 2 fault level, Level 3 fault level and Level 4 fault level. The fault level increases from Level 1 to Level 4 as the severity of the fault increases. Combine the fault diagnosis coefficient to match the corresponding evaluation threshold for each fault level. Step 502: Real-time acquisition of gas photoacoustic spectral data in transformer oil, and preprocessing of the acquired data. Using decoupling and separation methods, the mixed photoacoustic spectral data is decomposed into the spectral contributions of each gas component. Step 503: Extract fault diagnosis-related features from the decoupled spectral data, input the extracted features into the fault diagnosis model, identify the fault type and state, calculate the fault diagnosis coefficient, determine the corresponding fault level based on the changing trend of the fault diagnosis coefficient, and assess the severity of the fault.

8. The high-precision photoacoustic spectroscopy monitoring method for gas in transformer oil according to claim 7, characterized in that: Multiple fault levels correspond to multiple evaluation thresholds, wherein the evaluation thresholds include an upper threshold and a lower threshold; The multiple fault levels and the multiple evaluation thresholds satisfy the following relationship: Level 1 Fault ; Level 2 fault ; Level 3 Fault ; Level 4 Fault ; in, This is the fault diagnosis coefficient. These are the lower threshold for a level 2 fault and the upper threshold for a level 1 fault. These are the lower threshold for level 3 faults and the upper threshold for level 2 faults. These are the lower threshold values ​​corresponding to Level 4 fault level and the upper threshold values ​​corresponding to Level 3 fault level.

9. The high-precision photoacoustic spectroscopy monitoring method for gas in transformer oil according to claim 8, characterized in that: In step 6, the process of generating a transformer fault monitoring report and issuing an early warning signal is as follows: Step 601: Summarize the photoacoustic spectral data collected by the real-time monitoring system and the output results of the fault diagnosis model. Based on the analysis of the fault diagnosis model, determine the specific type of fault. Combine the fault diagnosis coefficient and fault level to assess the current severity of the fault, and then analyze the development speed and potential impact of the fault. Step 602: Summarize the results of fault monitoring and assessment, and compile a transformer fault monitoring report, including the report title, basic information, fault overview, fault severity assessment, development trend analysis, and recommended maintenance measures; Step 603: Based on the fault level and assessment results, automatically generate an early warning signal and issue an early warning message. The early warning message includes key information such as fault type, fault level, location of occurrence, and suggested preliminary response measures. Send the early warning signal and message to relevant personnel, including maintenance personnel, technical personnel, and management. Step 604: During the maintenance process, continuously track the maintenance progress and effect. After the maintenance is completed, evaluate the maintenance effect, verify the effectiveness of the maintenance measures and whether the fault has been completely resolved, and archive the transformer fault monitoring report for future reference.

Images