Method and system for detecting dissolved gas in transformer oil based on process optimization

Abstract

The invention relates to a method and system for detecting dissolved gas in transformer oil based on process optimization, and the method comprises the steps: collecting a transformer oil sample through multiple channels, and removing the dissolved gas in the transformer oil; the single laser source is controlled to be reused for the multiple collection channels in a time-sharing mode, gas released from the corresponding channels is sequentially excited according to a preset time sequence, and detection signals are generated; a gas component concentration analysis model based on a partial least squares regression algorithm is adopted to process the detection signals, predicted concentrations of all gas components are obtained, the gas component concentration analysis model screens key wavelengths based on a VIP value for cross absorption peaks, and an analysis sub-model special for the gas cross absorption peaks is established. Compared with the prior art, the performance bottleneck of the traditional photoacoustic spectrum technology in multi-point detection is broken through by adopting collaborative optimization of hardware multiplexing, algorithm innovation and process reconstruction, and an efficient and reliable technical path is provided for rapid and accurate monitoring of the dissolved gas in the transformer oil.

Description

Technical Field

[0001] The present invention relates to the technical field of detection of dissolved gas in transformer oil, and in particular to a minute-level detection method of dissolved gas in transformer oil based on process optimization. Background Art

[0002] In the field of power equipment condition monitoring, dissolved gas analysis (DGA) in transformer oil is a key method for diagnosing latent faults within transformers. With the development of smart grids and the increasing demand for equipment condition-based maintenance, traditional detection methods are no longer able to meet the real-time, portability, and automation requirements of modern power systems.

[0003] Currently, gas chromatography (GC), the classic method for DGA detection, has high accuracy and sensitivity, but its detection process is complex and requires multiple steps such as oil sample degassing, chromatographic separation, and standard gas calibration, resulting in a single detection cycle of several hours or even longer. In addition, this method relies on laboratory environments and professional operators, making it difficult to achieve rapid on-site detection and online monitoring. Although alternative technologies such as infrared spectroscopy and electrochemical sensors have shortened detection time, they still have shortcomings in terms of simultaneous detection of multi-component gases, anti-interference capabilities, and long-term stability.

[0004] Photoacoustic spectroscopy (PAS) is considered a potential breakthrough in DGA detection due to its high sensitivity, lack of carrier gas, and lack of pretreatment. However, existing PAS systems still face room for improvement in the detection process. Issues such as long gas equilibration times and insufficient signal acquisition efficiency limit their ability to achieve rapid detection within minutes. Furthermore, background noise suppression and trace gas quantification algorithms in complex oil sample environments require further improvement to ensure reliable detection results.

[0005] Chinese patent application CN111175232A discloses a method for parallel multi-chamber detection using multiple laser integrated light sources. Multiple oil and gas circulation loops share a common optical detection system. However, this solution suffers from cross-contamination caused by residual gas flow, long cycle times due to coupled degassing and detection, a fixed light source wavelength that is difficult to adapt to composition changes, a microphonic device susceptible to vibration interference and lacking a noise reduction algorithm, and insufficient resolution of overlapping characteristic peaks, resulting in low accuracy. These issues lead to a bulky system, poor time efficiency, weak adaptability, and limited reliability of results. The solution disclosed in Chinese patent application CN110361342A, which uses wavelength division multiplexing to generate multiple narrowband light channels of varying frequencies, suffers from fixed wavelength allocation that is difficult to adapt to gas composition changes; the photoacoustic cell structure is not optimized for response differences between different gases; a lack of noise reduction and spectral overlap resolution algorithms leads to crosstalk that can lead to quantitative errors; and insufficient coordination between degassing and detection, a long process time, and high fiber deployment costs. Summary of the Invention

[0006] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and provide a minute-level dissolved gas detection method in transformer oil based on process optimization.

[0007] The purpose of the present invention can be achieved by the following technical solutions:

[0008] As a first aspect of the present invention, a method for detecting dissolved gas in transformer oil based on process optimization is provided, comprising the following steps:

[0009] The transformer oil samples are collected through a multi-channel sampling device, and the transformer oil samples of each sampling channel are pre-treated and degassed respectively, and the gas degassed from each sampling channel is transmitted to the photoacoustic cell of the detection unit;

[0010] The light source is switched in time, and the photoacoustic cells corresponding to each sampling channel are stimulated in sequence according to a preset time sequence, so that the escaped gas reacts with the light source in the detection unit to generate a photoacoustic signal;

[0011] Photoacoustic signals are collected and the predicted concentration of each gas component is calculated using a partial least squares regression algorithm with a dedicated analytical sub-model for cross-absorption peaks.

[0012] As a preferred technical solution, the pretreatment and degassing treatment are specifically as follows:

[0013] The oil samples collected from each sampling channel are evenly distributed to multiple pre-treatment channels, and the oil sample temperature in each channel is independently controlled to be stable within the set temperature range;

[0014] The oil samples pretreated in each pretreatment channel are introduced into the parallel degassing device for degassing under set vacuum conditions. The vacuum and temperature parameters are dynamically adjusted by real-time monitoring of the degassing efficiency to ensure that degassing is completed within the set time.

[0015] As an optimal technical solution, the partial least squares regression algorithm is specifically as follows:

[0016] The photoacoustic signal of each gas channel is acquired by scanning the set wavelength band with a tunable quantum cascade laser to form the original absorption spectrum matrix X, where each row represents a sample and each column corresponds to the absorbance at a specific wavelength.

[0017] Use standard gas samples of known concentration to establish a reference concentration matrix Y, where each row corresponds to a sample and each column is a gas concentration.

[0018] The partial least squares regression algorithm is used to project the high-dimensional raw absorption spectrum data and reference concentration data into a low-dimensional latent variable space. The projection direction that can simultaneously explain the maximum covariance of the raw absorption spectrum matrix X and the reference concentration matrix Y is found. Multiple latent variables are extracted through iteration, and the residual error is gradually reduced.

[0019] For each latent variable, the weight vector w of the original absorption spectrum matrix X and the weight vector c of the reference concentration matrix Y are calculated, and the residual matrix of the original absorption spectrum matrix X and the reference concentration matrix Y is updated until convergence.

[0020] As a preferred technical solution, the partial least squares regression algorithm trains the PLSR sub-model separately for the cross-absorption peak, increasing the number of latent variables in the wavelength region corresponding to the cross-absorption peak, as follows:

[0021] Extract the cross-absorption peak band region sub-matrix X from the full-band spectrum sub And train the PLSR sub-model separately;

[0022] Filter key wavelengths based on VIP values ​​and remove irrelevant noise;

[0023] K-fold cross validation was used to optimize the number of latent variables to minimize the prediction error of the PLSR sub-model;

[0024] The PLSR sub-model was tested using an independent validation set to ensure that the correlation coefficient between the predicted concentration of each gas component and the reference value was greater than the set value and the relative error was less than the set error range.

[0025] As a preferred technical solution, the partial least squares regression algorithm pre-processes the collected photoacoustic signal, including: performing wavelet denoising on the collected photoacoustic signal, and standardizing and smoothing the spectrum after wavelet denoising.

[0026] As a second aspect of the present invention, a system for detecting dissolved gas in transformer oil is provided, the system comprising:

[0027] A multi-channel sampling device is used to collect transformer oil samples. The transformer oil samples collected by each sampling channel of the sampling device are processed in sequence through a pretreatment channel and a degassing device. The gas released by the degassing device is transmitted to a detection unit. A high-speed solenoid valve group is provided between the degassing device and the detection unit to achieve orderly switching of the gas released by each sampling channel.

[0028] The detection unit includes a plurality of photoacoustic cells corresponding to each sampling channel, each of which is provided with a signal acquisition unit; a gas buffer chamber is provided on both sides of the photoacoustic cell to balance the pressure difference between each channel;

[0029] The light source is used to generate a laser beam with a set detection wavelength; the high-speed optical switch performs time-sharing switching of the laser beam, and sequentially excites the photoacoustic cell corresponding to each sampling channel according to a preset timing;

[0030] The signal processing unit is respectively connected to the high-speed electromagnetic valve group, the high-speed optical switch and the signal acquisition unit, and is used to execute the above-mentioned method for detecting dissolved gas in transformer oil based on process optimization.

[0031] As a preferred technical solution, a filter membrane is provided between the multi-channel sampling device and the transformer for filtering and removing solid impurities.

[0032] As a preferred technical solution, the pre-processing channel is equipped with an independent constant temperature control unit for each collected transformer oil sample to stabilize the oil sample temperature within a set temperature range.

[0033] As a preferred technical solution, each degassing unit in the degassing device adopts a polytetrafluoroethylene breathable membrane structure.

[0034] As a preferred technical solution, the light source adopts a tunable quantum cascade laser.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] 1) The present invention proposes a method for detecting dissolved gases in transformer oil at the minute level based on process optimization. In order to achieve the minute-level detection goal, a time-multiplexing engineering control strategy is adopted. A single laser source and detection unit are time-multiplexed for multiple gas channels, and each channel completes full-component gas analysis within the detection window. The complexity and cost issues brought by multiple optical systems are avoided. At the signal processing level, the system integrates the multi-scale analysis capability of wavelet transform and reference cavity differential detection technology, effectively suppressing the background interference generated by the complex matrix of transformer oil. At the same time, a multivariate regression algorithm is used to analyze the data in the overlapping area of ​​the absorption spectrum, solving the problem of cross-interference of characteristic peaks of gases such as CH4 and C2H4.

[0037] 2) The present invention provides a method for detecting dissolved gases in transformer oil at the minute level based on process optimization. To achieve minute-level detection, a time-multiplexing engineering control strategy is employed. A single laser source and detection unit are time-multiplexed across multiple gas channels, with each channel completing full gas component analysis within a 20-30 second detection window. This design not only avoids the complexity and cost associated with multiple optical systems but also enables real-time adjustment of detection parameters by establishing a gas diffusion kinetics model, ensuring optimal signal-to-noise ratios for gas components of varying molecular weights.

[0038] 3) The present invention integrates the multi-scale analysis capability of wavelet transform, and eliminates mechanical vibration interference through wavelet denoising in the signal processing link. The present invention effectively suppresses the background interference generated by the complex matrix of transformer oil. PLSR is used to map high-dimensional spectral data and concentration data to a low-dimensional latent variable space through projection to solve the maximum covariance. The latent variables of the PLSR model are used to capture the main features of the full band (such as strong absorption peaks of other gases and background trends) and to describe global information. In addition, directional optimization selection is performed in combination with spectral characteristics and model performance. First, the sub-matrix of the region is extracted from the full-band spectrum and the model is trained separately. The key wavelength is screened based on the VIP value, and the number of latent variables is optimized using multi-fold cross-validation to construct a PLSR sub-model for the cross-absorption peak construction area. The newly added latent variables focus on the spectral details of the overlapping peak area, which can enhance the ability to capture subtle differences; ensure the accuracy of the model's analysis of overlapping peaks, and effectively solve the problem of cross-interference of PLSR characteristic peaks.

[0039] 4) During the sampling phase, the present invention removes solid impurities through filtration with a 0.45μm microporous membrane and uses a constant temperature control unit to stabilize the oil sample temperature within the range of 50±2°C to ensure sample consistency. During the detection phase, a tunable quantum cascade laser is used as the light source, and a high-speed optical switch is used to time-share the excitation of each channel photoacoustic cell, covering the 2.5-12μm band, to ensure efficient detection of all gas components. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 A schematic flow chart of a method for detecting dissolved gas in transformer oil at the minute level based on process optimization provided by an embodiment of the present invention;

[0041] Figure 2 A schematic flow chart of step 1 of a method for detecting dissolved gas in transformer oil at a minute level based on process optimization provided by an embodiment of the present invention;

[0042] Figure 3 A flow chart of step 4 of the method for detecting dissolved gas in transformer oil at a minute level based on process optimization provided by an embodiment of the present invention;

[0043] Figure 4 A flow chart of step 5 of the method for detecting dissolved gas in transformer oil at a minute level based on process optimization provided by an embodiment of the present invention;

[0044] Figure 5 A schematic structural diagram of a minute-level transformer oil dissolved gas detection system based on process optimization provided by an embodiment of the present invention;

[0045] The numbers in the figure are as follows: 1. Transformer; 2. Filter membrane; 3. Pretreatment channel; 4. Degassing device; 5. Breathable membrane; 6. Degassing controller; 7. High-speed solenoid valve group; 8. Gas buffer chamber; 9. Light source; 10. High-speed optical switch; 11. Photoacoustic cell; 12. Signal acquisition unit; 13. Signal processing unit. DETAILED DESCRIPTION

[0046] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0047] Example 1

[0048] The present invention provides a method for detecting dissolved gases in transformer oil at the minute level based on process optimization. In order to achieve the detection goal at the minute level, a time-multiplexing engineering control strategy is adopted to time-share multiplex a single laser source and a detection unit for multiple gas channels, and each channel completes full-component gas analysis within a detection window of 20-30 seconds. This design not only avoids the complexity and cost issues brought about by multiple optical systems, but also adjusts the detection parameters in real time by establishing a gas diffusion kinetics model to ensure that gas components of different molecular weights can obtain the best signal-to-noise ratio. At the signal processing level, the multi-scale analysis capability of wavelet transform and the reference cavity differential detection technology are integrated to effectively suppress the background interference generated by the complex matrix of transformer oil. At the same time, a multivariate regression algorithm is used to analyze the data in the overlapping area of ​​the absorption spectrum, solving the problem of cross-interference of characteristic peaks of gases such as CH4 and C2H4. Such as Figure 1 As shown, the specific steps include:

[0049] Step 1: The sampling device collects transformer oil samples into the pretreatment channel;

[0050] Step 2: introducing the pretreated oil sample into the degassing device 4 for degassing;

[0051] Step 3, the gas released in step 2 is transmitted to a detection unit;

[0052] Step 4, time-sharing switching of the light source 9 so that the gas in the detection unit interacts with the light source to generate a photoacoustic signal;

[0053] Step 5: The signal acquisition unit 12 collects the photoacoustic signal and transmits it to the signal processing unit 13 for processing and analysis;

[0054] Step 6: Generate a comprehensive report based on the processing results, including the concentration, growth rate and fault warning level of each gas component.

[0055] As one of the implementations of the present invention, Figure 2As shown, step 1 is specifically implemented as follows:

[0056] Step 11: Use a multi-channel sampling device to simultaneously collect transformer oil samples, and filter through a 0.45 μm microporous filter membrane 2 to remove solid impurities.

[0057] In step 12, the oil sample is evenly distributed to multiple pretreatment channels 3. Each channel is equipped with an independent constant temperature control unit to stabilize the oil sample temperature within the range of 50±2°C.

[0058] As one of the implementation methods of the present invention, in step 2, the pretreated oil sample is introduced into the parallel degassing device 4. Each degassing unit adopts a polytetrafluoroethylene breathable membrane 5 structure and performs degassing treatment under a vacuum condition of 0.08 MPa. The vacuum degree and temperature parameters are dynamically adjusted by real-time monitoring of the degassing efficiency to ensure that efficient degassing is completed within 90 seconds.

[0059] As one of the implementation methods of the present invention, in step 3, the orderly switching of the gases in each degassing channel is achieved through the high-speed solenoid valve group 7, the micro-flow control technology is used to ensure the smooth transmission of the gas to the detection unit, and a gas buffer chamber 8 is set to balance the pressure differences in each channel. The switching period is controlled at 25±5 seconds.

[0060] As one of the implementations of the present invention, Figure 3 As shown, step 4 is specifically implemented as follows:

[0061] Step 41 : Using a tunable quantum cascade laser as the light source 9 , and implementing time-sharing switching of the laser beam through a high-speed optical switch 10 .

[0062] Step 42 , the photoacoustic cell 11 corresponding to each channel is excited in sequence according to a preset time sequence, the detection wavelength covers the 2.5-12 μm band, and the detection time for each channel is allocated to 28 seconds.

[0063] As one of the implementations of the present invention, Figure 4 As shown, step 5 is specifically implemented as follows:

[0064] In step 51 , the signal acquisition unit 12 acquires the photoacoustic signals of each channel and transmits the acquired signals to the signal processing unit 13 .

[0065] In step 52 , the signal processing unit 13 performs wavelet noise reduction processing on the photoacoustic signal transmitted by the acquisition unit to eliminate mechanical vibration interference.

[0066] Step 53: calculate the concentration of each gas component using a partial least squares regression algorithm, and establish a dedicated analytical model specifically for the cross-absorption peak of CH4 and C2H4 at the 3.3 μm band.

[0067] Furthermore, in step 53 of the present invention, in the detection of dissolved gases in transformer oil, a partial least squares regression (PLSR) algorithm is used to calculate the concentration of each gas component. The PLSR algorithm is specifically improved for cross-absorption peaks (such as the overlapping peaks of CH4 and C2H4 in the 3.3μm band) to establish a dedicated analytical model. The specific implementation process is as follows:

[0068] Step 531 , a tunable quantum cascade laser is used to scan the 2.5-12 μm band to obtain the photoacoustic signals of each gas channel, forming an original absorption spectrum matrix X (each row represents a sample, and each column corresponds to the absorbance of a specific wavelength).

[0069] In step 532, a reference concentration matrix Y is established using standard gas samples of known concentration (such as H2, CH4, C2H2, C2H4, C2H6, etc.) (each row corresponds to a sample, and each column is a gas concentration).

[0070] In step 533 , the spectrum after wavelet denoising is normalized and Savitzky-Golay smoothing is performed to enhance the distinction of characteristic peaks.

[0071] In step 534, PLSR projects the high-dimensional spectral data X and concentration data Y into a low-dimensional latent variable space and finds the projection direction that can simultaneously explain the maximum covariance of X and Y. Its mathematical essence is to solve:

[0072] max cov(Xw,Yc)

[0073] Where: w represents the weight vector of X, w T w=1;c represents the weight vector of Y, c T c=1.

[0074] By iteratively extracting multiple latent variables (LVs), the residual error is gradually reduced.

[0075] Step 535 : For each latent variable, calculate the weight vector w of X and the weight vector c of Y, and update the residual matrices of X and Y until convergence.

[0076] Step 536: For the overlapping peaks in the 3.3 μm band, extract the sub-matrix X of the region from the full-band spectrum. sub , train the PLSR sub-model separately. By increasing the number of latent variables in this region (e.g., from 5 to 8), the model's ability to capture subtle differences is enhanced.

[0077] Step 537: Filter key wavelengths based on VIP (Variable Importance in Projection) values ​​to remove irrelevant noise. For example, the characteristic wavelength of CH4 may be concentrated at 3.31 μm, while C2H4 has a higher weight at 3.28 μm.

[0078] In step 538 , k-fold cross validation (eg, k=5) is used to evaluate the model performance, and the number of latent variables is selected to minimize the prediction error (eg, RMSE).

[0079] Step 539: Use an independent validation set to test the model to ensure that the correlation coefficient (R 2 )>0.98, and the relative error is <±2%.

[0080] The latent variables of the PLSR submodel in the present invention are selected by combining spectral characteristics with model performance for targeted optimization. Specifically, when processing the cross-absorption peak of CH4 and C2H4 in the 3.3μm band, the submatrix of this region is first extracted from the full-band spectrum and the model is trained separately. The first five latent variables of the original model mainly capture the main features of the full band (such as strong absorption peaks of other gases and background trends) to describe global information; the newly added three latent variables (bringing the total to eight) focus on the spectral details of the 3.3μm region. By constraining the weight vector to act only on the wavelength points of this band, screening the key wavelengths of CH4 (3.31μm) and C2H4 (3.28μm) based on the VIP value, and optimizing the number of latent variables using 5-fold cross-validation, the ability to capture subtle differences in overlapping peaks is enhanced, avoiding the introduction of interference from other regions, and ultimately minimizing the prediction error, effectively solving the problem of cross-interference of characteristic peaks, and improving the analysis accuracy and model generalization ability.

[0081] Step 540: Spectral data X new , first perform the same preprocessing, then project it into the latent variable space through the PLSR model, and finally output the concentration prediction value:

[0082] Y pred =X new ·B PLSR +Y offset

[0083] Among them: B PLSR Represents the model regression coefficient matrix; Y offset represents the intercept term during calibration.

[0084] PLSR effectively suppresses the background noise of the transformer oil matrix through latent variable extraction and is more stable than traditional least squares (OLS). This method achieves rapid and high-precision quantitative analysis of dissolved gases in transformer oil through the deep combination of PLSR and photoacoustic spectroscopy, providing a reliable data basis for fault diagnosis.

[0085] Through the optimized combination of the above steps, this detection method shortens the traditional gas detection process that takes several hours to less than 5 minutes, while ensuring that the detection accuracy meets the DL / T722 standard requirements, providing an efficient and reliable technical means for transformer condition monitoring.

[0086] Example 2

[0087] As one specific embodiment of the present invention, Figure 5 As shown, this embodiment provides a system for detecting dissolved gas in transformer oil, specifically comprising:

[0088] A multi-channel sampling device is used to collect transformer oil samples. The transformer oil samples collected by each sampling channel of the sampling device are processed in sequence through a pretreatment channel 3 and a degassing device 4. The gas released by the degassing device 4 is transmitted to the detection unit. A high-speed electromagnetic valve group 7 is provided between the degassing device 4 and the detection unit to achieve orderly switching of the gas released by each sampling channel.

[0089] The detection unit includes a plurality of photoacoustic cells 11 corresponding to each sampling channel, each of which is provided with a signal acquisition unit 12; a gas buffer chamber 8 is provided on both sides of the photoacoustic cell 11 for balancing the pressure difference between each channel;

[0090] The light source 9 is used to generate a laser beam with a set detection wavelength; the high-speed optical switch 10 performs time-sharing switching of the laser beam and sequentially excites the photoacoustic cell 11 corresponding to each sampling channel according to a preset timing;

[0091] The signal processing unit 13 is signal-connected to the high-speed electromagnetic valve group 7 , the high-speed optical switch 10 and the signal acquisition unit 12 , respectively, and is used to execute the method for detecting dissolved gas in transformer oil based on process optimization as described in Example 1.

[0092] Furthermore, a filter membrane 2 is provided between the multi-channel sampling device and the transformer 1 for filtering and removing solid impurities.

[0093] Furthermore, the pre-processing channel 3 is equipped with an independent constant temperature control unit for each collected transformer oil sample to stabilize the oil sample temperature within a set temperature range.

[0094] Furthermore, each degassing unit in the degassing device 4 adopts a polytetrafluoroethylene breathable membrane 5 structure.

[0095] Furthermore, the light source 9 adopts a tunable quantum cascade laser.

[0096] The process of detecting dissolved gas in transformer oil using the system provided in this embodiment is as follows:

[0097] A multi-channel sampling device simultaneously collects transformer oil samples. Solid impurities are removed by filtering through a 0.45μm microporous filter membrane 2, and the oil samples are evenly distributed across multiple pretreatment channels 3. Each channel is equipped with an independent thermostatic control unit to maintain the oil sample temperature within a range of 50±2°C. The pretreated oil samples are then introduced into a parallel degassing device 4. Each degassing unit utilizes a polytetrafluoroethylene breathable membrane 5 and performs degassing under a vacuum of 0.08MPa. Real-time monitoring of degassing efficiency dynamically adjusts vacuum and temperature parameters to ensure efficient degassing within 90 seconds. A high-speed solenoid valve assembly 7 enables orderly switching of gases between degassing channels. Micro-flow control technology ensures smooth gas delivery to the detection unit. A gas buffer chamber 8 is provided to balance pressure differences between channels, and the switching cycle is controlled within 25±5 seconds. A tunable quantum cascade laser is used as the light source 9, and the time-sharing switching of the laser beam is achieved through a high-speed optical switch 10. The photoacoustic cell 11 corresponding to each channel is excited in sequence according to the preset timing. The detection wavelength covers the 2.5-12μm band, and the detection time for each channel is allocated to 28 seconds. After the signal acquisition unit 12 collects the photoacoustic signals of each channel, it is transmitted to the signal processing unit 13. The signal processing unit 13 performs wavelet noise reduction processing on the photoacoustic signals transmitted by the acquisition unit to eliminate mechanical vibration interference, and uses the partial least squares regression algorithm to calculate the concentration of each gas component. In particular, a special analytical model is established for the cross-absorption peak of CH4 and C2H4 in the 3.3μm band. Step 6: Generate a comprehensive report containing the concentration of each gas component, growth rate and fault warning level based on the processing results.

[0098] The above describes in detail the preferred embodiments of the present invention. It should be understood that those skilled in the art can make numerous modifications and variations based on the concepts of the present invention without inventive effort. Therefore, any technical solutions that can be derived by those skilled in the art through logical analysis, reasoning, or limited experimentation based on the concepts of the present invention and the prior art should be within the scope of protection defined by the claims.

Claims

1. A method for detecting dissolved gas in transformer oil based on process optimization, characterized in that the steps include: Multi-channel collection of transformer oil samples and removal of dissolved gases in transformer oil; Control a single laser source to multiplex multiple acquisition channels in time-sharing mode, and stimulate the gases released from the corresponding channels in sequence according to the preset timing to generate detection signals; A gas component concentration analysis model based on the partial least squares regression algorithm is used to process the detection signals to obtain the predicted concentration of each gas component. The gas component concentration analysis model selects key wavelengths based on VIP values ​​for cross-absorption peaks and establishes a special analytical sub-model for gas cross-absorption peaks.

2. The method for detecting dissolved gas in transformer oil based on process optimization according to claim 1, characterized in that: The method of collecting multi-channel transformer oil sample gas is as follows: The oil samples collected by each sampling channel are evenly distributed to multiple pre-processing channels (3), and the temperature of the oil sample in each channel is independently controlled to be stable within a set temperature range; The oil samples pretreated in each pretreatment channel (3) are introduced into the parallel degassing device (4) for degassing under set vacuum conditions. The vacuum and temperature parameters are dynamically adjusted by real-time monitoring of the degassing efficiency to ensure that the degassing is completed within the set time.

3. The method for detecting dissolved gas in transformer oil based on process optimization according to claim 1, characterized in that: The partial least squares regression algorithm is specifically as follows: The photoacoustic signal of each gas channel is acquired by scanning the set wavelength band with a tunable quantum cascade laser to form the original absorption spectrum matrix X, where each row represents a sample and each column corresponds to the absorbance at a specific wavelength. Use standard gas samples of known concentration to establish a reference concentration matrix Y, where each row corresponds to a sample and each column is a gas concentration. The partial least squares regression algorithm is used to project the high-dimensional raw absorption spectrum data and reference concentration data into a low-dimensional latent variable space. The projection direction that can simultaneously explain the maximum covariance of the raw absorption spectrum matrix X and the reference concentration matrix Y is found. Multiple latent variables are extracted through iteration, and the residual error is gradually reduced. For each latent variable, the weight vector w of the original absorption spectrum matrix X and the weight vector c of the reference concentration matrix Y are calculated, and the residual matrix of the original absorption spectrum matrix X and the reference concentration matrix Y is updated until convergence.

4. The method for detecting dissolved gas in transformer oil based on process optimization according to claim 3, characterized in that: The partial least squares regression algorithm trains the PLSR sub-model separately for the cross-absorption peak, and increases the number of latent variables in the wavelength region corresponding to the cross-absorption peak, as follows: Extract the cross-absorption peak band region sub-matrix X from the full-band spectrum sub And train the PLSR sub-model separately; Filter key wavelengths based on VIP values ​​and remove irrelevant noise; K-fold cross validation was used to optimize the number of latent variables to minimize the prediction error of the PLSR sub-model; The PLSR sub-model was tested using an independent validation set to ensure that the correlation coefficient between the predicted concentration of each gas component and the reference value was greater than the set value and the relative error was less than the set error range.

5. The method for detecting dissolved gas in transformer oil based on process optimization according to claim 3, characterized in that: The partial least squares regression algorithm pre-processes the collected photoacoustic signal, including: performing wavelet noise reduction processing on the collected photoacoustic signal, and performing standardization and smoothing processing on the spectrum after wavelet noise reduction.

6. A system for detecting dissolved gas in transformer oil, characterized in that: The system comprises: The gas collection device is used to collect transformer oil samples through multiple channels, remove dissolved gases from the transformer oil, and input the gases removed from each channel into the detection unit in an orderly manner; The optical signal excitation device controls a single laser source to be time-divided and multiplexed into multiple collection channels, and excites the gases released from the corresponding channels in sequence according to a preset time sequence to generate detection signals; A signal processing unit (13) is configured to obtain a detection signal and execute the method for detecting dissolved gas in transformer oil based on process optimization as claimed in any one of claims 1 to 5.

7. A transformer oil dissolved gas detection system according to claim 6, characterized in that: The gas collection device comprises: A multi-channel sampling device is used for collecting transformer oil samples; the transformer oil samples collected by each sampling channel of the sampling device are processed in sequence through a pre-processing channel (3) and a degassing device (4); the gas degassed by the degassing device (4) is transmitted to a detection unit; a high-speed electromagnetic valve group (7) is provided between the degassing device (4) and the detection unit to achieve orderly switching of the gas degassed by each sampling channel.

8. A transformer oil dissolved gas detection system according to claim 7, characterized in that: A filter membrane (2) is provided between the multi-channel sampling device and the transformer (1) for filtering and removing solid impurities; the pretreatment channel (3) is equipped with an independent constant temperature control unit for each transformer oil sample collected, stabilizing the oil sample temperature within a set temperature range; and each degassing unit in the degassing device (4) adopts a polytetrafluoroethylene breathable membrane (5) structure.

9. A transformer oil dissolved gas detection system according to claim 6, characterized in that: The detection unit comprises a plurality of photoacoustic cells (11) corresponding to each sampling channel, each photoacoustic cell (11) being provided with a signal acquisition unit (12); and a gas buffer chamber (8) provided on both sides of the photoacoustic cell (11) for balancing the pressure difference between each channel.

10. A transformer oil dissolved gas detection system according to claim 6, characterized in that: The optical signal excitation device comprises: a light source (9) for generating a laser beam with a set detection wavelength; a high-speed optical switch (10) for performing time-sharing switching of the laser beam; the light source (9) adopts a tunable quantum cascade laser.

Images