Multicomponent Gas Detection Method in Transformer Oil Based on Multi-Spectrum Fusion

Abstract

The present invention discloses a method for detecting multi-component gases in transformer oil based on the fusion of multiple spectra, which belongs to the technical field of transformer oil and comprises the following steps: sampling transformer oil and pre-processing the same using spectral technology; collecting and processing multi-source spectral data of transformer oil to determine spectral fusion data of transformer oil; establishing a mathematical model for detecting multi-component gases in transformer oil, analyzing the spectral fusion data of transformer oil, and determining the detection results of multi-component gases in transformer oil. The present invention solves the existing problem that multi-component gases in transformer oil cannot be effectively detected and the detection accuracy, sensitivity and stability cannot be effectively improved. The present invention can effectively detect multi-component gases in transformer oil based on the fusion of multiple spectra, realize simultaneous online monitoring of multi-component gases in transformer oil, effectively improve the accuracy, sensitivity and stability of detection, and provide fast and accurate information support for transformer status assessment and fault warning.

Description

Technical Field

[0001] The present invention relates to the technical field of transformer oil, and in particular to a method for detecting multi-component gas in transformer oil based on fusion of multiple spectra. Background Art

[0002] Dissolved gases in transformer oil are a key indicator of internal transformer faults. Detecting dissolved gases in transformer oil is a crucial tool for assessing transformer operating conditions. Changes in the type and content of dissolved gases often indicate potential faults within the transformer, such as partial discharge, overheating, or insulation aging. While established, traditional detection methods such as gas chromatography suffer from long detection cycles and complex procedures.

[0003] Chinese patent publication number CN119939205A discloses a method and system for predicting the structure of dissolved gas graphs in transformer oil using multiple views. The method includes collecting a multi-element sequence of dissolved gases in the oil, extracting hidden features using a multi-head attention mechanism module, and obtaining multiple sets of RELI feature matrices. The RELI feature matrix is segmented and constructed using a multi-view graph generator. The feature graph is further processed using a graph convolutional network to obtain features of different subspace information. Subsequently, the RELI feature matrix is updated and re-entered into the multi-head attention mechanism module to capture the complex relationships between features. Finally, the processed feature matrix is output to generate a predicted gas sequence. However, this patent has the following drawbacks:

[0004] Existing technologies cannot effectively detect multi-component gases in transformer oil based on multiple spectral fusion, cannot effectively improve the accuracy, sensitivity and stability of detection, and cannot provide fast and accurate information support for transformer status assessment and fault warning. Summary of the Invention

[0005] The purpose of the present invention is to provide a multi-component gas detection method in transformer oil based on multiple spectral fusion, which can effectively detect multi-component gases in transformer oil based on multiple spectral fusion, realize simultaneous online monitoring of multi-component gases in transformer oil, effectively improve the accuracy, sensitivity and stability of detection, and provide fast and accurate information support for transformer status assessment and fault warning, thereby solving the problems raised in the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] The multi-component gas detection method in transformer oil based on multiple spectral fusion includes:

[0008] Transformer oil was sampled and pre-treated using spectroscopy techniques;

[0009] Collect and process multi-source spectral data of transformer oil to determine transformer oil spectral fusion data;

[0010] A mathematical model for multi-component gas detection in transformer oil is established, and the transformer oil spectrum fusion data is analyzed to determine the detection results of multi-component gas in transformer oil.

[0011] Preferably, collecting transformer oil multi-source spectral data includes:

[0012] Pre-process the transformer oil sample based on infrared spectroscopy to obtain transformer oil infrared spectral data;

[0013] Preprocess the transformer oil sample based on photoacoustic spectroscopy to obtain transformer oil photoacoustic spectroscopy data;

[0014] The transformer oil sample is pre-processed based on tunable semiconductor laser absorption spectroscopy to obtain transformer oil laser absorption spectrum data;

[0015] The transformer oil multi-source spectrum data is determined based on the transformer oil infrared spectrum data, transformer oil photoacoustic spectrum data and transformer oil laser absorption spectrum data.

[0016] Preferably, the transformer oil multi-source spectral data is processed, including:

[0017] Wavelet transform is used to denoise the transformer oil multi-source spectral data, remove the noise in the transformer oil multi-source spectral data, and reduce the impact of random noise on the detection of multi-component gases in transformer oil;

[0018] The least square method is used to perform baseline correction on transformer oil multi-source spectral data to eliminate fluorescence background interference in transformer oil multi-source spectral data.

[0019] Maximum value normalization is used to normalize the transformer oil multi-source spectral data to remove the differences in the transformer oil multi-source spectral data and make the transformer oil multi-source spectral data comparable.

[0020] Preferably, the transformer oil multi-source spectral data is processed, including:

[0021] Perform feature extraction on the transformer oil multi-source spectral data to obtain each feature corresponding to the transformer oil multi-source spectral data;

[0022] Retrieving gas types corresponding to the multi-component gas in the transformer oil;

[0023] Retrieve from the database the characteristic reference value corresponding to each characteristic of each gas type at the reference concentration;

[0024] Retrieve historical detection data of multiple groups of gas detection;

[0025] Retrieving the data value of each feature corresponding to different gas concentrations of each gas type from the plurality of groups of historical gas detection data;

[0026] performing ratio processing on the different gas concentrations of each gas type in the multiple groups of historical gas detection data with the reference concentration corresponding to the gas type to obtain a concentration ratio coefficient;

[0027] performing ratio processing on the data value of each feature corresponding to each gas type at different gas concentrations in the plurality of groups of historical gas detection data and its corresponding feature reference value to obtain a feature ratio coefficient;

[0028] The characteristic ratio coefficient of each feature corresponding to different gas concentrations of each gas type is compared with the concentration ratio coefficient to obtain the characteristic-concentration correlation coefficient;

[0029] Using the feature-concentration correlation coefficient generated by different gas concentrations corresponding to each feature in each gas type, the feature-concentration correlation coefficient variance of each feature in each gas type is obtained as the dynamic variance of each feature in each gas type;

[0030] Extract the dynamic variance of all gas species corresponding to each feature;

[0031] Perform average processing on the dynamic variances of all gas types corresponding to each feature to obtain the average dynamic variance;

[0032] The dynamic variance average is compared with the preset variance average threshold. When the dynamic variance average corresponding to the feature is not lower than the preset variance average threshold, the weight value of the feature is incremented by 3% of the initial weight value.

[0033] Preferably, processing transformer oil multi-source spectral data further includes:

[0034] Feature extraction is performed on the multi-source spectral data of transformer oil. Feature vectors valuable for multi-component gas detection in transformer oil are extracted from the multi-source spectral data of transformer oil, and the feature vectors are weighted fused to determine the transformer oil spectral fusion data.

[0035] Preferably, the features extracted from the transformer oil multi-source spectral data are screened to determine feature vectors valuable for multi-component gas detection in transformer oil, including:

[0036] After performing feature extraction on the transformer oil multi-source spectral data, extracting features corresponding to the transformer oil multi-source spectral data;

[0037] Get the Pearson correlation coefficient between each feature and other features;

[0038] The Pearson correlation coefficient between each feature and other features is used to obtain the average value of the Pearson correlation coefficient between each feature and other features;

[0039] Obtain the nonlinear correlation strength between each feature and the j-th gas concentration in multi-component gas detection;

[0040] Extract the dynamic variance of each feature at different concentrations of each gas;

[0041] Obtain the average dynamic variance corresponding to each feature based on the dynamic variance of each feature at different gas concentrations;

[0042] Obtaining a feature value index corresponding to each feature using the Pearson correlation coefficient corresponding to each feature, the dynamic variance mean, and the nonlinear correlation strength between each feature and the j-th gas concentration in multi-component gas detection;

[0043] Comparing the characteristic value index with a preset index reference value;

[0044] Features whose feature value index is not lower than a preset index reference value are retrieved as feature vector elements, and the feature vector elements are used to form a feature vector valuable for detecting multi-component gas in transformer oil.

[0045] Preferably, obtaining the nonlinear correlation strength between each feature and the j-th gas concentration in multi-component gas detection includes:

[0046] Retrieve the concentration value corresponding to the j-th gas concentration;

[0047] Retrieve the specific feature data corresponding to each feature;

[0048] Extract each feature and all grid divisions corresponding to the j-th gas concentration in multi-component gas detection from the database;

[0049] Dynamically divide the data range of each feature using each grid division method, and obtain m intervals corresponding to each feature after division corresponding to each grid division method;

[0050] Dynamically divide the data range of the j-th gas concentration using each grid division method, and obtain n intervals corresponding to the gas concentration corresponding to each grid division method;

[0051] Obtaining a nonlinear correlation strength data value between each feature in each grid division method and the j-th gas concentration in multi-component gas detection using the m intervals corresponding to each feature in each grid division method and the n intervals corresponding to the j-th gas concentration, combined with the concentration value corresponding to the j-th gas concentration and the specific feature data corresponding to each feature;

[0052] The maximum nonlinear correlation strength data value among the nonlinear correlation strength data values corresponding to all grid division modes is used as the nonlinear correlation strength between each feature and the j-th gas concentration in multi-component gas detection.

[0053] Preferably, analyzing the transformer oil spectrum fusion data includes:

[0054] Establish a mathematical model for multi-component gas detection in transformer oil, deploy the mathematical model in the actual multi-component gas detection environment;

[0055] The transformer oil spectrum fusion data is input into the mathematical model of multi-component gas detection in transformer oil. Based on the mathematical model of multi-component gas detection in transformer oil, the transformer oil spectrum fusion data is analyzed and identified. By analyzing the spectral characteristics of various gas molecules in transformer oil at different wavelengths, the multi-component gas in transformer oil is identified and quantitatively analyzed to determine the detection results of multi-component gas in transformer oil.

[0056] Preferably, a mathematical model for detecting multi-component gases in transformer oil is established, and the following operations are performed:

[0057] Collect historical data on multi-component gas detection in transformer oil, divide the collected historical data into training sets and test sets;

[0058] Based on machine learning technology, a training set is used to train the machine learning model, so that the machine learning model can autonomously learn the multi-component gas detection behavior in transformer oil from the training set, identify and quantitatively analyze the multi-component gas in transformer oil, and determine the mathematical model for multi-component gas detection in transformer oil based on machine learning;

[0059] A test set was used to perform a performance test on the machine learning-based mathematical model for multi-component gas detection in transformer oil. This was used to evaluate whether the model could achieve the expected results of identifying and quantitatively analyzing multi-component gases in transformer oil, and to determine the model test evaluation results.

[0060] According to the model test evaluation results, the mathematical model for multi-component gas detection in transformer oil based on machine learning is adjusted and optimized, and then the optimal mathematical model for multi-component gas detection in transformer oil is determined.

[0061] Preferably, the performance test of the mathematical model for detecting multi-component gas in transformer oil based on machine learning is performed, including:

[0062] The test set is input into the mathematical model for multi-component gas detection in transformer oil based on machine learning. The performance of the mathematical model for multi-component gas detection in transformer oil based on machine learning is judged according to the output results of the mathematical model, and whether the mathematical model for multi-component gas detection in transformer oil based on machine learning can achieve the expected effect of identifying and quantitatively analyzing multi-component gas in transformer oil is evaluated.

[0063] When the mathematical model for multi-component gas detection in transformer oil based on machine learning cannot achieve the expected effect of identifying and quantitatively analyzing the multi-component gas in transformer oil, the parameters of the mathematical model for multi-component gas detection in transformer oil based on machine learning are continuously adjusted, and the mathematical model for multi-component gas detection in transformer oil based on machine learning is continuously iteratively optimized until the mathematical model for multi-component gas detection in transformer oil based on machine learning can achieve the expected effect of identifying and quantitatively analyzing the multi-component gas in transformer oil, thereby determining the optimal mathematical model for multi-component gas detection in transformer oil.

[0064] Preferably, the multi-component gas detection result in transformer oil includes the type and concentration of the multi-component gas in transformer oil, and the multi-component gas detection result in transformer oil is displayed to management personnel in real time in a visual form, making it convenient for management personnel to maintain and manage the transformer according to the multi-component gas detection result in transformer oil.

[0065] Compared with the prior art, the present invention has the following beneficial effects:

[0066] The present invention samples transformer oil and preprocesses it using spectral technology to obtain multi-source spectral data of transformer oil, processes the multi-source spectral data of transformer oil, determines transformer oil spectral fusion data, establishes a mathematical model for multi-component gas detection in transformer oil, analyzes the transformer oil spectral fusion data, analyzes the spectral characteristics of various gas molecules in transformer oil at different wavelengths, and then identifies and quantitatively analyzes the multi-component gas in transformer oil to determine the detection result of the multi-component gas in transformer oil. The multi-component gas in transformer oil can be effectively detected based on the fusion of multiple spectra, and simultaneous online monitoring of the multi-component gas in transformer oil can be achieved, which can effectively improve the accuracy, sensitivity and stability of detection and provide fast and accurate information support for transformer status assessment and fault warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 The present invention is a flow chart of a method for detecting multi-component gases in transformer oil based on fusion of multiple spectra. DETAILED DESCRIPTION

[0068] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0069] In order to solve the problem that the existing multi-component gas in transformer oil cannot be effectively detected based on multiple spectral fusion, the accuracy, sensitivity and stability of detection cannot be effectively improved, and fast and accurate information support cannot be provided for transformer status assessment and fault warning, please refer to Figure 1 , this embodiment provides the following technical solutions:

[0070] The multi-component gas detection method in transformer oil based on multiple spectral fusion includes:

[0071] Transformer oil is sampled and preprocessed using spectral technology. Multi-source spectral data of transformer oil is collected and processed to determine the spectral fusion data of transformer oil.

[0072] In this embodiment, collecting transformer oil multi-source spectral data includes:

[0073] Pre-process the transformer oil sample based on infrared spectroscopy to obtain transformer oil infrared spectral data;

[0074] Preprocess the transformer oil sample based on photoacoustic spectroscopy to obtain transformer oil photoacoustic spectroscopy data;

[0075] The transformer oil sample is pre-processed based on tunable semiconductor laser absorption spectroscopy to obtain transformer oil laser absorption spectrum data;

[0076] The transformer oil multi-source spectrum data is determined based on the transformer oil infrared spectrum data, transformer oil photoacoustic spectrum data and transformer oil laser absorption spectrum data.

[0077] Specifically, through the integration of spectral technology, simultaneous online monitoring of multi-component gases in transformer oil can be achieved, which facilitates the subsequent effective detection of multi-component gases in transformer oil and can effectively improve the accuracy, sensitivity and stability of detection.

[0078] In this embodiment, the transformer oil multi-source spectral data is processed, including:

[0079] Wavelet transform is used to denoise transformer oil multi-source spectral data, remove noise from transformer oil multi-source spectral data, reduce the impact of random noise on the detection of multi-component gases in transformer oil, improve the quality of transformer oil multi-source spectral data, and avoid errors caused by noise in subsequent analysis;

[0080] The least squares method is used to perform baseline correction on the transformer oil multi-source spectral data to eliminate the fluorescence background interference in the transformer oil multi-source spectral data and ensure that the transformer oil multi-source spectral data can reflect the true gas composition information;

[0081] The maximum value normalization is used to normalize the transformer oil multi-source spectral data to remove the differences in the transformer oil multi-source spectral data and make the transformer oil multi-source spectral data comparable.

[0082] Feature extraction is performed on the multi-source spectral data of transformer oil. Feature vectors valuable for the detection of multi-component gases in transformer oil are extracted from the multi-source spectral data of transformer oil. The feature vectors are weightedly fused to determine the transformer oil spectral fusion data, which is convenient for the subsequent identification and quantitative analysis of multi-component gases in transformer oil.

[0083] A mathematical model for multi-component gas detection in transformer oil is established, and the transformer oil spectral fusion data is analyzed to determine the detection results of multi-component gas in transformer oil. The detection results of multi-component gas in transformer oil include the type and concentration of multi-component gas in transformer oil, and the detection results of multi-component gas in transformer oil are displayed to management personnel in real time in a visual form, which is convenient for management personnel to maintain and manage the transformer according to the detection results of multi-component gas in transformer oil.

[0084] Specifically, the multi-source spectral data of transformer oil is processed, including:

[0085] Perform feature extraction on the transformer oil multi-source spectral data to obtain each feature corresponding to the transformer oil multi-source spectral data;

[0086] Retrieving gas types corresponding to the multi-component gas in the transformer oil;

[0087] Retrieve from the database the characteristic reference value corresponding to each characteristic of each gas type at the reference concentration;

[0088] Retrieve historical detection data of multiple groups of gas detection;

[0089] Retrieving the data value of each feature corresponding to different gas concentrations of each gas type from the plurality of groups of historical gas detection data;

[0090] performing ratio processing on the different gas concentrations of each gas type in the multiple groups of historical gas detection data with the reference concentration corresponding to the gas type to obtain a concentration ratio coefficient;

[0091] performing ratio processing on the data value of each feature corresponding to each gas type at different gas concentrations in the plurality of groups of historical gas detection data and its corresponding feature reference value to obtain a feature ratio coefficient;

[0092] The characteristic ratio coefficient of each feature corresponding to different gas concentrations of each gas type is compared with the concentration ratio coefficient to obtain the characteristic-concentration correlation coefficient;

[0093] Using the feature-concentration correlation coefficient generated by different gas concentrations corresponding to each feature in each gas type, the feature-concentration correlation coefficient variance of each feature in each gas type is obtained as the dynamic variance of each feature in each gas type;

[0094] Extract the dynamic variance of all gas species corresponding to each feature;

[0095] Perform average processing on the dynamic variances of all gas types corresponding to each feature to obtain the average dynamic variance;

[0096] The dynamic variance average is compared with the preset variance average threshold. When the dynamic variance average corresponding to the feature is not lower than the preset variance average threshold, the weight value of the feature is incremented by 3% of the initial weight value.

[0097] The technical effect of the above technical solution is as follows: First, for the multi-component gases in transformer oil, features are extracted from spectral data and combined with historical detection data. A series of ratio processing (concentration ratio coefficient, feature ratio coefficient, and feature-concentration correlation coefficient) is used to quantify the variation of each feature at different gas concentrations. For example, if the ratio of the feature ratio coefficient to the concentration ratio coefficient (feature-concentration correlation coefficient) of a feature fluctuates significantly when gas concentration changes, it indicates that the feature is significantly affected by changes in gas concentration. The variance of each feature's feature-concentration correlation coefficient at different gas concentrations is calculated as the dynamic variance. Variance is a measure of data dispersion. A large dynamic variance indicates that the feature's feature-concentration correlation coefficient fluctuates significantly at different gas concentrations, indicating that the feature is more sensitive to changes in gas concentration and is more valuable in reflecting changes in gas composition. The average dynamic variance of all gas types corresponding to each feature is calculated and compared with a preset variance average threshold. If the average value of the dynamic variance is not lower than the threshold, it means that the comprehensive fluctuation degree of this feature under different gas concentrations is large, and its ability to reflect changes in gas composition is strong. By increasing its weight value (the increment is 3% of the initial weight value), this feature can play a greater role in the subsequent analysis of transformer oil data and fault diagnosis, and the analysis results can more accurately reflect the actual situation of the transformer oil.

[0098] By extracting features from multi-source spectral data of transformer oil and comprehensively considering the correlations between features and concentrations at varying concentrations of multiple gas components, a more accurate characterization of the actual gas composition in transformer oil can be achieved. Compared to methods that do not consider these complex correlations, this approach reduces misjudgments caused by simplistic data processing, improves the accuracy of gas composition detection in transformer oil, and consequently enhances the reliability of transformer internal fault diagnosis. Leveraging historical gas detection data from multiple sets of gas detections, the approach calculates feature-concentration correlation coefficients and dynamic variances, taking into account various relationships such as the ratios of different gas concentrations to baseline concentrations and the ratios of feature data values to feature baseline values. This approach dynamically reflects the complex relationships between features and concentrations based on varying historical data and current data characteristics, making the solution more adaptable to transformer oil data under different operating conditions and stages, rather than relying on fixed parameters for analysis. By comparing the average dynamic variance with a preset threshold and adjusting feature weights, features that are more "active" (i.e., have larger dynamic variances) at varying gas concentrations are given higher weights. In the subsequent analysis and fault diagnosis of transformer oil data, these more discriminative features can play a greater role, optimize the weight distribution of features in the entire analysis model, and improve the performance of the analysis model.

[0099] Specifically, the features extracted from the transformer oil multi-source spectral data are screened to determine the feature vectors that are valuable for multi-component gas detection in transformer oil, including:

[0100] After performing feature extraction on the transformer oil multi-source spectral data, extracting features corresponding to the transformer oil multi-source spectral data;

[0101] Get the Pearson correlation coefficient between each feature and other features;

[0102] The Pearson correlation coefficient between each feature and other features is used to obtain the average value of the Pearson correlation coefficient between each feature and other features;

[0103] Obtain the nonlinear correlation strength between each feature and the j-th gas concentration in multi-component gas detection;

[0104] Extract the dynamic variance of each feature at different concentrations of each gas;

[0105] Obtain the average dynamic variance corresponding to each feature based on the dynamic variance of each feature at different gas concentrations;

[0106] Obtaining a feature value index corresponding to each feature using the Pearson correlation coefficient corresponding to each feature, the dynamic variance mean, and the nonlinear correlation strength between each feature and the j-th gas concentration in multi-component gas detection;

[0107] The feature value index corresponding to each feature is obtained by the following formula:

[0108] ;

[0109] Among them, R represents the feature value index corresponding to each feature; k represents the total number of gas types in multi-component gas detection; w j represents the nonlinear correlation strength between each feature and the j-th gas concentration; P represents the Pearson correlation coefficient corresponding to each feature; σ represents the dynamic variance average corresponding to each feature; specifically, In, w j It represents the nonlinear correlation strength between each feature and the jth gas concentration, and measures the degree of complex and nonlinear correlation between the feature and different gas concentrations. j Multiplying and raising the values to the kth power comprehensively assesses the strength of a feature's association with different gas concentrations. This calculation yields an average value that reflects the feature's overall correlation with gas concentrations, demonstrating its comprehensive ability to correlate multi-component gas concentrations and assessing its value from a nonlinear perspective. In the equation, P is the Pearson correlation coefficient corresponding to each feature, reflecting the degree of linear correlation between the feature and other features. 1-P highlights the independence of the features, and σ is the dynamic variance average corresponding to each feature. The variance reflects the degree of data dispersion, and the dynamic variance average reflects the fluctuation of the feature under different gas concentrations. ln(1+σ 2 )+1 as the denominator, when σ is larger (that is, the characteristic fluctuation is large and the response is sensitive to the change of gas concentration), Larger values emphasize the importance of features with large fluctuations in feature value assessment, as these features better reflect changes in gas concentration. This formula comprehensively considers the nonlinear relationship between the feature and multi-component gas concentration, the linear correlation between the feature and other features, and the fluctuation of the feature under different gas concentrations. Comprehensively evaluating feature value from multiple dimensions, compared to evaluating from a single perspective, can more accurately reflect the true value of the feature to multi-component gas detection, making the selected features more representative and improving the accuracy and reliability of the multi-component gas detection model. By considering feature independence, the interference and information redundancy caused by inter-feature correlation are reduced, allowing the model to focus more on features that contribute unique information. Furthermore, considering dynamic variance can screen features sensitive to changes in gas concentration, reduce data fluctuation interference, enhance the model's anti-interference ability in the complex and changing environment of transformer oil multi-source spectral data, and stabilize detection performance. Calculating the feature value index based on this formula can accurately screen valuable features and optimize the feature set. Remove low-value or redundant features, reduce the amount of calculation and data processing complexity, improve the efficiency of multi-component gas detection, and at the same time enhance the stability and generalization ability of the detection model so that it can perform better under different working conditions and data conditions.

[0110] Comparing the characteristic value index with a preset index reference value;

[0111] Features whose feature value index is not lower than a preset index reference value are retrieved as feature vector elements, and the feature vector elements are used to form a feature vector valuable for detecting multi-component gas in transformer oil.

[0112] The technical advantages of the above-mentioned technical solution are as follows: By calculating the Pearson correlation coefficient and its average, features with weak correlations with other features can be screened out. Features with low correlation carry relatively independent information, avoiding interference from redundant information between features and detection, thereby improving the accuracy of multi-component gas detection. Combined with the dynamic variance average, features that change significantly at different gas concentrations are highlighted. These features are sensitive to changes in gas concentration and can more accurately reflect the gas composition in transformer oil. By considering the strength of nonlinear correlations with gas concentration and exploring the complex relationship between features and gas concentration, multi-component gas detection can be achieved with more comprehensive and accurate results, thereby improving detection accuracy. By screening out valuable feature vectors and removing redundant and low-value features, the amount of subsequent data processing is reduced. In the processing of multi-source spectral data and multi-component gas detection, there is no need to analyze and calculate a large number of irrelevant or low-value features, making the algorithm run faster, shortening detection time, and improving detection efficiency. This is particularly advantageous when processing large amounts of transformer oil spectral data. The feature vectors constructed through screening contain independent features that are highly correlated with gas detection. The detection model established based on this eigenvector is less affected by external interference factors, and the impact of data fluctuations in different detection scenarios on the model is reduced, making the model more stable in multi-component gas detection and the results more reliable.

[0113] Specifically, obtaining the nonlinear correlation strength between each feature and the j-th gas concentration in multi-component gas detection includes:

[0114] Retrieve the concentration value corresponding to the j-th gas concentration;

[0115] Retrieve the specific feature data corresponding to each feature;

[0116] Extract each feature and all grid divisions corresponding to the j-th gas concentration in multi-component gas detection from the database;

[0117] Dynamically divide the data range of each feature using each grid division method, and obtain m intervals corresponding to each feature after division corresponding to each grid division method;

[0118] Dynamically divide the data range of the j-th gas concentration using each grid division method, and obtain n intervals corresponding to the gas concentration corresponding to each grid division method;

[0119] Obtaining a nonlinear correlation strength data value between each feature in each grid division method and the j-th gas concentration in multi-component gas detection using the m intervals corresponding to each feature in each grid division method and the n intervals corresponding to the j-th gas concentration, combined with the concentration value corresponding to the j-th gas concentration and the specific feature data corresponding to each feature;

[0120] The maximum nonlinear correlation strength data value among the nonlinear correlation strength data values corresponding to all grid division modes is used as the nonlinear correlation strength between each feature and the j-th gas concentration in multi-component gas detection.

[0121] The nonlinear correlation strength between each feature and the j-th gas concentration in multi-component gas detection is obtained by the following formula:

[0122] ;

[0123] Among them, w j Mutual information (MIC) represents the strength of the nonlinear correlation between each feature and the jth gas concentration in multi-component gas detection. MIC(X, Y) quantifies the nonlinear relationship between features and gas concentrations using the maximum information coefficient (MIC). X represents the specific feature data corresponding to each feature; Y represents the concentration value corresponding to the jth gas concentration; I(X, Y) represents the mutual information between the specific feature data corresponding to each feature and the concentration value corresponding to the jth gas concentration; m represents the number of intervals corresponding to each feature; and n represents the number of intervals corresponding to the gas concentration. Specifically, mutual information (I(X, Y)) measures the degree of dependence between two random variables, X (feature data) and Y (gas concentration value). It is based on the concept of information entropy and essentially measures the amount by which information about Y can reduce uncertainty about X, or vice versa. In this scenario, the greater the mutual information, the more information is shared between the feature data and the gas concentration value, and the stronger the correlation between the two, regardless of whether the correlation is linear or nonlinear. It quantifies the mutual dependence between two variables from an information theory perspective and is a fundamental metric for determining whether a feature and gas concentration are associated. When calculating mutual information, its value may be affected by the way the data is divided (i.e. the number of intervals). , normalizing the mutual information I(X,Y) to a relatively stable range, making the correlation strength measurements between different feature and gas concentration combinations comparable. This method avoids bias due to differences in the number of partitioned intervals and more accurately reflects the true nonlinear correlation between features and gas concentrations. This formula, based on mutual information and normalization from information theory, captures the essential correlation and dependency between features and gas concentrations. Unlike traditional methods that only consider linear relationships, it has excellent detection capabilities for nonlinear relationships and does not miss complex nonlinear correlation information, thus more accurately determining the nonlinear correlation strength between features and gas concentrations, providing a reliable basis for subsequent multi-component gas detection and feature screening. By integrating multiple grid partitioning methods using the maximum information coefficient (MIC(X,Y)), it avoids the bias and errors that can result from a single partitioning method. Different partitioning methods may highlight different data features. Taking the optimal partitioning method (i.e., the maximum information coefficient) makes this method highly adaptable to various data distributions and variations, reduces the impact of factors such as data fluctuation and noise on the correlation strength calculation, and improves the stability and reliability of the results, thereby enhancing robustness. This formula is applicable to all types of feature data and gas concentration data, independent of the specific data distribution (e.g., normal distribution, uniform distribution, etc.). Whether continuous or discrete, this formula can be applied to calculate the strength of nonlinear correlations by using appropriate interval partitioning. This formula has broad applicability and can effectively measure the correlation between features and gas concentrations in a variety of transformer oil multi-source spectral data scenarios, enhancing the versatility of the technical solution across diverse application environments.

[0124] The technical solution described above achieves the following technical benefits: Dynamically partitioning the feature data range and gas concentration data range through multiple grid partitioning methods enables more detailed capture of the complex relationships between features and gas concentrations. Different grid partitioning methods allow for data analysis at varying scales and angles. After obtaining nonlinear correlation strength data values from multiple partitioning methods, the highest value is selected as the final nonlinear correlation strength. This maximizes the potential for strong correlations between features and gas concentrations, avoiding the omission of important correlation information due to the limitations of a single partitioning method. This improves the accuracy of the nonlinear correlation strength measurement between features and gas concentrations, enabling more accurate subsequent multi-component gas detection based on this approach. Dynamic grid partitioning adapts to the distribution characteristics of different feature and gas concentration data. Complex and fluctuating data distributions can be accurately described using finer grid partitioning methods. Similarly, relatively simple data distributions can be effectively addressed using appropriate partitioning methods. This flexibility enables the solution to address a variety of data scenarios, enhancing its adaptability to analyzing nonlinear correlations between features and gas concentrations in diverse transformer oil data scenarios and ensuring a relatively accurate assessment of correlation strength under various operating conditions. Considering multiple grid partitioning methods and selecting the maximum correlation strength data value reduces the risk of incorrectly assessing correlation strength due to unreasonable or inappropriate individual partitioning methods. By integrating multiple partitioning results, the correlation strength is determined in a more robust manner, making the resulting nonlinear correlation strength more reliable. This also increases the credibility of subsequent operations such as feature screening and multi-component gas detection, thereby improving the reliability of the entire multi-component gas detection system.

[0125] In this embodiment, the transformer oil spectrum fusion data is analyzed, including:

[0126] Establish a mathematical model for multi-component gas detection in transformer oil, deploy the mathematical model in the actual multi-component gas detection environment;

[0127] The transformer oil spectrum fusion data is input into the mathematical model of multi-component gas detection in transformer oil. Based on the mathematical model of multi-component gas detection in transformer oil, the transformer oil spectrum fusion data is analyzed and identified. By analyzing the spectral characteristics of various gas molecules in transformer oil at different wavelengths, the multi-component gas in transformer oil is identified and quantitatively analyzed to determine the detection results of multi-component gas in transformer oil.

[0128] In this embodiment, a mathematical model for detecting multi-component gases in transformer oil is established, and the following operations are performed:

[0129] Collect historical data on multi-component gas detection in transformer oil, divide the collected historical data into training sets and test sets;

[0130] Based on machine learning technology, a training set is used to train the machine learning model, so that the machine learning model can autonomously learn the multi-component gas detection behavior in transformer oil from the training set, identify and quantitatively analyze the multi-component gas in transformer oil, and determine the mathematical model for multi-component gas detection in transformer oil based on machine learning;

[0131] A test set was used to perform a performance test on the machine learning-based mathematical model for multi-component gas detection in transformer oil. This was used to evaluate whether the model could achieve the expected results of identifying and quantitatively analyzing multi-component gases in transformer oil, and to determine the model test evaluation results.

[0132] According to the model test evaluation results, the mathematical model for multi-component gas detection in transformer oil based on machine learning is adjusted and optimized, and then the optimal mathematical model for multi-component gas detection in transformer oil is determined.

[0133] Among them, the performance test of the mathematical model for multi-component gas detection in transformer oil based on machine learning was carried out, including:

[0134] The test set is input into the mathematical model for multi-component gas detection in transformer oil based on machine learning. The performance of the mathematical model for multi-component gas detection in transformer oil based on machine learning is judged according to the output results of the mathematical model, and whether the mathematical model for multi-component gas detection in transformer oil based on machine learning can achieve the expected effect of identifying and quantitatively analyzing multi-component gas in transformer oil is evaluated.

[0135] When the mathematical model for multi-component gas detection in transformer oil based on machine learning cannot achieve the expected effect of identifying and quantitatively analyzing the multi-component gas in transformer oil, the parameters of the mathematical model for multi-component gas detection in transformer oil based on machine learning are continuously adjusted, and the mathematical model for multi-component gas detection in transformer oil based on machine learning is continuously iteratively optimized until the mathematical model for multi-component gas detection in transformer oil based on machine learning can achieve the expected effect of identifying and quantitatively analyzing the multi-component gas in transformer oil, thereby determining the optimal mathematical model for multi-component gas detection in transformer oil.

[0136] In summary, through the fusion of spectral technology, multi-source spectral data of transformer oil is collected, and the multi-source spectral data of transformer oil is processed to determine the spectral fusion data of transformer oil. By establishing a mathematical model for multi-component gas detection in transformer oil, the spectral fusion data of transformer oil is analyzed. By analyzing the spectral characteristics of various gas molecules in transformer oil at different wavelengths, the multi-component gas in transformer oil is identified and quantitatively analyzed to determine the detection results of multi-component gas in transformer oil. Multi-component gas in transformer oil can be effectively detected based on the fusion of multiple spectra, and simultaneous online monitoring of multi-component gas in transformer oil can be realized, which can effectively improve the accuracy, sensitivity and stability of detection, and can provide fast and accurate information support for transformer status assessment and fault warning.

[0137] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0138] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A multi-component gas detection method in transformer oil based on multiple spectral fusion, characterized in that: include: Transformer oil was sampled and pre-treated using spectroscopy techniques; Collect and process multi-source spectral data of transformer oil, extract feature vectors valuable for multi-component gas detection in transformer oil, and determine transformer oil spectral fusion data; The characteristic value index is compared with a preset index reference value, and the features whose characteristic value index is not less than the preset index reference value are retrieved as feature vector elements, and the feature vector elements are used to form a feature vector valuable for multi-component gas detection in transformer oil; A mathematical model for multi-component gas detection in transformer oil is established, and the transformer oil spectrum fusion data is analyzed to determine the detection results of multi-component gas in transformer oil.

2. The method for detecting multi-component gases in transformer oil based on multiple spectrum fusion as claimed in claim 1, wherein: Collect multi-source spectral data of transformer oil, including: Pre-process the transformer oil sample based on infrared spectroscopy to obtain transformer oil infrared spectral data; Preprocess the transformer oil sample based on photoacoustic spectroscopy to obtain transformer oil photoacoustic spectroscopy data; The transformer oil sample is pre-processed based on tunable semiconductor laser absorption spectroscopy to obtain transformer oil laser absorption spectrum data; The transformer oil multi-source spectrum data is determined based on the transformer oil infrared spectrum data, transformer oil photoacoustic spectrum data and transformer oil laser absorption spectrum data.

3. The method for detecting multi-component gases in transformer oil based on multiple spectrum fusion as claimed in claim 1, wherein: Processing of transformer oil multi-source spectral data, including: Wavelet transform is used to denoise the transformer oil multi-source spectral data, remove the noise in the transformer oil multi-source spectral data, and reduce the impact of random noise on the detection of multi-component gases in transformer oil; The least square method is used to perform baseline correction on transformer oil multi-source spectral data to eliminate fluorescence background interference in transformer oil multi-source spectral data. The maximum value normalization is used to normalize the transformer oil multi-source spectral data to remove the differences in the transformer oil multi-source spectral data and make the transformer oil multi-source spectral data comparable. Feature extraction is performed on the multi-source spectral data of transformer oil. Feature vectors valuable for multi-component gas detection in transformer oil are extracted from the multi-source spectral data of transformer oil, and the feature vectors are weighted fused to determine the transformer oil spectral fusion data.

4. The method for detecting multi-component gases in transformer oil based on multiple spectrum fusion as claimed in claim 3, wherein: Processing of transformer oil multi-source spectral data, including: Perform feature extraction on the transformer oil multi-source spectral data to obtain each feature corresponding to the transformer oil multi-source spectral data; Retrieving gas types corresponding to the multi-component gas in the transformer oil; Retrieve from the database the characteristic reference value corresponding to each characteristic of each gas type at the reference concentration; Retrieve historical detection data of multiple groups of gas detection; Retrieving the data value of each feature corresponding to different gas concentrations of each gas type from the plurality of groups of historical gas detection data; performing ratio processing on the different gas concentrations of each gas type in the multiple groups of historical gas detection data with the reference concentration corresponding to the gas type to obtain a concentration ratio coefficient; performing ratio processing on the data value of each feature corresponding to each gas type at different gas concentrations in the plurality of groups of historical gas detection data and its corresponding feature reference value to obtain a feature ratio coefficient; The characteristic ratio coefficient of each feature corresponding to different gas concentrations of each gas type is compared with the concentration ratio coefficient to obtain the characteristic-concentration correlation coefficient; Using the feature-concentration correlation coefficient generated by different gas concentrations corresponding to each feature in each gas type, the feature-concentration correlation coefficient variance of each feature in each gas type is obtained as the dynamic variance of each feature in each gas type; Extract the dynamic variance of all gas species corresponding to each feature; Perform average processing on the dynamic variances of all gas types corresponding to each feature to obtain the average dynamic variance; The dynamic variance average is compared with the preset variance average threshold. When the dynamic variance average corresponding to the feature is not lower than the preset variance average threshold, the weight value of the feature is incremented by 3% of the initial weight value.

5. The method for detecting multi-component gases in transformer oil based on multiple spectrum fusion as claimed in claim 4, wherein: The features extracted from the transformer oil multi-source spectral data are screened to determine the feature vectors that are valuable for multi-component gas detection in transformer oil, including: After performing feature extraction on the transformer oil multi-source spectral data, extracting features corresponding to the transformer oil multi-source spectral data; Get the Pearson correlation coefficient between each feature and other features; The Pearson correlation coefficient between each feature and other features is used to obtain the average value of the Pearson correlation coefficient between each feature and other features; Obtain the nonlinear correlation strength between each feature and the j-th gas concentration in multi-component gas detection; Extract the dynamic variance of each feature at different concentrations of each gas; Obtain the average dynamic variance corresponding to each feature based on the dynamic variance of each feature at different gas concentrations; Obtaining a feature value index corresponding to each feature using the Pearson correlation coefficient corresponding to each feature, the dynamic variance mean, and the nonlinear correlation strength between each feature and the j-th gas concentration in multi-component gas detection; Comparing the characteristic value index with a preset index reference value; Features whose feature value index is not lower than a preset index reference value are retrieved as feature vector elements, and the feature vector elements are used to form a feature vector valuable for detecting multi-component gas in transformer oil.

6. The method for detecting multi-component gases in transformer oil based on multiple spectrum fusion as claimed in claim 5, characterized in that: Obtain the nonlinear correlation strength between each feature and the j-th gas concentration in multi-component gas detection, including: Retrieve the concentration value corresponding to the j-th gas concentration; Retrieve the specific feature data corresponding to each feature; Extract each feature and all grid divisions corresponding to the j-th gas concentration in multi-component gas detection from the database; Dynamically divide the data range of each feature using each grid division method, and obtain m intervals corresponding to each feature after division corresponding to each grid division method; Dynamically divide the data range of the j-th gas concentration using each grid division method, and obtain n intervals corresponding to the gas concentration corresponding to each grid division method; Obtaining a nonlinear correlation strength data value between each feature in each grid division method and the j-th gas concentration in multi-component gas detection using the m intervals corresponding to each feature in each grid division method and the n intervals corresponding to the j-th gas concentration, combined with the concentration value corresponding to the j-th gas concentration and the specific feature data corresponding to each feature; The maximum nonlinear correlation strength data value among the nonlinear correlation strength data values corresponding to all grid division modes is used as the nonlinear correlation strength between each feature and the j-th gas concentration in multi-component gas detection.

7. The method for detecting multi-component gases in transformer oil based on multiple spectrum fusion according to claim 1, wherein: Analysis of transformer oil spectral fusion data, including; Establish a mathematical model for multi-component gas detection in transformer oil, deploy the mathematical model in the actual multi-component gas detection environment; The transformer oil spectrum fusion data is input into the mathematical model of multi-component gas detection in transformer oil. Based on the mathematical model of multi-component gas detection in transformer oil, the transformer oil spectrum fusion data is analyzed and identified. By analyzing the spectral characteristics of various gas molecules in transformer oil at different wavelengths, the multi-component gas in transformer oil is identified and quantitatively analyzed to determine the detection results of multi-component gas in transformer oil.

8. The method for detecting multi-component gases in transformer oil based on multiple spectrum fusion according to claim 7, wherein: To establish a mathematical model for multi-component gas detection in transformer oil, perform the following operations: Collect historical data on multi-component gas detection in transformer oil, divide the collected historical data into training sets and test sets; Based on machine learning technology, a training set is used to train the machine learning model, so that the machine learning model can autonomously learn the multi-component gas detection behavior in transformer oil from the training set, identify and quantitatively analyze the multi-component gas in transformer oil, and determine the mathematical model for multi-component gas detection in transformer oil based on machine learning; A test set was used to perform a performance test on the machine learning-based mathematical model for multi-component gas detection in transformer oil. This was used to evaluate whether the model could achieve the expected results of identifying and quantitatively analyzing multi-component gases in transformer oil, and to determine the model test evaluation results. According to the model test evaluation results, the mathematical model for multi-component gas detection in transformer oil based on machine learning is adjusted and optimized, and then the optimal mathematical model for multi-component gas detection in transformer oil is determined.

9. The method for detecting multi-component gases in transformer oil based on multiple spectrum fusion according to claim 8, wherein: The performance of the machine learning-based mathematical model for multi-component gas detection in transformer oil was tested, including: The test set is input into the mathematical model for multi-component gas detection in transformer oil based on machine learning. The performance of the mathematical model for multi-component gas detection in transformer oil based on machine learning is judged according to the output results of the mathematical model, and whether the mathematical model for multi-component gas detection in transformer oil based on machine learning can achieve the expected effect of identifying and quantitatively analyzing multi-component gas in transformer oil is evaluated. When the mathematical model for multi-component gas detection in transformer oil based on machine learning cannot achieve the expected effect of identifying and quantitatively analyzing the multi-component gas in transformer oil, the parameters of the mathematical model for multi-component gas detection in transformer oil based on machine learning are continuously adjusted, and the mathematical model for multi-component gas detection in transformer oil based on machine learning is continuously iteratively optimized until the mathematical model for multi-component gas detection in transformer oil based on machine learning can achieve the expected effect of identifying and quantitatively analyzing the multi-component gas in transformer oil, thereby determining the optimal mathematical model for multi-component gas detection in transformer oil.

10. The method for detecting multi-component gases in transformer oil based on multiple spectrum fusion according to claim 1, wherein: The multi-component gas detection result in transformer oil includes the type and concentration of the multi-component gas in the transformer oil, and is displayed to management personnel in real time in a visual form, making it convenient for management personnel to maintain and manage the transformer based on the multi-component gas detection result in transformer oil.

Images