A transformer oil dissolved gas detection system

Abstract

This invention proposes a multi-sensor detection system for dissolved gases in transformer oil. First, the transformer oil is degassed. The degassed gas is then blown into a chromatographic column using a carrier gas to separate the components. The separated gases are then sequentially fed into a spectrometer and a chromatographic instrument to obtain independent spectral data. Finally, a Gaussian fusion algorithm or a neural network algorithm is used to derive and calculate the data. The two sets of data are then compared and fused to obtain the final dissolved gas concentration data. This invention combines chromatographic and spectroscopic techniques, overcoming the shortcomings of each individual detection method, and by fusing the data from both, it improves the reliability and accuracy of the final results.

Description

Technical Field

[0001] This invention relates to the field of gas detection technology, specifically to a chromatographic and spectroscopic detection system for dissolved gases in transformer oil. Background Technology

[0002] With the continuous construction of ultra-high voltage substations in my country, there are higher demands for transformer inspection. The time interval from the occurrence of a fault to the detection of anomalies is getting shorter and shorter, requiring faster problem detection and resolution. Large power plants typically use oil-immersed power transformers in their main transformers and high-voltage plants. Because they are located on the critical path of power transmission, their operational reliability is closely related to the economic efficiency of the power plant. Whether it's a thermal or electrical fault, oil-immersed power transformers will cause the insulation medium—transformer oil—to crack, producing some characteristic gases, especially hydrogen (H2) and acetylene (C2H2). These values ​​are used to identify early possible transformer operational faults, and there are now higher detection limits. Currently, the detection limit for C2H2 in the power system has been reduced from 0.5 ppm to 0.2 ppm.

[0003] Existing gas chromatography methods are highly sensitive to pressure and temperature. In actual field applications, they are prone to missed and false alarms due to aging and varying operating conditions, affecting the assessment of transformer status and potentially leading to hazards. Furthermore, traditional chromatography requires a large dynamic range for CO2 detection. At high CO2 concentrations, the linearity and accuracy of the data are low, and the accuracy of the chromatographic detector also decreases at high concentrations to accommodate the large detection range. Summary of the Invention

[0004] The purpose of this invention is to provide a transformer oil dissolved gas detection system that combines chromatography and spectroscopy to overcome the shortcomings of each technology. The gas extracted from the transformer oil is sent to a spectroscopic instrument and a chromatographic instrument respectively, and the data from both are cross-compared. A fusion algorithm is used to process the data from multiple sensors to improve the reliability of the detection results.

[0005] To achieve the above objectives, this invention provides a transformer oil dissolved gas detection system, comprising a degassing unit, a chromatographic column, a spectrometer, a chromatographic detector, and a data fusion processing unit. The degassing unit degasses the transformer oil, purges it with a carrier gas, and then separates the gas components using the chromatographic column. The separated gas then enters the spectrometer and chromatographic detector for qualitative and quantitative analysis of each gas component, generating corresponding spectral and chromatographic data. The data fusion processing unit fuses multi-sensor data based on a Gaussian fusion algorithm and a neural network fusion model to obtain fused data for gas detection.

[0006] Furthermore, the data fusion processing unit first performs peak height detection on the spectral data and chromatographic data. If the difference between the two data exceeds a threshold, it is determined that the sensor is abnormal and no further algorithm processing is performed. Instead, a standard gas is used for re-detection. If the data difference does not exceed the threshold, further data fusion processing is performed.

[0007] Furthermore, the Gaussian fusion processing algorithm includes the following steps:

[0008] S101, Gaussian distribution fitting, performs Gaussian distribution fitting on the data peaks of chromatographic and spectral data to obtain Gaussian distribution fitting parameters, including the mean, standard deviation, and amplitude of the Gaussian distribution of the data peaks.

[0009] S102, Gaussian mixture model construction, using the fitted Gaussian distribution parameters as initial parameters to construct a Gaussian mixture model;

[0010] S103, Model Fitting: Optimize the Gaussian mixture model using a data fitting algorithm to obtain the optimal mixture model parameters.

[0011] S104, Peak fusion: By combining the Gaussian mixture model, the peak fusion parameters of each gas are obtained. The peak fusion parameters include the mean, standard deviation and amplitude of the Gaussian distribution of the peak.

[0012] S105, Result Verification: Verify whether the fused result can accurately describe the actual gas component peaks.

[0013] Furthermore, in step 102, it is assumed that one or more gas peaks have been detected in the data from each sensor. For each peak, a Gaussian distribution can be used to fit the data, and the probability density function of the Gaussian distribution is as follows:

[0014]

[0015] Where: μ is the expected value of the Gaussian distribution, representing the center position of the peak; σ 2 It is the variance, representing the width of the peak.

[0016] Furthermore, in step 102, assuming there are multiple sensors, and a certain gas is detected by multiple sensors, that is, the same gas peaks are detected simultaneously, then the N peaks are fused. Taking H2 gas as an example:

[0017]

[0018] Where λ is the weight of the i-th peak, and satisfies And λ i >0, μ i and σ i 2These are the mean and variance of the i-th Gaussian distribution.

[0019] Furthermore, in step 103, the expectation-maximization algorithm is used to fit the parameters of the Gaussian mixture model, where:

[0020] The simplified notation of the Gaussian distribution is denoted as , which represents the k-th Gaussian distribution in sample x. i The probability density function at:

[0021] p(x i )=N(x i μ k ,σ k 2 )

[0022] Calculate the posterior probability γ(z) for each component of each sample data. ik ), where z ik This represents the probability that the i-th sample belongs to the k-th component:

[0023]

[0024] Update the model parameters based on the obtained posterior probabilities:

[0025]

[0026]

[0027]

[0028] Repeat the above steps until the parameters converge to obtain the optimal parameters.

[0029] Furthermore, in step 104, the peaks detected by each sensor are fused into one or more more accurate peaks, and the parameters of the fused peaks are obtained by fitting the parameters of the Gaussian mixture model.

[0030] Furthermore, when the Gaussian normal distribution characteristics cannot be used for fitting during the Gaussian fusion processing algorithm, a neural network fusion model is selected to fuse multi-sensor data.

[0031] Furthermore, the neural network fusion model extracts and learns features from spectral and chromatographic data respectively, fuses the features extracted in the middle or end layers of the network, and finally outputs the concentration value of each component.

[0032] Furthermore, the neural network fusion model includes:

[0033] The input layer receives preprocessed spectroscopic and chromatographic data, and performs convolution calculations on the spectral data using trained 1D convolution kernels to compress the data volume and extract the feature values ​​of the peaks.

[0034] Hidden layers are used to allow the neural network to further learn the relationships between different gas components and extract the data features of each component;

[0035] Feature fusion combines and integrates data features from different detectors.

[0036] The fully connected layer is used to further process and compress the features of the fused data;

[0037] The output layer outputs classification results based on different task types, including predicting gas concentration and fault type.

[0038] The present invention has the following beneficial effects and technological advancements.

[0039] 1. Highly efficient fault detection and monitoring

[0040] This invention employs a combined chromatographic and spectroscopic technique to detect dissolved gases in transformer oil, and uses Gaussian fusion and neural network fusion algorithms to process multi-sensor data. By combining chromatographic and spectroscopic detection techniques, the content of gases such as H2 and C2H2 in transformer oil can be detected rapidly and accurately, significantly improving the speed and accuracy of fault detection.

[0041] 2. Improve detection sensitivity and accuracy

[0042] This invention utilizes a combined chromatography and spectroscopy technique, along with Gaussian fusion and neural network fusion algorithms. This reduces the detection limit for gases from 0.5 ppm to 0.2 ppm, meeting the higher accuracy requirements of current power systems. This highly sensitive detection technology significantly improves the accuracy of early transformer fault detection and reduces false alarm and missed detection rates.

[0043] 3. Strong anti-interference ability and good adaptability

[0044] This invention effectively reduces the problems of missed and false alarms caused by pressure and temperature changes in actual field applications of gas chromatography. The system improves the accuracy of transformer condition assessment through real-time calibration using an adaptive algorithm.

[0045] 4. Improved linearity and accuracy

[0046] This invention uses a combination of chromatography and spectroscopy to detect CO2. It solves the problem of the large dynamic range required for traditional chromatography in CO2 detection, and ensures data linearity and accuracy even at high CO2 concentrations.

[0047] 5. Social and economic benefits

[0048] This invention improves the reliability of power system transformer operation, reduces accidents caused by inadequate detection, and enhances the safety and stability of the power system. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is a schematic diagram of the transformer oil dissolved gas detection system of the present invention.

[0051] Figure 2 This is a logical architecture diagram of the data fusion algorithm part of the present invention.

[0052] Figure 3 This is a schematic diagram of the Gaussian fusion algorithm of the present invention.

[0053] Figure 4 This is a network structure diagram of the neural network fusion model of the present invention. Detailed Implementation

[0054] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0055] To better understand the above technical solutions, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present invention and to fully convey the scope of the invention to those skilled in the art.

[0056] Figure 1 This is a system framework diagram of a transformer oil dissolved gas detection system according to the present invention. Figure 1The system disclosed includes a transformer oil dissolved gas detection system, comprising: a degassing unit, a chromatographic column, a spectrophotometer, a chromatographic detector, a standard gas control unit, and a data fusion processing unit. This system sequentially feeds the gases degassing from the transformer oil into both a spectrophotometer and a chromatographic detector, cross-compares the data from both, and uses a fusion algorithm to process data from multiple sensors, thereby improving the reliability of the detection results.

[0057] Specifically, this system first uses a degassing unit to degas the transformer oil. After quantification, the sample gas is then purged by carrier gas and passed through a chromatographic column. The chromatographic column is filled with a specialized stationary phase material, which can effectively separate different components in the gas sample.

[0058] Preferably, the chromatographic column separates gas components based on the different characteristics of gas molecules (such as molecular size, polarity, etc.), ensuring that each component can independently enter the subsequent sensor. The gas components separated by the chromatographic column are first sent to the spectrometer detector.

[0059] Preferably, the spectrometer uses a non-closed detection method, continuously detecting during ventilation, and the data is acquired and stored by a lower-level computer (data fusion processing unit). Specifically, the spectrometer utilizes a light source of a specific wavelength to perform qualitative and quantitative analysis of gas components through absorption or emission spectra.

[0060] Finally, the gas sample is fed into a chromatographic detector, which further separates and detects the gas components based on their distribution behavior between the stationary and mobile phases.

[0061] Preferably, the chromatographic detector is a thermal conductivity detector (TCD) or a flame ionization detector (FID), which obtains the concentration information of the gas components by detecting their conductivity or ionization degree, and the data is also collected by the lower-level computer (data fusion processing unit).

[0062] In this embodiment, both spectral and chromatographic data include detection spectra after sample gas purging over a period of time, with consistent time axes and corresponding peak positions, meeting the basic requirements of the fusion calculation data format.

[0063] Preferably, after obtaining spectral and chromatographic data, the data fusion processing unit first detects whether there is excessive data difference caused by sensor failure. When the difference between the spectral and chromatographic data is too large, it is determined that one or more sensors may be faulty. The data fusion processing unit sends a signal to the standard gas control unit, which performs a measurement using standard gas instead of the degassing section. The standard sample result is compared with the previous data, and the faulty sensor is calibrated according to the standard gas data (within the calibrable range) or the device fault is reported to remind on-site personnel to repair or replace the equipment.

[0064] like Figure 3 As shown, Gaussian fusion processing is performed first in the data fusion process. In this embodiment, both spectral and chromatographic data approximate a normal distribution, which aligns with the application scenario of the Gaussian fusion algorithm. It should be noted that the Gaussian Mixture Model (GMM) is a parameterized probabilistic model that assumes the data is composed of several Gaussian distributions. Each sensor's measured peak can be considered a Gaussian distribution, and the GMM is used to merge these distributions to obtain a more accurate description of the gas component peaks. The GMM is generally an effective method for fusing peak data of the same gas from different sensors. During data fusion, if two sensors measure the same gas, but the peak shape, width, and intensity differ due to instrument characteristics or environmental conditions, a Gaussian mixture model can be used to fuse these peaks.

[0065] Specifically, the Gaussian fusion processing algorithm includes the following steps:

[0066] S101, Gaussian distribution fitting, performs Gaussian distribution fitting on the data peaks of chromatographic and spectral data to obtain Gaussian distribution fitting parameters, including the mean, standard deviation, and amplitude of the Gaussian distribution of the data peaks.

[0067] S102, Gaussian mixture model construction, using the fitted Gaussian distribution parameters as initial parameters to construct a Gaussian mixture model;

[0068] S103, Model Fitting: Optimize the Gaussian mixture model using a data fitting algorithm to obtain the optimal mixture model parameters.

[0069] S104, Peak fusion: By combining the Gaussian mixture model, the peak fusion parameters of each gas are obtained. The peak fusion parameters include the mean, standard deviation and amplitude of the Gaussian distribution of the peak.

[0070] S105, Result Verification: Verify whether the fused result can accurately describe the actual gas component peaks.

[0071] The following describes how the Gaussian fusion algorithm implements data fusion using specific examples:

[0072] 1. Data preprocessing and peak detection

[0073] First, the data from the chromatograph and spectrometer are preprocessed, including peak detection and peak parameter extraction. Typically, classic methods from one-dimensional signal processing, such as finding local maxima or setting thresholds, can be used to detect peaks. For a single-gas sensitive spectrometer, only one peak will appear throughout the measurement process. The position of the peak for the same gas may vary slightly, so several data points need to be shifted before fitting to ensure that the peak positions of different sensors are consistent.

[0074] 2. Gaussian distribution fitting

[0075] Suppose we have detected one or more gas peaks in the data from each sensor. For each peak, we can fit it with a Gaussian distribution. The probability density function of the Gaussian distribution is as follows:

[0076]

[0077] Where: μ is the expected value of the Gaussian distribution, representing the center position of the peak; σ 2 It is the variance, representing the width of the peak.

[0078] 3. Construction of Gaussian Mixture Model

[0079] Suppose we have several sensors, and a certain gas is detected by multiple sensors, meaning that the same gas peaks are detected simultaneously. Then, we fuse the N peaks. The following example uses H2:

[0080]

[0081] Where λ is the weight of the i-th component (peak), and satisfies And λ i >0, μ i and σ i 2 These are the mean and variance of the i-th Gaussian distribution.

[0082] 4. Model parameter fitting

[0083] The Expectation Maximization (EM) algorithm is used to fit the parameters of the Gaussian mixture model. The steps of the EM algorithm are as follows:

[0084] First, the Gaussian distribution can be simplified as , representing the k-th Gaussian distribution in sample x. i The probability density function at:

[0085] p(x i )=N(x i μ k ,σ k 2 )

[0086] Exception Step:

[0087] Calculate the posterior probability γ(z) for each component of each sample data. ik ), where z ik This represents the probability that the i-th sample belongs to the k-th component:

[0088]

[0089] M-step (Maximization Step):

[0090] Update the model parameters based on the obtained posterior probabilities.

[0091]

[0092]

[0093]

[0094] Repeat the E-step and M-step until convergence (e.g., the parameter change is less than a certain threshold or the maximum number of iterations is reached) to obtain the optimal parameters.

[0095] 5. Peak Fusion

[0096] Ultimately, the peaks detected by each sensor are fused into one or more more accurate peaks. The parameters (mean, variance, weights) of the fused peaks can be obtained by fitting the parameters of the Gaussian mixture model.

[0097] 6. Result Verification

[0098] The results showed that the fused peaks could more accurately describe the actual gas composition peaks, and the fitting effect of the fitted Gaussian mixture model was compared with that of the original data.

[0099] In the Gaussian fusion processing algorithm, when the original data is disturbed and the Gaussian normal distribution characteristics are not obvious, Gaussian fitting may fail, that is, the parameters cannot converge or the difference between the fused spectral peaks and the original data is too large. In this case, the Gaussian fusion algorithm will not be applicable. In this case, a fusion neural network is selected for processing. It uses multimodal learning to extract and learn features from chromatographic and spectral data separately. The extracted features are fused in the middle or last layer of the network, and finally the concentration value of each component gas is directly output.

[0100] Specifically, the implementation steps of the neural network fusion model include:

[0101] S201. Network Construction: Design a multimodal neural network to process chromatographic and spectral data respectively.

[0102] S202. Feature Extraction: Extract chromatographic and spectral features from the first few layers of the network.

[0103] S203. Feature fusion: Feature fusion is performed in the middle or end layers of the network.

[0104] S204. Subsequent layers: Use fully connected layers, etc., for numerical regression tasks.

[0105] Using neural networks for data fusion enables comprehensive analysis and processing of gas component data detected by spectroscopic and chromatographic methods. This method can automatically learn the complex relationships between data from different sensors, thereby improving the accuracy and reliability of fault diagnosis. The following section explains the data fusion implementation using a neural network fusion model with specific examples.

[0106] like Figure 4 As shown, the construction of the neural network fusion model includes:

[0107] 1. Data Preparation

[0108] First, we need to prepare gas composition data from spectroscopic and chromatographic methods. Let's assume we have the following data:

[0109] Chromatographic data: Time series data of the concentrations of multiple gaseous components (e.g., hydrogen, acetylene, etc.) in each sample.

[0110] Spectral data: Each sample contains spectral intensity values ​​at multiple wavelengths.

[0111] 2. Data Preprocessing

[0112] Data standardization ensures that data from different sensors have the same scale. Typically, the concentration of each gas component can be standardized to a range with a mean of 0 and a standard deviation of 1, facilitating network computation and improving convergence. Filtering and smoothing remove noise, reducing the difficulty of model training.

[0113] 3. Construct a neural network model

[0114] a. Input layer

[0115] The input layer of the neural network receives preprocessed spectroscopic and chromatographic data, and performs convolution calculations on the spectral data using 1D convolution kernels. Unlike the Gaussian fusion algorithm, it is not necessary to separate the peaks of each component here, but to extract the feature values ​​of the peaks through the calculation of the convolution kernels.

[0116] b. Hidden layer

[0117] Hidden layers are used to further learn the complex relationships between different gas components, extract component data features, and provide effective data for feature fusion. One-dimensional convolutional kernels are still used for convolution calculations, and pooling layers are used to reduce data dimensionality. The more layers, the more complex the model, the higher the accuracy, but also the greater the convergence difficulty and the larger the dataset required.

[0118] c. Feature fusion

[0119] The fusion stage integrates data features from different modalities. Through fusion, the characteristics of multimodal data can be comprehensively utilized to improve model performance. The fused features contain information from data from different modalities, providing a more comprehensive and accurate representation. Common methods include concatenation, weighted averaging, and attention mechanisms.

[0120] d. Fully connected layer

[0121] Fully connected layers are used to further process and compress the fused features to facilitate the final decision or output. Fully connected layers map high-dimensional features to a low-dimensional space, aggregate information, and provide input to the output layer. They typically have a large number of nodes and use non-linear activation functions.

[0122] f. Output layer

[0123] The output layer can have different settings depending on the task. If the goal is to classify fault types, the number of nodes in the output layer is equal to the number of fault types, and the activation function can be Softmax. If the goal is to predict gas concentration, the number of nodes in the output layer is equal to the number of gas components, and the activation function can be a linear function.

[0124] 4. Model Training

[0125] Train the neural network using historical data. Choose an appropriate loss function and optimization algorithm, such as cross-entropy loss (for classification) or mean squared error (for regression). The optimization algorithm can be Adam or SGD (stochastic gradient descent).

[0126] 5. Data Fusion Results

[0127] a. Output fault type

[0128] If the output layer of the neural network is a fault type classification, the fusion result is the probability distribution of each fault type. Through comprehensive analysis of data on different gas components, the neural network can output the probability of each fault type, aiding in fault diagnosis.

[0129] b. Output gas concentration

[0130] If the output layer of the neural network is a gas concentration prediction, the fusion result is the combined gas component concentrations. These concentration values ​​can then be used for decision analysis or fault diagnosis.

[0131] 6. Mobile network deployment

[0132] Current neural network training typically relies on Python or TensorFlow, meaning training is performed on a Python platform, and the exported results need to be deployed on a suitable embedded platform. Current technical solutions include: directly selecting a board that can run Python; a computing card with an NPU; using FPGAs for parallel acceleration; and reconstructing the convolutional neural network using C language code based on the exported network parameters.

[0133] Although embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A transformer oil dissolved gas detection system, characterized in that, The system includes a degassing unit, a chromatographic column, a spectrometer detector, a chromatographic detector, and a data fusion processing unit. The degassing unit degasses the transformer oil. Sample gas, purged with carrier gas, is passed through the chromatographic column for gas component separation. The separated gas then enters the spectrometer and chromatographic detector for qualitative and quantitative analysis of each gas component, generating corresponding chromatographic and spectral data. The data fusion processing unit fuses multi-sensor data based on a Gaussian fusion algorithm and a neural network fusion model to obtain fused gas detection data. The Gaussian fusion algorithm includes the following steps: S101, Gaussian distribution fitting, performs Gaussian distribution fitting on the data peaks of chromatographic and spectral data to obtain Gaussian distribution fitting parameters, including the mean, standard deviation, and amplitude of the Gaussian distribution of the data peaks. S102, Gaussian mixture model construction, using the fitted Gaussian distribution parameters as initial parameters to construct a Gaussian mixture model; S103, Model Fitting: Optimize the Gaussian mixture model using a data fitting algorithm to obtain the optimal mixture model parameters. S104, Peak fusion: By combining the Gaussian mixture model, the peak fusion parameters of each gas are obtained. The peak fusion parameters include the mean, standard deviation and amplitude of the Gaussian distribution of the peak. S105, Result Verification: Verify whether the fused result can accurately describe the actual gas component peaks. When it is impossible to fit the spectrum using Gaussian normal distribution features during the Gaussian fusion processing algorithm, a neural network fusion model is selected to fuse multi-sensor data. The neural network fusion model extracts and learns features from spectral and chromatographic data respectively. It extracts features independently from spectral data from different sensors in the input and hidden layers of the neural network, fuses and processes the extracted features in the intermediate layer of the neural network, and finally summarizes the data in the output layer to output the concentration information of each gas component. The neural network fusion model includes: The input layer receives preprocessed spectroscopic and chromatographic data, and performs convolution calculations on the spectral data using different 1D convolution kernels to extract the feature values ​​of the peaks. Hidden layers are used to learn the relationships between different gas components and extract the data features of the components; Feature fusion integrates feature data from different modes, enabling data collected by different sensors to be calculated in the same dimension. The fully connected layer is used to process and compress the features of the fused data; The output layer outputs classification results based on different task types, including fault type and predicted gas concentration.

2. The transformer oil dissolved gas detection system as described in claim 1, characterized in that, The data fusion processing unit first performs peak height detection on the spectral data and chromatographic data. If the difference between the two data exceeds a threshold, it is determined that the sensor is abnormal and no further algorithm processing is performed. Instead, a standard gas is used for re-detection. If the data difference does not exceed the threshold, data fusion processing is performed.

3. The transformer oil dissolved gas detection system as described in claim 1, characterized in that, In step 101, it is assumed that one or more gas peaks have been detected in the data from each sensor; for each peak, a Gaussian distribution is used for fitting, and the probability density function of the Gaussian distribution is as follows: in: It is the mathematical expectation of the Gaussian distribution, representing the center position of the peak; 2 It is the variance, representing the width of the peak.

4. The transformer oil dissolved gas detection system as described in claim 1, characterized in that, In step 102, assuming there are multiple sensors, and a certain gas is detected by multiple sensors, that is, the same gas peak is detected at the same time, then the N peaks are fused.

5. The transformer oil dissolved gas detection system as described in claim 1, characterized in that, In step 103, the expectation-maximization algorithm is used to fit the parameters of the Gaussian mixture model, where: The simplified notation of the Gaussian distribution is denoted as , which represents the k-th Gaussian distribution in the sample. The probability density function at: Calculate the posterior probability of each component for each sample data. ,in This represents the probability that the i-th sample belongs to the k-th component: Update the model parameters based on the obtained posterior probabilities: Repeat the above steps until the parameters converge to obtain the optimal parameters.

6. The transformer oil dissolved gas detection system as described in claim 1, characterized in that, In step 104, the peaks detected by each sensor are fused into one or more more accurate peaks, and the parameters of the fused peaks are obtained by fitting the parameters of the Gaussian mixture model.

Images