A method and system for dynamically monitoring mixed gas in an electrical equipment cabinet

Abstract

The application relates to the technical field of gas monitoring, in particular to a method and system for dynamically monitoring mixed gas in an electrified equipment box. The method comprises the following steps: acquiring an original spectrum signal; performing baseline correction by polynomial fitting to eliminate background interference; removing cross-sensitivity of similar molecular weight gas by using an independent component analysis algorithm to separate independent gas absorption signals; determining target characteristic absorption peaks with absorption overlap according to characteristic frequencies; obtaining pure characteristic absorption peaks by using wavelet transform denoising; determining a concentration peak value capture window according to an extreme point position and a time offset; obtaining pure concentration data by multiplying a maximum absorption intensity value and a calibration coefficient; extracting peak concentration data by using a sliding window; and comparing the peak concentration data with a breakdown threshold to determine an insulation breakdown state. The application combines polynomial fitting baseline correction, independent component analysis separation and wavelet denoising processing, effectively eliminates background interference and cross-sensitivity problems, and realizes accurate analysis of the insulation breakdown state.

Description

Technical Field

[0001] This invention relates to the field of gas monitoring technology, specifically to a method and system for dynamic monitoring of mixed gases inside an electrical equipment box. Background Technology

[0002] With the rapid development of power systems, dynamic monitoring of mixed gases inside electrical equipment enclosures has become a key technical aspect of ensuring the safe and stable operation of power systems. Electrical equipment enclosures are one of the core pieces of equipment in power systems, and the health status of the insulation components within their enclosed environment directly affects the reliable operation of the entire power system. Timely and accurate detection of the characteristic gases released when internal insulation components break down plays an irreplaceable role in preventing catastrophic power outages.

[0003] Among these technologies, gas monitoring based on spectral analysis is widely used due to its high sensitivity and ability to simultaneously detect multiple components. This technology achieves accurate identification of gas composition and concentration by analyzing the absorption characteristics of gases to specific wavelengths of light, providing an effective technical means for condition monitoring of power equipment.

[0004] However, existing gas monitoring technologies have significant shortcomings when dealing with complex, enclosed environments. First, traditional monitoring methods rely on broad gas response threshold settings, making them highly susceptible to physical masking and chemical interference from non-faulty gases, leading to frequent false alarms and difficulty in obtaining pure and reliable concentration data. Second, there is a severe cross-sensitivity contradiction between background gases and faulty characteristic gases within the chamber. When a minor breakdown occurs in the insulation, the released carbon monoxide and hydrogen sulfide concentrations are extremely low, while the similar molecular weight gases filling the chamber cause drastic fluctuations in the sensor baseline, resulting in the weak fault gas signal being swallowed up by the large background noise. Furthermore, due to the spread of the cross-sensitivity effect, the characteristic absorption peaks of carbon monoxide and hydrogen sulfide severely overlap, causing the sensor to receive no longer two independent and clear concentration peaks, but rather a mixed waveform with blurred boundaries and intertwined patterns. This waveform overlap makes it impossible for the system to accurately pinpoint the instantaneous window of a trace gas concentration burst. Summary of the Invention

[0005] The purpose of this invention is to address the aforementioned shortcomings in the prior art by providing a method and system for dynamic monitoring of mixed gases inside an electrical equipment box.

[0006] The objective of this invention is achieved through the following technical solution: a method for dynamic monitoring of mixed gas inside an electrical equipment box, comprising the following steps: S1. Obtain the raw spectral signal within a continuous time period from the mixed gas monitoring node inside the power supply box; S2. The original spectral signal is subjected to baseline correction processing using a polynomial fitting method to eliminate baseline fluctuations caused by background interference, thereby obtaining a baseline-free spectral signal. S3. An independent component analysis algorithm is used on the baseline-de-spectral signal to eliminate the cross-sensitivity of gases with similar molecular weights and separate multiple independent gas absorption signals. S4. If the characteristic frequency of the gas absorption signal falls within the preset frequency range of carbon monoxide and hydrogen sulfide, then the gas absorption signal is determined to be a target characteristic absorption peak with absorption overlap. S5. Denoising is performed on the target feature absorption peak using wavelet transform algorithm to eliminate boundary ambiguity and obtain a pure feature absorption peak. S6. Determine the peak concentration capture window for trace gases based on the extreme point position of the pure characteristic absorption peak and the preset time offset. S7. Extract the maximum absorption intensity value from the concentration peak capture window and perform a product operation based on the maximum absorption intensity value and a preset calibration coefficient to calculate the pure concentration data; S8. Extract local maxima from the pure concentration data using a sliding window method to obtain peak concentration data for state assessment; S9. If the peak concentration data is greater than the preset breakdown threshold, insulation breakdown is determined to have occurred, and the insulation breakdown state analysis result is obtained.

[0007] Preferably, the baseline correction processing performed on the original spectral signal using a polynomial fitting method to eliminate baseline fluctuations caused by background interference, resulting in a baseline-free spectral signal, specifically involves: S21. The original spectral signal is segmented and the continuous time period is divided into multiple analysis intervals of preset length. S22. Within each analysis interval, the least squares method is used to determine the polynomial coefficients, and a polynomial fitting curve that can reflect the baseline drift trend within that interval is constructed. S23. Adaptively select the polynomial order based on the complexity of baseline fluctuations; S24. Subtract the fitted curve from the original spectral signal to obtain the baseline-de-baseline spectral signal.

[0008] Preferably, the specific principle of baseline correction using polynomial fitting is as follows: The core of polynomial fitting lies in utilizing the mathematical properties of polynomial functions to construct a fitting model that accurately reflects the changing trend of background interference. In the implementation process, firstly, features are extracted from the original spectral signal in the frequency domain to identify the characteristic frequency components of baseline drift. Then, based on the identified baseline features, a polynomial function of appropriate order is selected for fitting. The choice of polynomial order directly determines the balance between fitting accuracy and computational complexity; too low an order cannot accurately capture complex baseline fluctuations, while too high an order easily introduces overfitting problems. The system adopts an adaptive order selection strategy, dynamically adjusting the polynomial order according to the root mean square error of baseline fluctuations to ensure that the fitted curve accurately tracks baseline changes without overfitting noise. The fitted curve obtained after polynomial fitting represents the estimate of background interference. Accurately subtracting it from the original spectral signal can effectively eliminate the influence of baseline fluctuations on weak fault gas signals, significantly improving the signal-to-noise ratio.

[0009] Preferably, the step of employing an independent component analysis algorithm on the baseline-de-baseline spectral signal to eliminate cross-sensitivity of gases with similar molecular weights and separate multiple independent gas absorption signals specifically involves: S31. Center the baseline-de-baseline spectral signal to eliminate the DC component of the signal; S32. Whiten the centered signal and standardize the variance of the signal; S33. Construct the separation matrix for independent component analysis and solve the separation matrix using an iterative optimization algorithm; S34. Use the separation matrix to perform a linear transformation on the whitened signal to separate multiple independent source signals. S35. Estimate the number of independent components in the separated source signals and screen out the independent gas absorption signals with physical significance.

[0010] Preferably, the specific principle of using the Independent Component Analysis (ICA) algorithm to eliminate cross-sensitivity is as follows: The core assumption of the ICA algorithm is that the absorption signals of each independent gas component are statistically independent and non-Gaussian. Blind source separation is achieved by mining the higher-order statistical characteristics of the signals. In this application, the mixed gas in the chamber contains multiple components, and the spectral absorption signals of these components are superimposed on the detector to form a mixed signal. The ICA algorithm constructs a separation matrix to decompose the mixed signal into multiple independent source signals, each corresponding to the absorption characteristics of a gas component. The optimization objective of the algorithm is to minimize the mutual information between the separated signals, making each signal as independent as possible. During the iterative optimization process, a non-Gaussianity metric is used as the optimization criterion, and a fixed-point iterative algorithm is used for rapid convergence. The separated independent gas absorption signals represent the absorption profiles of different gas components in the chamber, providing a pure signal source for subsequent feature recognition and effectively solving the cross-sensitivity problem between gases with similar molecular weights.

[0011] Preferably, the step of determining that the gas absorption signal is a target characteristic absorption peak with absorption overlap if the characteristic frequency of the gas absorption signal falls within the preset frequency range of carbon monoxide and hydrogen sulfide specifically means: S41. Establish a preset characteristic frequency database for carbon monoxide and characteristic frequency database for hydrogen sulfide; S42. Extract the characteristic frequencies of each gas absorption signal. The characteristic frequencies are extracted using the spectral peak search method. S43. Determine whether the characteristic frequencies of each gas absorption signal fall within the frequency ranges of carbon monoxide and hydrogen sulfide simultaneously. S44. If the characteristic frequency of a gas absorption signal falls within the frequency range of two gases simultaneously, then it is determined that the signal has absorption overlap and is marked as the target characteristic absorption peak.

[0012] Preferably, the specific principle for determining the target characteristic absorption peak is as follows: Carbon monoxide and hydrogen sulfide each have their own unique characteristic absorption peaks in the spectral domain. However, due to their similar molecular weights and cross-sensitivity effects, the characteristic absorption peaks of the two gases overlap in the frequency domain. This invention establishes a precise characteristic frequency database to store the characteristic frequency ranges of each of the two gases. During the determination process, the system first extracts the spectral characteristics of the absorption signal of the gas to be detected and locates the main peak frequency of the signal using a peak search algorithm. Then, the main peak frequency is matched with a preset frequency range. If the frequency range conditions of carbon monoxide and hydrogen sulfide are simultaneously met, the signal is determined to have absorption overlap, which is the target characteristic absorption peak. This frequency range matching method can accurately locate the signal segment that needs to be processed, providing a clear processing target for subsequent wavelet denoising.

[0013] Preferably, the step of using wavelet transform algorithm to denoise the target feature absorption peak, eliminating boundary ambiguity, and obtaining a pure feature absorption peak specifically involves: S51. Perform multi-scale wavelet decomposition on the target feature absorption peak signal; S52. Calculate the noise energy distribution of each decomposition level and adaptively set the denoising threshold. S53. Threshold the wavelet coefficients to suppress noise-dominant coefficients; S54. Reconstruct the signal using the retained wavelet coefficients to obtain the pure characteristic absorption peak.

[0014] Preferably, the specific principle of using wavelet transform algorithm for denoising is as follows: Wavelet transform can provide good localization characteristics in both the time and frequency domains, making it particularly suitable for processing signals with blurred boundaries. The denoising process first decomposes the target feature absorption peak signal into detail coefficients and approximation coefficients at multiple scales. During decomposition, the signal energy is redistributed across different scales; the energy of the useful signal is usually concentrated in the large coefficients at a few scales, while the energy of the noise is distributed across all scales. By analyzing the energy distribution of coefficients at each scale, a denoising threshold can be adaptively set. A soft thresholding method is used to select the threshold, setting coefficients below the threshold to zero and shrinking coefficients above the threshold. The denoised signal is reconstructed through inverse wavelet transform, at which point the signal boundaries are clear and the features are obvious, providing a good data foundation for subsequent concentration peak capture.

[0015] Preferably, determining the peak concentration capture window for trace gases based on the extreme point position of the pure characteristic absorption peak and a preset time offset specifically involves: S61. Search for extreme points in the pure characteristic absorption peak signal to determine the moment when the gas concentration reaches its peak. S62. Extract the timestamps corresponding to the extreme points as the concentration peak times; S63. Set a preset time offset based on the gas diffusion characteristics and sensor response characteristics; S64. Superimpose the time offset onto the concentration peak time to obtain the start and end times of the concentration peak capture window.

[0016] Preferably, the specific principle for determining the concentration peak capture window is as follows: the extreme point position of the pure characteristic absorption peak directly reflects the moment when the target gas concentration reaches its peak during monitoring. However, since it takes a certain amount of time for the gas to diffuse from the insulation breakdown point to the sensor detection position, and the sensor itself has a response delay, there is a systematic time offset between the actual gas concentration peak moment and the detected signal peak moment. This invention corrects the detected peak moment by a preset time offset to obtain the true gas concentration peak moment. The width of the concentration peak capture window is set comprehensively based on the gas diffusion rate and the sensor's response characteristics to ensure that the complete shape of the concentration peak can be captured, providing sufficient data support for subsequent concentration calculations.

[0017] Preferably, the step of extracting the maximum absorption intensity value from the concentration peak capture window and multiplying the maximum absorption intensity value with a preset calibration coefficient to calculate the pure concentration data specifically involves: S71. Search for the maximum absorption intensity value within the concentration peak capture window; S72. Extract the maximum absorption intensity value as the concentration peak signal within the window; S73. Read the calibration coefficients for the corresponding gas type from the preset calibration coefficient database; S74. Multiply the maximum absorption intensity value with the calibration coefficient to convert it into the actual concentration and obtain the pure concentration data.

[0018] Preferably, the specific principle for calculating pure concentration data is as follows: there is a linear relationship between spectral absorption intensity and gas concentration described by the Lambert-Beer law, that is, absorbance is directly proportional to concentration. This invention establishes a mapping relationship between absorption intensity and concentration through calibration coefficients, which are obtained through standard gas calibration experiments, ensuring the accuracy of concentration calculation. Within the concentration peak capture window, the maximum absorption intensity value is extracted as the highest concentration of the target gas during the monitoring process. Multiplying this value by the corresponding calibration coefficient yields the pure concentration data. This concentration calculation method based on the peak capture window avoids interference from background noise, ensuring the purity and reliability of the concentration data.

[0019] Preferably, the step of extracting local maxima from the pure concentration data using a sliding window approach to obtain peak concentration data for state assessment specifically involves: S81. Set the length of the sliding window to a predetermined step size; S82. Move the window over the time series of pure concentration data with a preset sliding step size; S83. Calculate the maximum value of all data points within each window position as the local peak value at that position; S84. Connect the local peak values ​​at each window position to form a peak concentration data sequence.

[0020] Preferably, the principle behind using a sliding window method to extract peak concentration data is as follows: Pure concentration data exhibits random fluctuations over time, which may be caused by factors such as sensor noise and environmental disturbances. To obtain stable peak data for state assessment, this invention employs a sliding window method for local maximum extraction. The sliding window gradually slides across the time series, and at each preset step, the maximum value of all data points within the window is calculated as the local peak at that window position. This method effectively filters out random fluctuations in the concentration data, extracting a peak sequence that reflects the trend of gas concentration changes. The length and step size of the sliding window are optimized according to the characteristics of the monitored object, ensuring that sensitivity to real concentration changes is maintained while filtering out noise.

[0021] Preferably, if the peak concentration data is greater than a preset breakdown threshold, then an insulation breakdown is determined to have occurred, and the insulation breakdown state analysis result is obtained as follows: S91. Read the breakdown threshold of the corresponding insulation material from the preset breakdown threshold database; S92. Compare the peak concentration data with the breakdown threshold point by point; S93. If the peak concentration data is greater than the breakdown threshold, insulation breakdown is determined to have occurred. S94. Generate insulation breakdown state analysis results including breakdown time, location, and severity.

[0022] Preferably, the specific principle for determining the insulation breakdown state is as follows: When insulation breakdown occurs, a large amount of characteristic gas is released, and the concentration of these gases is significantly higher than the background concentration under normal conditions. This invention establishes a mapping relationship between concentration and breakdown state through a preset breakdown threshold. The breakdown threshold is set based on the breakdown characteristics of the insulating material and historical fault data. When the peak concentration data exceeds this threshold, it indicates that the insulation component has experienced a breakdown fault. The system simultaneously records the time point of breakdown and the corresponding peak concentration, providing detailed data support for subsequent fault diagnosis and maintenance decisions.

[0023] A dynamic monitoring system for mixed gas inside an electrical equipment enclosure, comprising: The spectral acquisition module is located inside the power supply box and is used to continuously acquire spectral signals inside the box and transmit the acquired raw spectral signals to the signal processing module. The signal processing module, connected to the spectral acquisition module, is used to perform baseline correction and independent component analysis on the raw spectral signal, and separate multiple independent gas absorption signals. The feature recognition module, connected to the signal processing module, is used to determine whether there is absorption overlap between the gas absorption signals, and to perform wavelet denoising processing on the signals with overlap to obtain pure feature absorption peaks. The concentration calculation module, connected to the feature recognition module, is used to calculate pure concentration data based on pure characteristic absorption peaks and extract peak concentration data using a sliding window method. The state determination module, connected to the concentration calculation module, is used to compare the peak concentration data with the breakdown threshold and output the insulation breakdown state analysis result.

[0024] The beneficial effects of this invention are as follows: This invention uses polynomial fitting for baseline correction, effectively eliminating baseline fluctuations caused by background interference and making the weak fault gas signal stand out from the massive background noise; it uses an independent component analysis algorithm to eliminate the cross-sensitivity of gases with similar molecular weights, separating multiple independent gas absorption signals and solving the problem of physical overlap of characteristic absorption peaks; it uses a wavelet transform algorithm to denoise the target characteristic absorption peaks, eliminating boundary ambiguity and obtaining pure characteristic absorption peaks; based on this, the concentration peak capture window is determined by the extreme point position and time offset, ensuring accurate capture of trace gas concentration peaks. In summary, this invention successfully isolates overlapping characteristic absorption peaks from complex in-chamber background gases, eliminates cross-interference of gases with similar molecular weights, accurately locks the concentration peak capture window of trace gases, obtains pure concentration data, and achieves accurate analysis of the breakdown state of insulating components. Attached Figure Description

[0025] The invention will be further illustrated with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the invention. For those skilled in the art, other drawings can be obtained based on the following drawings without any creative effort.

[0026] Figure 1 This is a flowchart of the method for dynamic monitoring of mixed gas inside the electrical equipment box of the present invention; Figure 2 This is a system schematic diagram of the dynamic monitoring method for mixed gas inside the equipment box of the present invention. Detailed Implementation

[0027] The present invention will be further described in conjunction with the following embodiments.

[0028] Example 1;

[0029] like Figure 1 and Figure 2As shown, this embodiment provides a detailed description of the method and system for dynamic monitoring of mixed gas inside an electrical equipment enclosure. In this embodiment, the system architecture includes a spectral acquisition module installed inside the electrical equipment enclosure. This module consists of a high-precision infrared spectral sensor, whose wavelength detection range covers a specific mid-infrared region. The wavelength range of this region is set to a specific numerical interval, the spectral resolution is set to a specific value, and the sampling frequency is set to a specific value. The spectral acquisition module contacts the gas inside the enclosure through an optical fiber probe to achieve continuous acquisition of the spectral signal of the mixed gas inside the enclosure. The optical fiber probe is made of corrosion-resistant quartz optical fiber, and its probe tip is located in the gas flow area inside the enclosure to ensure effective capture of gas components near the insulating components.

[0030] In continuous operation, the spectral acquisition module collects spectral data of the gas inside the chamber at preset sampling intervals, which are set to a specific time unit. During the acquisition process, the raw spectral signal output by the spectral sensor contains absorption intensity values ​​at multiple wavelength points, and the data at each wavelength point represents the gas absorption characteristics at that wavelength position. The raw spectral signal is converted into a digital signal by an analog-to-digital converter, with the conversion accuracy set to a specific number of bits to ensure that the data has sufficient quantization accuracy throughout the dynamic range. The digitized raw spectral signal is transmitted to the signal processing module for subsequent processing via the data bus.

[0031] After receiving the raw spectral signal, the signal processing module first performs the baseline correction processing described in step S2. During the baseline correction processing, the signal processing module segments the raw spectral signal, dividing the continuous time period into multiple analysis intervals of preset length. The length of each analysis interval is set to a specific time unit, and the spectral data collected within this time range is treated as an independent processing unit. The division of the analysis intervals takes into account the time-varying characteristics of baseline fluctuations. If the interval is too long, the baseline estimation will be inaccurate, and if the interval is too short, the statistical reliability will be reduced. Therefore, the length of the interval division needs to be balanced between system computing resources and estimation accuracy.

[0032] Within each analysis interval, the signal processing module uses the least squares method to determine the polynomial coefficients and constructs a polynomial fitting curve that reflects the baseline drift trend within that interval. The core of the least squares method is to find polynomial coefficients that minimize the sum of squared errors between the fitted curve and the original signal. In the implementation process, the signal processing module first extracts the absorption intensity values ​​of the original spectral signal at each wavelength point, and then constructs an objective function for fitting. This objective function characterizes the degree of deviation between the fitted value and the actual value. Through iterative solution, the system can obtain the optimal combination of polynomial coefficients.

[0033] The polynomial order is adaptively determined based on the complexity of the baseline fluctuations, with the order range set within a specific numerical interval. When the baseline fluctuations exhibit a simple linear trend, the system automatically selects a lower-order polynomial for fitting. When the baseline fluctuations exhibit complex nonlinear characteristics, the system automatically increases the polynomial order to improve fitting accuracy. The adaptive selection of the order is achieved by analyzing the root mean square error (RMSE) of the baseline fluctuations. When the RMSE exceeds a preset first threshold, the system increases the polynomial order; when the RMSE is below a preset second threshold, the system decreases the polynomial order. This adaptive strategy ensures that the fitted curve accurately tracks baseline changes without overfitting noise.

[0034] After subtracting the fitted curve from the original spectral signal, a baseline-degraded spectral signal is obtained. The baseline-degraded spectral signal eliminates baseline fluctuations caused by background interference, allowing the weak fault gas signal to stand out from the massive background noise. The signal-to-noise ratio of the signal after baseline degradation is significantly improved, creating favorable conditions for subsequent gas component separation.

[0035] After baseline correction is completed, the signal processing module performs independent component analysis as described in step S3. The core assumption of the independent component analysis algorithm is that the absorption signals of each independent gas component are statistically independent and non-Gaussian. Blind source separation is achieved by mining the higher-order statistical characteristics of the signal. In this invention, the mixed gas in the chamber contains multiple components, and the spectral absorption signals of these components are superimposed on the detector to form a mixed signal. The independent component analysis algorithm decomposes the mixed signal into multiple independent source signals by constructing a separation matrix, and each source signal corresponds to the absorption characteristics of a gas component.

[0036] Independent component analysis first involves centering the baseline-debasement spectral signal to eliminate the DC component. The centering process is implemented by calculating the average value of the baseline-debasement spectral signal at all wavelengths and time points, and then subtracting this average value from each data point in the signal. The mean of the signal after centering is zero, which lays the foundation for subsequent whitening processing.

[0037] After the centering process is completed, the signal processing module performs whitening on the centered signal to standardize the variance of the signal and make the variances of each dimension consistent. The purpose of whitening is to eliminate the correlation between the dimensions of the signal and make the statistical characteristics of each dimension of the processed signal more independent. Whitening is achieved by calculating the autocorrelation matrix of the signal and performing eigenvalue decomposition on it. The eigenvectors obtained after decomposition constitute the whitening matrix. Multiplying the original signal by the whitening matrix yields the whitened signal.

[0038] After whitening, the signal processing module constructs the separation matrix for independent component analysis and solves the separation matrix through an iterative optimization algorithm. The core objective of the iterative optimization algorithm is to minimize the mutual information between the separated signals, so that each signal is as independent as possible. During the iterative optimization process, the algorithm uses a non-Gaussianity metric as the optimization criterion and converges quickly through a fixed-point iterative algorithm. The fixed-point iterative algorithm has a fast convergence speed and can obtain a separation matrix that meets the accuracy requirements within a finite number of iterations.

[0039] After the separation matrix is ​​solved, the signal processing module uses the separation matrix to perform a linear transformation on the whitened signal to separate multiple independent source signals. Each source signal represents the absorption profile of a gas component in the chamber, and the number of source signals is equal to the preset number of independent components. The estimation of the number of independent components is based on information theory criteria, such as the Akaike information criterion or the Bayesian information criterion. The optimal number of components is selected by comparing the model complexity and goodness of fit under different numbers of components.

[0040] Finally, the number of independent components in the separated source signals is estimated, and independent gas absorption signals with physical significance are screened out. The screening process is based on the energy distribution, spectral characteristics, and matching degree with a known gas feature database. Signals with clear physical significance are retained, while meaningless noise signals are eliminated. The separated independent gas absorption signals represent the absorption profiles of different gas components in the chamber, providing a pure signal source for subsequent feature identification and effectively solving the problem of cross-sensitivity between gases with similar molecular weights.

[0041] After receiving the isolated independent gas absorption signals, the feature recognition module performs the target feature absorption peak determination process described in step S4. First, the feature recognition module establishes a preset carbon monoxide feature frequency database and a hydrogen sulfide feature frequency database. The feature frequency database stores the feature absorption peak positions of the two gases within a specific wavelength range. Each feature absorption peak position includes a center wavelength value and a feature bandwidth value. The center wavelength of the feature absorption peak of carbon monoxide is set to a specific value, and its feature bandwidth is set to a specific range. The center wavelength of the feature absorption peak of hydrogen sulfide is set to a specific value, and its feature bandwidth is set to a specific range.

[0042] The feature recognition module extracts the characteristic frequencies of each gas absorption signal. The extraction of characteristic frequencies adopts the spectral peak search method. The specific implementation process of the spectral peak search method is as follows: First, the gas absorption signal is subjected to Fourier transform to convert the signal from the time domain to the frequency domain. Then, the spectral points with local maxima are searched in the frequency domain. These spectral points correspond to the main frequency components in the signal. During the search process, a peak threshold is set. Only spectral points that exceed the threshold are identified as valid peaks.

[0043] After extracting the characteristic frequencies, the feature recognition module determines whether the characteristic frequencies of each gas absorption signal fall within the frequency ranges of both carbon monoxide and hydrogen sulfide. The determination process involves comparing the extracted characteristic frequencies with preset frequency ranges for carbon monoxide and hydrogen sulfide. The frequency ranges for carbon monoxide and hydrogen sulfide are set to specific ranges. If the characteristic frequency of a gas absorption signal falls within the frequency ranges of both gases, it is determined that the signal has absorption overlap and is marked as a target characteristic absorption peak. This frequency range matching method can accurately locate the signal segments that need to be processed, providing a clear processing target for subsequent wavelet denoising.

[0044] The determination of target characteristic absorption peaks solves the problem of physical overlap of characteristic absorption peaks. Since carbon monoxide and hydrogen sulfide have similar molecular weights, the characteristic absorption peaks of the two gases overlap in the frequency domain. Traditional single-gas detection methods cannot effectively distinguish overlapping absorption peaks. This invention can accurately identify target characteristic absorption peaks with absorption overlap by establishing a precise characteristic frequency database and performing interval matching, providing a clear target signal for subsequent processing.

[0045] After the feature recognition module completes the determination of the target feature absorption peak, it performs wavelet denoising processing as described in step S5. The core advantage of the wavelet transform algorithm is that it can provide good localization characteristics in both the time and frequency domains, and is particularly suitable for processing signals with blurred boundaries. The specific implementation process of denoising processing includes five main stages: multi-scale wavelet decomposition, noise energy distribution analysis, adaptive threshold setting, wavelet coefficient threshold processing, and signal reconstruction.

[0046] In the multi-scale wavelet decomposition stage, the feature recognition module performs multi-scale wavelet decomposition on the target feature absorption peak signal, with the number of decomposition levels set to a specific value. The wavelet decomposition uses a specific wavelet basis function, which has good compact support and orthogonality, and can effectively realize the multi-scale decomposition of the signal. During the decomposition process, the signal is decomposed into approximation coefficients and detail coefficients through a series of high-pass and low-pass filters. The approximation coefficients represent the low-frequency components of the signal, and the detail coefficients represent the high-frequency components of the signal. As the number of decomposition levels increases, the approximation coefficients are further decomposed into approximation coefficients and detail coefficients at smaller scales, realizing the redistribution of signal energy at different scales.

[0047] After completing the multi-scale wavelet decomposition, the feature recognition module calculates the noise energy distribution of each decomposition level and adaptively sets the denoising threshold. The specific method for noise energy distribution analysis is as follows: analyze the detail coefficients of each decomposition level and statistically analyze the statistical characteristics of the coefficients, such as mean and standard deviation. Based on the analysis results, the system automatically calculates the noise estimation level of that level and sets the denoising threshold accordingly. The threshold setting adopts the soft threshold method, which can better preserve the effective features of the signal while suppressing noise.

[0048] In the wavelet coefficient thresholding stage, thresholding is performed on the coefficients at each decomposition level to suppress noise-dominant coefficients and retain effective feature coefficients. During the processing, coefficients smaller than the denoising threshold are set to zero, and coefficients larger than the denoising threshold are shrunk to reduce their amplitude. This processing method can effectively eliminate the influence of noise on the signal while preserving the effective feature information of the signal to the greatest extent.

[0049] After thresholding, the feature recognition module uses the retained wavelet coefficients to reconstruct the signal and obtain the pure feature absorption peak. The signal reconstruction process uses inverse wavelet transform to reconstruct the processed wavelet coefficients in the reverse order of decomposition, resulting in the denoised pure feature absorption peak. At this point, the signal boundary is clear and the features are obvious, providing a good data foundation for subsequent concentration peak capture.

[0050] After receiving the pure characteristic absorption peak, the concentration calculation module performs the concentration peak capture window determination process described in step S6; the concentration peak capture window is determined based on the extreme point position of the pure characteristic absorption peak and the preset time offset.

[0051] The concentration calculation module first searches for extreme points in the pure characteristic absorption peak signal to determine the moment when the gas concentration reaches its peak. The extreme point search adopts a peak detection algorithm, which compares the signal point by point in the time series to identify the local maximum point. During the search process, a minimum peak height threshold is set, and only points exceeding this threshold are identified as valid extreme points.

[0052] After extracting the timestamps corresponding to the extreme points as the peak concentration times, the concentration calculation module sets a preset time offset based on the gas diffusion characteristics and sensor response characteristics. The gas diffusion characteristics depend on the molecular structure of the gas and the environmental conditions inside the chamber. The diffusion rate of hydrogen sulfide is set to a specific value, and the diffusion rate of carbon monoxide is set to a specific value. The sensor response characteristics include the sensor's response delay time, which is obtained through calibration experiments and set to a specific time unit.

[0053] The time offset is superimposed on the concentration peak time to obtain the start and end times of the concentration peak capture window; the setting of the window width takes into account the gas diffusion rate and sensor response characteristics to ensure that the complete shape of the concentration peak can be captured; the specific value of the window width is set to a specific time unit, which covers the complete process from the start of concentration rise to the concentration peak and then to the concentration decline.

[0054] After the concentration peak capture window is determined, the concentration calculation module performs the pure concentration data calculation and processing described in step S7. First, it searches for the maximum absorption intensity value within the concentration peak capture window and extracts the maximum absorption intensity value as the concentration peak signal within the window. The search for the maximum absorption intensity value traverses all data points within the window and finds the data point with the maximum value as the concentration peak signal.

[0055] After extracting the concentration peak signal, the concentration calculation module reads the calibration coefficient of the corresponding gas type from the preset calibration coefficient database. The calibration coefficient database stores the calibration coefficients of various gases in different concentration ranges. The calibration coefficients are obtained through standard gas calibration experiments, and the experimental conditions are consistent with the actual monitoring conditions. The calibration coefficient of carbon monoxide is set to a specific value, and the calibration coefficient of hydrogen sulfide is set to a specific value.

[0056] After obtaining the calibration coefficient, the concentration calculation module multiplies the maximum absorption intensity value with the calibration coefficient to convert it into the actual concentration and obtain the pure concentration data. This concentration calculation method based on the peak capture window avoids the interference of background noise and ensures the purity and reliability of the concentration data.

[0057] After the pure concentration data calculation is completed, the concentration calculation module performs the peak concentration data extraction processing described in step S8. First, the length of the sliding window is set to a predetermined step size, which is set to a specific time unit. The selection of the sliding window length needs to achieve a balance between smoothing effect and time resolution. If the length is too long, it will reduce the sensitivity to concentration changes, and if the length is too short, it will not be able to effectively suppress noise.

[0058] The window moves across the time series of pure concentration data with a preset sliding step size. The sliding step size is set to a specific time unit, which is smaller than the window length to ensure that there is overlap between windows. The maximum value of all data points within each window position is calculated as the local peak value at that position. The local peak values ​​of each window position are concatenated to form a peak concentration data sequence. This sliding window processing method can effectively filter out random fluctuations in the concentration data and extract the peak sequence that reflects the trend of gas concentration changes.

[0059] After receiving the peak concentration data, the status determination module executes the insulation breakdown status determination process described in step S9. First, it reads the breakdown threshold of the corresponding insulation material from the preset breakdown threshold database. The breakdown threshold database stores the breakdown thresholds of different insulation materials under different working conditions. The threshold is set based on the breakdown characteristics of the insulation material and historical fault data. The breakdown threshold for carbon monoxide is set to a specific concentration value, and the breakdown threshold for hydrogen sulfide is set to a specific concentration value.

[0060] The peak concentration data is compared point by point with the breakdown threshold. If the peak concentration data is greater than the breakdown threshold, insulation breakdown is determined to have occurred. The determination result generates an insulation breakdown status analysis result that includes the breakdown time, location and severity. The breakdown time is recorded based on the timestamp corresponding to the peak concentration data, the breakdown location is determined based on the concentration distribution comparison of multiple monitoring points in the box, and the severity is determined based on the magnitude by which the concentration peak exceeds the threshold.

[0061] This embodiment, through the above technical solution, successfully separated overlapping characteristic absorption peaks in the complex background gas inside the chamber, eliminated cross-interference of gases with similar molecular weights, accurately locked the concentration peak capture window of trace gases, obtained pure concentration data, and achieved accurate analysis of the breakdown state of the insulating components. The various technical means used in this embodiment work together to form a complete gas monitoring and state determination technical solution, and the processing accuracy and robustness of each link have reached a high level.

[0062] Example 2;

[0063] This embodiment analyzes and explains the second specific implementation method and system for dynamic monitoring of mixed gases in the equipment box. The main difference between this embodiment and the first embodiment is: the optimization of the technical solution for high-concentration background gas interference scenarios, and the complex monitoring environment where multiple characteristic gases coexist.

[0064] In this embodiment, the configuration of the spectral acquisition module is basically the same as in Embodiment 1, but the sampling frequency is set to a higher value to obtain a higher time resolution. A higher sampling frequency can capture faster changes in gas concentration, which is of great significance for the detection of sudden discharge faults. At the same time, the spectral resolution is also set to a higher value to obtain a more refined ability to distinguish spectral features.

[0065] In the baseline correction process, this embodiment adopts a hybrid correction strategy that combines piecewise polynomial fitting and trend term separation. The piecewise polynomial fitting is the same as in Embodiment 1, but a trend term separation processing step is added. The core of trend term separation is to separate the slowly changing trend components and the rapidly changing characteristic components in the spectral signal. The slowly changing trend components mainly represent the long-term changes of the background gas, while the rapidly changing characteristic components represent the instantaneous release of the fault gas.

[0066] The specific implementation process of trend term separation is as follows: First, the baseline-de-spectral signal is low-pass filtered. The cutoff frequency of the low-pass filter is set to a specific value. Signal components below this frequency are considered as trend terms. The signal obtained after filtering out trend terms mainly contains rapidly changing gas absorption characteristics, which are closely related to the release of fault gases. The introduction of trend terms provides additional reference information for baseline correction, enabling the system to distinguish between normal background fluctuations and abnormal fault signals.

[0067] In the independent component analysis process, this embodiment adopts a combined processing strategy that combines second-order moment analysis and higher-order moment analysis. Second-order moment analysis is mainly used to remove the correlation of the signal, while higher-order moment analysis is used to extract the independence information of the signal. The combined processing strategy can more accurately separate independent gas components from complex mixed signals, especially for gas mixtures with large concentration differences, it has a better separation effect.

[0068] In the target feature absorption peak determination process, this embodiment adds multi-feature fusion determination logic. In addition to frequency range matching determination, the system also adds time-domain waveform feature determination and frequency-domain morphological feature determination. Time-domain waveform features include parameters such as rise time, fall time, and pulse width, which can reflect the gas release characteristics. Frequency-domain morphological features include parameters such as the position of the main peak, the distribution of side peaks, and the spectral bandwidth, which can reflect the spectral fingerprint characteristics of the gas. Multi-feature fusion determination can more accurately distinguish between target feature absorption peaks that truly have overlapping absorption and interference signals that accidentally fall within the frequency range.

[0069] In the wavelet denoising process, this embodiment uses lifting wavelet transform instead of traditional wavelet transform. Lifting wavelet transform is an integer wavelet transform implementation that does not rely on Fourier transform, and has the advantages of high computational efficiency and low memory usage. In the denoising process, lifting wavelet transform adopts a specific lifting scheme, which can achieve lossless reconstruction of the signal and retain more signal detail information.

[0070] In the concentration peak capture window determination stage, this embodiment adopts an adaptive window width adjustment strategy. The window width is dynamically adjusted according to the drastic change in gas concentration. When a rapid increase in concentration is detected, the window width is automatically narrowed to improve time resolution, and when the concentration change tends to level off, the window width is automatically widened to improve detection accuracy. The adaptive adjustment strategy enables the system to better adapt to the detection needs of different fault types and different severity levels.

[0071] In the pure concentration data calculation stage, this embodiment adopts a dynamic calibration coefficient update mechanism. Traditional static calibration coefficients cannot adapt to the impact of changes in environmental conditions. The dynamic update mechanism automatically adjusts the calibration coefficients based on the real-time collected standard gas reference signals. The dynamic update process includes: periodically collecting standard gas samples of known concentrations, measuring their absorption intensity values, calculating correction factors based on the deviation between the measured values ​​and the standard values, and applying the correction factors to the real-time concentration calculation. This dynamic update mechanism ensures the accuracy of concentration calculations during long-term operation.

[0072] In the peak concentration data extraction stage, this embodiment adopts a variable-length sliding window strategy. The window length is dynamically adjusted according to the data quality. When obvious outliers are detected in the data, the window length is automatically increased to enhance the smoothing effect. When the data quality is good, the window length is automatically decreased to maintain sensitivity to real changes. The variable-length strategy achieves a better balance between noise suppression and change detection.

[0073] In the insulation breakdown state determination process, this embodiment adds multi-parameter fusion determination logic; in addition to concentration threshold determination, the system also adds concentration change rate determination, duration determination, and cumulative dose determination; concentration change rate determination is used to detect sudden discharge faults, duration determination is used to distinguish between continuous faults and transient interference, and cumulative dose determination is used to assess the overall impact of the fault; multi-parameter fusion determination significantly reduces false alarm rate and false negative rate, and improves the reliability of state determination.

[0074] This embodiment optimizes the above technical solution, further improving the system's adaptability and detection accuracy in complex monitoring environments, and can meet the higher requirements for power equipment condition monitoring.

[0075] Example 3;

[0076] This embodiment analyzes and explains the third specific implementation method and system for dynamic monitoring of mixed gas in the equipment box; the main difference between this embodiment and embodiment one is: the technical solution is adapted for extreme working environment conditions, and the multi-node collaborative processing optimization is optimized for large-scale monitoring networks.

[0077] In this embodiment, the spectral acquisition module uses a dedicated spectral sensor that is resistant to high temperatures and corrosion, and its operating temperature range is extended to a wider range of values ​​to adapt to high-temperature operating environments; the optical window of the sensor is made of sapphire material with a special coating treatment, which can resist the erosion of corrosive gases such as hydrogen sulfide; the length of the fiber optic probe is customized according to the cabinet structure to ensure that the probe tip can be inserted into the key monitoring position inside the cabinet.

[0078] In the baseline correction process, this embodiment adds a temperature compensation correction mechanism. Temperature changes can cause the response characteristics of the spectral sensor to drift, thereby affecting the baseline stability. The temperature compensation correction mechanism sets a temperature monitoring point near the sensor probe to collect ambient temperature data in real time and compensates and corrects the spectral signal based on the temperature data. The mathematical model for compensation correction is established through a pre-conducted temperature calibration experiment, and the model parameters are stored in the system's parameter database.

[0079] In the independent component analysis process, this embodiment adds a prior knowledge guidance mechanism. This mechanism uses known gas composition information to guide the separation process, improving separation efficiency and accuracy. Prior knowledge includes: the types of gases that may be present in the chamber, the typical concentration range of each gas, and the spectral characteristic database of each gas. The specific implementation of the guidance mechanism is as follows: a penalty term is added to the optimization objective function of the independent component analysis. This penalty term penalizes separation results that deviate from the prior knowledge, guiding the separation results to converge in a direction that conforms to the prior knowledge.

[0080] In the determination of target feature absorption peaks, this embodiment adds spatiotemporal correlation analysis logic. Spatiotemporal correlation analysis uses data from multiple monitoring points for joint determination. The same gas release event will generate related signal responses at multiple monitoring points. By analyzing the temporal correlation and spatial distribution characteristics between signals from different monitoring points, it is possible to more accurately distinguish between the real target feature absorption peaks and local interference signals. The specific implementation of spatiotemporal correlation analysis is as follows: calculate the correlation coefficient between signals from different monitoring points. Signals with correlation coefficients exceeding a set threshold are determined to have spatiotemporal correlation and belong to the real gas release signals.

[0081] In the wavelet denoising process, this embodiment employs a hybrid denoising strategy combining empirical mode decomposition (EMD) and wavelet transform. EMD adaptively decomposes the signal into multiple intrinsic mode functions (IMFs), each representing the signal's oscillation characteristics at a specific time scale. The hybrid denoising strategy first performs preliminary decomposition of the signal using EMD, then applies wavelet denoising to each IMF separately, and finally reconstructs the denoised signal from the processed IMFs. This hybrid denoising strategy combines the advantages of both methods, resulting in stronger denoising capabilities and better signal fidelity.

[0082] In the concentration peak capture window determination process, this embodiment adds a diffusion model prediction mechanism. The diffusion model prediction mechanism uses a gas diffusion physics model to predict the spatiotemporal distribution of gas concentration and optimizes the setting of window position and width based on the prediction results. The input parameters of the diffusion model include: the location of the gas release source, release time, release intensity, ambient temperature, and box geometry. The calculation results of the model give the spatial and temporal distribution characteristics of gas concentration, and the concentration peak capture window can be set more accurately based on these distribution characteristics.

[0083] In the pure concentration data calculation stage, this embodiment adds a cross-sensitivity correction mechanism. Cross-sensitivity correction is used to eliminate mutual interference between different gas components. When multiple gases are present at the same time, the absorption signal of each gas will be affected by the superposition of the absorption signals of other gases. The specific implementation of cross-sensitivity correction is as follows: establish a cross-response matrix of multiple gases. This matrix describes the influence weight of each gas on each other's detection channels. Perform matrix inversion operation on the original concentration calculation result according to the cross-response matrix to obtain the corrected pure concentration data.

[0084] In the peak concentration data extraction stage, this embodiment adopts a processing strategy combining Kalman filtering and sliding window. Kalman filtering is an optimal estimation method that can make the optimal estimation of the system state in a noisy environment. Combined with sliding window processing, Kalman filtering recursively estimates the data within the window to obtain a smooth and accurate peak concentration data sequence. During the processing, the model parameters of Kalman filtering are adaptively adjusted according to the statistical characteristics of the actual data to ensure the optimality of the estimation results.

[0085] In the insulation breakdown condition determination process, this embodiment adds a multi-level early warning mechanism. This mechanism sets multiple early warning levels based on the severity of the fault, with each level corresponding to different response measures. When the peak concentration data exceeds the first early warning threshold, the system generates a yellow early warning signal to alert maintenance personnel to monitor the equipment status. When the peak concentration data exceeds the second early warning threshold, the system generates an orange early warning signal to alert maintenance personnel to prepare for maintenance measures. When the peak concentration data exceeds the breakdown threshold, the system generates a red early warning signal and triggers an emergency shutdown procedure. This multi-level early warning mechanism enables early fault detection and graded response, improving the system's intelligent operation and maintenance level.

[0086] This embodiment optimizes the above technical solution, further enhancing the system's reliability in extreme working environments and its practicality in large-scale monitoring networks, thus meeting the comprehensive needs of power system status monitoring.

[0087] In summary, this invention uses polynomial fitting for baseline correction, effectively eliminating baseline fluctuations caused by background interference and making the weak fault gas signal stand out from the massive background noise. It eliminates the cross-sensitivity of gases with similar molecular weights through independent component analysis, separating multiple independent gas absorption signals and solving the problem of physical overlap of characteristic absorption peaks. It uses wavelet transform algorithm to denoise the target characteristic absorption peaks, eliminating boundary ambiguity and obtaining pure characteristic absorption peaks. Furthermore, it determines the concentration peak capture window by using the extreme point position and time offset, ensuring accurate capture of trace gas concentration peaks. Finally, it extracts peak concentration data and performs breakdown determination using a sliding window method, achieving accurate analysis of the insulation component's breakdown state.

[0088] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit the scope of protection of the present invention. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the essence and scope of the technical solutions of the present invention.

Claims

1. A method for dynamic monitoring of mixed gas inside an electrical equipment box, characterized in that: Includes the following steps: S1. Obtain the raw spectral signal within a continuous time period from the mixed gas monitoring node inside the power supply box; S2. The original spectral signal is subjected to baseline correction processing using a polynomial fitting method to eliminate baseline fluctuations caused by background interference, thereby obtaining a baseline-free spectral signal. S3. An independent component analysis algorithm is used on the baseline-de-spectral signal to eliminate the cross-sensitivity of gases with similar molecular weights and separate multiple independent gas absorption signals. S4. If the characteristic frequency of the gas absorption signal falls within the preset frequency range of carbon monoxide and hydrogen sulfide, then the gas absorption signal is determined to be a target characteristic absorption peak with absorption overlap. S5. Denoising is performed on the target feature absorption peak using wavelet transform algorithm to eliminate boundary ambiguity and obtain a pure feature absorption peak. S6. Determine the peak concentration capture window for trace gases based on the extreme point position of the pure characteristic absorption peak and the preset time offset. S7. Extract the maximum absorption intensity value from the concentration peak capture window and perform a product operation based on the maximum absorption intensity value and a preset calibration coefficient to calculate the pure concentration data; S8. Extract local maxima from the pure concentration data using a sliding window method to obtain peak concentration data for state assessment; S9. If the peak concentration data is greater than the preset breakdown threshold, insulation breakdown is determined to have occurred, and the insulation breakdown state analysis result is obtained.

2. The method for dynamic monitoring of mixed gas inside an electrical equipment box according to claim 1, characterized in that: The baseline correction process performed on the original spectral signal using a polynomial fitting method to eliminate baseline fluctuations caused by background interference, resulting in a baseline-free spectral signal, specifically involves: S21. The original spectral signal is segmented and the continuous time period is divided into multiple analysis intervals of preset length. S22. Within each analysis interval, the least squares method is used to determine the polynomial coefficients, and a polynomial fitting curve that can reflect the baseline drift trend within that interval is constructed. S23. Adaptively select the polynomial order based on the complexity of baseline fluctuations; S24. Subtract the fitted curve from the original spectral signal to obtain the baseline-de-baseline spectral signal.

3. The method for dynamic monitoring of mixed gas inside an electrical equipment box according to claim 1, characterized in that: The process of using an independent component analysis algorithm on the baseline-de-spectral signal to eliminate cross-sensitivity of gases with similar molecular weights and separate multiple independent gas absorption signals specifically involves: S31. Center the baseline-de-baseline spectral signal to eliminate the DC component of the signal; S32. Whiten the centered signal and standardize the variance of the signal; S33. Construct the separation matrix for independent component analysis and solve the separation matrix using an iterative optimization algorithm; S34. Use the separation matrix to perform a linear transformation on the whitened signal to separate multiple independent source signals. S35. Estimate the number of independent components in the separated source signals and screen out the independent gas absorption signals with physical significance.

4. The method for dynamic monitoring of mixed gas inside an electrical equipment box according to claim 1, characterized in that: If the characteristic frequency of the gas absorption signal falls within the preset frequency range of carbon monoxide and hydrogen sulfide, then the gas absorption signal is determined to be a target characteristic absorption peak with absorption overlap. Specifically: S41. Establish a preset characteristic frequency database for carbon monoxide and characteristic frequency database for hydrogen sulfide; S42. Extract the characteristic frequencies of each gas absorption signal. The characteristic frequencies are extracted using the spectral peak search method. S43. Determine whether the characteristic frequencies of each gas absorption signal fall within the frequency ranges of carbon monoxide and hydrogen sulfide simultaneously. S44. If the characteristic frequency of a gas absorption signal falls within the frequency range of two gases simultaneously, then it is determined that the signal has absorption overlap and is marked as the target characteristic absorption peak.

5. The method for dynamic monitoring of mixed gas inside an electrical equipment box according to claim 1, characterized in that: The specific steps involve using a wavelet transform algorithm to denoise the target feature absorption peak, eliminate boundary blurring, and obtain a pure feature absorption peak: S51. Perform multi-scale wavelet decomposition on the target feature absorption peak signal; S52. Calculate the noise energy distribution of each decomposition level and adaptively set the denoising threshold. S53. Threshold the wavelet coefficients to suppress noise-dominant coefficients; S54. Reconstruct the signal using the retained wavelet coefficients to obtain the pure characteristic absorption peak.

6. The method for dynamic monitoring of mixed gas inside an electrical equipment box according to claim 1, characterized in that: The determination of the peak concentration capture window for trace gases based on the extreme point position of the pure characteristic absorption peak and a preset time offset is specifically as follows: S61. Search for extreme points in the pure characteristic absorption peak signal to determine the moment when the gas concentration reaches its peak. S62. Extract the timestamps corresponding to the extreme points as the concentration peak times; S63. Set a preset time offset based on the gas diffusion characteristics and sensor response characteristics; S64. Superimpose the time offset onto the concentration peak time to obtain the start and end times of the concentration peak capture window.

7. The method for dynamic monitoring of mixed gas inside an electrical equipment box according to claim 1, characterized in that: The process of extracting the maximum absorption intensity value from the concentration peak capture window and multiplying it with a preset calibration coefficient to calculate the pure concentration data is as follows: S71. Search for the maximum absorption intensity value within the concentration peak capture window; S72. Extract the maximum absorption intensity value as the concentration peak signal within the window; S73. Read the calibration coefficients for the corresponding gas type from the preset calibration coefficient database; S74. Multiply the maximum absorption intensity value with the calibration coefficient to convert it into the actual concentration and obtain the pure concentration data.

8. The method for dynamic monitoring of mixed gas inside an electrical equipment box according to claim 1, characterized in that: The specific steps for extracting local maxima from the pure concentration data using a sliding window approach to obtain peak concentration data for state assessment are as follows: S81. Set the length of the sliding window to a predetermined step size; S82. Move the window over the time series of pure concentration data with a preset sliding step size; S83. Calculate the maximum value of all data points within each window position as the local peak value at that position; S84. Connect the local peak values ​​at each window position to form a peak concentration data sequence.

9. The method for dynamic monitoring of mixed gas inside an electrical equipment box according to claim 1, characterized in that: If the peak concentration data is greater than the preset breakdown threshold, insulation breakdown is determined to have occurred, and the insulation breakdown state analysis result is obtained as follows: S91. Read the breakdown threshold of the corresponding insulation material from the preset breakdown threshold database; S92. Compare the peak concentration data with the breakdown threshold point by point; S93. If the peak concentration data is greater than the breakdown threshold, insulation breakdown is determined to have occurred. S94. Generate insulation breakdown state analysis results including breakdown time, location, and severity.

10. A dynamic monitoring system for mixed gas inside an electrical equipment box, characterized in that: include: The spectral acquisition module is located inside the power supply box and is used to continuously acquire spectral signals inside the box and transmit the acquired raw spectral signals to the signal processing module. The signal processing module, connected to the spectral acquisition module, is used to perform baseline correction and independent component analysis on the raw spectral signal, and separate multiple independent gas absorption signals. The feature recognition module, connected to the signal processing module, is used to determine whether there is absorption overlap between the gas absorption signals, and to perform wavelet denoising processing on the signals with overlap to obtain pure feature absorption peaks. The concentration calculation module, connected to the feature recognition module, is used to calculate pure concentration data based on pure characteristic absorption peaks and extract peak concentration data using a sliding window method. The state determination module, connected to the concentration calculation module, is used to compare the peak concentration data with the breakdown threshold and output the insulation breakdown state analysis result.

Images