Online monitoring method and device for dissolved gas in oil of electrical equipment

Abstract

The invention provides an on-line monitoring method and device for dissolved gas in electrical equipment oil, and the method comprises the following steps: carrying out dynamic negative pressure constant temperature extraction treatment on insulating oil in electrical equipment to obtain a gas sample containing the dissolved gas; performing spectrum scanning treatment on the gas sample by using a multi-wavelength tunable laser to obtain absorption spectrum data of the multi-component gas; performing robust signal processing and abnormal value detection on the absorption spectrum data to obtain gas concentration data and an abnormal value mark; according to the gas concentration data and the abnormal value mark, intelligent diagnosis and early warning analysis are carried out on the operation state of the power equipment, and a monitoring result of the dissolved gas is obtained. By combining the laser spectrum technology and robust data processing, the abnormal values in the data can be identified and eliminated through robust signal processing and abnormal value detection, the reliability of the data is improved, high-precision and high-robustness data support is provided for subsequent intelligent diagnosis and early warning, and the accuracy of fault diagnosis and early warning is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of gas monitoring, and particularly to an on-line monitoring method and device for dissolved gases in the oil of power equipment. Background Art

[0002] Currently, the on-line monitoring technology for dissolved gases in power equipment is widely used in the fault diagnosis of transformers. The dissolved gases in transformer oil are mainly analyzed by chemical analysis methods or on-line monitoring methods based on spectral technology. Although these traditional technical methods can effectively detect the concentration and composition of dissolved gases, in practical applications, there is a significant defect: they often cannot provide sufficient robustness in complex environments, especially when the gas concentration is low, the operating state of the equipment fluctuates greatly, or it is affected by external interference, the accuracy and reliability of the data are poor.

[0003] Although the existing spectral monitoring technologies can achieve high precision, most of the solutions are sensitive to environmental factors (such as temperature, pressure, etc.) and system noise interference. As a result, in the complex operating environment of power equipment, it is impossible to effectively distinguish the changes in dissolved gases from the influence of the external environment, and false alarms or missed alarms often occur. Therefore, how to improve the accuracy and robustness of on-line monitoring of dissolved gases in the complex and dynamic power equipment environment has become a major challenge faced by the existing technology. Summary of the Invention

[0004] The main purpose of the present invention is to solve the technical problems of insufficient robustness and inability to effectively cope with complex environmental interference existing in the on-line monitoring technology for dissolved gases in existing power equipment; The first aspect of the present invention provides an on-line monitoring method for dissolved gases in the oil of power equipment. The on-line monitoring method for dissolved gases in the oil of power equipment includes: Performing dynamic negative pressure constant temperature extraction treatment on the insulating oil in the power equipment to obtain a gas sample containing dissolved gases; Performing spectral scanning treatment on the gas sample by using a multi-wavelength tunable laser to obtain absorption spectral data of multi-component gases; Performing robust signal processing and outlier detection on the absorption spectral data to obtain gas concentration data and outlier marks; According to the gas concentration data and the outlier marks, performing intelligent diagnosis and early warning analysis on the operating state of the power equipment to obtain the monitoring result of the dissolved gases.

[0005] Optionally, in the first implementation manner of the first aspect of the present invention, the performing spectral scanning treatment on the gas sample by using a multi-wavelength tunable laser to obtain absorption spectral data of multi-component gases includes: Select lasers with multiple different wavelengths according to the type of the preset target gas, and perform wavelength modulation processing on the gas sample to obtain a modulated optical signal; Perform second harmonic detection processing on the modulated optical signal to obtain a second harmonic signal; Perform adaptive wavelength locking processing on the second harmonic signal to obtain spectral data locked at the optimal absorption peak of the target gas; Calculate the absorption intensity of each target gas according to the locked spectral data to obtain absorption spectral data of the multi-component gas.

[0006] Optionally, in the second implementation manner of the first aspect of the present invention, the performing adaptive wavelength locking processing on the second harmonic signal to obtain spectral data locked at the optimal absorption peak of the target gas includes: Perform peak detection processing on the second harmonic signal to obtain the initial absorption peak position; Coarsely adjust the wavelength of the multi-wavelength tunable laser according to the initial absorption peak position to obtain a coarsely adjusted laser wavelength; Adjust the driving current of the multi-wavelength tunable laser based on the laser wavelength, and perform real-time measurement processing on the optical intensity after passing through the gas sample to obtain absorption spectral lines at different wavelengths; Calculate the spectral line center position according to the absorption spectral lines to obtain the laser parameters corresponding to the optimal absorption peak; Use a proportional-integral-derivative control algorithm to perform real-time adjustment processing on the driving parameters of the laser to obtain spectral data locked at the optimal absorption peak of the target gas.

[0007] Optionally, in the third implementation manner of the first aspect of the present invention, the performing robust signal processing and outlier detection on the absorption spectral data to obtain gas concentration data and outlier marks includes: Perform adaptive digital lock-in amplification processing on the absorption spectral data to obtain a demodulated second harmonic signal; Use the minimum covariance determinant method to perform iterative processing on the second harmonic signal to obtain a robust covariance estimator; According to the robust covariance estimator, perform Mahalanobis distance calculation processing on the second harmonic signal to obtain the outlier degree score of each data point; Classify and calculate the concentration of the data points according to the outlier degree score and a preset threshold to obtain gas concentration data and outlier marks.

[0008] Optionally, in the fourth implementation manner of the first aspect of the present invention, the performing adaptive digital lock-in amplification processing on the absorption spectral data to obtain a demodulated second harmonic signal includes: Perform analog-to-digital conversion processing on the absorption spectral data to obtain a digitized spectral signal, and generate digital sine and cosine reference signals with the same modulation frequency; Perform digital multiplication processing on the digitized spectral signal and the digital sine and cosine reference signals respectively to obtain two product signals; Perform low-pass filtering processing on the two product signals respectively to obtain two filtered signals, and use the two filtered signals as the in-phase component and the quadrature component respectively; Calculate the amplitude and phase of the signal according to the in-phase component and the quadrature component, and reconstruct the second harmonic signal according to the amplitude and phase to obtain the demodulated second harmonic signal.

[0009] Optionally, in the fifth implementation manner of the first aspect of the present invention, the iterative processing of the second harmonic signal by using the minimum covariance determinant method to obtain a robust covariance estimator includes: Randomly select h sample points from the second harmonic signal, calculate the initial mean vector and the initial covariance matrix, and calculate the determinant of the initial covariance matrix as the initial determinant value; Calculate the Mahalanobis distance of all data points by using the initial mean vector and the initial covariance matrix, and select the h points with the smallest distance as the new sample set; Recalculate the mean vector and the covariance matrix for the new sample set to obtain the updated mean vector and the covariance matrix; Calculate the determinant of the updated covariance matrix, and compare it with the initial determinant value to obtain a convergence judgment result; If the convergence judgment result does not converge, then use the updated mean vector and the covariance matrix as the new initial mean vector and the initial covariance matrix for iteration. If it converges, output the updated mean vector and the covariance matrix as the robust covariance estimator.

[0010] Optionally, in the sixth implementation manner of the first aspect of the present invention, the intelligent diagnosis and early warning analysis of the operating state of the power equipment according to the gas concentration data and the outlier mark to obtain the monitoring result of the dissolved gas includes: Use a pre-trained deep learning model to perform fault mode recognition processing on the gas concentration data to obtain a preliminary fault diagnosis result; Input the preliminary fault diagnosis result into a preset fuzzy logic multi-expert system, and perform comprehensive analysis processing in combination with the outlier mark to obtain the current fault diagnosis result; Perform time series analysis on the gas concentration data to obtain the gas concentration change trend, and based on the gas concentration change trend and the current fault diagnosis result, use a pre-trained prediction model for analysis and processing to obtain a potential fault prediction result; Perform comprehensive evaluation on the current fault diagnosis result and the potential fault prediction result to obtain the monitoring result of the dissolved gas.

[0011] The second aspect of the present invention provides an on-line monitoring device for dissolved gases in insulating oil of power equipment. The on-line monitoring device for dissolved gases in insulating oil of power equipment includes: A dynamic negative pressure constant temperature extraction module for performing dynamic negative pressure constant temperature extraction on the insulating oil in the power equipment to obtain a gas sample containing dissolved gases; A multi-wavelength spectral scanning module for performing spectral scanning on the gas sample using a multi-wavelength tunable laser to obtain absorption spectral data of multi-component gases; A robust signal processing module for performing robust signal processing and outlier detection on the absorption spectral data to obtain gas concentration data and outlier marks; An intelligent diagnosis and warning module for performing intelligent diagnosis and warning analysis on the operating state of the power equipment based on the gas concentration data and the outlier marks to obtain the monitoring result of the dissolved gas.

[0012] The above on-line monitoring method and device for dissolved gases in insulating oil of power equipment obtain a gas sample containing dissolved gases by performing dynamic negative pressure constant temperature extraction on the insulating oil in the power equipment; perform spectral scanning on the gas sample using a multi-wavelength tunable laser to obtain absorption spectral data of multi-component gases; perform robust signal processing and outlier detection on the absorption spectral data to obtain gas concentration data and outlier marks; perform intelligent diagnosis and warning analysis on the operating state of the power equipment based on the gas concentration data and the outlier marks to obtain the monitoring result of the dissolved gas. By combining laser spectroscopy technology and robust data processing, the present invention can identify and eliminate outliers in the data through robust signal processing and outlier detection, improve the reliability of the data, provide high-precision and high-robustness data support for subsequent intelligent diagnosis and warning, and improve the accuracy of fault diagnosis and warning.

[0013] Other features and advantages of the present invention will be described in the following description, and in part will be obvious from the description, or will be understood by implementing the present invention. The objectives and other advantages of the present invention are achieved and obtained by the structures specifically pointed out in the description, claims and drawings.

[0014] To make the above objects, features, and advantages of the present invention more apparent and understandable, the following provides preferred embodiments in conjunction with the accompanying drawings and describes them in detail as follows. Description of the Drawings

[0015] Figure 1 It is a schematic diagram of the first embodiment of the on-line monitoring method for dissolved gases in the oil of power equipment in the embodiment of the present invention; Figure 2 It is a schematic diagram of an embodiment of the on-line monitoring device for dissolved gases in the oil of power equipment in the embodiment of the present invention. Detailed Embodiment

[0016] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0017] The terms "including" and "having" and any variations thereof mentioned in the embodiments of the present invention are intended to cover non-exclusive inclusion. For example, a process, method, device, product, or equipment that includes a series of steps or units is not limited to the listed steps or units, but may optionally further include other unlisted steps or units, or may optionally further include other steps or units inherent to these processes, methods, products, or equipment.

[0018] To facilitate the understanding of this embodiment, first, a detailed introduction is given to an on-line monitoring method for dissolved gases in the oil of power equipment disclosed in the embodiments of the present invention. As Figure 1 shown, this method includes the following steps: 101. Perform dynamic negative pressure constant temperature extraction treatment on the insulating oil in the power equipment to obtain a gas sample containing dissolved gases; In one embodiment of the present invention, during the dissolved gas monitoring of power equipment, the dynamic negative pressure constant temperature extraction treatment of insulating oil is a key link in obtaining gas samples. First of all, the dissolved gases in the insulating oil have a certain concentration. These gases gradually dissolve in the oil during the long-term operation of the transformer, including gases such as hydrogen, carbon monoxide, and methane. In order to obtain these dissolved gas samples, it is necessary to rely on the negative pressure constant temperature extraction technology. This method can accurately extract gases from the oil while ensuring the gas dissolution state, and ensure the representativeness and stability of the gases. The insulating oil is guided into a specially designed extraction device, which is equipped with a heating system, a negative pressure system, and a gas sampling device. The heating system is used to maintain the oil in a constant temperature state, avoiding fluctuations in gas solubility caused by temperature changes. The oil enters the extraction chamber through a pipeline and undergoes constant temperature treatment in this chamber. By controlling the temperature of the heater, the oil temperature is maintained at a set constant value to ensure a stable release process of the dissolved gases. Temperature control is crucial for the solubility of gases in the oil and the release of dissolved gases, because temperature changes will directly affect the gas solubility and release efficiency. Too high or too low temperatures will cause deviations in gas concentration. Next, the negative pressure system starts. The negative pressure treatment uses an external gas pump to extract the gas in the extraction chamber, thus creating an environment below atmospheric pressure. Under the action of negative pressure, the gases dissolved in the oil will quickly be released and enter the gas sample collection container. By precisely controlling the magnitude of the negative pressure, the gas extraction process is ensured to proceed smoothly without too fast or too slow gas release speed. The use of negative pressure avoids problems such as gas volatilization and unstable concentration that may be caused by direct heating or atmospheric pressure release, ensuring that the extracted gas can accurately reflect the true concentration of the dissolved gas in the oil. The constant temperature and negative pressure treatment throughout the process can enable the gases dissolved in the oil to be completely extracted without being damaged, ensuring the representativeness and effectiveness of the extracted samples. At this time, the extracted gas sample has been fully processed, and the gas concentration and composition can reflect the internal operating state of the transformer. The negative pressure constant temperature extraction technology can maximize the retention of the original characteristics of the gas sample, avoiding gas loss or changes during the extraction process, thus providing high-quality samples for subsequent gas analysis. Finally, through a specific gas sampling device, the extracted gas sample is sent to the analysis system for component analysis. This process not only needs to ensure that the extracted gas concentration meets the requirements, but also avoid interference from external air or other pollutants. Each step is strictly controlled to ensure that no other impurities are introduced during the extraction process, affecting the accuracy of the gas sample. Therefore, the negative pressure constant temperature extraction treatment technology provides an accurate and reliable sample basis for subsequent dissolved gas analysis, ensuring the efficiency and accuracy of transformer operating state monitoring.

[0019] 102. Use a multi-wavelength tunable laser to perform spectral scanning on the gas sample to obtain absorption spectral data of multi-component gases; In an embodiment of the present invention, the use of a multi-wavelength tunable laser to perform spectral scanning on the gas sample to obtain absorption spectral data of multi-component gases includes: selecting a plurality of lasers with different wavelengths according to the types of preset target gases, and performing wavelength modulation processing on the gas sample to obtain a modulated optical signal; performing second harmonic detection processing on the modulated optical signal to obtain a second harmonic signal; performing adaptive wavelength locking processing on the second harmonic signal to obtain spectral data locked at the best absorption peak of the target gas; calculating the absorption intensity of each target gas according to the locked spectral data to obtain absorption spectral data of multi-component gases.

[0020] Specifically, during the spectral scanning process, it is first necessary to select a plurality of lasers with different wavelengths according to the types of preset target gases. Each gas molecule has its specific absorption wavelength, and the absorption characteristics of different gases vary with their molecular structures and vibration modes. For example, hydrogen usually has a specific absorption peak in the near-infrared region, while methane has obvious absorption characteristics in a slightly longer wavelength range. In order to accurately analyze multiple gas components in a gas sample, it is necessary to select lasers with multiple wavelengths according to the absorption characteristics of the target gases. The wavelengths of these lasers should be matched according to the molecular absorption peaks of the gases to ensure that the absorption peaks of each target gas can be fully covered by the laser signal. In actual operation, when selecting a suitable laser, not only the wavelength range but also the power, frequency stability, and output beam quality of the laser need to be considered. For example, when detecting hydrogen, selecting a laser within a specific wavelength range is to accurately match the absorption peak of hydrogen to ensure that the signal intensity is sufficient to detect minute changes in the gas. In addition, the laser also needs to be able to provide a stable output under the specified working conditions, which is crucial for improving the detection sensitivity and accuracy. Different target gases have different absorption characteristics, so selecting a suitable laser can not only enhance the signal intensity but also improve the resolution of gas components.

[0021] Specifically, after selecting the laser, the gas sample needs to be subjected to wavelength modulation processing. Wavelength modulation refers to controlling the wavelength output of the laser so that it produces periodic wavelength changes within a certain time range to achieve the purpose of scanning different wavelength ranges of the gas absorption characteristics. The key to this process lies in adjusting the wavelength of the laser so that it can cover the absorption bands of all target gases in the gas sample. The wavelength modulation of the laser is usually achieved through a precise electronic control system, which can accurately control the output of the laser to make it change smoothly within the set range. Through modulation, the laser signal will pass through multiple absorption peak regions, thereby obtaining absorption data at multiple wavelengths. During this process, the molecules in the gas sample will have different absorptions according to the laser intensity at different wavelengths, generating signal changes proportional to the gas concentration. These changes can be recorded by a photodetector and converted into digital signals for subsequent analysis. By modulating the wavelength of the optical signal, the reaction characteristics of different gas molecules within their absorption intervals can be accurately obtained, improving the response ability to gas concentration changes. During this process, wavelength modulation can not only improve the detection sensitivity but also reduce the measurement error caused by a single wavelength, ensuring that the absorption characteristics of all target gases can be accurately captured.

[0022] Specifically, the modulated optical signal will then undergo second-harmonic detection processing. When the gas molecules have a non-linear interaction with the laser light wave, the frequency of the optical signal will have a frequency doubling effect, generating a second-harmonic signal. The intensity of this second-harmonic signal is closely related to the laser intensity, gas concentration, and the absorption characteristics of the molecules. By generating the second-harmonic, an enhanced signal can be obtained on the basis of the original signal, thereby improving the detection ability for weak absorption signals. The frequency of the second-harmonic signal is a multiple of the laser frequency, usually twice the original frequency, which makes the second-harmonic signal more obvious and easier to distinguish compared to the original laser signal. During this process, a high-precision detector and filter are used to extract the second-harmonic signal. By accurately capturing the second-harmonic signal, the signal-to-noise ratio of the system can be improved, reducing the interference of background noise on the data. The detection of the second-harmonic signal is particularly important for the accurate identification of low-concentration gases because the absorption characteristics of low-concentration gases are often weak, and the second-harmonic signal can effectively enhance these weak signals, making them clearly distinguishable in a complex background.

[0023] Specifically, after obtaining the second harmonic signal, it is necessary to perform adaptive wavelength locking processing on it. Since the absorption characteristics of gas molecules may be affected by factors such as ambient temperature and pressure, resulting in a slight shift in the position of the absorption peak, it is necessary to dynamically adjust the wavelength of the laser to ensure that the laser wavelength is always located at the optimal absorption peak of the target gas. Adaptive wavelength locking technology usually relies on a feedback control system. By continuously monitoring the changes in the second harmonic signal, the system can automatically adjust the wavelength of the laser to ensure that it always coincides with the absorption peak of the target gas. During this process, the photodetector will continuously monitor the absorption of the gas sample and compare the detected absorption peak with the preset wavelength. According to the feedback information, the wavelength of the laser will be slightly adjusted to compensate for the influence of the external environment on the position of the absorption peak. Through this dynamic adjustment process, the system can ensure that the laser is always locked at the optimal absorption peak position of the target gas, thereby maximizing the accuracy and stability of the measurement signal. Wavelength locking technology is crucial for improving the accuracy of spectral scanning and reducing environmental interference. Especially when detecting multiple target gases, it can ensure that the absorption peaks of each gas can be accurately identified.

[0024] Specifically, based on the locked spectral data, the absorption intensities of each target gas can be calculated. The absorption intensity is proportional to the gas concentration. By analyzing the intensity of each absorption peak in the spectral data, the concentration of each target gas in the sample can be calculated. The absorption characteristics of each target gas at different wavelengths are unique. By measuring the position and intensity of these absorption peaks, accurate concentration values can be obtained. The multi-wavelength laser and precise wavelength modulation processing enable the simultaneous acquisition of absorption data of multiple gases in one scanning process, providing strong support for multi-component gas analysis. Through these absorption data, the concentration and type of dissolved gases in the sample can be effectively judged, providing accurate data basis for the health diagnosis of equipment such as transformers.

[0025] Further, the process of performing adaptive wavelength locking processing on the second harmonic signal to obtain spectral data locked at the optimal absorption peak of the target gas includes: performing peak detection processing on the second harmonic signal to obtain the initial absorption peak position; performing coarse adjustment processing on the wavelength of the multi-wavelength tunable laser according to the initial absorption peak position to obtain the coarsely adjusted laser wavelength; adjusting the driving current of the multi-wavelength tunable laser based on the laser wavelength and performing real-time measurement processing on the light intensity after passing through the gas sample to obtain absorption spectra at different wavelengths; calculating the center position of the spectrum according to the absorption spectra to obtain the laser parameters corresponding to the optimal absorption peak; and using a proportional-integral-derivative control algorithm to perform real-time adjustment processing on the driving parameters of the laser to obtain spectral data locked at the optimal absorption peak of the target gas.

[0026] Specifically, when performing adaptive wavelength locking of the second harmonic signal, peak detection of the second harmonic signal is first required. The second harmonic signal is the result of nonlinear interactions between gas molecules and laser light waves. This signal typically manifests as a frequency doubling of the fundamental frequency, exhibiting a characteristic waveform. To accurately lock onto the absorption peak of the target gas, the initial absorption peak position of the second harmonic signal must first be determined using a precise peak detection algorithm. Peak detection is based on real-time scanning of the second harmonic signal, utilizing digital signal processing algorithms to track amplitude changes within the signal. By performing a Fourier transform or sliding window method on the signal, it is possible to effectively extract portions of the signal with dramatic amplitude changes, which typically correspond to the absorption peak positions of the gas molecules. Determining this position is crucial for subsequent wavelength adjustment, as it provides a preliminary reference for laser wavelength tuning. Accurate peak detection not only determines the initial absorption peak position but also provides the necessary information for wavelength locking of the laser.

[0027] Specifically, after obtaining the initial absorption peak position, the next step is to coarsely tune the wavelength of the multi-wavelength tunable laser based on this position. Since the wavelength range of the laser output has a certain adjustment room, it is necessary to coarsely adjust the wavelength of the laser based on the initially detected absorption peak position to ensure that the output wavelength of the laser is roughly in the absorption peak region of the target gas. The coarse tuning process adjusts the laser wavelength from the initial state to an absorption peak position close to the target gas by adjusting the laser's wavelength tuning system. At this time, the laser's wavelength adjustment system can be a voltage-controlled optical element, such as a piezoelectric crystal or a liquid crystal grating. By controlling the current of these elements, the laser's wavelength can be precisely changed. In this coarse tuning stage, the laser's wavelength deviation is large, but through simple adjustments, the wavelength can be guided to near the absorption peak of the target gas, laying the foundation for subsequent fine adjustment.

[0028] Specifically, the coarsely adjusted laser wavelength provides a basis for further fine adjustment. Next, the driving current of the laser is adjusted through fine adjustment based on the laser wavelength. In this process, by precisely controlling the driving current of the laser, the wavelength of the laser is further adjusted to the optimal absorption peak of the target gas. The adjustment of the driving current directly affects the output wavelength of the laser, and thus affects the frequency of the light. By adjusting the current, the laser wavelength can be finely tuned, and the output wavelength of the laser signal can be accurately locked on the absorption peak of the target gas. At this stage, the system measures the light intensity transmitted through the gas sample in real time by performing a light transmission measurement on the gas sample, and compares it with the theoretical value to further correct the laser wavelength. The change in light intensity can reflect the interaction between the laser and gas molecules, and thus help to determine whether the current wavelength is accurately locked on the absorption peak of the target gas. At this time, through precise light intensity measurement and feedback adjustment, the accuracy of wavelength adjustment can be greatly improved, ensuring that the laser wavelength has the maximum absorption intensity at the target absorption peak.

[0029] Specifically, once the accurate wavelength value is obtained, the center position of the spectral line needs to be calculated by analyzing the absorption spectrum line next. The center position of the spectral line is determined by analyzing the shape and intensity of the absorption peak. The shape of the absorption spectrum line reflects the absorption characteristics of gas molecules to light of a specific wavelength, usually showing a symmetric or asymmetric absorption curve. To determine the center position of the spectral line, it is first necessary to measure the absorbance (A) at different wavelengths to plot the absorption spectrum line. The absorbance is proportional to the gas concentration and the laser power. The absorbance is calculated by the following formula to obtain the center position of the spectral line: ; where, is the light intensity transmitted through the gas sample, is the initial intensity of the incident light, is the light wavelength. The center position of the spectral line can be determined by curve fitting algorithms such as Gaussian fitting or Lorentz fitting. At this time, the laser wavelength corresponding to the optimal absorption peak is the wavelength of the absorption peak obtained by fitting. Finally, the proportional-integral-derivative (PID) control algorithm is used to adjust the driving parameters of the laser in real time to lock the optimal absorption peak of the target gas. The PID control algorithm is an algorithm widely used in automatic control systems. It can adjust the control parameters according to the current state error, historical error and error change rate of the system. In this solution, the PID control algorithm automatically adjusts the wavelength of the laser by calculating the difference between the real-time measured absorbance and the target value. Specifically, the PID controller dynamically adjusts the laser wavelength by performing proportional, integral and differential processing on the error, so that it remains at the optimal position at the target absorption peak. The formula is as follows: ; where, is the error value, representing the deviation between the current measured value and the target value. , , are the proportional, integral, and derivative gains respectively is the control output, used to adjust the wavelength of the laser. Through the real-time feedback and adjustment of the PID algorithm, the system can accurately lock the absorption peak of the target gas and continuously optimize the working parameters of the laser, thereby providing stable and high-precision spectral data. Finally, after the adaptive wavelength locking process, the obtained spectral data not only accurately reflects the absorption characteristics of the target gas but also provides a reliable basis for subsequent gas concentration analysis.

[0030] 103. Perform robust signal processing and outlier detection on the absorption spectral data to obtain gas concentration data and outlier markers; In one embodiment of the present invention, the performing robust signal processing and outlier detection on the absorption spectral data to obtain gas concentration data and outlier markers includes: performing adaptive digital lock-in amplification processing on the absorption spectral data to obtain the demodulated second harmonic signal; using the minimum covariance determinant method to perform iterative processing on the second harmonic signal to obtain a robust covariance estimator; according to the robust covariance estimator, performing Mahalanobis distance calculation processing on the second harmonic signal to obtain the outlier degree score of each data point; classifying and performing concentration calculation processing on the data points according to the outlier degree score and a preset threshold to obtain gas concentration data and outlier markers.

[0031] Specifically, when processing the absorption spectral data, it is first necessary to perform adaptive digital lock-in amplification processing on the data to obtain the demodulated second harmonic signal. Adaptive digital lock-in amplification technology is an important method in signal processing, which can effectively extract the target signal from a complex noise environment. In this solution, the absorption spectral data contains useful information from the absorption characteristics of the target gas but is also affected by environmental noise and instrument errors. Through lock-in amplification processing, the intensity of the second harmonic signal can be significantly enhanced, and the influence of interference signals can be reduced. The working principle of adaptive digital lock-in amplification is to demodulate the collected signal by using a reference signal (usually a sine signal with the same modulation frequency as the laser signal). During this process, the signal is first sampled and analog-to-digital converted, and then the digitized spectral signal is multiplied by the reference signal to obtain two product signals. These product signals are then processed by a low-pass filter to remove irrelevant high-frequency noise, thereby obtaining the low-frequency part containing the target information. Finally, by reconstructing these signals, a clear second harmonic signal can be obtained, enhancing the signal-to-noise ratio of the signal.

[0032] Specifically, the minimum covariance determinant (MCD) method is used to iteratively process the demodulated second harmonic signal to obtain a robust covariance estimator. The MCD method is a robust statistical method used to estimate the covariance matrix from data and identify outliers in the data. For the second harmonic signal, due to its complex source and the fact that the data may be affected by factors such as environmental changes and noise interference, traditional covariance estimation methods may be affected by unstable data, resulting in biased estimation results. The MCD method randomly selects a subset from the entire dataset in an iterative manner and calculates the mean and covariance matrix of this subset. Then, the Mahalanobis distance from each data point to this mean vector is calculated, and the sample set is updated based on these distances. This process continues until the determinant of the covariance matrix converges to a stable value. Through the MCD method, outliers that do not conform to the data distribution can be effectively removed, thereby obtaining a robust covariance estimator to ensure the stability and reliability of subsequent analysis results. The robust covariance estimator can handle noise and abnormal fluctuations in the data, making the estimation of gas concentration more accurate and avoiding the defect of being sensitive to abnormal data in traditional methods.

[0033] Specifically, based on the robust covariance estimator, the Mahalanobis distance of the second harmonic signal is calculated next to obtain the outlier degree score of each data point. The Mahalanobis distance is a method for measuring the difference between a sample point and the mean of a dataset. It takes into account the covariance structure of the data and is therefore more suitable for processing multi-dimensional data with correlations than the Euclidean distance. In this process, each data point is first standardized using the covariance matrix, and then the Mahalanobis distance from each data point to the mean is calculated. This distance value represents the degree of deviation of the data point from the overall data distribution. By calculating the Mahalanobis distances of all data points, the outlier degree score of each data point can be obtained. The higher the score, the greater the difference between the data point and the normal dataset, and the more likely it is to be abnormal data. This process can effectively identify outliers and provide a basis for subsequent classification and concentration calculation. The use of the Mahalanobis distance avoids the over-simplification of the data distribution in traditional methods and makes outlier detection more accurate.

[0034] Specifically, according to the abnormality degree score of each data point and a preset threshold, the data points are classified and concentration calculation processing is performed to obtain gas concentration data and outlier marks. In this step, first, a threshold is set to distinguish normal data from abnormal data. Those points with a Mahalanobis distance score higher than the threshold are marked as outliers, and the existence of these data points may be caused by instrument failures, environmental changes, sampling errors, etc. The normal data is used for further concentration calculation. For the normal data, using their absorption intensities and known gas absorption characteristics, the concentration of the gas can be calculated. This process usually relies on the previous analysis of absorption spectrum data. By establishing a mathematical relationship between gas concentration and absorption intensity, the concentration of each target gas in the gas sample is calculated. In concentration calculation, an algorithm based on the Lambert-Beer law is usually used. By measuring the absorbance at different wavelengths and the absorption characteristics of known gases, the gas concentration is deduced. Finally, the obtained gas concentration data will serve as an important basis for transformer operation status monitoring and fault warning. In this way, not only can the gas concentration be accurately calculated, but also abnormal data can be effectively identified and excluded, ensuring the high precision and reliability of the final monitoring results.

[0035] Further, the process of performing adaptive digital lock-in amplification processing on the absorption spectrum data to obtain the demodulated second harmonic signal includes: performing analog-to-digital conversion processing on the absorption spectrum data to obtain a digitized spectral signal, and generating digital sine and cosine reference signals with the same modulation frequency; respectively performing digital multiplication processing on the digitized spectral signal and the digital sine and cosine reference signals to obtain two product signals; respectively performing low-pass filtering processing on the two product signals to obtain two filtered signals, and using the two filtered signals as the in-phase component and the quadrature component respectively; calculating the amplitude and phase of the signal according to the in-phase component and the quadrature component, and reconstructing the second harmonic signal according to the amplitude and phase to obtain the demodulated second harmonic signal.

[0036] Specifically, when performing adaptive digital lock-in amplification processing on absorption spectral data, first, the absorption spectral data is subjected to analog-to-digital conversion processing to convert the analog signal into a digital signal. This process is achieved through an analog-to-digital converter (ADC), with the aim of converting the analog signal into discrete digital values. The analog signal is usually a continuous wave signal generated by the interaction between the laser and the gas sample, and the analog-to-digital converter converts these signals into digital signals at a certain sampling rate, thereby providing data support for subsequent digital signal processing. The digital signal after analog-to-digital conversion contains the frequency and amplitude information of the absorption spectrum, and the changes therein reflect the concentration and absorption characteristics of the gas. However, the digitized signal may still contain noise and interference. Therefore, before signal demodulation, it is necessary to further enhance the intensity of the useful signal. To ensure that the subsequent processing can effectively extract the target signal, digital sine and cosine reference signals with the same modulation frequency need to be generated. The frequencies of these reference signals are consistent with the laser modulation frequency and can be synchronized with the digitized spectral signal. By generating these reference signals, they can be precisely paired with the spectral signal for subsequent demodulation processing.

[0037] Specifically, the key to performing digital multiplication processing on the digitized spectral signal with the digital sine and cosine reference signals lies in performing point-by-point multiplication of the two signals, with the aim of matching the frequency characteristics of the modulation signal and the reference signal and extracting the useful signal components. During the digital multiplication processing, multiplying the digital spectral signal by the sine reference signal will generate a product signal containing the useful signal and frequency-related components. Similarly, multiplying the digital spectral signal by the cosine reference signal will also generate another product signal. These two product signals respectively reflect the components of the signal in the sine and cosine directions, and can enhance the absorption information of the target gas and suppress the noise components irrelevant to the laser modulation frequency. This processing method can not only extract the frequency components of the signal but also effectively remove the irrelevant high-frequency noise, thereby improving the accuracy and stability of subsequent signal processing. By separately extracting the two product signals, it lays the foundation for subsequent filtering processing.

[0038] Specifically, low-pass filtering is performed on these two product signals respectively to obtain the two filtered signals. The role of the low-pass filter is to remove the high-frequency components in the product signal and retain the low-frequency components. Since the second harmonic signal itself is a low-frequency signal, while the high-frequency signals mainly come from noise and other interference sources, low-pass filtering can effectively remove these unwanted high-frequency components and retain the key low-frequency information of the target signal. The low-pass filtering process uses a filter, and its design needs to ensure that it can filter out the high-frequency noise in the signal while retaining the low-frequency components in the target signal. The two low-pass filtered signals are used as the in-phase component and the quadrature component for subsequent processing. In this way, the filtered signal can clearly reflect the basic characteristics of the target signal, suppress the interference of external noise, and ensure the high quality of the signal.

[0039] Specifically, the amplitude and phase of the signal are calculated based on the in-phase component and the quadrature component. Amplitude and phase are two important parameters of the signal, which respectively reflect the intensity and phase difference of the signal. In this process, first, the in-phase component and the quadrature component need to be processed. Through mathematical tools such as Fourier transform, the complex form of the signal is calculated, that is, the complex amplitude and phase. The complex amplitude represents the intensity of the signal, while the complex phase describes the time characteristics of the signal. By calculating these parameters, the absorption characteristics of the target gas can be accurately captured, and thus a reliable basis can be provided for subsequent signal reconstruction and gas concentration calculation. The amplitude reflects the light absorption intensity of gas molecules, while the phase can provide time information to help analyze the change pattern of the gas under the laser spectrum. These calculation results directly affect the subsequent signal reconstruction process, ensuring that the demodulated signal can accurately reflect the absorption characteristics of the gas. Finally, based on the amplitude and phase calculated from the in-phase component and the quadrature component, the second harmonic signal is reconstructed to obtain the demodulated second harmonic signal. By combining the changes in amplitude and phase and using the form of complex numbers, the signal can be accurately reconstructed. The signal reconstruction can not only restore the absorption characteristics of the target gas but also enhance the clarity of the target signal on the basis of removing noise and interference. At this time, the demodulated second harmonic signal can be used for subsequent gas concentration calculation and fault diagnosis to ensure the accuracy and reliability of the spectral data. Through this process, the demodulation of the second harmonic signal can effectively improve the quality of the signal, avoid the influence brought by noise and environmental factors, and provide more stable and accurate results for gas sample analysis.

[0040] Further, the iterative processing of the second harmonic signal using the minimum covariance determinant method to obtain a robust covariance estimator includes: randomly selecting h sample points from the second harmonic signal, calculating an initial mean vector and an initial covariance matrix, and calculating the determinant of the initial covariance matrix as an initial determinant value; calculating the Mahalanobis distance of all data points using the initial mean vector and the initial covariance matrix, and selecting the h points with the smallest distance as a new sample set; recalculating the mean vector and the covariance matrix for the new sample set to obtain an updated mean vector and an updated covariance matrix; calculating the determinant of the updated covariance matrix and comparing it with the initial determinant value to obtain a convergence judgment result; if the convergence judgment result is non-convergent, then using the updated mean vector and the updated covariance matrix as new initial mean vector and initial covariance matrix for iteration, and if convergent, outputting the updated mean vector and the updated covariance matrix as the robust covariance estimator.

[0041] Specifically, when implementing the minimum covariance determinant (MCD) method to process the second harmonic signal, it is first necessary to randomly select h sample points from the collected second harmonic signal for initial calculation. This process is the starting step of the MCD method, and its purpose is to avoid the influence of outliers on the data analysis process by randomly selecting sample points. The h randomly selected sample points should be able to represent the characteristics of the entire data set. Therefore, the number of selected points needs to be appropriate, which can not only cover the main changes of the signal but also prevent the interference of extreme points. In practical applications, these sample points will be used to calculate the initial mean vector and the covariance matrix. The initial mean vector is the average of all sample points, which reflects the central tendency of the signal. The covariance matrix describes the correlation and variation degree of the signal. For the second harmonic signal, the covariance matrix reflects the dependence relationship between variables in the signal. The calculated value of the determinant of the covariance matrix provides a quantitative measure of the scale of the matrix, which can reflect the distribution range of the data points. Through the determinant value of the initial covariance matrix, a reference benchmark can be provided for subsequent iterative processing to determine whether there are abnormalities or deviations in the covariance matrix.

[0042] Next, using the initial mean vector and covariance matrix, calculate the Mahalanobis distance for all data points. The Mahalanobis distance is a distance metric in multi-dimensional space that, by considering the covariance relationships in the dataset, can more accurately measure the difference between each data point and the mean. The Mahalanobis distance takes into account the distribution structure of the data points, avoiding the defect of the Euclidean distance that ignores covariance. In this step, by calculating the Mahalanobis distance for all data points, data points with significant deviations from the overall data distribution can be identified, and these points are usually considered potential outliers. Based on the calculated Mahalanobis distance, the h points with the smallest distance from the mean can be selected as the new sample set for subsequent processing. This step effectively reduces the impact of outliers on the overall analysis result by removing data points with large distances. After selecting the new sample set, use these points to recalculate the mean vector and covariance matrix. The new mean vector and covariance matrix will reflect the characteristics of the dataset after removing outliers, and these new statistics provide a more accurate data basis for further analysis. The new covariance matrix will more precisely reflect the true distribution of the signal, while the mean vector can better represent the central tendency of the data. This process is equivalent to optimizing the preliminary statistical estimation and gradually adjusting the selection of the sample set and the accuracy of data analysis. When calculating the updated mean vector and covariance matrix, the Mahalanobis distance is still used as the measure of how much a data point deviates from the mean, which ensures the robustness of the results. Subsequently, it is necessary to calculate the determinant of the updated covariance matrix and compare it with the initial determinant value. The purpose of this comparison process is to judge the stability of the dataset and the convergence of the covariance matrix. If the difference between the updated determinant and the initial value is small, it indicates that the sample set has converged and the data distribution tends to be stable, and the iteration can be stopped. If the difference is large, it means that the current estimated value of the covariance matrix is not yet stable and further iteration is required. During this process, the formula for calculating the determinant is: ; where, is the eigenvalue of the covariance matrix. The smaller the determinant value, the more concentrated the distribution of data points in that direction; the larger the determinant value, the more dispersed the data is. By comparing the initial determinant and the updated determinant, it can be determined whether the current estimate is accurate and further processing can be decided. If the updated mean vector and covariance matrix have not converged, the new mean vector and covariance matrix are used as the new initial values and the iteration continues. Each iteration updates the mean and covariance matrix based on the new sample set and new statistics until the determinant converges. This iterative processing method ensures that the final estimated value accurately reflects the actual data distribution by gradually optimizing the covariance estimate. Once the convergence judgment shows that the updates of the covariance matrix and mean vector have stabilized, the final robust covariance estimator can be output. This estimator represents the result of multiple iterations and optimizations, which can effectively resist the influence of outliers and noise and provide accurate statistical analysis. In practical applications, this robust covariance estimator can be used for further outlier detection and gas concentration calculation to ensure high precision and high reliability of signal analysis.

[0043] 104. Based on the gas concentration data and the outlier markers, perform intelligent diagnosis and early warning analysis on the operating status of the power equipment to obtain the monitoring results of the dissolved gas.

[0044] In an embodiment of the present invention, the performing intelligent diagnosis and early warning analysis on the operating status of the power equipment based on the gas concentration data and the outlier markers to obtain the monitoring results of the dissolved gas includes: using a pre-trained deep learning model to perform fault mode recognition processing on the gas concentration data to obtain a preliminary fault diagnosis result; inputting the preliminary fault diagnosis result into a preset fuzzy logic multi-expert system and performing comprehensive analysis processing in combination with the outlier markers to obtain the current fault diagnosis result; performing time series analysis processing on the gas concentration data to obtain the gas concentration change trend, and based on the gas concentration change trend and the current fault diagnosis result, using a pre-trained prediction model to perform analysis processing to obtain a potential fault prediction result; performing comprehensive evaluation processing on the current fault diagnosis result and the potential fault prediction result to obtain the monitoring results of the dissolved gas.

[0045] Specifically, when performing intelligent diagnosis and early warning analysis on the operating status of power equipment, it is first necessary to use a pre-trained deep learning model to perform fault mode recognition processing on gas concentration data. Deep learning models, especially convolutional neural networks (CNNs) and long short-term memory networks (LSTMs), have powerful feature extraction and sequence modeling capabilities and are very suitable for processing dissolved gas concentration data from power equipment. These concentration data not only contain the instantaneous concentration values of the gases but may also contain non-linear relationships and time dependencies related to equipment failures. Through training on a large amount of historical fault data, the deep learning model can automatically identify potential fault modes and abnormal signals. For example, when overload or partial discharge occurs inside a transformer, the concentration of dissolved gases in the oil will exhibit specific patterns. By training a deep learning network, the model can identify these patterns and gradually adjust the network weights through error backpropagation so that the final output can accurately predict different fault types. The objective function during the training process usually combines a loss function, such as mean squared error (MSE) or cross-entropy loss function, and is optimized according to the deviation between the actual prediction result and the true fault type. After completing the training of the deep learning model, when real-time gas concentration data is input, the model can output preliminary fault diagnosis results, which provide a basic judgment of the fault category for subsequent comprehensive analysis.

[0046] Specifically, the preliminary fault diagnosis results are input into a preset fuzzy logic multi-expert system for comprehensive analysis and processing. The fuzzy logic system can perform reasoning and decision-making under uncertain conditions by converting expert experience and theoretical models into fuzzy rules. In this solution, the fuzzy logic system uses the knowledge input by multiple experts to comprehensively evaluate the faults and takes into account the outlier marks in the gas concentration data. The outlier marks are provided by the previous processing process, and these marks can help the system identify which data points may not conform to the normal gas concentration change pattern, further improving the accuracy of the diagnosis. The fuzzy system will perform reasoning through a series of fuzzy rules based on the input fault mode, concentration change, and outlier marks to obtain a comprehensive fault diagnosis result. For example, if the gas concentration suddenly increases and the outlier marks indicate that the signal may be affected by external interference, the fuzzy system will output a high-confidence early warning signal, indicating that there may be a problem with the state of the transformer. The core of this process is to perform weighted processing on various types of information through fuzzy reasoning to obtain a comprehensive fault diagnosis result, which will be used as an important reference in subsequent analysis.

[0047] Specifically, after obtaining the preliminary fault diagnosis results, it is necessary to perform time series analysis on the gas concentration data to further explore its changing trend. Time series analysis techniques can extract potential trends and periodic characteristics from the historical changes of gas concentration data, so as to predict future changing trends. In this step, sequence analysis methods such as autoregressive integrated moving average (ARIMA) model or long short-term memory network (LSTM) can be used to identify the changing patterns of gas concentration in different time periods. For example, the LSTM network can capture the regular fluctuations of gas concentration within a certain time range by memorizing the long-term dependence of historical data, and predict future concentration changing trends. This is crucial for early fault detection because some fault modes gradually evolve over a long time, and the LSTM model can capture these changes and issue warning signals. Through time series analysis of gas concentration data, the system can monitor the changes of gas concentration in real time and predict its future changing direction. The prediction results usually show whether the gas concentration will further increase or return to the normal level, which provides strong evidence for judging whether the equipment is in a potential fault state.

[0048] Specifically, based on the changing trend of gas concentration and the current fault diagnosis results, use a pre-trained prediction model to conduct further potential fault prediction analysis. The core of this stage is to combine known fault modes and time series prediction results to predict the future health status of the equipment. The prediction model may include support vector machine (SVM) or deep neural network (DNN). These models can predict whether the equipment has the risk of major faults by learning historical fault data and equipment operation parameters. By combining the future trend of gas concentration with the current fault diagnosis results, the prediction model can judge whether the equipment is in the early stage of a fault or has reached the critical point of a fault. To make predictions, the system usually needs a dynamic update mechanism to receive new gas concentration data in real time and re-evaluate potential faults based on this. Based on this, the system will output the prediction results of potential faults, such as predicting that the transformer will experience overload or partial discharge within the next few hours. Finally, through comprehensive consideration of the current fault diagnosis results and potential fault prediction results, a comprehensive assessment of the equipment's operating status is carried out. This comprehensive assessment process will output a monitoring result of dissolved gas, providing specific fault diagnosis information and warning signals for maintenance personnel. This monitoring result not only provides the current fault type of the equipment but also predicts the possible future fault risks, thus helping decision-makers formulate more effective maintenance and repair plans. In this process, deep learning models, fuzzy logic systems, time series analysis, and prediction models cooperate with each other to jointly achieve intelligent monitoring and warning of the state of power equipment.

[0049] In this embodiment, by performing dynamic negative pressure constant temperature extraction on the insulating oil in the power equipment, a gas sample containing dissolved gas is obtained; a multi-wavelength tunable laser is used to perform spectral scanning on the gas sample to obtain absorption spectral data of multi-component gas; robust signal processing and outlier detection are performed on the absorption spectral data to obtain gas concentration data and outlier marks; according to the gas concentration data and outlier marks, intelligent diagnosis and early warning analysis are performed on the operating state of the power equipment to obtain the monitoring result of the dissolved gas. By combining laser spectroscopy technology and robust data processing, the present invention can identify and eliminate outliers in the data through robust signal processing and outlier detection, improve the reliability of the data, provide high-precision and high-robustness data support for subsequent intelligent diagnosis and early warning, and improve the accuracy of fault diagnosis and early warning.

[0050] The online monitoring method for dissolved gas in power equipment oil in the embodiments of the present invention has been described above. Next, the online monitoring device for dissolved gas in power equipment oil in the embodiments of the present invention will be described. Please refer to Figure 2 , an embodiment of the online monitoring device for dissolved gas in power equipment oil in the embodiments of the present invention includes: A dynamic negative pressure constant temperature extraction module 201, configured to perform dynamic negative pressure constant temperature extraction on the insulating oil in the power equipment to obtain a gas sample containing dissolved gas; A multi-wavelength spectral scanning module 202, configured to use a multi-wavelength tunable laser to perform spectral scanning on the gas sample to obtain absorption spectral data of multi-component gas; A robust signal processing module 203, configured to perform robust signal processing and outlier detection on the absorption spectral data to obtain gas concentration data and outlier marks; An intelligent diagnosis and early warning module 204, configured to perform intelligent diagnosis and early warning analysis on the operating state of the power equipment according to the gas concentration data and the outlier marks to obtain the monitoring result of the dissolved gas.

[0051] In the embodiments of the present invention, the on-line monitoring device for dissolved gases in the insulating oil of the power equipment operates the on-line monitoring method for dissolved gases in the insulating oil of the power equipment. The on-line monitoring device for dissolved gases in the insulating oil of the power equipment obtains a gas sample containing dissolved gases through dynamic negative-pressure constant-temperature extraction of the insulating oil in the power equipment; performs spectral scanning on the gas sample by using a multi-wavelength tunable laser to obtain absorption spectral data of multi-component gases; performs robust signal processing and outlier detection on the absorption spectral data to obtain gas concentration data and outlier marks; and performs intelligent diagnosis and early warning analysis on the operating state of the power equipment according to the gas concentration data and outlier marks to obtain the monitoring result of the dissolved gases. By combining laser spectroscopy technology and robust data processing, the present invention can identify and eliminate outliers in the data through robust signal processing and outlier detection, improve the reliability of the data, provide high-precision and high-robustness data support for subsequent intelligent diagnosis and early warning, and improve the accuracy of fault diagnosis and early warning.

[0052] Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the above-described systems, devices, or units can refer to the corresponding processes in the foregoing method embodiments and will not be described herein again.

[0053] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the embodiments of the present invention. The foregoing storage medium includes various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs.

[0054] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An on-line monitoring method for dissolved gases in the oil of power equipment, characterized in that, The on-line monitoring method for dissolved gases in the insulating oil of power equipment includes: Performing dynamic negative-pressure constant-temperature extraction on the insulating oil in the power equipment to obtain a gas sample containing dissolved gases; Performing spectral scanning on the gas sample using a multi-wavelength tunable laser to obtain absorption spectral data of multi-component gases; Performing robust signal processing and outlier detection on the absorption spectral data to obtain gas concentration data and outlier marks; According to the gas concentration data and the outlier marks, performing intelligent diagnosis and early warning analysis on the operating state of the power equipment to obtain the monitoring result of the dissolved gases.

2. The on-line monitoring method for dissolved gases in oil of power equipment according to claim 1, characterized in that, The performing spectral scanning on the gas sample using a multi-wavelength tunable laser to obtain absorption spectral data of multi-component gases includes: Selecting lasers with multiple different wavelengths according to the types of preset target gases, and performing wavelength modulation on the gas sample to obtain a modulated optical signal; Performing second harmonic detection on the modulated optical signal to obtain a second harmonic signal; Performing adaptive wavelength locking on the second harmonic signal to obtain spectral data locked at the best absorption peak of the target gas; Calculating the absorption intensity of each target gas according to the locked spectral data to obtain absorption spectral data of multi-component gases.

3. The on-line monitoring method for dissolved gases in oil of electrical equipment according to claim 2, characterized in that, The performing adaptive wavelength locking on the second harmonic signal to obtain spectral data locked at the best absorption peak of the target gas includes: Performing peak detection on the second harmonic signal to obtain the initial absorption peak position; Performing coarse adjustment on the wavelength of the multi-wavelength tunable laser according to the initial absorption peak position to obtain the coarsely adjusted laser wavelength; Adjusting the driving current of the multi-wavelength tunable laser based on the laser wavelength, and performing real-time measurement on the light intensity after passing through the gas sample to obtain absorption spectral lines at different wavelengths; Calculating the center position of the spectral line according to the absorption spectral line to obtain the laser parameters corresponding to the best absorption peak; Using a proportional-integral-derivative control algorithm to perform real-time adjustment on the driving parameters of the laser to obtain spectral data locked at the best absorption peak of the target gas.

4. The on-line monitoring method for dissolved gases in oil of electrical equipment according to claim 1, characterized in that, The performing robust signal processing and outlier detection on the absorption spectral data to obtain gas concentration data and outlier marks includes: Performing adaptive digital lock-in amplification on the absorption spectral data to obtain a demodulated second harmonic signal; Performing iterative processing on the second harmonic signal using the minimum covariance determinant method to obtain a robust covariance estimator; Calculating the Mahalanobis distance of the second harmonic signal according to the robust covariance estimator to obtain the outlier degree score of each data point; Classifying and calculating the concentration of data points according to the outlier degree score and a preset threshold to obtain gas concentration data and outlier marks.

5. The on-line monitoring method for dissolved gases in oil of electrical equipment according to claim 4, characterized in that, The performing adaptive digital lock-in amplification on the absorption spectral data to obtain a demodulated second harmonic signal includes: Performing analog-to-digital conversion on the absorption spectral data to obtain a digitized spectral signal, and generating digital sine and cosine reference signals with the same modulation frequency; The digitized spectral signals are respectively subjected to digital multiplication processing with the digital sine and cosine reference signals to obtain two product signals; The two product signals are respectively subjected to low-pass filtering processing to obtain two filtered signals, and the two filtered signals are respectively used as the in-phase component and the quadrature component; The amplitude and phase of the signal are calculated according to the in-phase component and the quadrature component, and the second harmonic signal is reconstructed according to the amplitude and phase to obtain the demodulated second harmonic signal.

6. The on-line monitoring method for dissolved gases in oil of electrical equipment according to claim 4, characterized in that, The iterative processing of the second harmonic signal by using the minimum covariance determinant method to obtain a robust covariance estimator includes: Randomly select h sample points from the second harmonic signal, calculate the initial mean vector and the initial covariance matrix, and calculate the determinant of the initial covariance matrix as the initial determinant value; Calculate the Mahalanobis distance of all data points by using the initial mean vector and the initial covariance matrix, and select the h points with the smallest distance as the new sample set; Recalculate the mean vector and the covariance matrix for the new sample set to obtain the updated mean vector and covariance matrix; Calculate the determinant of the updated covariance matrix and compare it with the initial determinant value to obtain a convergence judgment result; If the convergence judgment result does not converge, the updated mean vector and covariance matrix are used as the new initial mean vector and initial covariance matrix for iteration. If it converges, the updated mean vector and covariance matrix are output as the robust covariance estimator.

7. The on-line monitoring method for dissolved gases in oil of electrical equipment according to claim 1, characterized in that, The intelligent diagnosis and early warning analysis of the operating state of the power equipment according to the gas concentration data and the outlier mark to obtain the monitoring result of the dissolved gas includes: Use a pre-trained deep learning model to perform fault mode recognition processing on the gas concentration data to obtain a preliminary fault diagnosis result; Input the preliminary fault diagnosis result into a preset fuzzy logic multi-expert system, and perform comprehensive analysis processing in combination with the outlier mark to obtain the current fault diagnosis result; Perform time series analysis processing on the gas concentration data to obtain the gas concentration change trend, and perform analysis processing by using a pre-trained prediction model according to the gas concentration change trend and the current fault diagnosis result to obtain a potential fault prediction result; Perform comprehensive evaluation processing on the current fault diagnosis result and the potential fault prediction result to obtain the monitoring result of the dissolved gas.

8. An on-line monitoring device for dissolved gases in the oil of a power equipment, characterized in that, The on-line monitoring device for dissolved gases in the oil of the power equipment includes: A dynamic negative pressure constant temperature extraction module for performing dynamic negative pressure constant temperature extraction processing on the insulating oil in the power equipment to obtain a gas sample containing dissolved gases; A multi-wavelength spectral scanning module for performing spectral scanning processing on the gas sample by using a multi-wavelength tunable laser to obtain absorption spectral data of multi-component gases; A robust signal processing module for performing robust signal processing and outlier detection on the absorption spectral data to obtain gas concentration data and outlier marks; An intelligent diagnosis and early warning module, which is used to perform intelligent diagnosis and early warning analysis on the operating state of the power equipment according to the gas concentration data and the outlier marks, so as to obtain the monitoring results of the dissolved gas.

Images