Transformer oil metal impurity intelligent identification method based on capillary electrophoresis spectrogram

Abstract

The invention discloses a transformer oil metal impurity intelligent identification method based on a capillary electrophoresis spectrogram, and relates to the technical field of impurity detection and identification. According to the invention, a detection environment based on improved miniaturized CE-C4D is constructed and signal data is collected; performing multi-stage preprocessing on the signal to obtain a pure spectrogram; performing elastic matching and time axis nonlinear correction on the pure spectrogram and a standard fingerprint spectrogram library by using a DTW algorithm; extracting a peak area based on the aligned spectrogram and carrying out inversion calculation on the absolute concentration of each metal element; evaluating the confidence coefficient of the concentration value, and if the concentration value does not reach the standard, adaptively optimizing parameters and reprocessing; constructing a fault fingerprint feature vector based on the final concentration value, and inputting a diagnosis model to output fault type and severity evaluation; according to the method, identification errors caused by electrophoresis migration time drift are overcome, high-sensitivity and accurate quantification of trace metal and intelligent fault diagnosis are achieved, and a reliable solution is provided for on-site monitoring of the state of the transformer.

Description

Technical Field

[0001] This invention belongs to the field of impurity detection and identification technology, and specifically relates to an intelligent identification method for metallic impurities in transformer oil based on capillary electrophoresis spectroscopy. Background Technology

[0002] Metallic impurities (such as copper, iron, and zinc) in transformer oil are important indicators reflecting the internal insulation condition and wear of transformers. Existing detection methods mainly rely on inductively coupled plasma atomic emission spectrometry (ICP-AES). Although this method is accurate, the equipment is expensive, bulky, and complex to operate, and it depends on offline analysis by professional personnel, making it difficult to meet the needs of rapid on-site diagnosis and maintenance.

[0003] In search of on-site solutions, the capillary electrophoresis-capacitive coupling non-contact conductivity detector (CE-C4D) has attracted attention due to its potential cost and size advantages. However, existing CE-C4D technology faces the following problems when dealing with complex transformer oil sample testing: 1. Insufficient basic detection performance: conventional designs have low signal-to-noise ratios for trace metal detection and weak anti-interference capabilities against complex oil sample matrices; 2. Defects in the core algorithm: electrophoretic migration time drifts with environmental conditions, while existing identification methods mostly use fixed "time windows" for matching. Once the peak position shifts, it leads to misjudgment or missed detection, resulting in poor reliability; 3. Lack of intelligence: the analysis process stops at providing concentration values ​​and fails to intelligently correlate multi-metal concentration spectra with the internal fault mechanisms of transformers, making it impossible to form conclusions that can directly guide operation and maintenance.

[0004] Therefore, the problems of existing technologies can be summarized as follows: 1. At the hardware level, there is a lack of a miniaturized detection device that combines high sensitivity, strong anti-interference capability, and suitability for field use; 2. At the algorithm level, there is a lack of robust identification technology that can effectively overcome time drift; 3. At the intelligence level, there is a lack of an end-to-end intelligent analysis system from raw data to fault diagnosis. This hinders the effective application of CE-C4D technology in the field of transformer condition monitoring. Summary of the Invention

[0005] (a) Technical problems to be solved To address the problems in related technologies, this invention provides an intelligent identification method for metallic impurities in transformer oil based on capillary electrophoresis spectroscopy, thereby overcoming the aforementioned technical problems existing in the prior art.

[0006] (II) Technical Solution To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution: This invention provides a method for intelligent identification of metallic impurities in transformer oil based on capillary electrophoresis spectroscopy, comprising the following steps: S1. Construct a detection environment based on an improved miniaturized capillary electrophoresis-capacitive coupling non-contact conductivity detector; based on the detection environment, collect the time-domain conductivity signal data and environmental context data of the current transformer oil sample to be tested; S2. The time-domain conductance signal data and environmental context data collected in S1 are preprocessed through multi-level digital signal processing and baseline correction processing to obtain clean electrophoresis spectrum data. S3. Construct a standard metal ion fingerprint spectrum library. Using the DTW algorithm, calculate the optimal normalization path between the pure electrophoresis spectrum data and the standard metal ion fingerprint spectrum library. Perform time axis nonlinear correction and characteristic peak alignment to obtain the aligned standard time domain spectrum. S4. Based on the aligned standard time-domain spectrum, the peak area characteristics of each metal ion are extracted by an adaptive integral window algorithm, and the absolute concentration values ​​of each metal impurity element are calculated by inversion using a preset standard curve equation. S5. Calculate whether the confidence level of the absolute concentration value in S4 meets the standard. If it does, use the absolute concentration value in S4 as the final concentration value. Otherwise, adjust the preprocessing parameters in S2 or the parameters of the DTW algorithm in S3, and repeat S2 to S4 until the confidence level meets the standard. S6. Construct a multidimensional fault fingerprint feature vector based on the final concentration value and input it into the transformer insulation condition diagnosis model to output the evaluation results of the potential fault types and severity inside the transformer. Preferably, step S1 includes the following steps: S11. Prepare an electrophoresis buffer solution system; the buffer solution system includes L-histidine at a concentration of 0.1-10 mM, lactic acid at a concentration of 0.1-10 mM, and 18-crown-6 ether at a concentration of 0.1-10 mM, and adjust the pH value to between 2 and 6 using acetic acid to obtain the buffer solution to be used; S12. Configure an improved miniaturized CE-C4D device; the device has a built-in 0-20 MHz function generator, a 16-bit high-precision data acquisition card and a step-up transformer; the step-up transformer is used to control the detection channel and the reference channel to maintain a precise 180° phase difference; S13. The transformer oil sample to be tested is subjected to high-temperature ashing or microwave digestion to convert it into an aqueous metal ion solution suitable for electrophoresis. S14. Fill the buffer solution prepared in S11 into the capillary of the CE-C4D device configured in S12, and introduce the aqueous metal ion solution obtained in S13 for separation; record the conductivity changes during the separation process through the acquisition card to form the original time-domain conductivity signal data, and record the ambient temperature and capillary column temperature during the detection as environmental context data. Preferably, step S2 includes the following steps: S21. Select Symlets wavelet as the basis function, decompose the original time-domain conductance signal data and environmental context data output by S1 into several layers; extract the high-frequency detail coefficients of each layer and perform soft thresholding to reconstruct the signal to obtain the denoised conductance signal; S22. Using an asymmetric least squares smoothing algorithm, the denoised conductance signal obtained in S21 is used as input. The background drift trend of the signal is calculated by iterative weighted least squares method to obtain the fitting baseline. S23. Subtract the fitted baseline obtained in S22 from the denoised conductivity signal obtained in S21 to eliminate the influence of background drift and obtain pure electrophoresis spectrum data with the baseline zero. Preferably, step S22 includes the following steps: S221. Construct a first objective function; the first objective function includes the sum of squares of the deviations between the denoised conductance signal output in S21 and the baseline to be determined, as well as a smoothness penalty term for the baseline to be determined; S222. Initialize the weight vector of the first objective function and set the smoothing coefficient; S223. Solve for the baseline vector by minimizing the first objective function, and update the weight vector according to the relationship between the current baseline vector and the denoised conductance signal. S224. Repeat S223 until the weight vector converges, and output the final fitted baseline. Preferably, step S3 includes the following steps: S31. Under standard experimental conditions, standard solutions and mixed solutions of metal ions were measured separately. Standard spectral sequences containing standard retention time, peak shape information, and standard relative position information were obtained as reference templates to obtain a standard metal ion fingerprint spectral library. S32. Define the pure electrophoresis spectrum data obtained in S2 as the test sequence, and define the reference template in the standard metal ion fingerprint spectrum library obtained in S31 as the reference sequence. S33. Based on the test sequence and the reference sequence, calculate the Euclidean distance between each point in the test sequence and each point in the reference sequence, and construct a cumulative distance matrix; the row length of the cumulative distance matrix is ​​the length of the test sequence, and the column length is the length of the reference sequence; S34. Based on the cumulative distance matrix, use the DTW algorithm to find an optimal regular path in the cumulative distance matrix constructed in S33 that minimizes the total cumulative distance. S35. Based on the mapping relationship of the optimal regularized path obtained in S34, the time axis of the test sequence is nonlinearly stretched or compressed to make its characteristic peak position strictly aligned with the reference sequence, so as to obtain the aligned standard time domain spectrum. Preferably, step S34 includes the following steps: S341. Based on the environmental context data obtained in S14, set the constraints for the DTW algorithm. S342. Calculate the value of each element in the cumulative distance matrix using the recursive formula; S343. Backtrack from the end of the cumulative distance matrix to the starting point to obtain the optimal regularized path; Preferably, step S4 includes the following steps: S41. In the standard time-domain spectrum obtained after alignment in S3, the integration interval of each metal ion is locked according to the preset position of the standard metal ion fingerprint spectrum library. S42. Perform numerical integration on the signal within the integration interval determined in S41 to obtain the response peak area of ​​each metal ion. S43. Obtain the linear regression equations for the concentration and peak area of ​​each element in advance; S44. Substitute the response peak area obtained in S42 into the linear regression equation in S43 to calculate the absolute concentration of elements in the transformer oil.

[0007] Preferably, step S5 includes the following steps: S51. Extract the total cumulative distance of the optimal regularized path obtained in S34 and the signal-to-noise ratio when calculating the peak area in S42. Calculate the confidence score of the absolute concentration values ​​of elements in transformer oil obtained in S44. S52. Compare the confidence score obtained in S51 with the preset threshold; if the confidence score is greater than or equal to the confidence threshold, use the absolute concentration value in S4 as the final concentration value; if the confidence score is less than the confidence threshold, trigger the adaptive parameter optimization mechanism to maximize the cross-correlation coefficient of the spectral alignment and construct a second objective function. S53. Based on the second objective function, the particle swarm optimization algorithm is used to search for the combination of wavelet transform and asymmetric least squares smoothing algorithm parameters in S2 and window parameters of DTW algorithm in S3 within the preset search space to obtain the optimal parameter combination. S54. Feed back the parameters in the optimal parameter combination to S2 and S3, and repeat S2-S5 until the confidence level is met to obtain the final concentration value. Preferably, S53 includes the following steps: S531. Construct a particle set; based on the search space and particle set of wavelet transform, the parameters of the asymmetric least squares smoothing algorithm and the window parameters of the DTW algorithm in S3, construct an initial particle position set; use the position of each particle in the initial particle position set as a combination of the parameters of wavelet transform, the asymmetric least squares smoothing algorithm and the window parameters of the DTW algorithm. S532. In each iteration, the fitness value of the position of each particle in the particle set is calculated according to the second objective function, the fitness of each particle in the particle set is updated, and the best individual particle and the global best particle are obtained in each iteration. S533, repeat S532, stop iterating when the maximum number of iterations is reached, and use the globally optimal particle as the optimal parameter combination; Preferably, step S6 includes the following steps: S61. Extract the absolute concentration values ​​of each element from the final concentration values ​​in S5, and construct a fault fingerprint feature vector containing the absolute concentration and the concentration ratio between elements. S62. Pre-construct a transformer insulation condition diagnostic model; input the fault fingerprint feature vector constructed in S61 into the transformer insulation condition diagnostic model, and output the assessment results of fault type and severity.

[0008] (III) Beneficial Effects The present invention has the following beneficial effects: At the hardware level, this invention constructs an improved miniaturized CE-C4D device based on a specific buffer solution system and precise phase difference technology. This effectively suppresses background noise and matrix interference, significantly improving the signal-to-noise ratio and sensitivity for detecting trace metal ions in complex transformer oil samples. While ensuring high performance, the device achieves miniaturization and cost control, overcoming the bottleneck of traditional large laboratory instruments (such as ICP-AES) being difficult to deploy on-site. This provides a reliable hardware foundation for rapid and accurate on-site detection of transformer oil.

[0009] At the algorithm level, this invention innovatively introduces the Dynamic Time Warping (DTW) algorithm to perform elastic matching and nonlinear correction of electrophoresis spectra, completely abandoning the fixed window identification method which is susceptible to time drift, and ensuring accurate and robust identification of the characteristic peaks of each metal ion under different operating conditions. Furthermore, by constructing a confidence evaluation mechanism that integrates spectrum alignment quality and signal-to-noise ratio, and driving the formation of an adaptive optimization closed loop for preprocessing and matching parameters, the system has the ability to self-verify and optimize for complex samples, greatly improving the reliability and stability of the overall analysis process.

[0010] At the intelligent level, this invention surpasses the traditional analysis mode that only provides concentration data. By constructing a multi-dimensional fault fingerprint feature vector from precisely quantified multi-element concentration information and inputting it into a diagnostic model that integrates industry standards and expert knowledge, it achieves an integrated leap from "chemical composition detection" to "intelligent diagnosis of equipment health status." It can automatically analyze the internal fault types (such as overheating, discharge, wear, etc.) and their severity corresponding to the metal impurity fingerprint, providing direct and efficient decision support for transformer condition assessment and preventive maintenance, and has significant engineering application value.

[0011] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0012] To more clearly illustrate the technical solutions of the embodiments of the invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the invention. For those skilled in the art, the drawings can be obtained from these drawings without creative effort.

[0013] Figure 1 This is a flowchart illustrating the intelligent identification method for metal impurities in transformer oil based on capillary electrophoresis spectroscopy according to the present invention. Figure 2 This is a CE-C4D schematic diagram of the intelligent identification method for metal impurities in transformer oil based on capillary electrophoresis spectroscopy of the present invention. Figure 3 This is an example of the intelligent identification method for metal impurities in transformer oil based on capillary electrophoresis spectroscopy of the present invention, which uses CE-C4D to determine the standard solutions of six metal elements; Figure 4 This is an example of the intelligent identification method for metal impurities in transformer oil based on capillary electrophoresis spectroscopy of the present invention, which measures the spectral data of metal elements contained in a mineral oil sample. Detailed Implementation

[0014] The technical solutions of the embodiments of the invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the invention, and not all embodiments. Based on the embodiments of the invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the invention.

[0015] To resolve the above issues, please refer to [link / reference]. Figure 1 This invention discloses an intelligent identification method for metallic impurities in transformer oil based on capillary electrophoresis spectroscopy, comprising the following steps: S1. Construct a detection environment based on an improved miniaturized capillary electrophoresis-capacitive coupling non-contact conductivity detector (CE-C4D); based on the detection environment, collect the time-domain conductivity signal data and environmental context data of the current transformer oil sample to be tested; S2. The time-domain conductance signal data and environmental context data collected in S1 are preprocessed through multi-level digital signal processing and baseline correction processing to obtain clean electrophoresis spectrum data. S3. Construct a standard metal ion fingerprint spectrum library. Using the DTW algorithm, calculate the optimal normalization path between the pure electrophoresis spectrum data and the standard metal ion fingerprint spectrum library. Perform time axis nonlinear correction and characteristic peak alignment to obtain the aligned standard time domain spectrum. S4. Based on the aligned standard time-domain spectrum, the peak area characteristics of each metal ion are extracted by an adaptive integral window algorithm, and the absolute concentration values ​​of each metal impurity element are calculated by inversion using a preset standard curve equation. S5. Calculate whether the confidence level of the absolute concentration value in S4 meets the standard. If it does, use the absolute concentration value in S4 as the final concentration value. Otherwise, adjust the preprocessing parameters in S2 or the parameters of the DTW algorithm in S3, and repeat S2 to S4 until the confidence level meets the standard. S6. Construct a multidimensional fault fingerprint feature vector based on the final concentration value and input it into the transformer insulation condition diagnosis model to output the evaluation results of the potential fault types and severity inside the transformer. The above embodiments, through the improved miniaturized CE-C4D device combined with a specific buffer system and phase difference technology, significantly improve the signal-to-noise ratio and anti-interference capability of trace detection while achieving low cost and miniaturization, laying a reliable hardware foundation for field applications. The innovative introduction of the Dynamic Time Warping (DTW) algorithm for spectral elastic alignment overcomes identification errors caused by electrophoretic migration time drift. Combined with confidence assessment and parameter adaptive optimization closed-loop, the robustness and reliability of quantitative analysis results are ensured. By constructing the precise quantitative results into a fault fingerprint feature vector and inputting it into a diagnostic model integrating expert knowledge, an intelligent diagnostic leap from "metal concentration data" to "transformer fault type and severity" is achieved. This streamlines the entire chain from sample introduction to maintenance decision-making, providing a high-precision, high-reliability, and field-suitable innovative tool for transformer condition monitoring.

[0016] S1. Construct a detection environment based on an improved miniaturized capillary electrophoresis-capacitive coupling non-contact conductivity detector (CE-C4D); based on the detection environment, collect the time-domain conductivity signal data and environmental context data of the current transformer oil sample to be tested; The above embodiment S1 includes the following steps: S11. Prepare an electrophoresis buffer solution system; the buffer solution system includes L-histidine (L-His) at a concentration of 0.1-10 mM, lactic acid at a concentration of 0.1-10 mM, and 18-crown-6 ether at a concentration of 0.1-10 mM, and adjust the pH value to between 2 and 6 using acetic acid to obtain the buffer solution to be used; In specific implementation, the above embodiment S11 is as follows: The embodiment employs a buffer system optimized for metal cations, with L-His as a ligand to form a weak complex with the metal ion, thereby regulating the effective ion mobility and achieving separation; lactic acid is used as an auxiliary ligand to further optimize the separation selectivity; and 18-crown-6 ether utilizes its cavity structure to specifically complex K. + Na + Alkali metal ions are used to prevent their rapid elution from interfering with other metal peaks. Within this pH range, the metal ions are positively charged, and the electroosmotic flow (EOF) on the inner wall of the capillary is suppressed to a certain extent, which is beneficial for the separation and detection of cations. In this embodiment, an aqueous solution containing 9 mM (millimoles / liter) L-His, 4.6 mM lactic acid, and 0.38 mM 18-crown-6 was prepared and the pH was adjusted to 4.25. This buffer solution is not only used to provide the electrophoresis medium, but its specific conductivity characteristics are also the chemical basis for the effective operation of the phase difference technique in S12. S12. Configure an improved miniaturized CE-C4D device; the device has a built-in 0-20 MHz function generator, a 16-bit high-precision data acquisition card and a step-up transformer; the step-up transformer is used to control the detection channel and the reference channel to maintain a precise 180° phase difference; In specific implementation, the above embodiment S12 is as follows: Please refer to Figure 2 In the figure, a represents an adjustable high-voltage power supply, b represents positive and negative electrodes, c represents a conductivity cell, d represents a capillary tube, e represents a metal impurity sample cell, f represents an adjustable sine wave function generator, g represents a 16-bit high-precision data acquisition card, and h represents a display. The adjustable sine wave function generator serves as the excitation source; the 16-bit high-precision data acquisition card is used for signal digitization; the 0-30 kV adjustable high-voltage power supply provides the electrophoretic separation voltage; and the C4D conductivity cell serves as the detection probe. The device for achieving phase differential includes a step-up transformer with a secondary coil having a center tap or two reverse windings, capable of generating two sinusoidal excitation signals with equal amplitude but strictly 180° phase difference. One signal is applied to the excitation electrode on the detection capillary, and the other is applied to the reference channel (or a reference impedance simulated by a circuit). At the receiving end, the two signals converge at the current addition point. Since the capacitive current generated by the background buffer is equal in magnitude and opposite in direction in the two paths, they cancel each other out (differential). However, when metal ions pass through the detection electrode, the resulting minute impedance change is not canceled out and is thus detected. Through the above differential control, the background noise amplitude is reduced by more than 97%, making the weak signals of trace metals (10-20 nM level) stand out, and it can operate in the field without a complex shielded room. S13. The transformer oil sample to be tested is subjected to high-temperature ashing or microwave digestion to convert it into an aqueous metal ion solution suitable for electrophoresis. In specific implementation, the above embodiment S13 is as follows: The specific steps of high-temperature ashing are as follows: accurately weigh 25.00g of transformer oil sample into a porcelain crucible, keep it at 350℃ for 4 hours to slowly carbonize the sample, and then keep it at 650℃ for 3 hours for ashing until the residue is white; after cooling, add an appropriate amount of hydrochloric acid to dissolve the residue, and finally make up to 10mL with nitric acid; The specific steps of microwave digestion are as follows: weigh 0.5g of oil sample into a polytetrafluoroethylene digestion vessel, add 5mL of HNO3 and 2mL of H2O2, seal it and put it into a microwave digester, set the program to heat up to 180℃ in 10 minutes, keep it for 20 minutes, cool it to remove the acid, and make up to 10mL with ultrapure water; both methods can effectively extract metals from oil, the microwave digestion method is faster, and the dry ashing method has a large processing capacity and is suitable for trace enrichment; S14. Fill the buffer solution prepared in S11 into the capillary of the CE-C4D device configured in S12, and introduce the aqueous metal ion solution obtained in S13 for separation; record the conductivity changes during the separation process through the acquisition card to form the original time-domain conductivity signal data, and record the ambient temperature and capillary column temperature during the detection as environmental context data. In specific implementation, the above embodiment S14 is as follows: In this embodiment, the total length of the quartz capillary is 60cm, the effective length is 45cm, the inner diameter is 75μm, and the separation voltage is +15 kV; gravity injection is used (height difference 10cm, time 30s); the acquisition card is started to record the time-domain conductivity signal, and at the same time, the ambient temperature sensor data is read; if the ambient temperature fluctuation exceeds ±2℃, an anomaly is recorded, providing prior information of "large drift" for the alignment algorithm in subsequent S3; The above embodiments, through hardware improvements, utilize a miniaturized CE-C4D device combined with precise 180° phase difference technology to effectively cancel background noise, thereby highlighting trace signals and achieving a balance between device portability and high sensitivity. In terms of the chemical system, an optimized buffer formulation for metal cations (L-His, lactic acid, and 18-crown-6 ether working synergistically) ensures effective separation of multiple elements. Combined with mature high-temperature ashing or microwave digestion pretreatment methods, these components together form a robust and reliable detection platform suitable for rapid on-site analysis and output of raw data and environmental information.

[0017] S2. The time-domain conductance signal data and environmental context data collected in S1 are preprocessed through multi-level digital signal processing and baseline correction processing to obtain clean electrophoresis spectrum data. The above embodiment S2 includes the following steps: S21. Select Symlets wavelet as the basis function, decompose the original time-domain conductance signal data and environmental context data output by S1 into several layers; extract the high-frequency detail coefficients of each layer and perform soft thresholding to reconstruct the signal to obtain the denoised conductance signal; In specific implementation, the above embodiment S21 specifically involves: using the original time-domain conductance signal data and environmental context data collected in S14 as input, and selecting the Symlets wavelet; the Symlets wavelet has good symmetry and tight support, and has a high similarity to the Gaussian shape of the electrophoresis peak, effectively preserving the peak shape characteristics; performing 5-level wavelet decomposition on the original time-domain conductance signal data and environmental context data to obtain one low-frequency approximation coefficient (cA5) and five high-frequency detail coefficients (cD1-cD5); noise is mainly concentrated in cD1-cD3; and using a soft thresholding function to process the detail coefficients; thresholding a 1. The calculation formula is: Where σ is the estimated standard deviation of noise, and ln represents the natural logarithm, i.e., a mathematical constant. e A logarithmic function with a base of approximately 2.71828. N ln is the signal length. N Indicates the signal length N Perform with constant e Logarithmic calculation with base 0; soft thresholding avoids signal oscillations caused by hard thresholding; inverse wavelet transform is performed using the processed coefficients to reconstruct the signal; even with a signal-to-noise ratio as low as 3:1, it can effectively remove glitches with a peak height loss of less than 2%; S22. Using the asymmetric least squares smoothing algorithm (AsLS), the denoised conductance signal obtained in S21 is used as input, and the background drift trend of the signal is calculated by iterative weighted least squares method to obtain the fitting baseline. The above embodiment S22 includes the following steps: S221. Construct a first objective function; the first objective function includes the sum of squares of the deviations between the denoised conductance signal output in S21 and the baseline to be determined, as well as a smoothness penalty term for the baseline to be determined; S222. Initialize the weight vector of the first objective function and set the smoothing coefficient λ; S223. Solve for the baseline vector by minimizing the first objective function, and update the weight vector according to the relationship between the current baseline vector and the denoised conductance signal. S224. Repeat S223 until the weight vector converges, and output the final fitted baseline. In specific implementation, the above embodiment S22 is specifically as follows: the first objective function is expressed as ,in, S This represents quantifying and minimizing the total deviation between the fitted baseline and the ideal baseline. w i Indicates the first i The weight of each data point y i Indicates the first i The original signal values ​​of each data point z i Indicates the first i The fitted baseline value for each data point, where λ represents the smoothing coefficient, is set to 10 in this embodiment. -5 , (Δ z i ) 2 Indicates the fitted baseline z exist i The second difference at the point; in the iteration, if y i > z i (representing the signal peak region), then assigns a minimal weight. w i =0.001, if y i ≤ z i (Representing the baseline region), then assign weights. w i =1; Fix this weight, and solve for the new baseline vector that minimizes the first objective function. z new Update the weights with the new baseline and repeat this process; when the L2 norm of the difference between the baseline vectors obtained from two adjacent iterations is less than the preset tolerance (e.g., 10), the remaining weights are updated. -6 When the weights and baseline have converged, the iteration terminates. S23. Subtract the fitted baseline obtained in S22 from the denoised conductivity signal obtained in S21 to eliminate the influence of background drift and obtain pure electrophoresis spectrum data with the baseline returned to zero. At this time, the signal values ​​of all non-peak regions return to near 0, laying the foundation for subsequent peak identification. The above embodiments significantly improve the quality and usability of electrophoresis spectrum data through a multi-level digital signal preprocessing process; Symlets wavelet transform effectively filters out high-frequency random noise while preserving the original shape of characteristic peaks to the greatest extent; combined with an asymmetric least squares smoothing algorithm, it can intelligently distinguish between signal peaks and baseline regions, accurately fit and subtract complex background drift through iterative calculation; this preprocessing combination can work stably under low signal-to-noise ratio conditions, obtaining a clean spectrum with zero baseline, laying a key data foundation for subsequent accurate peak identification, alignment and quantitative analysis.

[0018] S3. Construct a standard metal ion fingerprint spectrum library. Using the DTW algorithm, calculate the optimal normalization path between the pure electrophoresis spectrum data and the standard metal ion fingerprint spectrum library. Perform time axis nonlinear correction and characteristic peak alignment to obtain the aligned standard time domain spectrum. Please see Figure 3 The above embodiment S3 includes the following steps: S31. Under standard experimental conditions, determine the concentrations of metal ions (Cu) respectively. 2+ Fe 2+ Zn 2+ Mg 2+ K + Na + Standard solutions and their mixtures were used to obtain standard spectral sequences containing standard retention times, peak shapes, and relative positions as reference templates to create a standard metal ion fingerprint spectral library. In specific implementation, the above embodiment S31 is as follows: Under constant temperature conditions of 25°C, the spectrum of the standard mixed solution is measured. The fingerprint spectrum characteristics obtained in this embodiment are as follows: the first peak is K. + (Migration time approximately 2.1 min), the second peak is Na + (Migration time approximately 2.3 min), the third peak is Mg 2+ (Migration time approximately 2.6 min), the fourth peak is Mn 2+ / Zn 2+ (Migration time approximately 2.8 min, depending on the specific buffer solution); the fifth peak is Fe. 2+ (Migration time approximately 3.1 min); the sixth peak is Cu. 2+ (Migration time approximately 3.4 min); save this sequence as a reference template. C = { c 1, c2, ..., c i ,..., c p};in, c i Indicates the first in the reference template i Complete characteristic data of each ion peak. p This indicates the total number of ion peaks in the reference template. c p Indicates the first in the reference template p All characteristic data of each ion peak; this embodiment sets p= 6; S32. Define the purified electrophoresis pattern data obtained in S2 as the test sequence. D The reference template in the standard metal ion fingerprint spectrum library obtained in S31 is defined as the reference sequence. C f ; In specific implementation, the above embodiment S32 is as follows: Assume the pure spectral sequence of the sample to be tested is... D = { d 1, d 2,..., d j ,..., d q}, d j Indicates the first j Complete characteristic data of each ion peak. q This represents the total number of ion peaks in the pure spectral sequence. d q Indicates the first pure spectral sequence q Complete characteristic data of each ion peak; due to differences in oil sample viscosity or changes in ambient temperature. D The Cu²⁺ peak in the image may have drifted to 3.6 min, which would lead to misjudgment if identified using an absolute time window. S33. Based on the test sequence and the reference sequence, calculate the test sequence. D Each point in the reference sequence C f The Euclidean distance of each point in the array is used to construct a dimension of . p × q Cumulative distance matrix F The row length of the cumulative distance matrix is ​​the test sequence. D length q The column length is the reference sequence. C f length p ;in F ( i , j ) = (d j - c i )²; where, ( i , j ) indicates that the cumulative distance matrix contains the first digit. i The first ion peak and the first ion peak j The characteristic points of each ion peak; F ( i , j ) represents a point ( i , j The local distance of the test spectrum is calculated; the cumulative distance matrix reflects the similarity between each point in the test spectrum and each point in the standard spectrum. S34. Based on the cumulative distance matrix, use the Dynamic Time Warping (DTW) algorithm to find an optimal warping path in the cumulative distance matrix constructed in S33 that minimizes the total cumulative distance. The above embodiment S34 includes the following steps: S341. Based on the environmental context data obtained in S14, set the constraints for the DTW algorithm; if the ambient temperature fluctuates greatly, widen the search window. S342. Calculate the value of each element in the cumulative distance matrix using the recursive formula; S343. Backtrack from the end of the cumulative distance matrix to the starting point to obtain the optimal regularized path; In specific implementation, the following constraints are set: These constraints include, based on the ambient temperature data recorded in S14, if the temperature deviates significantly from the standard value (e.g., the difference between the temperature and the standard value exceeds a preset temperature difference threshold), the local window restriction is relaxed during the search; the path must start from (1, 1) and end at (...). p , q End; path steps can only be between adjacent points, cannot jump, and time cannot be reversed; the calculation formula of the Dynamic Time Warping (DTW) algorithm is as follows: ;in, β ( i , j () indicates cumulative distance; This indicates a state transition, in order to reach ( i , j The previous step can only come from three adjacent points: β ( i -1, j ) indicates that only one step is taken forward on the D sequence (in C (Stay above), corresponding expansion D Timeline; β ( i , j -1) indicates that only whenC The sequence moves forward one step (in) D (Stay above), corresponding to compression D Timeline; β ( i- 1, j -1) indicates moving one step forward in both sequences simultaneously, representing a one-to-one alignment; the optimal path is obtained through the above formula; the optimal path includes the test sequence. D With reference sequence C f Time mapping relationship; S35. Based on the mapping relationship of the optimal regularized path obtained in S34, nonlinearly stretch or compress the time axis of the test sequence D so that its characteristic peak position is aligned with that of the reference sequence. C f Strict alignment yields the aligned standard time-domain spectrum; In specific implementation, the above embodiment S35 specifically involves: based on the mapping relationship of the optimal path, determining the sequence to be tested. D The time axis is subjected to nonlinear transformation; for example, when a peak in the spectrum under test is detected at 3.6 min, it is similar to that of the standard spectrum. Figure 3 The peak at 0.4 min has the highest matching degree (similar waveform), which will compress the time axis of that segment of the spectrum to be tested, so that its peak value returns to 3.4 min; after DTW processing, no matter how the migration time of the sample to be tested drifts, as long as the relative order and shape of the peaks exist, they can be accurately "captured" and aligned to the standard position; this allows features to be extracted directly at fixed time points without manual intervention. The above embodiments effectively solve the problem of identification errors caused by migration time drift in capillary electrophoresis by constructing a standard metal ion fingerprint spectrum library and innovatively introducing a dynamic time warping algorithm. This method can automatically find the optimal mapping path of feature points between the spectrum to be tested and the standard spectrum, and intelligently adjust the search strategy according to environmental factors (such as temperature), thereby achieving nonlinear correction and precise alignment of the time axis. This process does not require manual setting of a fixed time window, significantly improving the automation level and anti-interference ability of peak position identification, and providing a high-quality spectrum data foundation with consistent time reference and high comparability for subsequent accurate quantification and fault diagnosis.

[0019] S4. Based on the aligned standard time-domain spectrum, the peak area characteristics of each metal ion are extracted by an adaptive integral window algorithm, and the absolute concentration values ​​of each metal impurity element are calculated by inversion using a preset standard curve equation. The above embodiment S4 includes the following steps: S41. In the aligned standard time-domain spectrum obtained in S3, the integration interval of each metal ion is locked according to the preset position in the standard metal ion fingerprint spectrum library; taking Cu...2+ For example, in this embodiment, Cu is set 2+ The integration interval is ±0.1 min; similarly, the integration intervals for other ions are set. S42. Perform numerical integration on the signal within the integration interval determined in S41 to obtain the response peak area of ​​each metal ion. S43. Obtain the linear regression equations for the concentration and peak area of ​​each element in advance; In specific implementation, the above embodiment S43 specifically includes: This embodiment contains linear regression equations for the concentrations and peak areas of each element (Cu, Fe, Zn, Mg, K) and their detection limits; as shown in Table 1 below, in the table, y Indicates the peak area. x molar concentration:

[0020] S44. Substitute the response peak area obtained in S42 into the linear regression equation in S43 to calculate the absolute concentration (mass concentration) of each element Cu, Fe, Zn, Mg, K and Na in the transformer oil. For specific implementation details, please refer to [link / reference]. Figure 4 Specifically, in the above embodiment S44: This embodiment takes Cu as an example, such as the linear equation of Cu as follows: y =643.26 x +1.22, the measured peak area is 65.5, so the concentration is calculated to be approximately 0.1 mol / L (unit conversion required); the system automatically converts the calculated concentration from molar concentration (mM) to mass fraction in transformer oil (mg / kg), which needs to be combined with the oil sample mass (25g) and fixed volume (10mL) weighed in S13 for conversion; the content of metal elements in the main transformer oil after conversion is shown in Table 2 below, and compared with the content of metal elements measured by traditional ICP-OES:

[0021] Furthermore, to avoid the randomness caused by a single oil product, a different oil product than that in Table 2 was used for testing. The steps were the same as S1-S4, resulting in Table 3:

[0022] The absolute concentration values ​​in the tables of the above embodiments are all data that meet the confidence level in S5. Based on the high-quality spectrum aligned by the preceding steps, the above embodiments achieve accurate quantification of the concentration of various metal impurities through adaptive integral window and standard curve inversion. By combining the linear regression equation of preset position and peak area of ​​the ion fingerprint library, the spectral signal can be automatically and accurately converted into absolute concentration values. As shown in the embodiments, its quantitative results (CE-C4D method) are in high agreement with the traditional authoritative method (ICP-OES method), and the error is within a reasonable range. This fully verifies the accuracy and reliability of the entire process from sample processing to quantitative calculation, and provides a solid data foundation for subsequent fault diagnosis.

[0023] S5. Calculate whether the confidence level of the absolute concentration value in S4 meets the standard. If it does, use the absolute concentration value in S4 as the final concentration value. Otherwise, adjust the preprocessing parameters in S2 or the parameters of the DTW algorithm in S3, and repeat S2 to S4 until the confidence level meets the standard. The above embodiment S5 includes the following steps: S51. Extract the total cumulative distance of the optimal regularized path obtained in S34 and the signal-to-noise ratio when calculating the peak area in S42. Calculate the confidence score of the absolute concentration values ​​of elements in transformer oil obtained in S44. In specific implementation, the above embodiment S53 is specifically as follows: using the formula Score=(1 / Distance)+SNR; where Score represents the confidence score of the absolute concentration value of elements in transformer oil; Distance represents a "minimum cumulative distance" value generated when DTW alignment is performed in S34. The smaller this value is, the more similar the spectrum to be tested is to the standard spectrum, and the more reliable the identification is; SNR represents the signal-to-noise ratio of the combined peak in S42, that is, the ratio of the signal peak height to the baseline noise standard deviation within the integration window in S41. S52. Compare the confidence score obtained in S51 with the preset threshold; if the confidence score is greater than or equal to the confidence threshold, use the absolute concentration value in S4 as the final concentration value; if the confidence score is less than the confidence threshold, trigger the adaptive parameter optimization mechanism to maximize the cross-correlation coefficient of the spectral alignment and construct a second objective function. The above embodiment S52 includes the following steps: In specific implementation, the above embodiment S52 is as follows: If a test is found to be suspicious due to severe oil sample contamination and extreme baseline fluctuations, resulting in the DTW-aligned score being lower than the threshold, then an adaptive parameter optimization mechanism is triggered, aiming to maximize the cross-correlation coefficient of the spectral alignment, and constructing a second objective function: collecting the aligned test sequences. D With reference sequence C f Between Pearson Cross-correlation number Pdc Obtain the baseline noise standard deviation of the signal after S2 processing. B dc The average peak height of the sum of the peaks A dc The reciprocal of the signal purity is obtained. B dc / A dc The second objective function is then: ;in F dc An evaluation metric representing the alignment effect of quantized spectra and the quality of signal processing; S53. Based on the second objective function, the particle swarm optimization algorithm (PSO) is used to search for the combination of wavelet transform, asymmetric least squares smoothing algorithm parameters in S2 and window parameters of DTW algorithm in S3 within the preset search space to obtain the optimal parameter combination. The above embodiment S53 includes the following steps: S531. Construct a particle set, and set the size of the particle set to... r The particle set is then represented as: ,in, z Represents a set of particles; z i Represents the first in the set of particles i One particle; z r Represents the first particle in the set of particles r One particle; based on the search space and particle set of wavelet transform, asymmetric least squares smoothing algorithm parameters, and window parameters of the DTW algorithm in S3, an initial particle position set is constructed. ,in, u Represents the initial position set of the particles. u i Represents the first particle in the set of particles i The initial position of each particle. u r Represents the first particle in the set of particles r The initial position of each particle; the position of each particle in the initial position set is used as a combination of wavelet transform, asymmetric least squares smoothing algorithm parameters and DTW algorithm window parameters; S532. Perform iterative update operations on the particles in the particle set, with each update occurring within a preset search space; and calculate the fitness value of each particle's position according to the second objective function. Update the fitness of each particle in the particle set according to its fitness value from high to low, and obtain the best individual particle and the globally best particle in each iteration. During the update process, retain the positions of particles with the top 5% fitness, continue to retain the middle positions of the top 46% fitness positions (i.e., 45% of the positions enter the next round), and obtain the remaining 50% through random particle movement. The best individual particle is the particle with the highest fitness value in each iteration, and the globally best particle is the particle with the highest fitness value in all iterations. S533, repeat S532, stop iterating when the maximum number of iterations is reached, and use the globally optimal particle as the optimal parameter combination; Example of S53 in the above embodiment: Adjust the parameters in S2 to increase the number of wavelet decomposition layers from 5 to 7 (enhancing denoising), or increase the smoothing coefficient of the asymmetric least squares smoothing algorithm (enhancing baseline correction). Adjust the parameters in S3 to increase the time window limit of DTW to limit the degree of alignment distortion and prevent noise from being mistaken for signal peaks. Reprocess the original data using the new parameters. If the score increases to ≥ the confidence threshold after 3 iterations, output the result. If it is still below the confidence threshold, mark "spectral abnormality" and prompt manual intervention to check the original spectrum. S54. Feed back the parameters in the optimal parameter combination to S2 and S3, and repeat S2-S5 until the confidence level is met to obtain the final concentration value. The above embodiments construct an intelligent analysis method capable of self-evaluation and self-correction by introducing confidence scoring and adaptive optimization closed loop. The reliability of each detection result is quantified by using spectral alignment quality and signal-to-noise ratio as core indicators. When the confidence level is insufficient, the system automatically triggers a particle swarm optimization algorithm to intelligently adjust the pre-processing and alignment parameters, iteratively optimizing until a reliable result is output. This design significantly improves the robustness of the method when dealing with complex and contaminated oil samples, ensuring the accuracy and reliability of the final quantitative data, and is a key guarantee for achieving automated and highly reliable on-site detection.

[0024] S6. Construct a multidimensional fault fingerprint feature vector based on the final concentration value and input it into the transformer insulation condition diagnosis model to output the evaluation results of the potential fault types and severity inside the transformer. The above embodiment S6 includes the following steps: S61. Extract the absolute concentration values ​​of each element from the final concentration values ​​in S5, and construct a fault fingerprint feature vector containing the absolute concentration and the concentration ratio between elements. In specific implementation, the above embodiment S61 is as follows: In a certain detection in this embodiment, the data obtained are Cu=0.8mg / kg, Fe=0.1mg / kg, Zn=0.2mg / kg, and total metals=1.1mg / kg; construct the vector V=[0.8, 0.1, 0.2, 8.0(Cu / Fe), 0.25 (Zn / Cu), ...]; S62. Pre-construct a transformer insulation condition diagnostic model; input the fault fingerprint feature vector constructed in S61 into the transformer insulation condition diagnostic model, and output the assessment results of fault type and severity; In specific implementation, the above embodiment S62 includes the following steps: S621. Obtain the fault determination logic from the expert rule base data generated based on GB / T 7597 and historical transformer fault cases to obtain the transformer insulation status diagnostic model; alternatively, a machine learning model can be trained based on GB / T 7597 and historical transformer fault case data to obtain the transformer insulation status diagnostic model. Training is a routine operation for machine learning model training, which will not be elaborated further in this embodiment; this embodiment continues to execute S622 with fault determination logic as an example. S622. Input the fault fingerprint feature vector into the transformer insulation condition diagnostic model and execute the following judgment rules: Rule G1: Is the total metal content > 0.5 mg / kg? If yes, proceed to fault subdivision G2; otherwise, output "Oil quality normal, insulation good"; Rule G2 (dominant element judgment): Case H1: If the Cu content in the vector is significantly high and Cu / Fe > 5, the diagnosis conclusion is that there is an overheating fault in the transformer winding or leads, or friction in the bare copper parts. It is recommended to check the load record and perform a winding DC resistance test; Case H2: If the Fe content is high and Cu / Fe < 1, the diagnosis conclusion is that there is corrosion or partial discharge in the core, tank, or clamps. It is recommended to perform dissolved gas analysis (DGA) in the oil to help confirm the discharge type; Case H3: If the Zn content is abnormal (e.g., > 0.5 mg / kg), the diagnosis conclusion is that the cooler (copper pipe zinc plating) is corroded or the submersible pump bearing is worn. It is recommended to check the cooling system efficiency and submersible pump vibration; Case H4: If a high concentration of Na is detected... + or K + The diagnosis conclusion is that there may be moisture due to poor sealing (environmental dust mixing in) or deep aging of the insulation paper (cellulose degradation products); In this embodiment, based on the aforementioned hypothetical data (Cu-dominant), the output is "Warning: Abnormal copper content detected (0.8 mg / kg), high Cu / Fe ratio, suspected winding overheating, it is recommended to shorten the detection cycle and conduct electrical tests." The above embodiments extract precise metal concentration values ​​and construct a multidimensional fault fingerprint feature vector containing concentration ratios. This vector is then input into a diagnostic model integrated with an expert rule base, enabling automated identification and severity assessment of typical internal transformer faults (such as winding overheating, core discharge, and cooling system wear). This method surpasses the traditional analysis mode that only provides concentration data, and can directly output diagnostic conclusions with clear operational and maintenance guidance significance, providing efficient and intelligent decision support for preventive maintenance and condition-based repair of equipment.

[0025] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0026] The preferred embodiments of the invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention.

Claims

1. A method for intelligent identification of metallic impurities in transformer oil based on capillary electrophoresis spectroscopy, characterized in that, Includes the following steps: S1. Construct a detection environment based on an improved miniaturized capillary electrophoresis-capacitive coupling non-contact conductivity detector; based on the detection environment, collect the time-domain conductivity signal data and environmental context data of the current transformer oil sample to be tested; S2. The time-domain conductance signal data and environmental context data collected in S1 are preprocessed through multi-level digital signal processing and baseline correction processing to obtain clean electrophoresis spectrum data. S3. Construct a standard metal ion fingerprint spectrum library. Using the DTW algorithm, calculate the optimal normalization path between the pure electrophoresis spectrum data and the standard metal ion fingerprint spectrum library. Perform time axis nonlinear correction and characteristic peak alignment to obtain the aligned standard time domain spectrum. S4. Based on the aligned standard time-domain spectrum, the peak area characteristics of each metal ion are extracted by an adaptive integral window algorithm, and the absolute concentration values ​​of each metal impurity element are calculated by inversion using a preset standard curve equation. S5. Calculate whether the confidence level of the absolute concentration value in S4 meets the standard. If it does, use the absolute concentration value in S4 as the final concentration value. Otherwise, adjust the preprocessing parameters in S2 or the parameters of the DTW algorithm in S3, and repeat S2 to S4 until the confidence level meets the standard. S6. Construct a multidimensional fault fingerprint feature vector based on the final concentration value and input it into the transformer insulation condition diagnosis model to output the assessment results of potential fault types and severity inside the transformer.

2. The intelligent identification method for metal impurities in transformer oil based on capillary electrophoresis spectroscopy as described in claim 1, characterized in that, S1 includes the following steps: S11. Prepare an electrophoresis buffer solution system; the buffer solution system includes L-histidine at a concentration of 0.1-10 mM, lactic acid at a concentration of 0.1-10 mM, and 18-crown-6 ether at a concentration of 0.1-10 mM, and adjust the pH value to between 2 and 6 using acetic acid to obtain the buffer solution to be used; S12. Configure an improved miniaturized CE-C4D device; the device has a built-in 0-20 MHz function generator, a 16-bit high-precision data acquisition card and a step-up transformer; the step-up transformer is used to control the detection channel and the reference channel to maintain a precise 180° phase difference; S13. The transformer oil sample to be tested is subjected to high-temperature ashing or microwave digestion to convert it into an aqueous metal ion solution suitable for electrophoresis. S14. Fill the buffer solution prepared in S11 into the capillary of the CE-C4D device configured in S12, and introduce the aqueous metal ion solution obtained in S13 for separation. Record the conductivity changes during the separation process through the acquisition card to form the original time-domain conductivity signal data. At the same time, record the ambient temperature and capillary column temperature during the detection as environmental context data.

3. The intelligent identification method for metal impurities in transformer oil based on capillary electrophoresis spectroscopy as described in claim 1, characterized in that, S2 includes the following steps: S21. Select Symlets wavelet as the basis function, decompose the original time-domain conductance signal data and environmental context data output by S1 into several layers; extract the high-frequency detail coefficients of each layer and perform soft thresholding to reconstruct the signal to obtain the denoised conductance signal; S22. Using an asymmetric least squares smoothing algorithm, the denoised conductance signal obtained in S21 is used as input. The background drift trend of the signal is calculated by iterative weighted least squares method to obtain the fitting baseline. S23. Subtract the fitted baseline obtained in S22 from the denoised conductivity signal obtained in S21 to eliminate the influence of background drift and obtain pure electrophoresis spectrum data with zero baseline.

4. The intelligent identification method for metal impurities in transformer oil based on capillary electrophoresis spectroscopy as described in claim 3, characterized in that, S22 includes the following steps: S221. Construct a first objective function; the first objective function includes the sum of squares of the deviations between the denoised conductance signal output in S21 and the baseline to be determined, as well as a smoothness penalty term for the baseline to be determined; S222. Initialize the weight vector of the first objective function and set the smoothing coefficient; S223. Solve for the baseline vector by minimizing the first objective function, and update the weight vector according to the relationship between the current baseline vector and the denoised conductance signal. S224. Repeat S223 until the weight vector converges, and output the final fitted baseline.

5. The intelligent identification method for metal impurities in transformer oil based on capillary electrophoresis spectroscopy as described in claim 1, characterized in that, S3 includes the following steps: S31. Under standard experimental conditions, standard solutions and mixed solutions of metal ions were measured separately. Standard spectral sequences containing standard retention time, peak shape information, and standard relative position information were obtained as reference templates to obtain a standard metal ion fingerprint spectral library. S32. Define the pure electrophoresis spectrum data obtained in S2 as the test sequence, and define the reference template in the standard metal ion fingerprint spectrum library obtained in S31 as the reference sequence. S33. Based on the test sequence and the reference sequence, calculate the Euclidean distance between each point in the test sequence and each point in the reference sequence, and construct a cumulative distance matrix; the row length of the cumulative distance matrix is ​​the length of the test sequence, and the column length is the length of the reference sequence; S34. Based on the cumulative distance matrix, use the DTW algorithm to find an optimal regular path in the cumulative distance matrix constructed in S33 that minimizes the total cumulative distance. S35. Based on the mapping relationship of the optimal regularized path obtained in S34, the time axis of the test sequence is nonlinearly stretched or compressed to make its characteristic peak position strictly aligned with the reference sequence, thus obtaining the aligned standard time domain spectrum.

6. The intelligent identification method for metal impurities in transformer oil based on capillary electrophoresis spectroscopy as described in claim 5, characterized in that, S34 includes the following steps: S341. Based on the environmental context data obtained in S14, set the constraints for the DTW algorithm. S342. Calculate the value of each element in the cumulative distance matrix using the recursive formula; S343. Backtrack from the end point of the cumulative distance matrix to the starting point to obtain the optimal regular path.

7. The intelligent identification method for metal impurities in transformer oil based on capillary electrophoresis spectroscopy as described in claim 5, characterized in that, S4 includes the following steps: S41. In the standard time-domain spectrum obtained after alignment in S3, the integration interval of each metal ion is locked according to the preset position of the standard metal ion fingerprint spectrum library. S42. Perform numerical integration on the signal within the integration interval determined in S41 to obtain the response peak area of ​​each metal ion. S43. Obtain the linear regression equations for the concentration and peak area of ​​each element in advance; S44. Substitute the response peak area obtained in S42 into the linear regression equation in S43 to calculate the absolute concentration of elements in the transformer oil.

8. The intelligent identification method for metal impurities in transformer oil based on capillary electrophoresis spectroscopy as described in claim 7, characterized in that, S5 includes the following steps: S51. Extract the total cumulative distance of the optimal regularized path obtained in S34 and the signal-to-noise ratio when calculating the peak area in S42. Calculate the confidence score of the absolute concentration values ​​of elements in transformer oil obtained in S44. S52. Compare the confidence score obtained in S51 with the preset threshold; if the confidence score is greater than or equal to the confidence threshold, use the absolute concentration value in S4 as the final concentration value; if the confidence score is less than the confidence threshold, trigger the adaptive parameter optimization mechanism to maximize the cross-correlation coefficient of the spectral alignment and construct a second objective function. S53. Based on the second objective function, the particle swarm optimization algorithm is used to search for the combination of wavelet transform and asymmetric least squares smoothing algorithm parameters in S2 and window parameters of DTW algorithm in S3 within the preset search space to obtain the optimal parameter combination. S54. Feed back the parameters in the optimal parameter combination to S2 and S3, and repeat S2-S5 until the confidence level is met to obtain the final concentration value.

9. The intelligent identification method for metal impurities in transformer oil based on capillary electrophoresis spectroscopy as described in claim 8, characterized in that, S53 includes the following steps: S531. Construct a particle set; based on the search space and particle set of wavelet transform, the parameters of the asymmetric least squares smoothing algorithm and the window parameters of the DTW algorithm in S3, construct an initial particle position set; use the position of each particle in the initial particle position set as a combination of the parameters of wavelet transform, the asymmetric least squares smoothing algorithm and the window parameters of the DTW algorithm. S532. In each iteration, the fitness value of the position of each particle in the particle set is calculated according to the second objective function, the fitness of each particle in the particle set is updated, and the best individual particle and the global best particle are obtained in each iteration. S533, repeat S532. When the maximum number of iterations is reached, stop the iteration and use the globally optimal particle as the optimal parameter combination.

10. The intelligent identification method for metal impurities in transformer oil based on capillary electrophoresis spectroscopy as described in claim 8, characterized in that, S6 includes the following steps: S61. Extract the absolute concentration values ​​of each element from the final concentration values ​​in S5, and construct a fault fingerprint feature vector containing the absolute concentration and the concentration ratio between elements. S62. Pre-construct a transformer insulation condition diagnostic model; input the fault fingerprint feature vector constructed in S61 into the transformer insulation condition diagnostic model, and output the assessment results of fault type and severity.

Images