A detection method for furfural in insulating oil based on terahertz time-domain spectroscopy

Abstract

The present invention is a method for detecting furfural in insulating oil based on terahertz time-domain spectroscopy, which relates to the technical field of aging detection of insulating oil of power equipment, and comprises the following steps: for insulating oils of different aging states, the furfural content therein is determined by a chromatographic analysis method, and samples of corresponding concentrations are prepared; according to the liquid properties and terahertz spectral properties of the insulating oil, a transmission terahertz time-domain spectroscopy system is used to scan the sample and obtain the time-domain spectral data of the sample; a quantum chemical simulation method is used to simulate the absorption and reflection characteristics of furfural molecules in the terahertz frequency band, and the vibration mode, rotation energy level and terahertz band absorption characteristics in the real spectral signal are matched; by reconstructing the terahertz time-domain signal, the furfural content in the insulating oil is quantitatively detected. The present invention realizes rapid and non-contact detection of trace furfural in insulating oil by analyzing the terahertz spectral characteristics, and provides an effective evaluation method for the degree of transformer insulation aging.

Description

Technical Field

[0001] The invention relates to the technical field of aging detection of insulating oil of power equipment, and in particular to a method for detecting furfural in insulating oil based on terahertz time-domain spectroscopy. Background Art

[0002] Traditional methods for detecting furfural in insulating oil mainly rely on chemical analysis techniques, such as high performance liquid chromatography (HPLC) or gas chromatography (GC), which usually require complex sample pretreatment, long analysis cycles and high instrument costs. In addition, with the increasing number of power equipment, the demand for insulating oil aging detection is also growing, and traditional detection methods are difficult to meet the needs of efficient and large-scale detection. The emerging terahertz time-domain spectroscopy technology has shown unique advantages in material identification and component analysis. Its non-contact and non-destructive characteristics make it one of the important research directions in the field of analytical chemistry. However, when using the currently common terahertz detection technology to detect furfural in insulating oil, there are the following problems:

[0003] 1. The furfural content in insulating oil is extremely low, and the oil itself and the trace amount of water in it will interfere with the terahertz wave. The accuracy of the measurement cannot be guaranteed when using the current terahertz detection technology for detection; 2. Similar bond groups in the ternary mixture (methanol, furfural and water) in insulating oil may cause spectral peak overlap, increasing the difficulty of quantitative analysis; 3. The relationship between the terahertz characteristic spectrum of the insulating oil aging marker and the aging state of the oil paper is complex, and the aging marker is difficult to accurately characterize. Summary of the invention

[0004] The present invention aims to solve the deficiencies of the prior art and provides a method for detecting furfural in insulating oil based on terahertz time-domain spectroscopy.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] A method for detecting furfural in insulating oil based on terahertz time-domain spectroscopy comprises the following steps:

[0007] S1. For insulating oils in different aging states, the furfural content is determined by chromatography and samples of corresponding concentrations are prepared;

[0008] S2. According to the liquid properties and terahertz spectrum characteristics of the insulating oil, a transmission terahertz time-domain spectroscopy system is used to scan the sample, obtain the time-domain spectrum data of the sample, and calculate the characteristic parameters of absorption coefficient, dielectric constant, and transmittance. The absorption coefficient, dielectric constant, and transmittance reflect the characteristic information of the furfural aging marker in the sample;

[0009] S3. Use quantum chemical simulation methods to simulate the absorption and reflection characteristics of furfural molecules in the terahertz band, and match the vibration mode, rotation energy level and terahertz band absorption characteristics in the real spectral signal;

[0010] S31, establishing a three-dimensional molecular model for the furfural molecule according to the molecular formula, and for water containing hydroxyl groups, supplementing its adjacent hydroxyl groups to form a furfural molecular structure model containing hydrogen bonds;

[0011] S32. Based on the Gaussian orbital or plane wave function assumption, density functional theory quantum chemical calculations were performed to calculate the vibration modes, rotational energy levels and corresponding terahertz absorption characteristics of furfural molecules;

[0012] S33. In-depth analysis of how the chemical bonds and functional groups in the furfural molecule affect the terahertz spectrum, analysis of the specific absorption characteristics of hydroxyl and aldehyde functional groups in the terahertz frequency band, and how these characteristics change with slight changes in the molecular structure;

[0013] S34. Compare the experimentally obtained terahertz spectral data with the quantum chemical simulation results. Through this comparison, verify the accuracy of the simulation model and explain the specificity of the characteristic peaks observed in the experiment. By increasing the weight of the characteristic peaks, improve the predictive ability and generalization of the quantitative analysis model;

[0014] S4. Quantitatively detect the furfural content in insulating oil by reconstructing the terahertz time domain signal;

[0015] S41. For the sample containing only furfural obtained in S2, terahertz characteristic parameters, i.e., absorption coefficient, dielectric constant, and transmittance, are extracted to accurately characterize the corresponding furfural content; furfural concentration is taken as the dependent variable, and spectral characteristics are taken as independent variables, a relationship model is established, and quantum chemical simulation terahertz characteristic spectrum is introduced to assist in determining the characteristic spectral peak corresponding to its unique chemical vibration as an auxiliary variable;

[0016] S42, constructing a pure insulating oil sample containing a gradient distribution of furfural, and then performing terahertz time-domain spectroscopy detection on the sample;

[0017] S43, using the neural network model ANN model to quantitatively analyze the furfural content in insulating oil;

[0018] S431, model construction, using radial basis function neural network to build a model, the model has a three-layer network consisting of an input layer, a hidden layer and an output layer; the number of neurons in the input layer is equal to the dimension of the spectral data, the hidden layer uses a Gaussian function as a radial basis function, and the number of neurons in the output layer is equal to the number of target variables to be predicted, that is, furfural content;

[0019] S432, model training, the experimental data is divided into a training set and a test set; 5-fold cross validation is used to evaluate the stability and generalization ability of the model; all spectral data in the sample are introduced into the ANN model as input neurons, and passed to the hidden layer through the input layer. The hidden layer uses the radial basis function to calculate the value of the activation function; the output layer also provides the prediction results, calculates the error of the output layer, and adjusts the weights and center points through back propagation;

[0020] S433, model validation, using cross-validation statistical methods to determine the accuracy and skewness of the model;

[0021] S434. Estimate the effectiveness of furfural terahertz characteristic spectrum in the model;

[0022] S435. Apply the trained model to new insulating oil samples to predict the furfural content; interpret the furfural content in the sample based on the model's prediction results; adjust the model parameters based on the deviation between the model's prediction results and the actual concentration to further optimize the model's prediction ability.

[0023] In particular, S1 specifically includes the following steps:

[0024] S11, sample collection, collecting insulating oil samples from transformers in different aging states;

[0025] S12, pre-treatment, pre-treating the collected samples to remove impurities that affect chromatographic analysis;

[0026] S13, chromatographic analysis, using gas chromatography or liquid chromatography to determine the furfural content in the sample;

[0027] S14, data analysis, recording and analyzing the chromatogram to determine the furfural content in the insulating oil in different aging states;

[0028] S15, standard solution preparation, preparing a series of furfural standard solutions with known concentrations for establishing a standard curve;

[0029] S16. Sample dilution: Based on the chromatographic analysis results, the insulating oil samples in different aging states are diluted to a suitable concentration range to ensure an obvious response within the terahertz spectrum measurement range, and the furfural content in the samples covers the range of furfural content from new oil to highly aged oil products, so as to simulate the various aging states encountered by the insulating oil in actual use.

[0030] In particular, S2 specifically includes the following steps:

[0031] S21. Instrument preparation: ensure that the transmission terahertz time-domain spectroscopy system is in good working condition and check whether all components are operating normally;

[0032] S22, sample loading, placing the sample prepared in S16 into the sample pool of the terahertz spectroscopy system, ensuring that the sample is positioned correctly and stably;

[0033] S23, sample measurement, performing terahertz time-domain spectroscopy scanning on samples of different concentrations in turn, recording the time-domain spectral data of each sample, and repeating the measurement for each sample multiple times to improve the reliability and accuracy of the data;

[0034] S24, data preprocessing, preprocessing the collected raw time-domain spectral data, including baseline correction and noise removal, to improve data quality;

[0035] S25, time domain to frequency domain conversion, using Fourier transform to convert the time domain signal into a frequency domain signal;

[0036] S26, calculating the transmittance by the ratio of the frequency domain signals;

[0037] S27, calculating the complex refractive index by the transmittance and the sample thickness;

[0038] S28, the dielectric constant is calculated by the square of the complex refractive index;

[0039] S29. The absorption coefficient is calculated by the imaginary part of the complex refractive index.

[0040] In particular, S34 includes the following steps:

[0041] S341, data comparison and analysis;

[0042] Frequency matching: Match the characteristic peak frequency in the experimental data with the characteristic frequency in the simulation results; compare the relative intensity of the characteristic peaks in the experimental data and simulation results, and analyze the consistency between the two;

[0043] Spectral overlap: Draw the spectra of experimental data and simulation results to visually compare the similarities and differences between the two; calculate the correlation coefficient between experimental data and simulation results, evaluate the correlation between the two, and perform correlation analysis;

[0044] S342, specific interpretation of characteristic peaks;

[0045] Sources of characteristic peaks: Based on simulation results, explain the vibration mode corresponding to each characteristic peak in the experimental data; analyze the contribution of different functional groups to the characteristic peaks;

[0046] Effects of small changes: Analyze the effects of small changes in molecular structure, i.e. rotation, stretching, asynchronous stretching and vibration, on characteristic peaks to explain subtle differences in experimental data;

[0047] S343, increasing the weight of the characteristic peak;

[0048] Characteristic peak selection: According to the experimental data and simulation results, the most representative characteristic peaks for the detection of furfural content, i.e. important characteristic peaks, are selected; higher weights are assigned to these characteristic peaks to increase their importance in the quantitative analysis model;

[0049] Model construction: Use machine learning methods to build quantitative analysis models; in feature engineering, increase the weights of important feature peaks to ensure that the model can better capture these features;

[0050] S344, Model verification and optimization;

[0051] Cross-validation: Divide the experimental data into training sets and test sets for cross-validation; evaluate the predictive ability and generalization of the model, and calculate the mean square error and determination coefficient indicators;

[0052] Hyperparameter tuning: Use the grid search method to optimize the model's hyperparameters and improve the model's performance; select the model with the best performance for practical applications.

[0053] In particular, in S33, the small changes in the molecular structure include rotation, stretching, asynchronous stretching, and vibration.

[0054] The beneficial effects of the present invention are:

[0055] 1. Use the terahertz time-domain spectroscopy system to scan samples, without the need to extract samples or use chemical reagents, avoiding the sample processing link. It has the advantages of non-destructive and rapid detection, can achieve real-time on-site monitoring, and is helpful for early warning and status assessment of equipment failures.

[0056] 2. The furfural content was determined by chromatography, and a series of samples of known concentration were prepared, which could be used as standard references for subsequent terahertz spectroscopy analysis. Quantum chemical simulation was used to analyze the characteristic spectra of different aging markers, thereby further improving the measurement accuracy of the characteristic spectra of the mixture and ensuring the accuracy and theoretical support of the experimental data.

[0057] 3. By preparing gradient samples and adjusting the concentration ratio, combining quantum chemical simulation and building a neural network model, the characteristic information in the overlapping spectra can be separated and extracted, thereby achieving accurate analysis of the mixture. It has strong anti-interference ability and can effectively eliminate the interference of other aging products (such as methanol and water) on the test results through algorithm correction, with high detection accuracy.

[0058] 4. Based on the propagation theory of terahertz waves in the medium, a characteristic parameter model of oil-paper aging was established, and the accuracy of samples with different aging degrees was verified through experiments. The quantitative analysis of multivariate mixtures was combined with a multi-layer neural network algorithm to achieve accurate characterization of aging markers. The degree of aging can be directly analyzed. The terahertz time-domain spectrum can directly identify the characteristic absorption peaks of aging products such as furfural, and can accurately reflect the aging state of insulating oil.

[0059] 5. By combining experimental measurement data with deep learning models, a prediction model for furfural concentration in insulating oil is constructed, which realizes the quantitative conversion from spectral data to furfural content, and improves the automation and accuracy of analysis. It can accurately evaluate the life of insulating oil, capture subtle changes in furfural content in insulating oil, and provide accurate assessment of the health status of insulating oil. Combined with trace water content detection, it can fully understand the aging and life status of the transformer. DETAILED DESCRIPTION

[0060] The present invention will be further described below in conjunction with embodiments:

[0061] A method for detecting furfural in insulating oil based on terahertz time-domain spectroscopy comprises the following steps:

[0062] S1. For insulating oils in different aging states, the furfural content is determined by a chromatographic analysis method, and samples of corresponding concentrations are prepared; specifically comprising the following steps:

[0063] S11, sample collection, collecting insulating oil samples from transformers in different aging states;

[0064] S12, pre-treatment, pre-treating the collected samples to remove impurities that affect chromatographic analysis; wherein the pre-treatment includes filtration, dilution, etc.;

[0065] S13, chromatographic analysis, using gas chromatography or liquid chromatography to determine the furfural content in the sample; wherein the selection of chromatographic column, the ratio of mobile phase, the type of detector and other parameters can be optimized according to specific experimental conditions;

[0066] S14, data analysis, recording and analyzing the chromatogram to determine the furfural content in the insulating oil in different aging states;

[0067] S15, standard solution preparation, preparing a series of furfural standard solutions with known concentrations for establishing a standard curve;

[0068] S16. Sample dilution: Based on the chromatographic analysis results, the insulating oil samples in different aging states are diluted to a suitable concentration range to ensure an obvious response within the terahertz spectrum measurement range, and the furfural content in the samples covers the range of furfural content from new oil to highly aged oil products, so as to simulate the various aging states encountered by the insulating oil in actual use.

[0069] S2. According to the liquid properties and terahertz spectrum characteristics of the insulating oil, a transmission terahertz time-domain spectroscopy system is used to scan the sample, obtain the time-domain spectrum data of the sample, and calculate the characteristic parameters of absorption coefficient, dielectric constant, and transmittance. The absorption coefficient, dielectric constant, and transmittance reflect the characteristic information of the furfural aging marker in the sample;

[0070] The specific steps include:

[0071] S21. Instrument preparation: ensure that the transmission terahertz time-domain spectroscopy system is in good working condition and check whether all components are operating normally;

[0072] S22, sample loading, placing the sample prepared in S16 into the sample pool of the terahertz spectroscopy system, ensuring that the sample is positioned correctly and stably;

[0073] S23, sample measurement, performing terahertz time-domain spectroscopy scanning on samples of different concentrations in turn, recording the time-domain spectral data of each sample, and repeating the measurement for each sample multiple times to improve the reliability and accuracy of the data; wherein, before measuring the sample, the background spectrum of the blank sample can be measured first, so as to perform background subtraction in subsequent data processing;

[0074] S24, data preprocessing, preprocessing the collected raw time-domain spectral data, including baseline correction and noise removal, to improve data quality;

[0075] S25, time domain to frequency domain conversion, using Fourier transform to convert the time domain signal into a frequency domain signal;

[0076] E s ( oh )= F [ E s ( t )];

[0077] E r ( oh )= F [ E r ( t )];

[0078] In the formula, Es ( t ) is the time domain transmission signal;

[0079] E r ( t ) is the reference signal;

[0080] E s ( oh ) is the frequency domain transmission signal,

[0081] E r ( oh ) is the frequency domain reference signal;

[0082] S26. Calculate the transmittance by the ratio of the frequency domain signal T ( oh );

[0083] T ( oh )= E s ( oh ) / E r ( oh );

[0084] S27, complex refractive index calculated by transmittance and sample thickness n ( oh )+ I ( oh );

[0085] ;

[0086] In the formula, d is the thickness of the sample;

[0087] c It is the speed of light;

[0088] T 0( oh ) is the transmittance of the empty sample cell;

[0089] n ( oh ) is the real part of the refractive index;

[0090] k ( oh ) is the imaginary part of the refractive index;

[0091] i is an imaginary unit;

[0092] S28, the dielectric constant is calculated by the square of the complex refractive index;

[0093] ϵ( oh ) = ( n ( oh )+ I ( oh ) 2 ;

[0094] Then we get the real part of the dielectric constant ϵ′( oh ) and the imaginary part ϵ″ ( oh );

[0095] S29, calculate the absorption coefficient by the imaginary part of the complex refractive index ;

[0096] ;

[0097] In the formula, l is the wavelength, ;

[0098] Through the above steps, characteristic parameters such as absorption coefficient, dielectric constant and transmittance can be extracted from the time-domain spectral data, and then the content and aging degree of furfural in the sample can be analyzed.

[0099] S3. Use quantum chemical simulation methods to simulate the absorption and reflection characteristics of furfural molecules in the terahertz frequency band, and match the vibration mode, rotation energy level and terahertz band absorption characteristics in the real spectral signal.

[0100] Specifically, the quantum chemical method is a computational method based on the principles of quantum mechanics, which is used to study the electronic structure and properties of molecules; through quantum chemical simulation, the absorption and reflection characteristics of furfural molecules in the terahertz band can be predicted. These characteristics include absorption coefficient, dielectric constant, etc. Simulation can reveal the vibration modes of furfural molecules in the terahertz band, which correspond to the atomic motion inside the molecule, such as stretching and bending. The simulation results can be used as a theoretical reference to help explain the terahertz spectral data observed in the experiment. By comparing the experimental data and simulation results, the accuracy and reliability of the experiment can be verified; simulation can provide the characteristic frequencies of furfural molecules in the terahertz band, which can be used to identify and quantitatively analyze the furfural content in the sample. Through simulation, we can understand which frequency ranges are most sensitive to the detection of furfural molecules, thereby optimizing the experimental conditions and selecting the best measurement frequency. The simulation results can also guide the preparation of samples to ensure that the response of the sample at a specific frequency is maximized. In practical applications, insulating oil samples may contain a variety of aging products. Simulation can help analyze the contributions of different components in these complex systems and separate the characteristic signals of furfural. Simulation can consider the effects of environmental factors such as temperature and pressure on the terahertz properties of furfural molecules, helping to understand the impact of experimental conditions on the results.

[0101] The specific steps are as follows:

[0102] S31, establishing a three-dimensional molecular model for the furfural molecule according to the molecular formula, and for water containing hydroxyl groups, supplementing its adjacent hydroxyl groups to form a furfural molecular structure model containing hydrogen bonds;

[0103] S32. Based on the Gaussian orbital or plane wave function assumption, density functional theory quantum chemical calculations were performed to calculate the vibration modes, rotational energy levels and corresponding terahertz absorption characteristics of furfural molecules;

[0104] Vibrational modes: Calculate the vibrational modes of molecules and obtain vibrational frequencies and infrared activities;

[0105] Rotational energy level: calculate the rotational energy level of a molecule and obtain the rotational constant;

[0106] Terahertz absorption characteristics: Calculate the absorption spectrum of molecules in the terahertz band;

[0107] The calculated vibration mode and absorption spectrum are used as theoretical references and compared with experimental data to verify the accuracy and reliability of the experiment. The characteristic frequency can be used to identify and quantitatively analyze the furfural content in the sample. According to the calculation results, the optimal measurement frequency and experimental conditions are selected to improve the sensitivity and accuracy of the detection.

[0108] Through the above steps, quantum chemical calculations can be performed based on density functional theory to obtain the vibration modes, rotational energy levels and their corresponding terahertz absorption characteristics of furfural molecules, providing theoretical support for subsequent experimental analysis.

[0109] S33. In-depth analysis of how the chemical bonds and functional groups in the furfural molecule affect the terahertz spectrum, analysis of the specific absorption characteristics of hydroxyl and aldehyde functional groups in the terahertz frequency band, and how these characteristics change with slight changes in the molecular structure;

[0110] Small changes in molecular structure include rotation, stretching, asynchronous stretching, and vibration;

[0111] Specific absorption characteristics of hydroxyl and aldehyde functional groups in the terahertz frequency band:

[0112] Hydroxyl (-OH);

[0113] Vibration mode: The vibration mode of hydroxyl group mainly includes stretching vibration (OH stretching) and bending vibration (OH bending).

[0114] Absorption in the terahertz band: The stretching vibration of hydroxyl groups is usually at higher frequencies (such as the infrared region), but in the terahertz band, the bending vibration of hydroxyl groups and the vibration mode of hydrogen bonds formed with water molecules can produce significant absorption.

[0115] Characteristic frequency: The bending vibration of hydroxyl groups is usually in the range of 1-2 THz.

[0116] Aldehyde group (-CHO);

[0117] Vibration modes: The vibration modes of aldehyde groups mainly include CH stretching, CO stretching and CH bending.

[0118] Absorption in the terahertz band: The CH stretching and CO stretching of the aldehyde group also have significant absorption in the terahertz band, especially the CH stretching vibration.

[0119] Characteristic frequencies: The CH stretching vibration of the aldehyde group is usually in the range of 2-4 THz, and the CO stretching vibration is in the range of 1-2 THz.

[0120] Effects of small changes in molecular structure:

[0121] Rotation: Influence, the rotation of molecules will affect their absorption characteristics in the terahertz band; the rotation constant (B) is related to the rotational inertia of the molecule. A smaller rotational inertia will lead to a higher rotational energy level spacing, thus affecting the distribution of the absorption spectrum. Analysis, by calculating the rotational constant of the molecule, the effect of rotation on terahertz absorption can be predicted.

[0122] Stretching: Influence, the stretching vibration of chemical bonds in molecules will directly affect their absorption characteristics in the terahertz band; for example, the stretching vibration of CH and OH bonds will produce absorption peaks at specific frequencies. Analysis, by calculating the frequencies of different stretching vibration modes, the impact of these modes on terahertz absorption can be predicted.

[0123] Asynchronous stretching: Influence, asynchronous stretching refers to the stretching vibration of multiple atoms in a molecule with different amplitudes at the same time; this vibration mode will produce complex absorption spectra in the terahertz band. Analysis, by calculating the frequency and intensity of asynchronous stretching vibration modes, the influence of these modes on terahertz absorption can be predicted.

[0124] Vibration: Influence, the vibration of chemical bonds in a molecule affects its absorption characteristics in the terahertz band; for example, the vibration of CC, CO, and CH bonds will produce absorption peaks at specific frequencies. Analysis, by calculating the frequency and intensity of different vibration modes, the impact of these modes on terahertz absorption can be predicted.

[0125] S34. Compare the experimentally obtained terahertz spectrum data with the quantum chemical simulation results, verify the accuracy of the simulation model through such comparison, explain the specificity of the characteristic peaks observed in the experiment, and improve the predictive ability and generalization of the quantitative analysis model by increasing the weight of the characteristic peaks; comprising the following steps:

[0126] Experimental data: The terahertz spectrum data of the furfural sample was re-obtained from the terahertz time-domain spectroscopy system, and the experimental data was preprocessed, including baseline correction and noise removal, to improve the data quality;

[0127] Simulation data: Extract the vibration mode and absorption spectrum data of furfural molecules in the terahertz band from the output files of quantum chemistry simulation software (such as Gaussian, ORCA, etc.), and convert the vibration frequency from cm -1 Convert to THz;

[0128] S341, data comparison and analysis;

[0129] Frequency matching: Match the characteristic peak frequency in the experimental data with the characteristic frequency in the simulation results; compare the relative intensity of the characteristic peaks in the experimental data and simulation results, and analyze the consistency between the two;

[0130] Spectral overlap: Draw the spectra of experimental data and simulation results to visually compare the similarities and differences between the two; calculate the correlation coefficient between experimental data and simulation results, evaluate the correlation between the two, and perform correlation analysis;

[0131] S342, specific interpretation of characteristic peaks;

[0132] Sources of characteristic peaks: Based on the simulation results, explain the vibration mode corresponding to each characteristic peak in the experimental data (such as bending vibration of hydroxyl group, CH stretching vibration of aldehyde group, etc.); analyze the contribution of different functional groups (such as hydroxyl group, aldehyde group) to the characteristic peaks;

[0133] Effects of small changes: Analyze the effects of small changes in molecular structure, i.e. rotation, stretching, asynchronous stretching and vibration, on characteristic peaks to explain subtle differences in experimental data;

[0134] S343, increasing the weight of the characteristic peak;

[0135] Characteristic peak selection: According to the experimental data and simulation results, the most representative characteristic peaks for the detection of furfural content, i.e. important characteristic peaks, are selected; higher weights are assigned to these characteristic peaks to increase their importance in the quantitative analysis model;

[0136] Model construction: Use machine learning methods (such as support vector machines (SVMs), neural networks, etc.) to build quantitative analysis models; in feature engineering, increase the weights of important feature peaks to ensure that the model can better capture these features;

[0137] S344, Model verification and optimization;

[0138] Cross-validation: Divide the experimental data into training sets and test sets for cross-validation; evaluate the predictive ability and generalization of the model, and calculate the mean square error and determination coefficient indicators;

[0139] Hyperparameter tuning: Use the grid search method to optimize the model's hyperparameters and improve the model's performance; select the model with the best performance for practical applications.

[0140] S4. Quantitatively detect the furfural content in insulating oil by reconstructing the terahertz time domain signal;

[0141] S41. For the sample containing only furfural obtained in S2, terahertz characteristic parameters, i.e., absorption coefficient, dielectric constant, and transmittance, are extracted to accurately characterize the corresponding furfural content; furfural concentration is taken as the dependent variable, spectral characteristics are taken as the independent variable, a relationship model is established, and quantum chemical simulation terahertz characteristic spectrum is introduced to assist in determining the characteristic spectral peak corresponding to its unique chemical vibration as an auxiliary variable; specifically, the following steps are included:

[0142] Signal processing and reconstruction:

[0143] Raw data: obtain time domain signals of furfural samples with different concentrations;

[0144] Fourier transform: convert time domain signal into frequency domain signal;

[0145] Preprocessing: baseline correction and noise removal are performed on frequency domain signals.

[0146] Feature parameter extraction:

[0147] Absorption Coefficient: Calculate the absorption coefficient;

[0148] Dielectric Constant: Calculate the dielectric constant;

[0149] Transmittance: Calculates the transmittance.

[0150] Quantum Chemistry Simulation:

[0151] Molecular modeling: construct a three-dimensional model of furfural molecules;

[0152] Geometry optimization: perform geometry optimization;

[0153] Vibration Mode Calculation: Calculate vibration modes and extract characteristic frequencies and intensities.

[0154] To model the relationship:

[0155] Data preparation: organize experimental data and simulation data;

[0156] Feature selection: select important features and auxiliary variables;

[0157] Model selection: Build regression models using either SVM or neural networks.

[0158] Model training and validation:

[0159] Data segmentation: Divide the experimental data into training set and test set;

[0160] Standardization: standardize the feature data;

[0161] Training model: Use training set data to train the model and adjust model parameters;

[0162] Hyperparameter tuning: Optimize the model’s hyperparameters using grid search;

[0163] Test set evaluation: Use the test set data to evaluate the model’s predictive ability and generalization.

[0164] Quantitative detection:

[0165] Model application: Apply the trained model to new insulating oil samples to predict the furfural content;

[0166] Result interpretation: Based on the model's predictions, explain the furfural content in the sample.

[0167] S42, constructing a pure insulating oil sample containing a gradient distribution of furfural, and then performing terahertz time-domain spectroscopy detection on the sample;

[0168] The specific steps include:

[0169] Sample preparation:

[0170] Pure insulating oil: Choose a pure insulating oil as the base;

[0171] Furfural solution: prepare furfural solutions of different concentrations such as 0.1%, 0.5%, 1.0%, 2.0%, and 3.0%;

[0172] Mixed samples: Add furfural solutions of different concentrations into pure insulating oil in proportion to prepare samples with gradient distribution;

[0173] Stirring and Standing: Make sure each sample is stirred thoroughly during the mixing process and allowed to stand for a period of time.

[0174] Terahertz time-domain spectroscopy detection:

[0175] Sample pool: Choose a suitable sample pool;

[0176] Instrument preparation: Ensure that the terahertz time-domain spectroscopy system is in good working condition;

[0177] Background measurement: measure the background spectrum of blank samples;

[0178] Sample measurement: Furfural insulating oil samples with different concentrations were scanned by terahertz time-domain spectroscopy in turn, and the time-domain spectral data of each sample was recorded.

[0179] Data processing and analysis:

[0180] Fourier transform: convert time domain signal into frequency domain signal;

[0181] Characteristic parameter extraction: calculation of absorption coefficient, dielectric constant and transmittance.

[0182] S43, using the neural network model ANN model to quantitatively analyze the furfural content in insulating oil;

[0183] S431, model construction, using radial basis function neural network to build a model, the model has a three-layer network consisting of an input layer, a hidden layer and an output layer; the number of neurons in the input layer is equal to the dimension of the spectral data, the hidden layer uses a Gaussian function as a radial basis function, and the number of neurons in the output layer is equal to the number of target variables to be predicted, that is, furfural content;

[0184] S432, model training, the experimental data is divided into a training set and a test set; 5-fold cross validation is used to evaluate the stability and generalization ability of the model; all spectral data in the sample are introduced into the ANN model as input neurons, and passed to the hidden layer through the input layer. The hidden layer uses the radial basis function to calculate the value of the activation function; the output layer also provides the prediction results, calculates the error of the output layer, and adjusts the weights and center points through back propagation;

[0185] S433, model validation, using cross-validation statistical methods to determine the accuracy and skewness of the model;

[0186] S434. Estimate the effectiveness of furfural terahertz characteristic spectrum in the model;

[0187] S435. Apply the trained model to new insulating oil samples to predict the furfural content; interpret the furfural content in the sample based on the model's prediction results; adjust the model parameters based on the deviation between the model's prediction results and the actual concentration to further optimize the model's prediction ability.

[0188] In the above, a furfural concentration prediction model in insulating oil is constructed by combining experimental measurement data with deep learning models. The model predicts the furfural content in unknown samples by comparing the data of known samples. For the test results, the concentration value of furfural content is obtained through model analysis, which provides a reference for the status evaluation of insulating oil.

[0189] The accuracy and theoretical support of the experimental data were ensured by chromatography and quantum chemical simulation. Terahertz spectroscopy provides the optical properties of the sample in a wide frequency band, which helps to fully understand the properties of the sample. Terahertz spectroscopy is a non-destructive detection method suitable for multiple detections and long-term monitoring. By establishing a neural network model, the quantitative conversion from spectral data to furfural content was achieved, which improved the automation and accuracy of the analysis.

[0190] The above is an exemplary description of the present invention. Obviously, the specific implementation of the present invention is not limited to the above-mentioned method. As long as various improvements are made using the method concept and technical solution of the present invention, or they are directly applied to other occasions without improvement, they are all within the protection scope of the present invention.

Claims

1. A method for detecting furfural in insulating oil based on terahertz time-domain spectroscopy, characterized in that: The following steps are involved: S1. For insulating oils in different aging states, the furfural content is determined by chromatography and samples of corresponding concentrations are prepared; S2. According to the liquid properties and terahertz spectrum characteristics of the insulating oil, a transmission terahertz time-domain spectroscopy system is used to scan the sample, obtain the time-domain spectrum data of the sample, and calculate the characteristic parameters of absorption coefficient, dielectric constant, and transmittance. The absorption coefficient, dielectric constant, and transmittance reflect the characteristic information of the furfural aging marker in the sample; S3. Use quantum chemical simulation methods to simulate the absorption and reflection characteristics of furfural molecules in the terahertz band, and match the vibration mode, rotation energy level and terahertz band absorption characteristics in the real spectral signal; S31, establishing a three-dimensional molecular model for the furfural molecule according to the molecular formula, and for water containing hydroxyl groups, supplementing its adjacent hydroxyl groups to form a furfural molecular structure model containing hydrogen bonds; S32. Based on the Gaussian orbital or plane wave function assumption, density functional theory quantum chemical calculations were performed to calculate the vibration modes, rotational energy levels and corresponding terahertz absorption characteristics of furfural molecules; S33. In-depth analysis of how the chemical bonds and functional groups in the furfural molecule affect the terahertz spectrum, analysis of the specific absorption characteristics of hydroxyl and aldehyde functional groups in the terahertz frequency band, and how these characteristics change with slight changes in the molecular structure, including rotation, stretching, asynchronous stretching and vibration; S34. Compare the experimentally obtained terahertz spectral data with the quantum chemical simulation results. Through this comparison, verify the accuracy of the simulation model and explain the specificity of the characteristic peaks observed in the experiment. By increasing the weight of the characteristic peaks, improve the predictive ability and generalization of the quantitative analysis model; S4. Quantitatively detect the furfural content in insulating oil by reconstructing the terahertz time domain signal; S41. For the sample containing only furfural obtained in S2, terahertz characteristic parameters, i.e., absorption coefficient, dielectric constant, and transmittance, are extracted to accurately characterize the corresponding furfural content; furfural concentration is taken as the dependent variable, and spectral characteristics are taken as independent variables, a relationship model is established, and quantum chemical simulation terahertz characteristic spectrum is introduced to assist in determining the characteristic spectral peak corresponding to its unique chemical vibration as an auxiliary variable; S42, constructing a pure insulating oil sample containing a gradient distribution of furfural, and then performing terahertz time-domain spectroscopy detection on the sample; S43, using the neural network model ANN model to quantitatively analyze the furfural content in insulating oil; S431, model construction, using radial basis function neural network to build a model, the model has a three-layer network consisting of an input layer, a hidden layer and an output layer; the number of neurons in the input layer is equal to the dimension of the spectral data, the hidden layer uses a Gaussian function as a radial basis function, and the number of neurons in the output layer is equal to the number of target variables to be predicted, that is, furfural content; S432, model training, the experimental data is divided into a training set and a test set; 5-fold cross validation is used to evaluate the stability and generalization ability of the model; all spectral data in the sample are introduced into the ANN model as input neurons, and passed to the hidden layer through the input layer. The hidden layer uses the radial basis function to calculate the value of the activation function; the output layer also provides the prediction results, calculates the error of the output layer, and adjusts the weights and center points through back propagation; S433, model validation, using cross-validation statistical methods to determine the accuracy and skewness of the model; S434. Estimate the effectiveness of furfural terahertz characteristic spectrum in the model; S435. Apply the trained model to new insulating oil samples to predict the furfural content; interpret the furfural content in the sample based on the model's prediction results; adjust the model parameters based on the deviation between the model's prediction results and the actual concentration to further optimize the model's prediction ability.

2. The method for detecting furfural in insulating oil based on terahertz time-domain spectroscopy according to claim 1, characterized in that: S1 specifically includes the following steps: S11, sample collection, collecting insulating oil samples from transformers in different aging states; S12, pre-treatment, pre-treating the collected samples to remove impurities that affect chromatographic analysis; S13, chromatographic analysis, using gas chromatography or liquid chromatography to determine the furfural content in the sample; S14, data analysis, recording and analyzing the chromatogram to determine the furfural content in the insulating oil in different aging states; S15, standard solution preparation, preparing a series of furfural standard solutions with known concentrations for establishing a standard curve; S16. Sample dilution: Based on the chromatographic analysis results, the insulating oil samples in different aging states are diluted to a suitable concentration range to ensure an obvious response within the terahertz spectrum measurement range, and the furfural content in the samples covers the range of furfural content from new oil to highly aged oil products, so as to simulate the various aging states encountered by the insulating oil in actual use.

3. The method for detecting furfural in insulating oil based on terahertz time-domain spectroscopy according to claim 2, characterized in that: S2 specifically includes the following steps: S21. Instrument preparation: ensure that the transmission terahertz time-domain spectroscopy system is in good working condition and check whether all components are operating normally; S22, sample loading, placing the sample prepared in S16 into the sample pool of the terahertz spectroscopy system, ensuring that the sample is positioned correctly and stably; S23, sample measurement, performing terahertz time-domain spectroscopy scanning on samples of different concentrations in turn, recording the time-domain spectral data of each sample, and repeating the measurement for each sample multiple times to improve the reliability and accuracy of the data; S24, data preprocessing, preprocessing the collected raw time-domain spectral data, including baseline correction and noise removal, to improve data quality; S25, time domain to frequency domain conversion, using Fourier transform to convert the time domain signal into a frequency domain signal; S26, calculating the transmittance by the ratio of the frequency domain signals; S27, calculating the complex refractive index by the transmittance and the sample thickness; S28, the dielectric constant is calculated by the square of the complex refractive index; S29. The absorption coefficient is calculated by the imaginary part of the complex refractive index.

4. The method for detecting furfural in insulating oil based on terahertz time-domain spectroscopy according to claim 1, characterized in that: S34 includes the following steps: S341, data comparison and analysis; Frequency matching: Match the characteristic peak frequency in the experimental data with the characteristic frequency in the simulation results; compare the relative intensity of the characteristic peaks in the experimental data and simulation results, and analyze the consistency between the two; Spectral overlap: Draw the spectra of experimental data and simulation results to visually compare the similarities and differences between the two; calculate the correlation coefficient between experimental data and simulation results, evaluate the correlation between the two, and perform correlation analysis; S342, specific interpretation of characteristic peaks; Sources of characteristic peaks: Based on simulation results, explain the vibration mode corresponding to each characteristic peak in the experimental data; analyze the contribution of different functional groups to the characteristic peaks; Effects of small changes: Analyze the effects of small changes in molecular structure, i.e. rotation, stretching, asynchronous stretching and vibration, on characteristic peaks to explain subtle differences in experimental data; S343, increasing the weight of the characteristic peak; Characteristic peak selection: According to the experimental data and simulation results, the most representative characteristic peaks for the detection of furfural content, i.e. important characteristic peaks, are selected; higher weights are assigned to these characteristic peaks to increase their importance in the quantitative analysis model; Model construction: Use machine learning methods to build quantitative analysis models; in feature engineering, increase the weights of important feature peaks to ensure that the model can better capture these features; S344, Model verification and optimization; Cross-validation: Divide the experimental data into training sets and test sets for cross-validation; evaluate the predictive ability and generalization of the model, and calculate the mean square error and determination coefficient indicators; Hyperparameter tuning: Use the grid search method to optimize the model's hyperparameters and improve the model's performance; select the model with the best performance for practical applications.