Cable overheating fault detection method and device based on electronic nose technology

Abstract

The invention discloses a cable overheat fault detection method and device based on an electronic nose technology. The method comprises the following steps: acquiring gas production results of various organic volatile gases generated by various cables of different materials and models in a power cable overheat fault simulation device along with temperature changes; performing feature extraction on the gas production result, and determining feature gas of an overheating fault by using a clustering algorithm according to a feature extraction result; respectively measuring by utilizing a gas sensor array to obtain quantitative concentration data of all optimal characteristic gas compounds in all cable samples along with temperature change, and obtaining a training sample; and training a neural network by using the training sample to obtain a cable overheating fault early recognition model which is used for outputting a corresponding real-time temperature prediction value according to the type of the cable to be detected and the type and concentration of the generated optimal characteristic gas so as to perform early recognition on the cable overheating fault.

Description

Technical Field

[0001] This application relates to methods and apparatus for detecting overheating faults, and more particularly to a method and apparatus for detecting cable overheating faults based on electronic nose technology. Background Technology

[0002] Cable overheating is a major hidden danger to the safe operation of power systems, and its early and accurate detection is crucial for preventing equipment damage and power grid accidents. Cable overheating will accelerate the aging of insulation materials, which may lead to single-phase grounding or even phase-to-phase short circuits. In addition, due to the narrow space of cable trenches, poor ventilation, and high cable laying density, cable overheating faults may further evolve into more serious fire accidents.

[0003] The current industry's traditional sampling methods generally include the following process: (1) discrete sampling to obtain the gas concentrations of 5 to 7 preset fixed gases; (2) based on the sampling results, diagnosing the fault type based on rules / thresholds; (3) indirectly inferring the approximate temperature range of the fault based on the sampling results. This traditional method has significant limitations, mainly reflected in:

[0004] 1. Analyzing only 5 to 7 preset fixed gases (such as H2, CH4, C2H2, etc.) may ignore the key gases of new insulating materials, and it relies heavily on historical thresholds or fixed ratio rules, making it difficult to accurately predict real-time temperature.

[0005] 2. In terms of monitoring coverage, traditional methods are not good at identifying key chemical signals. Since they mainly rely on the concentration data of the generated gas and the comparison of thresholds to diagnose the fault type, they have low sensitivity to oxygen-containing polar compounds (such as aldehydes and ketones). These substances are early markers of cellulose insulation materials in the low-temperature overheating stage, which can easily lead to inaccurate monitoring results. Summary of the Invention

[0006] The purpose of this invention is to solve the aforementioned problems in the prior art and application, and to provide a cable overheating fault detection method and device based on electronic nose technology. This method accurately determines the optimal characteristic gaseous compounds that can be used for early detection of cable overheating faults by extracting temperature response characteristics and chemical characteristics. By detecting the concentration of the optimal characteristic gaseous compounds released by the cable due to excessive temperature, an early, accurate and effective warning of cable overheating faults can be achieved.

[0007] The present invention adopts the following technical solution.

[0008] According to a first aspect of the present invention, a method for detecting cable overheating faults based on electronic nose technology is provided. The method includes the following steps:

[0009] To obtain the gas production results of various organic volatile gases generated by different cables of different materials and models in a power cable overheating fault simulation device as a function of temperature;

[0010] Temperature response features and chemical features are extracted from the gas production results. Based on the feature extraction results, a clustering algorithm is used to determine the optimal set of characteristic gas compounds for overheating faults.

[0011] The concentration of all optimal characteristic gaseous compounds in all cable samples was quantitatively measured with temperature using a gas sensor array to obtain training samples.

[0012] The neural network is trained using the training samples to obtain an early identification model for cable overheating faults. This model is used to output corresponding real-time temperature prediction values ​​based on the type of cable under test and the type and concentration of the optimal characteristic gas compounds generated, so as to identify cable overheating faults at an early stage.

[0013] Furthermore, the step of extracting features from the gas production results and determining the characteristic gas set of the overheating fault using a clustering algorithm based on the feature extraction results includes:

[0014] Obtain gas production results, including total ion chromatograms and mass spectra obtained by GC-MS and infrared absorption spectra obtained by FTIR;

[0015] The compound set of gaseous compounds produced by each cable sample was determined based on the total ion chromatogram, mass spectrum and infrared absorption spectrum, and then the quantitative data of the concentration of all gaseous compounds in each cable sample as a function of temperature, the characteristic ion abundance ratio and the functional group index were determined.

[0016] Temperature response vectors are extracted from concentration-quantitative data to construct a temperature-sensitive feature matrix. The temperature-sensitive feature matrix is ​​then clustered to select a set of temperature-sensitive compounds.

[0017] A chemical feature matrix is ​​constructed using the characteristic ion abundance ratio and functional group index of each temperature-sensitive compound at multiple temperature points. The chemical feature matrix is ​​then subjected to secondary clustering to determine the optimal set of characteristic gaseous compounds for overheating faults.

[0018] Furthermore, the determination of the compound set of gaseous compounds generated by each cable sample based on the total ion chromatogram, mass spectrum, and infrared absorption spectrum, and the subsequent determination of the quantitative data set of concentrations of all gaseous compounds in each cable sample as a function of temperature, includes:

[0019] Based on the total ion chromatogram and mass spectrum, the GC-MS concentration matrix of the i-th gaseous compound in the k-th cable sample was obtained by calibration using peak area ratio.

[0020] Based on FTIR quantitative analysis and using Beer-Lambert's law, the FTIR concentration matrix of the i-th gaseous compound in the k-th cable sample was obtained.

[0021] The GC-MS concentration matrix and FTIR concentration matrix of the i-th compound in the k-th cable sample are weighted and summed to obtain the weighted concentration as the corresponding quantitative concentration data.

[0022]

[0023] The weighting parameter w can be adjusted based on the signal-to-noise ratio.

[0024] Furthermore, a temperature response vector is extracted from the concentration quantification data to construct a temperature-sensitive feature matrix. This temperature-sensitive feature matrix is ​​then clustered to identify a set of temperature-sensitive compounds, including:

[0025] Multiple temperature response features of the i-th gaseous compound in the k-th cable sample are extracted from the quantitative concentration data to form the corresponding temperature response vector. Furthermore, based on Constructing the temperature-sensitive feature matrix M T ;

[0026] Based on the temperature-sensitive feature matrix M T Determine the neighborhood radius and the minimum number of minimum neighborhood points for M. T Perform DBSCAN clustering to obtain cluster label vectors; these cluster label vectors are used to indicate the cluster to which each temperature response vector belongs.

[0027] Based on the clustering label vector and the preset screening thresholds corresponding to the multiple temperature response features, temperature-sensitive clusters are identified;

[0028] Select the current gaseous compound to be screened, and count the number of cable samples in all cable samples that are identified as belonging to the temperature-sensitive cluster. When the ratio of this number to the total number of cable samples is greater than the set screening threshold, the current gaseous compound is considered to be a set of temperature-sensitive compounds.

[0029] Furthermore, based on the clustering label vector and the preset screening thresholds corresponding to the multiple temperature response features, temperature-sensitive clusters are identified, including:

[0030] The centroid vector is constructed by calculating the average value of each temperature response feature in all temperature response vectors belonging to the current cluster based on the cluster label vector. When the centroid vector meets the corresponding preset conditions, the current cluster is considered to be a temperature-sensitive cluster.

[0031] Furthermore, the step of considering the current cluster as a temperature-sensitive cluster when the centroid vector satisfies the corresponding preset condition includes:

[0032] When all temperature response features in the centroid vector are greater than the corresponding set threshold, the current cluster is considered to be a temperature-sensitive cluster.

[0033] Furthermore, the temperature response characteristics are set to include: temperature sensitivity coefficient, concentration change rate at a set temperature point T0, and linear significance determination coefficient.

[0034] Furthermore, a chemical feature matrix is ​​constructed using the characteristic ion abundance ratios and functional group indices of each temperature-sensitive compound at multiple temperature points. Secondary clustering is then performed on this chemical feature matrix to determine the optimal set of characteristic gaseous compounds for overheating faults, including:

[0035] The chemical feature vectors of each temperature-sensitive compound are constructed by using the characteristic ion abundance ratios and functional group indices at multiple temperature points, and then the chemical feature matrix is ​​obtained.

[0036] The silhouette coefficient is used to evaluate the clustering quality, and the clustering value that maximizes the silhouette coefficient is selected as the optimal clustering value.

[0037] K-means clustering is performed on the chemical feature matrix based on the optimal clustering value to determine the cluster affiliation label of each temperature-sensitive compound and the center vector of each cluster;

[0038] Based on the assigned tags and the center vectors of each cluster, and combined with the physicochemical properties of each temperature-sensitive compound, the optimal set of characteristic gas compounds for overheating faults is determined.

[0039] According to a second aspect of the present invention, a cable overheating fault detection device based on electronic nose technology, employing the method described in the first aspect of the present invention, is provided. The device includes:

[0040] The acquisition module is used to acquire the gas production results of various organic volatile gases generated by different materials and models of cables in the power cable overheating fault simulation device as a function of temperature.

[0041] The clustering module is used to extract temperature response features and chemical features from the gas production results, and to determine the optimal set of characteristic gas compounds for overheating faults based on the feature extraction results using a clustering algorithm.

[0042] The sample acquisition module is used to measure the quantitative data of the concentration of all the optimal characteristic gas compounds in all cable samples as a function of temperature using a gas sensor array, thereby obtaining training samples.

[0043] The training module is used to train the neural network using the training samples to obtain an early identification model for cable overheating faults. This model is used to output corresponding real-time temperature prediction values ​​based on the type of cable under test and the type and concentration of the optimal characteristic gas compound produced, so as to identify cable overheating faults at an early stage.

[0044] According to a third aspect of the present invention, a terminal is provided, comprising a processor and a storage medium. The storage medium is used to store instructions; the processor is used to operate according to the instructions to perform the steps of the method according to the first aspect of the present invention.

[0045] The beneficial effects of this invention are compared with those of the prior art:

[0046] 1. By acquiring the gas production results of various organic volatile gases generated by different cables of different materials and models in a power cable overheating fault simulation device as a function of temperature, and extracting temperature response features and chemical characteristics, the optimal characteristic gas compounds for overheating faults can be determined using a clustering algorithm based on the extracted features. This is crucial for identifying key characteristic gases of novel insulation materials. Furthermore, by using these optimal characteristic gas compounds to construct training samples and train a neural network, an early identification model for cable overheating faults can be obtained. Real-time temperature monitoring and accurate estimation can then be achieved through gas sensors in conjunction with the trained early identification model, thus enabling better early warning of cable overheating faults.

[0047] 2. By integrating data from two independent analytical techniques, GC-MS and FTIR, the systematic errors of a single technique can be effectively offset, improving the accuracy of concentration measurement. At the same time, the trace detection advantages of GC-MS and the high concentration stability of FTIR can be combined to expand the effective detection range. In addition, cross-validation of the two techniques can reduce the risk of misjudgment and ensure the reliability of the results.

[0048] 3. By extracting temperature response vectors from quantitative concentration data to construct a temperature-sensitive feature matrix for primary clustering, temperature-sensitive compounds are selected. This process quickly identifies a core temperature-sensitive subset from a massive dataset, significantly reducing the scale of subsequent analyses and avoiding redundant calculations. A secondary clustering is then performed using the characteristic ion abundance ratios and functional group indices of each temperature-sensitive compound at multiple temperature points to determine the optimal characteristic gaseous compounds for overheating faults. Cross-validation using chemical information from different dimensions enhances the robustness and accuracy of the results.

[0049] 4. Furthermore, by using DBSCAN clustering in the first clustering and Kmeans clustering in the second clustering, both feature reliability (DBSCAN denoising) and engineering practicality (Kmeans efficient grouping and purification) are taken into account, ensuring accurate and efficient screening of the optimal feature gas compounds.

[0050] 5. The chemical characteristic matrix includes characteristic ion abundance and functional group index. Characteristic ion abundance can capture molecular fragment signals (such as mass spectrometry fragment peaks), while functional group index can resolve the chemical nature of molecules (such as oxygen-containing group characteristics). Through the synergistic application of characteristic ion abundance and functional group index, trace substances below the instrument's detection limit can still be identified, significantly improving the response sensitivity of oxygen-containing polar compounds (such as aldehydes, ketones, and alcohols).

[0051] 6. In the early stages of a cable fire, volatile gases appear before smoke particles. Therefore, using a gas sensor can provide a faster early warning of a fire than a smoke sensor. Gas sensors also have advantages such as low cost and strong resistance to environmental interference. Attached Figure Description

[0052] Figure 1 This is a flowchart of the cable overheating fault detection method based on electronic nose technology of the present invention;

[0053] Figure 2 This is a flowchart of step S2 of the present invention;

[0054] Figure 3 This is a schematic diagram of the cable overheating fault detection device based on electronic nose technology of the present invention. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The embodiments described in this application are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, other embodiments obtained by those skilled in the art without creative effort are all within the protection scope of this invention.

[0056] In the early overheating stage, cable insulation materials release complex mixed vapors when heated, mainly consisting of volatile or semi-volatile organic compounds. This patent proposes using gas sensors to accurately detect early cable fires by identifying characteristic gases in the mixed vapors. This application aims to detect the composition of gases generated by PVC insulation when heated using GC-MS (Gas Chromatography-Mass Spectrometry) and TG-FTIR (Thermogravimetric-Fourier Transform Infrared Spectrometry) systems; identify suitable characteristic gases for early detection of cable fires and quantitatively determine their concentrations; and screen characteristic gas compounds with good gas-sensing properties through sample gas testing for use in cable fire detection and early warning.

[0057] Example 1

[0058] This embodiment provides a method for detecting cable overheating faults based on electronic nose technology.

[0059] like Figure 1 As shown, in one embodiment, the cable overheating fault detection method based on electronic nose technology of the present invention includes the following steps:

[0060] S1. Obtain the gas production results of various organic volatile gases generated by different materials and models of cables in the power cable overheating fault simulation device as a function of temperature.

[0061] In this step, the selected cables should cover as many typical cable materials as possible, as well as different voltage levels and different cross-sectional area specifications, with at least three different models of each material included.

[0062] The power cable overheating fault simulation device can be a conventional thermal fault simulation device in this field, which typically uses a closed reaction vessel, filled with air and with air flow and temperature control to simulate the real cable overheating fault environment.

[0063] The overheating fault simulation process can employ a stepped heating simulation method, with a temperature range of 50–300℃ ± 1℃ and a heating rate of 3℃ / min. Multiple temperature points can be set during the heating process for data sampling and analysis. Furthermore, the generated gas can be analyzed using gas chromatography-mass spectrometry (GC-MS) and / or Fourier transform infrared spectroscopy (FTIR) to obtain the gas production results. GC-MS can obtain the total ion chromatogram and mass spectrum of the generated gas, while FTIR can obtain the infrared absorption spectrum of the generated gas.

[0064] S2. Extract temperature response features and chemical features from the gas production results, and use clustering algorithms to determine the optimal set of characteristic gas compounds for overheating faults based on the feature extraction results.

[0065] The following example, using the simultaneous feature extraction of gas production results from GC-MS and FTIR, illustrates this step in detail. Figure 2 This step specifically includes:

[0066] S21. Obtain gas production results including the total ion chromatogram and mass spectrum obtained by GC-MS and the infrared absorption spectrum obtained by FTIR;

[0067] S22. Based on the total ion chromatogram, mass spectrum and infrared absorption spectrum, determine the compound set of gaseous compounds generated by each cable sample, and then determine the quantitative data of the concentration of all gaseous compounds in each cable sample as a function of temperature, the characteristic ion abundance ratio and the functional group index.

[0068] The compound set can be determined as follows: Based on the GC-MS total ion chromatogram, a list of compounds identified by GC-MS can be obtained through retention time index (RTI) calibration and NIST library matching. Functional groups verified by FTIR can then be obtained through second derivative peak identification and standard library matching, such as C=O, OH, Ar-H, and C=C. The compounds in the GC-MS-identified list are then screened based on the FTIR-verified functional groups to determine the compound set of the generated gas.

[0069] Concentration quantification data can be determined in the following ways:

[0070] First, based on the GC-MS internal standard method (e.g., using deuterated toluene as an internal standard), the GC-MS concentration matrix of the i-th gaseous compound in the k-th cable sample is obtained through peak area ratio calibration.

[0071] Next, based on FTIR quantitative analysis and using Beer-Lambert's law, the FTIR concentration matrix of the i-th gaseous compound in the k-th cable sample was obtained.

[0072] Finally, the GC-MS concentration matrix and FTIR concentration matrix of the i-th compound in the k-th cable sample are weighted and summed to obtain the weighted concentration as the corresponding quantitative concentration data:

[0073]

[0074] The weighting parameter w can be adjusted based on the signal-to-noise ratio.

[0075] By integrating data from two independent analytical techniques, GC-MS and FTIR, the systematic errors of a single technique can be effectively offset, improving the accuracy of concentration measurement. At the same time, the trace detection advantages of GC-MS and the high concentration stability of FTIR can be combined to expand the effective detection range. In addition, cross-validation of the two techniques can reduce the risk of misjudgment and ensure the reliability of the results.

[0076] The characteristic ion abundance ratio can be extracted from the mass spectrum and is expressed as:

[0077]

[0078] Among them, R m / z(i,k) The characteristic ion abundance ratio of the i-th compound in the k-th cable sample. Let m be the ion abundance of the i-th compound in the k-th cable sample at the selected characteristic ion m / z1. Let be the ion abundance of the i-th compound in the k-th cable sample at the selected characteristic ion m / z2.

[0079] The functional group index can be extracted from the infrared absorption spectrum. The specific extraction method can be found in existing technologies and will not be elaborated here.

[0080] S23. Extract the temperature response vector from the concentration quantification data to construct a temperature-sensitive feature matrix, perform a clustering on the temperature-sensitive feature matrix, and filter out a set of temperature-sensitive compounds from the clustering.

[0081] In this step, the DBSCAN clustering method is preferred for primary clustering. It can automatically identify clusters of arbitrary shapes and effectively remove noise points (such as unstable responses), retaining only compounds that are sensitive to temperature changes and have reliable response patterns.

[0082] Step S23 specifically includes:

[0083] S231. Extract multiple temperature response features of the i-th gaseous compound from the k-th cable sample from the concentration quantitative data to form the corresponding temperature response vector. Furthermore, based on Constructing the temperature-sensitive feature matrix M T .

[0084] As an example, this temperature response characteristic may include a temperature sensitivity coefficient, the rate of concentration change at a set temperature, and a linear significance coefficient of determination. Thus, It can be represented as:

[0085]

[0086] Where, β i,k , and These are the standardized temperature sensitivity coefficient, concentration change rate at a set temperature point T0 (T0 can be set as the critical temperature point of 180°C), and linear significance coefficient of determination, respectively, for the i-th compound in the k-th cable sample. This can be achieved by performing a concentration-temperature linear fit on each compound. Standardization can be achieved, for example, using the Z-score standardization method.

[0087] Constructed matrix M T Let M be a (K×I)×3 matrix, where I is the total number of gaseous compounds and K is the total number of cable samples. T Each row is the temperature response vector.

[0088] S232, Based on the temperature-sensitive feature matrix M T Determine the neighborhood radius and the minimum number of minimum neighborhood points for M. T Perform DBSCAN clustering to obtain cluster label vectors; these cluster label vectors are used to indicate the cluster to which each temperature response vector belongs.

[0089] Specifically, the cluster label vector is actually a K×I vector, and each element in the vector corresponds one-to-one with a temperature response vector. As an example, when the cluster label value corresponding to a certain temperature response vector is -1, it indicates that the temperature response vector is a noise point and does not belong to any cluster; when the label value is a non-negative integer 0, 1, 2, ..., it indicates that the temperature response vector belongs to the cluster with the corresponding number.

[0090] S233. Based on the clustering label vector and the preset screening thresholds corresponding to the multiple temperature response features, identify temperature-sensitive clusters.

[0091] This step specifically includes: constructing a centroid vector by calculating the average value of each temperature response feature among all temperature response vectors belonging to the current cluster based on the cluster label vector. When this centroid vector meets the corresponding preset conditions, the current cluster is considered a temperature-sensitive cluster. For example:

[0092] The sensitivity of the centroid vector (i.e., the mean temperature sensitivity) must be greater than the set threshold.

[0093] The rate of change of concentration along the centroid vector (i.e., the mean rate of change of concentration) must be greater than a set threshold; and

[0094] The coefficient of determination (i.e. the mean of the coefficient of determination) of the centroid vector must be greater than the set threshold.

[0095] S234. Select the current gas compound to be screened, and count the number of cable samples in all cable samples that are determined to belong to the temperature-sensitive cluster. When the ratio of this number to the total number of cable samples is greater than the set screening threshold, the current gas compound is considered to be a set of temperature-sensitive compounds.

[0096] As an example, when the ratio of the number of cable samples identified as belonging to the temperature-sensitive cluster to the total number of cable samples is greater than 0.7, the current gaseous compound is considered a temperature-sensitive compound, thus obtaining a set of temperature-sensitive compounds. It should be noted that this screening threshold can be adaptively adjusted based on actual circumstances, expert experience, or optimization algorithms.

[0097] Step S23 involves extracting the temperature response vector from the concentration quantification data to construct a temperature-sensitive feature matrix for clustering. This clustering process identifies temperature-sensitive compounds and allows for the rapid selection of a core temperature-sensitive subset from a vast number of compounds (e.g., selecting a few gaseous compounds from dozens of gaseous compounds). This significantly reduces the scale of subsequent analyses and avoids redundant calculations.

[0098] S24. Construct a chemical feature matrix using the characteristic ion abundance ratio and functional group index of each temperature-sensitive compound at multiple temperature points, and perform secondary clustering on the chemical feature matrix to determine the characteristic gas set of overheating fault.

[0099] In this step, the secondary clustering preferably adopts the K-means clustering method, which can efficiently classify the screened temperature-sensitive compounds. The number of final characteristic gases can be directly controlled by the preset number of clusters (K), and representative compounds (such as typical gases with similar characteristic ion abundance ratios and functional groups) can be quickly selected using the cluster centroids to form a concise and optimal set that covers key chemical characteristics.

[0100] Step S24 specifically includes:

[0101] S241. Construct chemical feature vectors for each temperature-sensitive compound using the characteristic ion abundance ratio and functional group index at multiple temperature points, and then obtain the chemical feature matrix.

[0102] Among them, the constructed chemical feature matrix M C For an I T An N×2 matrix, where each row represents the chemical feature vector of the corresponding temperature-sensitive compound. Where I... T This indicates the number of temperature-sensitive compounds, where N is the number of multiple temperature points, and 2 represents the I at the corresponding temperature point. TThese multiple temperature points need to cover the entire temperature range when set. Additionally, to eliminate dimensional differences, each element in the chemical characteristic matrix is ​​a standardized value (e.g., Z-score normalized).

[0103] S242. Use the silhouette coefficient to evaluate the clustering quality and select the clustering value that maximizes the silhouette coefficient as the optimal clustering value.

[0104] This step may specifically include:

[0105] First, choose a reasonable range of K values, such as 2≤K≤10.

[0106] Next, K-means clustering is performed on each K value, and the silhouette coefficient is calculated. The silhouette coefficient is an indicator of clustering quality, with a value ranging from [-1, 1]. A larger value indicates a better clustering effect. The specific calculation formula can be found in existing techniques and will not be elaborated here.

[0107] Furthermore, a line graph of K value versus silhouette coefficient is plotted, the curve shape is observed, and the K value with the largest silhouette coefficient is selected as the optimal clustering value.

[0108] S243. Perform K-means clustering on the chemical feature matrix based on the optimal clustering value to determine the cluster affiliation label of each temperature-sensitive compound and the center vector of each cluster.

[0109] The center vector of each cluster is the vector formed by the arithmetic mean of the elements in the chemical characteristic vectors of all temperature-sensitive compounds in that cluster.

[0110] S244. Based on the assigned tags and the center vectors of each cluster, and combined with the physicochemical properties of each temperature-sensitive compound, determine the optimal set of characteristic gas compounds for overheating faults.

[0111] As an example, this step may include:

[0112] First, based on the center vector of each cluster, select clusters that satisfy any of the following conditions:

[0113] 1. The central vector of this cluster exhibits significant characteristic changes in the low-temperature region (e.g., <200℃);

[0114] 2. The eigenvalues ​​of the cluster’s center vector (i.e., the characteristic ion abundance ratio and functional group index) exhibit strong monotonicity with increasing temperature (e.g., the slope is greater than a set threshold).

[0115] 3. The characteristic abrupt change temperature indicated by the center vector of this cluster matches the decomposition temperature of the cable material.

[0116] A list of key cluster labels is obtained based on the cluster labels that meet the conditions.

[0117] Then, based on the key cluster label list and cluster affiliation label, all gaseous compounds belonging to the key clusters are extracted to determine the set of candidate feature gases;

[0118] Next, by combining the physicochemical properties (such as molecular weight, stability, toxicity, etc.) of all candidate characteristic gas compounds, interference items (such as ambient background gases) are eliminated to obtain the optimal characteristic gas compound for overheating faults, and thus obtain the optimal set of characteristic gas compounds.

[0119] Step S24 involves constructing a chemical feature matrix using the characteristic ion abundance ratios and functional group indices of each temperature-sensitive compound at multiple temperature points, followed by secondary clustering to determine the optimal characteristic gaseous compound for overheating faults. This allows for cross-validation using chemical information from different dimensions, improving the robustness and accuracy of the results. Furthermore, this step incorporates characteristic ion abundance and functional group indices when constructing the chemical feature matrix. Characteristic ion abundance captures molecular fragment signals (e.g., mass spectrometry fragment peaks), while functional group indices elucidate the chemical nature of molecules (e.g., oxygen-containing group characteristics). The synergistic application of characteristic ion abundance and functional group indices enables the identification of trace substances below the instrument's detection limit, significantly improving the response sensitivity of oxygen-containing polar compounds (e.g., aldehydes and ketones). Moreover, by employing DBSCAN clustering in the primary clustering stage and K-means clustering in the secondary clustering stage, both feature reliability (DBSCAN noise reduction) and engineering practicality (K-means efficient grouping and purification) are balanced, ensuring accurate and efficient screening of the optimal characteristic gaseous compound.

[0120] S3. Using a gas sensor array, quantitative data on the concentration of all optimal characteristic gas compounds in all cable samples as a function of temperature are obtained to obtain training samples.

[0121] The input features for each training sample are the type of cable under test and the concentration vector of all optimal characteristic gaseous compounds in the corresponding cable. The output feature is the corresponding real-time temperature value. Both the input and output features have been standardized.

[0122] S4. The neural network is trained using the training samples to obtain an early identification model for cable overheating faults. This model is used to output corresponding real-time temperature prediction values ​​based on the type of cable under test and the type and concentration of the optimal characteristic gas compounds generated, so as to identify cable overheating faults at an early stage.

[0123] This step specifically includes:

[0124] Based on the cable type features of the training samples, a one-hot encoding conversion strategy is adopted to convert categorical variables into numerical vectors, which are then concatenated with the optimal feature gas compound concentration vector to form an input feature matrix that integrates cable type and gas concentration, ensuring that the model can simultaneously capture equipment attributes and gas dynamic information.

[0125] Based on the input feature dimension and the characteristics of the prediction target, the neural network topology is designed: for example, a fully connected layer stacked architecture can be adopted, the number of input layer nodes matches the fusion feature dimension, 3-5 hidden layers are set (64-128 nodes per layer), and the ReLU activation function is used to enhance the nonlinear fitting ability. The output layer uses a linear activation function to directly regress the real-time temperature value, and the mean squared error (MSE) is selected as the loss function to quantify the prediction bias.

[0126] Based on the training-validation-test dataset partitioning strategy, the training samples are divided into training set, validation set and test set. The training set is used for model weight update, the validation set is used for hyperparameter tuning (such as learning rate and batch size), and the test set is used for final model performance evaluation, ensuring the independence and representativeness of the data partitioning.

[0127] Based on gradient descent optimization algorithms (such as the Adam optimizer), batch iterative training is performed on the training set: the batch size is set to control the number of samples updated in a single parameter update, the gradient is calculated and the network weights are adjusted through backpropagation, the loss change on the validation set is monitored, and an early stopping mechanism is adopted (such as terminating training if the validation loss does not improve after several consecutive rounds) to prevent overfitting. Finally, the model parameters with the minimum validation loss are saved as the optimal model.

[0128] Based on independent samples in the test set, the generalization performance of the optimal model is evaluated: the root mean square error (RMSE) and mean absolute error (MAE) of the predicted temperature and the actual temperature are calculated to verify the prediction accuracy of the model on unknown data; at the same time, the model residual distribution is analyzed to ensure that the error is randomly distributed without systematic bias, thus confirming the reliability of the model.

[0129] Based on the optimal model deployment requirements, the trained neural network model is encapsulated into a callable interface, which supports real-time input of the cable type under test and the optimal characteristic gas compound concentration vector, outputs the standardized temperature prediction value, and restores it to the actual temperature value through destandardization operation, ultimately realizing the early real-time identification and temperature warning function of cable overheating fault.

[0130] Example 2

[0131] This embodiment provides a cable overheating fault detection device based on electronic nose technology. This device employs the cable overheating fault detection method based on electronic nose technology as described in Embodiment 1. Figure 3 The device includes:

[0132] The acquisition module is used to acquire the gas production results of various organic volatile gases generated by different materials and models of cables in the power cable overheating fault simulation device as a function of temperature.

[0133] The clustering module is used to extract features from the gas production results and to determine the characteristic gases of the overheating fault based on the feature extraction results using a clustering algorithm.

[0134] The sample acquisition module is used to measure the quantitative data of the concentration of all the optimal characteristic gas compounds in all cable samples as a function of temperature using a gas sensor array, thereby obtaining training samples.

[0135] The training module is used to train the neural network using the training samples to obtain an early identification model for cable overheating faults. This model is used to output corresponding real-time temperature prediction values ​​based on the type of cable under test and the type and concentration of the optimal characteristic gas compound produced, so as to identify cable overheating faults at an early stage.

[0136] Example 3

[0137] Embodiment 3 of the present invention provides a terminal.

[0138] A terminal includes a processor and a storage medium storing a program that can run on the processor. When the processor executes the program, it implements the steps in the cable overheating fault detection method based on electronic nose technology as described in Embodiment 1 of the present invention.

[0139] The beneficial effects of this invention are that, compared with the prior art,

[0140] 1. By extracting features from the gas production results of various organic volatile gases generated by different materials and models of cables in a power cable overheating fault simulation device as temperature changes, the optimal characteristic gas compounds of overheating faults can be determined by clustering algorithms based on the feature extraction results. Then, training samples can be constructed using these optimal characteristic gas compounds, and the neural network can be trained using the training samples to obtain an early identification model for cable overheating faults, thereby realizing early warning of cable overheating faults.

[0141] 2. By integrating data from two independent analytical techniques, GC-MS and FTIR, the systematic errors of a single technique can be effectively offset, improving the accuracy of concentration measurement. At the same time, the trace detection advantages of GC-MS and the high concentration stability of FTIR can be combined to expand the effective detection range. In addition, cross-validation of the two techniques can reduce the risk of misjudgment and ensure the reliability of the results.

[0142] 3. By extracting temperature response vectors from quantitative concentration data to construct a temperature-sensitive feature matrix for primary clustering, temperature-sensitive compounds are selected. This process quickly identifies a core temperature-sensitive subset from a massive dataset, significantly reducing the scale of subsequent analyses and avoiding redundant calculations. A secondary clustering is then performed using the characteristic ion abundance ratios and functional group indices of each temperature-sensitive compound at multiple temperature points to determine the optimal characteristic gaseous compounds for overheating faults. Cross-validation using chemical information from different dimensions enhances the robustness and accuracy of the results.

[0143] 4. Furthermore, by using DBSCAN clustering in the first clustering and Kmeans clustering in the second clustering, both feature reliability (DBSCAN denoising) and engineering practicality (Kmeans efficient grouping and purification) are taken into account, ensuring accurate and efficient screening of the optimal feature gas compounds.

[0144] 5. The chemical characteristic matrix includes characteristic ion abundance and functional group index. Characteristic ion abundance can capture molecular fragment signals (such as mass spectrometry fragment peaks), while functional group index can resolve the chemical nature of molecules (such as oxygen-containing group characteristics). Through the synergistic application of characteristic ion abundance and functional group index, trace substances below the instrument's detection limit can still be identified, significantly improving the response sensitivity of oxygen-containing polar compounds (such as aldehydes, ketones, and alcohols).

[0145] 6. In the early stages of a cable fire, volatile gases appear before smoke particles. Therefore, using a gas sensor can provide a faster early warning of a fire than a smoke sensor. Gas sensors also have advantages such as low cost and strong resistance to environmental interference.

[0146] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0147] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0148] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0149] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing the status information of the computer-readable program instructions to implement various aspects of this disclosure.

[0150] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A method for detecting cable overheating faults based on electronic nose technology, characterized in that, Includes the following steps: To obtain the gas production results of various organic volatile gases generated by different cables of different materials and models in a power cable overheating fault simulation device as a function of temperature; Temperature response characteristics and chemical characteristics were extracted from the gas production results. Based on the feature extraction results, a clustering algorithm is used to determine the optimal set of characteristic gas compounds for overheating faults; The concentration of all optimal characteristic gas compounds in all cable samples was quantitatively measured with temperature using a gas sensor array to obtain training samples. The neural network is trained using the training samples to obtain an early identification model for cable overheating faults. This model is used to output corresponding real-time temperature prediction values ​​based on the type of cable under test and the type and concentration of the optimal characteristic gas compounds generated, so as to identify cable overheating faults at an early stage.

2. The cable overheating fault detection method based on electronic nose technology according to claim 1, characterized in that, The temperature response characteristics and chemical characteristics were extracted from the gas production results. Based on the feature extraction results, a clustering algorithm is used to determine the optimal set of characteristic gaseous compounds for overheating faults, including: Obtain gas production results, including total ion chromatograms and mass spectra obtained by GC-MS and infrared absorption spectra obtained by FTIR; The compound set of gaseous compounds produced by each cable sample was determined based on the total ion chromatogram, mass spectrum and infrared absorption spectrum, and then the quantitative data of the concentration of all gaseous compounds in each cable sample as a function of temperature, the characteristic ion abundance ratio and the functional group index were determined. Temperature response vectors are extracted from concentration-quantitative data to construct a temperature-sensitive feature matrix. The temperature-sensitive feature matrix is ​​then clustered to select a set of temperature-sensitive compounds. A chemical feature matrix is ​​constructed using the characteristic ion abundance ratio and functional group index of each temperature-sensitive compound at multiple temperature points. The chemical feature matrix is ​​then subjected to secondary clustering to determine the optimal set of characteristic gaseous compounds for overheating faults.

3. The cable overheating fault detection method based on electronic nose technology according to claim 2, characterized in that, The process of determining the compound set of gaseous compounds generated by each cable sample based on the total ion chromatogram, mass spectrum, and infrared absorption spectrum, and then determining the quantitative data of the concentration of all gaseous compounds in each cable sample as a function of temperature, includes: Based on the total ion chromatogram and mass spectrum, the GC-MS concentration matrix of the i-th gaseous compound in the k-th cable sample was obtained by calibration using peak area ratio. Based on FTIR quantitative analysis and using Beer-Lambert's law, the FTIR concentration matrix of the i-th gaseous compound in the k-th cable sample was obtained. The GC-MS concentration matrix and FTIR concentration matrix of the i-th compound in the k-th cable sample are weighted and summed to obtain the weighted concentration as the corresponding quantitative concentration data. The weighting parameter w can be adjusted based on the signal-to-noise ratio.

4. The cable overheating fault detection method based on electronic nose technology according to claim 2, characterized in that, Temperature response vectors are extracted from concentration-quantitative data to construct a temperature-sensitive feature matrix. This matrix is ​​then subjected to clustering to identify temperature-sensitive compounds, including: Multiple temperature response features of the i-th gaseous compound in the k-th cable sample are extracted from the quantitative concentration data to form the corresponding temperature response vector. Furthermore, based on Constructing the temperature-sensitive feature matrix M T ; Based on the temperature-sensitive feature matrix M T Determine the neighborhood radius and the minimum number of minimum neighborhood points for M. T Perform DBSCAN clustering to obtain cluster label vectors; these cluster label vectors are used to indicate the cluster to which each temperature response vector belongs. Based on the clustering label vector and the preset screening thresholds corresponding to the multiple temperature response features, temperature-sensitive clusters are identified; Select the current gaseous compound to be screened, and count the number of cable samples in all cable samples that are identified as belonging to the temperature-sensitive cluster. When the ratio of this number to the total number of cable samples is greater than the set screening threshold, the current gaseous compound is considered to be a temperature-sensitive compound.

5. The cable overheating fault detection method based on electronic nose technology according to claim 4, characterized in that, Based on the clustering label vector and the preset screening thresholds corresponding to the multiple temperature response features, temperature-sensitive clusters are identified, including: The centroid vector is constructed by calculating the average value of each temperature response feature in all temperature response vectors belonging to the current cluster based on the cluster label vector. When the centroid vector meets the corresponding preset conditions, the current cluster is considered to be a temperature-sensitive cluster.

6. The cable overheating fault detection method based on electronic nose technology according to claim 5, characterized in that, The condition that the centroid vector satisfies the corresponding preset condition, indicating that the current cluster is a temperature-sensitive cluster, includes: When all temperature response features in the centroid vector are greater than the corresponding set threshold, the current cluster is considered to be a temperature-sensitive cluster.

7. The cable overheating fault detection method based on electronic nose technology according to claim 4, characterized in that, The temperature response characteristics are set to include: temperature sensitivity coefficient, concentration change rate at a set temperature point T0, and linear significance determination coefficient.

8. The cable overheating fault detection method based on electronic nose technology according to claim 2, characterized in that, A chemical feature matrix is ​​constructed using the characteristic ion abundance ratios and functional group indices of each temperature-sensitive compound at multiple temperature points. Secondary clustering is then performed on this chemical feature matrix to determine the optimal set of characteristic gaseous compounds for overheating faults, including: The chemical feature vectors of each temperature-sensitive compound are constructed by using the characteristic ion abundance ratios and functional group indices at multiple temperature points, and then the chemical feature matrix is ​​obtained. The silhouette coefficient is used to evaluate the clustering quality, and the clustering value that maximizes the silhouette coefficient is selected as the optimal clustering value. K-means clustering is performed on the chemical feature matrix based on the optimal clustering value to determine the cluster affiliation label of each temperature-sensitive compound and the center vector of each cluster; Based on the assigned tags and the center vectors of each cluster, and combined with the physicochemical properties of each temperature-sensitive compound, the optimal set of characteristic gas compounds for overheating faults is determined.

9. A cable overheating fault detection device based on electronic nose technology, employing the method of any one of claims 1 to 8, characterized in that, include: The acquisition module is used to acquire the gas production results of various organic volatile gases generated by different materials and models of cables in the power cable overheating fault simulation device as a function of temperature. The clustering module is used to extract temperature response features and chemical features from gas production results. Based on the feature extraction results, a clustering algorithm is used to determine the optimal set of characteristic gas compounds for overheating faults; The sample acquisition module is used to measure the quantitative data of the concentration of all the optimal characteristic gas compounds in all cable samples as a function of temperature using a gas sensor array, thereby obtaining training samples. The training module is used to train the neural network using the training samples to obtain an early identification model for cable overheating faults. This model is used to output corresponding real-time temperature prediction values ​​based on the type of cable under test and the type and concentration of the optimal characteristic gas compound produced, so as to identify cable overheating faults at an early stage.

10. A terminal, comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method according to any one of claims 1-7.

Images