Transformer oil chromatography state evaluation method, device and equipment and storage medium

Abstract

The invention provides a transformer oil chromatography state evaluation method and device, equipment and a storage medium, and the method comprises the steps: carrying out the preprocessing of a transformer oil chromatography sample data set, including data cleaning and feature extraction; then, a plurality of clustering starting points are selected as initial clustering centers, the membership degree of the clustering centers is calculated, and the clustering centers are updated; and calculating the adjustment amount according to the updated clustering center, and updating the clustering center again according to the chaotic sequence mapping. And repeating the above steps until the clustering process converges to preset precision or reaches a preset maximum number of iterations. And finally, calculating the correlation degree between the obtained trajectory clustering data and the fault type according to the iterated clustering center, and determining the fault type of the transformer according to the correlation degree. According to the method, the optimization process of the clustering algorithm is intervened by using the chaos sequence, the search process is improved, and the sensitivity of the fuzzy clustering algorithm to the initial value is relieved. And meanwhile, a plurality of iteration tracks are set for parallel calculation, so that the iterative global optimization capability of the algorithm is improved.

Description

Technical Field

[0001] The present application relates to the field of wire adjustment, and particularly to a method, device, equipment and storage medium for evaluating the chromatographic state of transformer oil. Background Art

[0002] Dissolved gas analysis (DGA) in transformer oil is a very important technology. This technology is carried out without powering off the equipment, regularly diagnoses the internal conditions of the operating transformer, discovers early latent faults and abnormalities inside the transformer in a timely manner, and ensures the safe and reliable operation of the equipment. DGA has a wide range of applications in engineering practice and is usually used as the first technical means for transformer monitoring.

[0003] The methods for fault type identification can be divided into traditional DGA analysis methods and modern intelligent identification methods. Traditional methods include the characteristic gas method, the three-ratio method, the David triangle method, the Duval Pentagon method, etc. These methods are easy to use. Only by simply calculating the DGA data according to the corresponding fault diagnosis rules and then referring to the corresponding charts can the corresponding fault types be obtained. However, these traditional methods have defects such as too absolute fault boundaries and missing codes in the ratio method. The fault diagnosis accuracy rate of the above traditional DGA methods is usually lower than 80%.

[0004] In practice, the abnormality of the transformer state is reflected by monitoring the oil chromatographic data, that is, the content of each gas in the transformer. However, the existing methods do not fully consider the correlation, distribution of each index of the transformer oil chromatographic data and the trend items caused by transformer aging and other problems. These problems are very likely to affect the final evaluation result of the transformer operation state.

[0005] With the development of computer technology, intelligent algorithms such as neural networks, genetic algorithms, and support vector machines have also been applied to DGA fault diagnosis and achieved good results. Among them, the fuzzy clustering algorithm uses the idea of fuzzy mathematics to give the degree to which DGA data belongs to various faults according to the membership degree, rather than a binary judgment, which is more in line with the actual situation of transformer faults. For example, in practice, usually taking 300°C and 700°C as the boundaries, the overheating faults of transformers are divided into three types: low-temperature overheating, medium-temperature overheating, and high-temperature overheating. However, there is no such clear boundary in reality, and low, medium, and high are fuzzy descriptions. Using the membership degree to divide the types of faults belongs to will be more in line with the actual situation. The same is true for discharge faults. Spark discharge and arc discharge (often referred to as low-energy-density discharge and high-energy-density discharge abroad), both essentially represent the breakdown of the insulating medium, and the difference mainly lies in the level of discharge energy. There is no clear and strict boundary between the two, and it is more appropriate to describe discharge faults with the membership degree. At the same time, there may also be various forms of composite fault types inside the transformer, and the fuzzy clustering algorithm also has great advantages in dealing with such problems.

[0006] However, the iterative optimization process of the fuzzy clustering algorithm belongs to a local search method based on gradient calculation, and its objective function is a typical non-convex function. Its optimization process is prone to falling into local extreme points. The classification results are sensitive to the initial values. Different iterative initial values often result in different classification results, seriously affecting the application of this algorithm in engineering practice. The algorithm is sensitive to the iterative initial values, and the situation of converging to local extreme points still exists. The algorithm still cannot stably converge to the global optimal solution. Summary of the Invention

[0007] The purpose of this application is to overcome the above technical problems in the prior art and provide a method, device, equipment, and storage medium for evaluating the chromatographic state of transformer oil.

[0008] This application provides a method for evaluating the chromatographic state of transformer oil, including:

[0009] S1 Preprocess the transformer oil chromatographic sample data set and set the parameters for parallel computing, including the number of trajectories, the length of the chaotic sequence, the iterative accuracy, and the maximum number of iterations;

[0010] S2 Select multiple clustering starting points corresponding to the number of trajectories as the initial clustering centers;

[0011] S3 Calculate the membership degree of the clustering centers and update the clustering centers according to the membership degree;

[0012] S4 Calculate the adjustment amount of the clustering centers according to the updated clustering centers;

[0013] S5 Obtain a chaotic sequence according to the length of the chaotic sequence, and map the chaotic sequence to the corresponding value range of the adjustment amount;

[0014] S6 Update the cluster center again according to the adjustment amount obtained by mapping the chaotic sequence;

[0015] S7 Repeat steps S2 - S6 until the clustering process converges to a preset accuracy or reaches a preset maximum number of iterations;

[0016] S8 Calculate the correlation degree between the obtained trajectory clustering data and the fault type according to the cluster center after iteration, and determine the transformer fault type according to the correlation degree.

[0017] Optionally, calculate the membership degree of the cluster center, and the expression is as follows:

[0018]

[0019] where, u ij is the membership degree, i represents the trajectory data number, j represents the number of updates of the cluster center of each trajectory data, x represents the trajectory data, v represents the initial cluster center, k represents the slope, h represents, and H is the symbol of the product formula.

[0020] Optionally, update the cluster center according to the membership degree, and the expression is as follows:

[0021] Substitute u ij into the following formula to obtain the cluster center V(L) = (v1`(L), v2`(L),..., vω`(L));

[0022]

[0023] where, L represents the number of iterations, m is a constant, usually taking the value of 2, and n represents the number of elements μij which is the membership degree of the sample Xi to the cluster center Vj;

[0024] Obtain the objective function value F(L) = (F1(L), F2(L),....., Fω(L)), and obtain the historical minimum value Fmin of each trajectory = (F1 - min, F2 - min,....... Fω - min);

[0025] where the historical minimum value Fi - min of the i - th trajectory = min(Fi(t)), t = [1, 2,..., L], and the cluster center vi - min corresponding to Fi - min;

[0026] Determine the historical minimum value of all trajectories up to the current step position, \(F_{global}=\min(F_i - \min)\), \(t = [1, 2, \cdots, \omega]\), and the clustering center \(v_{global}\) corresponding to \(F_{global}\).

[0027] Optionally, calculate the adjustment amount of the clustering center, and the expression is as follows:

[0028] \(\Delta V=\lambda_1\Delta V_1+\lambda_2\Delta V_2+\lambda_3\Delta V_3\);

[0029] where \(\Delta V_1\) is the current adjustment amount of each trajectory, \(\Delta V_2\) is the difference between the current clustering center of each trajectory and the historical optimal center of the trajectory, \(\Delta V_3\) is the difference between the current clustering center of each trajectory and \(\Delta V_2 =

[0030] \(V_{global}(L)-V(L)\), and \(\lambda_1\), \(\lambda_2\) and \(\lambda_3\) are weight factors, \(\lambda_1+\lambda_2+\lambda_3 = 1\).

[0031] Optionally, the expressions of \(\lambda_1\), \(\lambda_2\) and \(\lambda_3\) are as follows:

[0032]

[0033] \(\lambda_2=\lambda_1(1 - \lambda_1)\)

[0034] \(\lambda_3=(1 - \lambda_1)*(1 - \lambda_1)\).

[0035] Optionally, obtain the chaotic sequence according to the length of the chaotic sequence, and map the chaotic sequence to the corresponding value range of the adjustment amount, including:

[0036] Update \(n\) chaotic sequences using a preset formula;

[0037] Map the \(n\) chaotic sequences to the corresponding value range of the adjustment amount;

[0038] Obtain the tentative clustering centers corresponding to \(n\) adjustment amounts between \(V(L)\) and \(V(L + 1)\) through mapping;

[0039] Compare whether the tentative clustering center is better than the existing clustering center obtained by the formula \(V(L + 1)=V(L)+\Delta V\);

[0040] If so, update the clustering center.

[0041] Optionally, it further includes:

[0042] Set multiple clustering trajectories for parallel calculation, and realize the sharing and utilization of information between the trajectories.

[0043] This application also provides a transformer oil chromatogram state evaluation device, including:

[0044] A setting module, configured to preprocess a transformer oil chromatogram sample data set and set parameters for parallel computing, including the number of trajectories, the length of the chaotic sequence, the iteration precision, and the maximum number of iterations;

[0045] A selection module, configured to select a plurality of clustering starting points corresponding to the number of trajectories as initial clustering centers;

[0046] A first update module, configured to calculate the membership degree of the clustering centers and update the clustering centers according to the membership degree;

[0047] A calculation module, configured to calculate the adjustment amount of the clustering centers according to the updated clustering centers;

[0048] A mapping module, configured to obtain a chaotic sequence according to the length of the chaotic sequence and map the chaotic sequence to the value range corresponding to the adjustment amount;

[0049] A second update module, configured to update the clustering centers again according to the adjustment amount obtained by mapping the chaotic sequence;

[0050] Repeat the above modules until the clustering process converges to a preset precision or reaches a preset maximum number of iterations;

[0051] A judgment module, configured to calculate the correlation degree between the obtained trajectory clustering data and the fault type according to the clustering centers after iteration, and determine the transformer fault type according to the correlation degree.

[0052] This application further provides a transformer oil chromatogram state evaluation device, including:

[0053] A memory, configured to store a computer executable program of the above-mentioned transformer oil chromatogram state evaluation method;

[0054] A processor is used to retrieve the computer-executable program and execute the following steps: S1 Preprocess the transformer oil chromatogram sample dataset and set the parameters for parallel computing, including the number of trajectories, the length of the chaotic sequence, the iteration accuracy, and the maximum number of iterations; S2 Select multiple clustering starting points corresponding to the number of trajectories as the initial clustering centers; S3 Calculate the membership degree of the clustering centers and update the clustering centers according to the membership degree; S4 Calculate the adjustment amount of the clustering centers based on the updated clustering centers; S5 Obtain the chaotic sequence according to the length of the chaotic sequence and map the chaotic sequence to the value range corresponding to the adjustment amount; S6 Update the clustering centers again according to the adjustment amount obtained by mapping the chaotic sequence; S7 Repeat steps S2 - S6 until the clustering process converges to the preset accuracy or reaches the preset maximum number of iterations; S8 Calculate the correlation degree between the obtained trajectory clustering data and the fault type based on the clustering centers after iteration, and determine the transformer fault type according to the correlation degree.

[0055] The present application also provides a storage medium storing a computer-executable program, which is used to be retrieved by a processor and execute the steps of the above-mentioned method for evaluating the state of transformer oil chromatogram.

[0056] The beneficial effects of the present application are:

[0057] The present application provides a method for evaluating the state of transformer oil chromatogram, including: S1 Preprocess the transformer oil chromatogram sample dataset and set the parameters for parallel computing, including the number of trajectories, the length of the chaotic sequence, the iteration accuracy, and the maximum number of iterations; S2 Select multiple clustering starting points corresponding to the number of trajectories as the initial clustering centers; S3 Calculate the membership degree of the clustering centers and update the clustering centers according to the membership degree; S4 Calculate the adjustment amount of the clustering centers based on the updated clustering centers; S5 Obtain the chaotic sequence according to the length of the chaotic sequence and map the chaotic sequence to the value range corresponding to the adjustment amount; S6 Update the clustering centers again according to the adjustment amount obtained by mapping the chaotic sequence; S7 Repeat steps S2 - S6 until the clustering process converges to the preset accuracy or reaches the preset maximum number of iterations; S8 Calculate the correlation degree between the obtained trajectory clustering data and the fault type based on the clustering centers after iteration, and determine the transformer fault type according to the correlation degree. The present application uses the chaotic sequence to intervene in the optimization process of the clustering algorithm, improves the search process, and alleviates the sensitivity of the fuzzy clustering algorithm to the initial value. At the same time, in the iteration, multiple iteration trajectories are set for parallel computing, and information sharing and utilization are realized among the trajectories, improving the global optimization ability of the algorithm iteration. Description of the Drawings

[0058] Figure 1 It is a schematic diagram of the process for evaluating the state of transformer oil chromatogram in the present application; Figure 2 It is a reference slope schematic diagram of the data to be inspected and the fault type in this application; Figure 3 It is a schematic diagram of the fault type in this application. Detailed implementation manners

[0059] The following further describes this application in conjunction with the accompanying drawings and specific embodiments, so that those skilled in the art can better understand this application and be able to implement it.

[0060] The following content is all examples of the specific implementation process provided to detail the technical solution to be protected by this application. However, this application can also be implemented in other ways different from the described ones. Those skilled in the art can implement this application by using different technical means under the guidance of the concept of this application. Therefore, this application is not limited by the following specific embodiments.

[0061] As Figure 1 shown, the steps of a method for evaluating the chromatographic state of transformer oil include:

[0062] S1. Preprocess the transformer oil chromatographic sample data set, and set the parameters for parallel computing, including the number of trajectories, the length of the chaotic sequence, the iteration accuracy, and the maximum number of iterations.

[0063] Before performing fault diagnosis, it is first necessary to initialize the sample data set. This includes collecting, cleaning, and preprocessing the data of the dissolved gases in the transformer oil, so as to provide accurate basic data for subsequent fault diagnosis.

[0064] Specifically, the dissolved gases in the transformer oil are collected regularly or continuously through professional sensors and equipment. Ensure that the collected data is accurate and complete, and record information such as the collection time and location. Then clean the collected data to remove outliers, missing values, and abnormal data. This is achieved through data filtering, interpolation, smoothing and other techniques. Preprocess the cleaned data, including operations such as data standardization and normalization, so that subsequent fault diagnosis algorithms can better process the data.

[0065] Next, the following settings are made:

[0066] Set the number of trajectories ω for parallel computing.

[0067] During the fault diagnosis process, in order to improve the efficiency and accuracy of diagnosis, a parallel computing method is adopted. In this application, a number of trajectories ω for parallel computing is set. This value is adjusted according to the actual computing resources and diagnostic requirements.

[0068] Set the chaotic sequence n.

[0069] A chaotic sequence is a sequence with high randomness and complexity, which is used to generate pseudo-random numbers. In fault diagnosis, chaotic sequences are used to generate pseudo-random numbers for initializing the positions and directions of clustering centers. The length n of the specific chaotic sequence is set according to the actual situation and algorithm requirements.

[0070] Set the iteration accuracy.

[0071] During the fault diagnosis process, the algorithm needs to perform multiple iterations to find the optimal solution. In this application, an iteration accuracy is set to control the convergence speed and accuracy of the algorithm iteration. The iteration accuracy is adjusted according to the actual requirements and algorithm performance.

[0072] Set the maximum number of iterations.

[0073] To avoid the algorithm falling into a local optimal solution or failing to converge, a maximum number of iterations is set in this application. When the algorithm reaches the maximum number of iterations, even if it has not converged to the optimal solution, the iteration is terminated in advance. This ensures the stability and reliability of the algorithm.

[0074] S2. Select multiple clustering starting points corresponding to the number of trajectories as the initial clustering centers.

[0075] This application selects ω clustering starting points as the initial clustering centers of l trajectories respectively, V(L) = {v1, v2,..., v ω}.

[0076] The starting points are randomly selected or selected according to some prior knowledge or experience.

[0077] The initial clustering center V(L) will be used as the starting point of the algorithm iteration. Through the optimization search in the iteration process, the positions and directions of the clustering centers are continuously updated to achieve the classification of sample data and fault diagnosis.

[0078] In the iteration process, pseudo-random numbers need to be generated using chaotic sequences at each step of the iteration, and the positions and directions of the clustering centers are updated according to the pseudo-random numbers. At the same time, this application also needs to set an iteration accuracy. When the algorithm reaches the iteration accuracy, the iteration is terminated in advance to ensure the stability and reliability of the algorithm.

[0079] By setting the number of trajectories ω for parallel computing, this application performs the calculations and searches for multiple trajectories simultaneously, thereby improving the efficiency and accuracy of the algorithm. Each trajectory uses its own initial clustering center and chaotic sequence, and jointly completes the task of fault diagnosis through information sharing and collaborative computing.

[0080] S3. Calculate the membership degrees of the clustering centers and update the clustering centers according to the membership degrees.

[0081] Substitute U(L) (the membership degree of the L-th iteration) into the following formula to obtain the cluster center V(L) = (v1(L), v2(L),..., v ω (L)):

[0082]

[0083] where u ij is the membership degree, i represents the trajectory data number, j represents the number for updating the cluster center of each trajectory data, x represents the trajectory data, v represents the initial cluster center, k represents the slope, and H is the symbol of the product formula.

[0084] After obtaining the membership degree matrix U(L), the present application substitutes it into the following formula to calculate the new cluster center V(L). The specific calculation method is as follows:

[0085]

[0086] where X i represents the i-th sample, Σ represents the summation symbol, L represents the number of iterations, m is a constant, usually taking the value of 2, and n represents the number of elements μij which is the membership degree of the sample Xi to the cluster center Vj.

[0087] Through the above calculation, the present application obtains ω cluster centers V(L) of the L-th iteration. The specific steps are as follows:

[0088] Obtain the objective function value F(L) = (F1(L), F1(L),....., Fω(L));

[0089] Obtain the historical minimum value Fmin of each trajectory = (F1-min, F2-min,....... Fω-min), where the historical minimum value Fi-min of the i-th trajectory = min(Fi(t)), t = [1, 2,..., L], and the corresponding cluster center vi-min of Fi-min;

[0090] To determine the historical minimum value of each trajectory, the present application needs to compare the objective function values of each trajectory. The specific calculation method is as follows:

[0091] Fi-min = min(Fi(t)), t = [1, 2,..., L];

[0092] where Fi(t) represents the objective function value of the i-th trajectory at the t-th iteration. Through the above calculation, the present application obtains the historical minimum value Fi-min of each trajectory and the corresponding cluster center vi-min.

[0093] Determine the historical minimum value Fglobal = min(Fi - min) of all trajectories up to the current step position, where t = [1, 2,..., ω], and the corresponding cluster center vglobal of Fglobal.

[0094] To determine the historical minimum value of all trajectories, this application needs to compare the historical minimum values of each trajectory. The specific calculation method is as follows:

[0095] Fglobal = min(Fi - min), t = [1, 2,..., ω];

[0096] Among them, Fi - min represents the historical minimum value of the i-th trajectory. Through the above calculation, this application obtains the historical minimum value Fglobal of all trajectories and the corresponding cluster center vglobal.

[0097] S4 Calculate the adjustment amount of the cluster center according to the updated cluster center.

[0098] Specifically, calculate the current adjustment amount ΔV1 of each trajectory:

[0099] ΔV1 = V′(L) - V(L);

[0100] Among them, V′(L) represents the cluster center of each trajectory in the L-th iteration, and V(L) represents the actual cluster center of each trajectory in the L-th iteration.

[0101] Calculate the difference ΔV2 between the current cluster center of each trajectory and the historical optimal center of this trajectory:

[0102] ΔV2 = Vi - min(L) - V(L);

[0103] Among them, Vi - min(L) represents the historical minimum value of the cluster center of each trajectory in the L-th iteration.

[0104] Calculate the difference ΔV3 between the current cluster center of each trajectory and the global optimal cluster center:

[0105] ΔV3 = Vglobal(L) - V(L);

[0106] Among them, Vglobal(L) represents the historical minimum value of the cluster center of all trajectories in the L-th iteration.

[0107] Calculate the adjustment amount ΔV of the cluster center:

[0108] ΔV = λ1ΔV1 + λ2ΔV2 + λ3ΔV3;

[0109] Among them, λ1, λ2, and λ3 are weight factors, where λ1 + λ2 + λ3 = 1, representing the proportions of the three adjustment amounts in the total adjustment amount. These weight factors are adjusted according to the actual situation to ensure the stability and accuracy of the algorithm.

[0110] Calculate the cluster center V(L + 1) for the (L + 1)-th iteration:

[0111] V(L + 1) = V(L) + ΔV;

[0112] Through the above steps, the present application obtains the cluster center V(L + 1) for the (L + 1)-th iteration.

[0113] S5 Obtain the chaotic sequence according to the length of the chaotic sequence, and map the chaotic sequence to the corresponding value range of the adjustment amount.

[0114] Specifically, use the following formula to update n chaotic sequences to obtain the chaotic sequence N(L) = (x1(L), x1(L),..., xn(L)):

[0115] x(n + 1) = μx(n)(1 - x(n));

[0116] In order to update n chaotic sequences, the present application first needs to calculate using the following formula:

[0117] N(L) = (x1(L), x2(L),..., xn(L));

[0118] Among them, x1(L), x2(L),..., xn(L) are the updated values of the n chaotic sequences.

[0119] Map the above n chaotic sequences to the corresponding value range of the adjustment amount, N(L)ΔV = (x1(L)ΔV, x2(L)ΔV,..., x n (L)ΔV), to obtain the tentative cluster center points corresponding to these n adjustment amounts between V(L) and V(L + 1).

[0120] S6 Update the cluster center again according to the adjustment amount mapped from the chaotic sequence;

[0121] Next, the present application needs to compare the objective function values of these tentative cluster centers with the objective function values of the existing cluster centers. If the objective function value of a certain tentative cluster center is smaller, it means that this center point is better than the existing center point, and the cluster center corresponding to the corresponding trajectory needs to be updated.

[0122] The specific update process is as follows:

[0123] Calculate the objective function values corresponding to these n points FN(L) = (Fx1, Fx2,,... Fxn )

[0124] Among them, represents the objective function value of the j-th tentative clustering center.

[0125] For each trajectory i, this application needs to compare its existing objective function value with the objective function value of the tentative clustering center. If the objective function value F of the j-th tentative clustering center of the i-th trajectory i-xj is better than the existing objective function value of the i-th trajectory corresponding in V(L + 1), then the clustering center of this trajectory is updated from Δv determined by the formula to x i to j (L)Δv i .

[0126] Correspondingly, the clustering center of the (L + 1)-th iteration of the i-th trajectory in the formula V(L + 1) = V(L) + ΔV also needs to be updated. The specific update formula is:

[0127] V i (L + 1) = V i (L) + x j (L)Δv i ;

[0128] Through the above steps, this application obtains the updated clustering center V i (L + 1), and continues the next round of iterative calculation.

[0129] S7 Repeat steps S2 - S6 until the clustering process converges to the preset accuracy or reaches the preset maximum number of iterations.

[0130] Until the clustering process converges to the preset accuracy or reaches the preset maximum number of iterations, this application ends the iteration of the algorithm. At this time, this application needs to analyze and organize the clustering results.

[0131] The process of analyzing the clustering results includes the following aspects:

[0132] By comparing the clustering centers at different numbers of iterations, this application observes the changing trend of the clustering centers. If the clustering centers gradually converge to a stable state, then this application considers that the clustering process has converged.

[0133] This application calculates the objective function values at different numbers of iterations and observes their changing trends. If the objective function values gradually decrease and tend to be stable, then this application considers that the clustering effect is good. In addition, this application also calculates indexes such as the silhouette coefficient and the CH index of the clustering to evaluate the clustering effect.

[0134] By observing the clustering results, this application analyzes the distribution of samples in each cluster. If the number of samples in a certain cluster is significantly more than that in other clusters, then this application deems that the sample distribution in this cluster is relatively concentrated.

[0135] Through the analysis of the clustering results, this application discovers some potential patterns or regularities. For example, if the samples in a certain cluster have similar features or attributes, then this application deems that there is a certain potential pattern or regularity in this cluster.

[0136] S8 Calculate the correlation degree between the trajectory clustering data obtained according to the iterated cluster centers and the fault types, and determine the transformer fault types according to the correlation degree.

[0137] Calculate the correlation degree between the trajectory clustering data obtained according to the iterated cluster centers and the fault types. The calculation of the correlation degree is achieved by comparing the slope of the trajectory clustering data with the reference slope of the fault type. The specific steps are as follows:

[0138] Determine the reference slope of the fault type. This reference slope is calculated according to the abscissa and ordinate of the data points in the dataset of this fault type.

[0139] Calculate the slope of the trajectory clustering data. This slope is calculated according to the abscissa and ordinate of the trajectory clustering data.

[0140] Calculate the correlation degree between the trajectory clustering data and the fault types. If the signs of the slope of the trajectory clustering data and the reference slope of the fault type are the same, then the correlation degree is equal to the difference between the slope of the trajectory clustering data and the reference slope of the fault type divided by the absolute value of the difference plus a positive number ε (ε > 0); if the signs of the slope of the trajectory clustering data and the reference slope of the fault type are opposite, then the correlation degree is equal to -1 multiplied by the difference between the slope of the trajectory clustering data and the reference slope of the fault type divided by the absolute value of the difference plus a positive number ε (ε > 0).

[0141] Determine the transformer fault types according to the magnitude of the correlation degree. Specifically, if the correlation degree between the trajectory clustering data and a certain fault type is the largest, then classify this trajectory clustering data into this fault type. In this way, the fault types of the transformer are determined by calculating the correlation degrees between the trajectory clustering data and different fault types.

[0142] Furthermore, when the signs of ki and k0 (ki and k0 represent the slope of the data to be detected and the reference slope of the fault type respectively) are the same, the correlation coefficient increases monotonically with the increase of the slope difference between the two, as Figure 2 shown. At this time, if the slope difference between k0 and ki is larger, then the correlation coefficient between these two slopes is larger.

[0143] When the signs of ki and k0 are opposite, the correlation coefficient decreases monotonically as the slope difference between the two increases, as Figure 2 shown. At this time, if the slope difference between k0 and ki is larger, the correlation coefficient between the two slopes is smaller instead.

[0144] For the calculation of the correlation degree of each piece of data in the dataset with respect to this type of fault, the following steps are taken:

[0145] First, it is necessary to determine the slope difference between each piece of data in the dataset and the reference slope k0. This slope difference is calculated based on the abscissa and ordinate of the data points.

[0146] Then, the correlation coefficient is calculated based on the slope difference obtained in the previous step. If the signs of ki and k0 are the same, the correlation coefficient is equal to the slope difference divided by the absolute value of the slope difference plus a positive number ε (ε>0); if the signs of ki and k0 are opposite, the correlation coefficient is equal to -1 multiplied by the slope difference divided by the absolute value of the slope difference plus a positive number ε (ε>0).

[0147] Finally, the correlation coefficients corresponding to each piece of data are sorted, and the fault type corresponding to the largest correlation coefficient is selected as the fault type of this data.

[0148] For the calculation of the characteristic correlation degree of the fault, the following steps are taken:

[0149] First, it is necessary to determine the reference slope k0 corresponding to each type of fault. This reference slope is calculated based on the abscissa and ordinate of the data points in the dataset of this type of fault.

[0150] Then, based on the reference slope k0 of each type of fault, the slope difference between each piece of data in this type of fault and the reference slope k0 is calculated.

[0151] Finally, the characteristic correlation degree of this type of fault is calculated based on the slope difference obtained in the previous step. If the signs of all the data of this type of fault and the reference slope k0 are the same, the characteristic correlation degree is equal to the average value of the correlation coefficients corresponding to all the data; if the signs of all the data of this type of fault and the reference slope k0 are opposite, the characteristic correlation degree is equal to the average value of the absolute values of the correlation coefficients corresponding to all the data.

[0152] For the calculation of the correlation degree of the fault for the data to be inspected, the following steps are taken:

[0153] First, it is necessary to determine the reference slope ki corresponding to the data to be inspected. This reference slope is calculated based on the abscissa and ordinate of the data to be inspected.

[0154] Then, according to the reference slope ki of the data to be detected, the slope difference between the data to be detected and the reference slope k0 of each fault type is calculated respectively.

[0155] Finally, the correlation of the data to be tested to each fault type is calculated based on the slope difference obtained in the previous step. If the reference slope ki of the data to be tested has the same sign as the reference slope k0 of a certain fault type, then the correlation of the data to be tested to this fault type is equal to the slope difference between the reference slope ki of the data to be tested and the reference slope k0 of this fault type divided by the absolute value of the slope difference plus a positive number ε (ε>0); if the reference slope ki of the data to be tested has the opposite sign to the reference slope k0 of a certain fault type, then the correlation of the data to be tested to this fault type is equal to -1 multiplied by the slope difference between the reference slope ki of the data to be tested and the reference slope k0 of this fault type divided by the absolute value of the slope difference plus a positive number ε (ε>0).

[0156] A concrete example:

[0157] More than 2,000 transformer oil chromatographic data were collected and organized from the literature published in recent years. Each data included the content of five gases: H2, CH4, C2H6, C2H4, and C2H2, as well as the corresponding fault category.

[0158] like Figure 3 As shown, the data covers six types of faults: low temperature fault, medium temperature fault, high temperature fault, partial discharge, low energy discharge and arc discharge.

[0159] These data come from many sources and are highly dispersed, so they need to be screened and sorted. First, the data is initialized to calculate the percentage of hydrogen in hydrogen hydrocarbons and the percentage of various hydrocarbon gases in total hydrocarbons; then the data is classified according to the fault type, the mean and standard deviation of each fault data are obtained, and the data outside the mean twice the standard deviation are eliminated. After multiple rounds of screening, 60 DGA data are left for clustering test analysis, including 10 data for each fault type, and each data contains 5 attributes, namely the gas content of H2, CH4, C2H6, C2H4, and C2H2.

[0160] The clustering test includes the following steps:

[0161] (1) Data initialization. The data initialization process is to calculate the proportion of various fault gases. The H2 content in DGA data is usually large. During the initialization process, the percentage of hydrogen in hydrogen hydrocarbons and the percentage of various hydrocarbon gases in total hydrocarbons are calculated.

[0162] (2) The DGA data set is divided into six categories. The iteration stopping condition is set as the cluster center and membership adjustment amount of two consecutive iterations are both less than the preset parameter ε, that is:

[0163]

[0164]

[0165] In the formula, ε = 1×10-3.

[0166] (3) Statistically determine the fault types represented by each category (subset), and conduct discriminant analysis on the classification results.

[0167] Considering the problems that the indicators of the dissolved gas content in transformer oil are correlated, non-normal, and have a cumulative trend, introduce the above-mentioned transformer oil chromatogram state evaluation method. For the data after the one-dimensional outlier function transformation, what is commonly used in statistics to reflect the degree of outliers is: observe whether the new observed data falls into the normal operation interval of the transformer constructed by this one-dimensional data:

[0168] (Q1 - 1.5×IQR, Q3 + 1.5×IQR);

[0169] Where Q1 and Q3 represent the 0.25 and 0.75 quantiles of the data respectively, and IQR = Q3 - Q1. Observed data exceeding this interval is usually considered outlier data.

[0170]

[0171] Among them, med(Zn) represents the median of Zn. (w1, w3) represents different ranges in the following cases of right skew and left skew. Since the definition of the one-dimensional outlier function AO1(z) mainly involves indicators such as the median, upper and lower quantiles of Zn, it has good robustness. However, considering the 7 different dimensions of transformer oil chromatogram data, it is necessary to further adopt the projection pursuit method to convert it into one-dimensional data. Specifically, a 7-dimensional projection outlier function applicable to oil chromatogram data can be used.

[0172] Collect the oil chromatogram data of n periods before the collection time T and store it as Xn.

[0173] Calculate the minimum outlier value point x0 (or the center point), the bag area (that is, the area containing half of the observed values of Xn), the fence area (that is, the area centered on x0 and expanding the bag by 1.8 times), and the safety area (that is, the area centered on x0 and expanding the bag until its boundary touches a certain upper limit specified in Q / GDW 1168—2013) respectively.

[0174] (3) Calculate the AO value corresponding to the oil chromatogram data x at time T, and determine whether the state of x is in: (i) the area bag (absolutely safe), (ii) outside the area bag but within the fence (generally safe), (iii) outside the fence but within the safety area (safe but attention needed); and (iv) outside the safety area (abnormal); and make corresponding early warning evaluations according to (i)-(iv).

[0175] (4) For the new state evaluation at time T+1: First, remove X1 from Xn, then add the x at time T, and obtain the updated Xn, and repeat the evaluation.

Claims

1. A method for evaluating the chromatographic state of transformer oil, characterized in that Including: S1 Preprocess the transformer oil chromatogram sample data set and set the parameters for parallel computing, including the number of trajectories, the length of the chaotic sequence, the iteration accuracy, and the maximum number of iterations; S2 Select multiple clustering starting points corresponding to the number of trajectories as the initial clustering centers; S3 Calculate the membership degrees of the clustering centers and update the clustering centers according to the membership degrees; S4 Calculate the adjustment amount of the clustering centers according to the updated clustering centers; S5 Obtain the chaotic sequence according to the length of the chaotic sequence and map the chaotic sequence to the corresponding value range of the adjustment amount; S6 Update the clustering centers again according to the adjustment amount obtained by mapping the chaotic sequence; S7 Repeat steps S2 - S6 until the clustering process converges to the preset accuracy or reaches the preset maximum number of iterations; S8 Calculate the correlation degree between the obtained trajectory clustering data and the fault type according to the clustering centers after iteration, and determine the transformer fault type according to the correlation degree.

2. The method for evaluating the chromatographic state of transformer oil according to claim 1, wherein, Calculate the membership degree of the clustering center, and the expression is as follows: Among them, u ij is the membership degree, i represents the trajectory data number, j represents the number of updates of the clustering center for each trajectory data, x represents the trajectory data, v represents the initial clustering center, k represents the slope, and H is the symbol of the product formula.

3. The method for evaluating the chromatographic state of transformer oil according to claim 2, characterized in that Update the clustering center according to the membership degree, and the expression is as follows: Substitute u ij into the following formula to obtain the clustering center V(L) = (v1`(L), v2`(L),..., vω`(L)); Where L represents the number of iterations, m is a constant, usually taking the value of 2, n represents the number of elements μij which is the membership degree of the sample Xi to the clustering center Vj; Obtain the objective function value F(L) = (F1(L), F2(L),....., Fω(L)), and obtain the historical minimum value Fmin = (F1 - min, F2 - min,....... Fω - min) of each trajectory; Where the historical minimum value Fi - min of the i - th trajectory is Fi - min = min(Fi(t)), t = [1, 2,..., L], and the corresponding clustering center vi - min of Fi - min; Determine that up to the current step position, the historical minimum value Fglobal = min(Fi - min), t = [1, 2,..., ω], and the corresponding clustering center vglobal of Fglobal.

4. The method for evaluating the chromatographic state of transformer oil according to claim 1, wherein Calculate the adjustment amount of the clustering center, and the expression is as follows: ΔV = λ1ΔV1 + λ2ΔV2 + λ3ΔV3; Where ΔV1 is the current adjustment amount of each trajectory, ΔV2 is the difference between the current clustering center of each trajectory and the historical optimal center of this trajectory, ΔV3 is the difference between the current clustering center of each trajectory and ΔV2 = Vglobal(L) - V(L), and λ1, λ2, and λ3 are weight factors, λ1 + λ2 + λ3 = 1.

5. The method for evaluating the chromatographic state of transformer oil according to claim 4, characterized in that The expressions of the said λ1, λ2, and λ3 are as follows: λ2 = λ1(1 - λ1) λ3 = (1 - λ1)*(1 - λ1).

6. The method for evaluating the chromatographic state of transformer oil according to claim 1, characterized in that Obtain the chaotic sequence according to the length of the chaotic sequence and map the chaotic sequence to the corresponding value range of the adjustment amount, including: Update n chaotic sequences using a preset formula; Map the n chaotic sequences to the corresponding value range of the adjustment amount; Obtain the tentative clustering centers corresponding to n adjustment amounts between V(L) and V(L + 1) through mapping; Compare whether the tentative clustering centers are better than the existing clustering centers obtained by the formula V(L + 1) = V(L) + ΔV; If so, update the cluster center.

7. A method for evaluating the chromatographic state of transformer oil according to any one of claims 1 to 6, characterized in that It further includes: Set multiple clustering trajectories for parallel computing, and realize the shared utilization of information between the trajectories.

8. A transformer oil chromatogram state evaluation device, characterized in that It includes: A setting module, used to preprocess the transformer oil chromatogram sample data set and set the parameters of parallel computing, including the number of trajectories, the length of the chaotic sequence, the iteration accuracy, and the maximum number of iterations; A selection module, used to select multiple clustering starting points corresponding to the number of trajectories as the initial cluster centers; A first update module, used to calculate the membership degree of the cluster center and update the cluster center according to the membership degree; A calculation module, used to calculate the adjustment amount of the cluster center according to the updated cluster center; A mapping module, used to obtain a chaotic sequence according to the length of the chaotic sequence and map the chaotic sequence to the value range corresponding to the adjustment amount; A second update module, used to update the cluster center again according to the adjustment amount obtained by mapping the chaotic sequence; Repeat the above modules until the clustering process converges to a preset accuracy or reaches a preset maximum number of iterations; A judgment module, used to calculate the correlation degree between the obtained trajectory clustering data and the fault type according to the cluster center after iteration, and determine the transformer fault type according to the correlation degree.

9. A transformer oil chromatogram state evaluation device, characterized in that, It includes: A memory, used to store the computer-executable program of the method for evaluating the state of transformer oil chromatogram according to any one of claims 1 to 7; A processor, used to retrieve the computer-executable program and execute: S1 preprocess the transformer oil chromatogram sample data set and set the parameters of parallel computing, including the number of trajectories, the length of the chaotic sequence, the iteration accuracy, and the maximum number of iterations; S2 select multiple clustering starting points corresponding to the number of trajectories as the initial cluster centers; S3 calculate the membership degree of the cluster center and update the cluster center according to the membership degree; S4 calculate the adjustment amount of the cluster center according to the updated cluster center; S5 obtain a chaotic sequence according to the length of the chaotic sequence and map the chaotic sequence to the value range corresponding to the adjustment amount; S6 update the cluster center again according to the adjustment amount obtained by mapping the chaotic sequence; S7 repeat steps S2 - S6 until the clustering process converges to a preset accuracy or reaches a preset maximum number of iterations; S8 calculate the correlation degree between the obtained trajectory clustering data and the fault type according to the cluster center after iteration, and determine the transformer fault type according to the correlation degree.

10. A storage medium, characterized in that, Store a computer-executable program, which is used to be retrieved by a processor and execute the steps of the method for evaluating the state of transformer oil chromatogram according to any one of claims 1 to 7.

Images