Transformer diagnosis method and system based on insulating oil detection

Abstract

The invention discloses a transformer diagnosis method and system based on insulating oil detection. The diagnosis method comprises the following steps: acquiring gas concentrations of gas in transformer oil at a plurality of different time nodes within a time range; fitting according to the gas concentration of each component gas under the plurality of time nodes to obtain a gas dissipation curve of each component; calculating the gas production rate of each component gas in the time range according to the gas dissipation curve; obtaining a ratio matrix of the gas production rates of the component gases; and predicting a transformer operation fault corresponding to the ratio matrix through a neural network model. According to the method, the neural network model is utilized to predict the ratio matrix established among the multiple component gases, so that the comprehensive operation state of the transformer corresponding to the ratio matrix is obtained.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power detection, and in particular to a transformer diagnosis method and system based on insulating oil detection. Background Art

[0002] Common transformer faults, such as partial discharge and transformer overheating, often produce a variety of fault characteristic gases. The type and concentration of the fault gas produced when the transformer fails are closely related to whether the transformer fails or even the type of fault, and are important monitoring objects for transformer fault early warning. Especially for oil-immersed transformers, the internal oil and paper insulation structure can easily age and deteriorate due to the above faults, producing a variety of fault characteristic gases, including: hydrogen (H2), methane (CH4), acetylene (C2H2), ethylene (C2H4), ethane (C2H6), carbon monoxide (CO), carbon dioxide (CO2), etc. Unlike other types of transformers, the fault gases produced by oil-immersed transformers are mostly directly dissolved in the transformer oil, making detection difficult.

[0003] The multi-component transformer oil dissolved gas detection device developed based on laser photoacoustic spectroscopy detection technology can detect multi-component dissolved gases in transformer oil, but comprehensive diagnosis and confirmation of the transformer based on the results after detecting dissolved gases is still a technical challenge.

[0004] In addition, the existing oil and gas detection equipment is mainly used to detect the gas components after the transformer oil is degassed at a certain moment. In the process of long-term and multiple transformer monitoring, the existing oil and gas detection equipment is only used to detect the gas components at different moments and evaluate the operating status of the transformer at each moment, but lacks evaluation means for the comprehensive operating status of the transformer over a period of time. Summary of the invention

[0005] Based on this, a first aspect of an embodiment of the present invention discloses a transformer diagnosis method based on insulating oil detection.

[0006] The diagnostic method comprises:

[0007] Obtaining the gas concentration of gas in transformer oil at multiple different time nodes within a time range;

[0008] The gas escape curve of each component is obtained by fitting the gas concentration of each component gas at multiple time nodes;

[0009] Calculate the gas production of each component gas within the time range according to the gas escape curve;

[0010] Obtain the ratio matrix of gas production between each component gas;

[0011] The transformer operation fault corresponding to the ratio matrix is ​​predicted through a neural network model.

[0012] In the embodiments disclosed in the present invention,

[0013] The component gases include one or more of acetylene C2H2, ethylene C2H4, methane CH4, ethane C2H6, hydrogen H2, carbon monoxide CO, carbon dioxide CO2, trace water H2O, and oxygen O2.

[0014] In the embodiments disclosed in the present invention,

[0015] The gas escape curve configuration is fitted as follows:

[0016] The gas emission curve of each component gas is fitted by the least square method.

[0017] In the embodiments disclosed in the present invention,

[0018] Fitting the gas concentration of each component gas at multiple time points to obtain the gas escape curve of each component, and selecting an abnormal interval of the gas escape curve of at least one of the component gases;

[0019] The gas production of each component gas in the abnormal interval is calculated according to the gas escape curve.

[0020] In the embodiments disclosed in the present invention,

[0021] The abnormal interval configuration is selected as follows:

[0022] The maximum slope of the gas escape curve traversing each component gas;

[0023] comparing the maximum slopes of the various component gases;

[0024] Selecting the slope change starting point and slope change end point of the maximum slope with the largest value;

[0025] The abnormal interval within the time range is obtained according to the slope change starting point and the slope change end point.

[0026] In the embodiments disclosed in the present invention,

[0027] Obtaining the gas production includes calculating one or more of a relative gas production rate, an absolute gas production rate or a total gas production rate of each component gas within the time range according to the gas escape curve;

[0028] Acquiring the ratio matrix includes acquiring the ratio matrix of relative gas production rates, absolute gas production rates or total gas production between the component gases.

[0029] In the embodiments disclosed in the present invention,

[0030] Training the neural network model includes,

[0031] Obtaining sample gas concentrations of gas in sample transformer oil at multiple different time points within a sample time range;

[0032] The sample gas dissipation curve of each component is obtained by fitting the sample gas concentration of each component gas at multiple time nodes;

[0033] Calculate the sample gas production of each component gas within the sample time range according to the sample gas escape curve;

[0034] Obtain a sample matrix of sample gas production between each component gas;

[0035] Acquire a sample label of the sample transformer oil within the time range, wherein the sample label represents a fault state of the sample transformer;

[0036] The neural network model is trained according to a combination of the sample matrix and the sample labels.

[0037] In the embodiments disclosed in the present invention,

[0038] The neural network model is a BP neural network model with a fully connected layer.

[0039] In the embodiments disclosed in the present invention,

[0040] Improved BP neural network model based on particle population.

[0041] Based on this, a second aspect of an embodiment of the present invention discloses a transformer diagnosis system based on insulating oil detection.

[0042] The diagnostic system comprises,

[0043] The acquisition module obtains the gas concentration of the gas in the transformer oil at multiple different time points within a time range;

[0044] A fitting module is used to obtain the gas escape curve of each component according to the gas concentration of each component gas at multiple time nodes;

[0045] A calculation module, for calculating the gas production of each gas component within the time range according to the gas escape curve; and obtaining a ratio matrix of the gas production between each gas component;

[0046] The prediction module predicts the transformer operation fault corresponding to the ratio matrix through a neural network model.

[0047] Compared with the prior art, the embodiments of the present invention, in a first aspect, use a neural network model to predict a ratio matrix established between multiple component gases to obtain a comprehensive operating state of the transformer corresponding to the ratio matrix; in a second aspect, improve the initial weight and threshold of the neural network model based on the particle population, optimize the training of the neural network model, and improve the reliability of the model prediction.

[0048] In view of the above-mentioned solutions, the present invention will describe the disclosed exemplary embodiments in detail below with reference to the accompanying drawings, so as to make other features and advantages of the embodiments of the present invention clear. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0050] Figure 1 It is a flowchart of a transformer diagnosis method based on insulating oil detection;

[0051] Figure 2 It is a structural schematic diagram of the transformer diagnosis method based on insulating oil detection. DETAILED DESCRIPTION

[0052] In order to facilitate understanding of the present application, the present application will be described more fully below with reference to the relevant drawings. Embodiments of the present application are provided in the drawings. However, the present application can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present application more thorough and comprehensive.

[0053] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application.

[0054] The embodiment of the present invention discloses a transformer diagnosis method based on insulating oil detection. The diagnosis method can predict transformer operation failures based on transformer insulating oil detection, solving the technical problem of the lack of evaluation means for the comprehensive operation status of the transformer within a period of time in the prior art.

[0055] Figure 1 A schematic flow chart of the transformer diagnosis method based on insulating oil detection according to this embodiment is shown.

[0056] Figure 1 The diagnostic method is shown to include steps 10 to 50 .

[0057] 10. Obtain the gas concentration of the gas in the transformer oil at multiple different time nodes within a time range.

[0058] Among them, the real-time gas concentration of each component gas released by the transformer oil at multiple different time nodes within a time range can be obtained through high-precision online monitoring equipment (such as photoacoustic spectroscopy gas detection equipment).

[0059] Preferably, each component gas includes but is not limited to acetylene C2H2, ethylene C2H4, methane CH4, ethane C2H6, hydrogen H2, carbon monoxide CO, carbon dioxide CO2, trace water H2O, oxygen O2, etc. The detection range of the high-precision online monitoring equipment in the prior art for acetylene C2H is 0.1-200 μL / L; the detection range of ethylene C2H4, methane CH4, and ethane C2H6 is 0.5-1000 μL / L; the detection range of hydrogen H2 is 2-2000 μL / L; the detection range of carbon monoxide CO is 25-5000 μL / L; the detection range of carbon dioxide CO2 is 25-15000 μL / L.

[0060] 20 The gas escape curves of each component gas are obtained by fitting the gas concentration of each component gas at multiple time nodes.

[0061] Among them, the gas escape curve of each component gas can be obtained by least square fitting using the gas concentration at multiple time nodes:

[0062] Taking acetylene C2H2 as an example, the gas concentrations at n different time nodes are known: (x1, y1)(x2, y2)...(xn, yn).

[0063] Through research and analysis, it is found that the change in the gas concentration of acetylene C2H2 can be expressed by a curve equation approximating a parabola y=a 2 x 2 +a 1 x+a 0 . Among them, a0, a1, and a2 are unknown coefficients. If (x1, y1) is substituted into the equation, we can get That is Then (xi,yi), we have:

[0064] The combined matrix is: Assumptions For A, For T, For X, then AX = T. Then x = (A T A) -1 A TT, and then obtain the curve equation of the gas escape curve of acetylene C2H2 within a period of time.

[0065] 30 Calculate the gas production of each component gas within the time range based on the gas escape curve.

[0066] Among them, the gas production is the accumulation of gas production in the gas dissipation curve within the abnormal interval. The abnormal interval is an interval of an abnormal component gas in the gas dissipation curve. Selecting the abnormal interval includes traversing the maximum slope of the gas dissipation curve of each component gas, comparing the maximum slope of each component gas, and selecting the slope change starting point and slope change end point of the maximum slope with the largest value, wherein the slope change starting point is the time starting point of the maximum slope, and the slope change end point is the time end point from the maximum slope to the time when it gradually becomes less than a slope threshold. According to the slope change starting point and slope change end point, the abnormal interval corresponding to the gas dissipation curve of the component gas is obtained. The gas production of each component gas is calculated based on the abnormal interval of the same component gas as the time basis.

[0067] In some embodiments, the gas production of each component gas is calculated based on the time when the respective maximum slope is the respective abnormal interval.

[0068] Furthermore, obtaining the gas production includes calculating one or more of the relative gas production rate, the absolute gas production rate or the total gas production of each component gas within a time range according to the gas escape curve.

[0069] 40 obtains the ratio matrix of gas production between each component gas.

[0070] Wherein, obtaining the ratio matrix includes obtaining the ratio matrix of the relative gas production rate, absolute gas production rate or total gas production between each component gas. The vertical direction of the ratio matrix is ​​each component gas, and the horizontal direction is the ratio of the relative gas production rate, absolute gas production rate or total gas production to other component gases.

[0071] 50 A neural network model is used to predict the transformer operation fault corresponding to the ratio matrix.

[0072] Among them, the neural network model is a BP neural network model with a fully connected layer.

[0073] The BP neural network includes an input layer, a hidden layer and an output layer, wherein tansig is selected as the transfer function of the hidden layer, purelin is selected as the transfer function of the output layer, and learngdm is selected as the learning function; the number of neural nodes in the hidden layer is determined by an empirical formula method and a trial method, for example, the range of the number of nodes in the hidden layer is determined according to an empirical formula, and then the number within the range is tried one by one, and finally the number of neural nodes corresponding to the minimum error is selected. The input data enters the network from the input layer, and is processed layer by layer through the hidden layer until the output layer outputs. The state of the neurons in each layer only affects the state of the neurons in the next layer. If the expected output is not obtained in the output layer, the signal transmission process is transferred to back propagation, and the weights and thresholds are continuously adjusted according to the prediction error, so that the predicted output result of the BP neural network is constantly close to the expected output value. In this embodiment, the input of the BP neural network is a ratio matrix, and the output is a sample label that characterizes the comprehensive operating state of the transformer.

[0074] Furthermore, this embodiment obtains the initial weight and threshold of the BP neural network based on the particle population, including:

[0075] Initialize various parameters of various particle swarm algorithms, including the number of particle swarms, maximum number of updates, etc.

[0076] Initialize the velocity (V) and position (X) of each particle and calculate the fitness value of the particle.

[0077] Configure the update function for particles.

[0078] Formula (1)Vk+1=wVK+c1r1(q1-xk)+c2r2(q2-xk),

[0079] Formula (2) Xk+1=Xk+Vk+1,

[0080] Where V represents the velocity of the i-th particle in the d-dimensional space, X represents the position of the i-th particle in the d-dimensional space, q1 represents the historical optimal position of the i-th particle in the d-dimensional space, q2 represents the historical optimal position of all particles in the d-dimensional space, r1 represents the first random number between 0 and 1, r2 represents the second random number between 0 and 1, c1 represents the first acceleration factor, c2 represents the second acceleration factor, k is the number of particle swarm updates, and ω is the inertia weight ranging from 0.5 to 1.

[0081] Update the velocity (V) and position (X) of each particle according to equations (1) and (2).

[0082] Perform genetic operations on the particle swarm after updating the speed V and position X, treat each particle as a chromosome, evaluate the fitness of each chromosome again, and retain the optimal chromosome;

[0083] Selection operation: Roulette method is used, and the probability of individual i being selected is pi.

[0084] N is the total number of particles, and Fi is the fitness value of the i-th particle.

[0085] Crossover operation: According to the crossover probability Pc (preconfigured), two individuals (parents) are randomly selected from the chromosome group after the selection operation to exchange partial information, thereby generating two new chromosomes. Assume that the i-th chromosome (ai) and the k-th chromosome (ak) cross at the position of gene j.

[0086] a*i and a*k represent two new offspring chromosome individuals after crossover, and b represents a random number between 0 and 1.

[0087] Mutation operation: The mutation operation is similar to gene mutation in reality, that is, some chromosomes are selected according to the set mutation probability Pm (preconfigured), and some gene information in the chromosome is modified to increase the possibility of obtaining the optimal particle.

[0088] g*i is the mutated gene information, and rand is a random number between 0 and 1.

[0089] The fitness of the generated optimal chromosome is evaluated again, and the most optimal chromosome is retained; then the process returns to step 4) to loop until the maximum number of particle swarm updates Maxiter is reached, and the global optimal particle is obtained.

[0090] The obtained global optimal particle is decoded according to the network structure of the configured BP neural network and the length of the initial weights and thresholds to obtain the current excellent network initial weights and thresholds.

[0091] Furthermore, a training set for training the BP neural network model is obtained, including obtaining the sample gas concentration of the gas in the sample transformer oil at multiple different time nodes within a sample time range. The sample gas dispersion curve of each component gas is fitted according to the sample gas concentration of each component gas at multiple time nodes. The sample gas production of each component gas within the sample time range is calculated according to the sample gas dispersion curve. A sample matrix of the sample gas production between each component gas is obtained. The sample label of the sample transformer oil within the time range is obtained, and the sample label represents the fault state of the sample transformer. According to the combination of the sample matrix and the sample label as a sample set, for example,

[0092]

[0093] And, the improved BP neural network model is trained according to multiple training sets, network initial weights, and thresholds until the comprehensive transformer comprehensive operating state output by the improved BP neural network model corresponding to the test set after training is consistent with the actual comprehensive transformer comprehensive operating state or the deviation is less than the threshold, and the training of the improved BP neural network model is terminated.

[0094] Based on this, the transformer diagnosis method based on insulating oil detection in this embodiment uses a neural network model to predict the ratio matrix established between multiple component gases to obtain the comprehensive operating state of the transformer corresponding to the ratio matrix.

[0095] Furthermore, an embodiment of the present invention includes a transformer diagnosis system based on insulating oil detection.

[0096] Figure 2 A schematic structural diagram of a transformer diagnosis system based on insulating oil detection in this embodiment is shown.

[0097] Figure 2 The diagnostic system is shown to include an acquisition module 100, a fitting module 200, a calculation module 300 and a prediction module 400. The acquisition module 100 obtains the gas concentration of the gas in the transformer oil at multiple different time nodes within a time range. The fitting module 200 obtains the gas escape curve of each component according to the gas concentration of each component gas at multiple time nodes. The calculation module 300 calculates the gas production of each component gas within the time range according to the gas escape curve. The ratio matrix of the gas production between the components of gas is obtained; the prediction module 400 predicts the transformer operation fault corresponding to the ratio matrix through a neural network model.

[0098] Based on this understanding, the technical solution of the present invention can essentially or the part that contributes to the prior art can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of various embodiments of the present invention.

[0099] Note that the above are only preferred embodiments of the present invention and the technical principles used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the appended claims.

Claims

1. A transformer diagnosis method based on insulating oil detection, characterized in that: The diagnostic method comprises: Obtaining the gas concentration of gas in transformer oil at multiple different time nodes within a time range; The gas escape curve of each component is obtained by fitting the gas concentration of each component gas at multiple time nodes; Calculate the gas production of each component gas within the time range according to the gas escape curve; Obtain the ratio matrix of gas production between each component gas; The transformer operation fault corresponding to the ratio matrix is ​​predicted through a neural network model.

2. The transformer diagnosis method based on insulating oil detection according to claim 1 is characterized in that: The component gases include one or more of acetylene C2H2, ethylene C2H4, methane CH4, ethane C2H6, hydrogen H2, carbon monoxide CO, carbon dioxide CO2, trace water H2O, and oxygen O2.

3. The transformer diagnosis method based on insulating oil detection according to claim 1 is characterized in that: The gas escape curve configuration is fitted as follows: The gas emission curve of each component gas is fitted by the least square method.

4. The transformer diagnosis method based on insulating oil detection according to claim 1 is characterized in that: Fitting the gas concentration of each component gas at multiple time points to obtain the gas escape curve of each component, and selecting an abnormal interval of the gas escape curve of at least one of the component gases; The gas production of each component gas in the abnormal interval is calculated according to the gas escape curve.

5. The transformer diagnosis method based on insulating oil detection according to claim 4 is characterized in that: The abnormal interval configuration is selected as follows: The maximum slope of the gas escape curve traversing each component gas; comparing the maximum slopes of the various component gases; Selecting the slope change starting point and slope change end point of the maximum slope with the largest value; The abnormal interval within the time range is obtained according to the slope change starting point and the slope change end point.

6. The transformer diagnosis method based on insulating oil detection according to claim 1 is characterized in that: Obtaining the gas production includes calculating one or more of a relative gas production rate, an absolute gas production rate or a total gas production rate of each component gas within the time range according to the gas escape curve; Acquiring the ratio matrix includes acquiring the ratio matrix of relative gas production rates, absolute gas production rates or total gas production between the component gases.

7. The transformer diagnosis method based on insulating oil detection according to claim 1 is characterized in that: Training the neural network model includes, Obtaining sample gas concentrations of gas in sample transformer oil at multiple different time points within a sample time range; The sample gas dissipation curve of each component is obtained by fitting the sample gas concentration of each component gas at multiple time nodes; Calculate the sample gas production of each component gas within the sample time range according to the sample gas escape curve; Obtain a sample matrix of sample gas production between each component gas; Acquire a sample label of the sample transformer oil within the time range, wherein the sample label represents a fault state of the sample transformer; The neural network model is trained according to a combination of the sample matrix and the sample labels.

8. The transformer diagnosis method based on insulating oil detection according to claim 7 is characterized in that: The neural network model is a BP neural network model with a fully connected layer.

9. The transformer diagnosis method based on insulating oil detection according to claim 8, characterized in that: Improved BP neural network model based on particle population.

10. A transformer diagnostic system based on insulating oil detection, characterized in that: The diagnostic system comprises, The acquisition module obtains the gas concentration of the gas in the transformer oil at multiple different time points within a time range; A fitting module is used to obtain the gas escape curve of each component according to the gas concentration of each component gas at multiple time nodes; A calculation module, for calculating the gas production of each gas component within the time range according to the gas escape curve; and obtaining a ratio matrix of the gas production between each gas component; The prediction module predicts the transformer operation fault corresponding to the ratio matrix through a neural network model.

Images