Transformer fault detection method based on multi-point oil chromatographic data

Abstract

The present application relates to the technical field of transformer fault identification, and in particular to a transformer fault detection method based on multi-point oil chromatogram data, comprising: obtaining oil chromatogram monitoring data of a plurality of oil chromatogram sampling points in a transformer; determining spatial distribution deviation of a same gas component of each sampling point, and determining spatial distribution eligibility; in response to the spatial distribution eligibility being eligible, eliminating abnormal sampling points from the oil chromatogram sampling points to obtain a set of trusted point positions; based on the oil chromatogram monitoring data, determining time sequence change rates of each point position in the set of trusted point positions, determining a synchronous jump rate according to the number of point positions and the total number of point positions in the set of trusted point positions; updating the set of trusted point positions; and in response to the number of trusted point positions in the updated set of trusted point positions being greater than or equal to a preset number threshold, performing transformer fault detection. The present application improves the reliability of multi-point oil chromatogram data in transformer fault detection.

Description

Technical Field

[0001] This invention relates to the field of transformer fault identification technology, and more specifically, to a transformer fault detection method based on multi-point oil chromatography data. Background Technology

[0002] Dissolved gas analysis in oil is an important method for transformer fault detection. Online oil chromatography monitoring devices can promptly detect potential faults by periodically detecting the composition and content of dissolved gases in insulating oil.

[0003] Traditional transformer oil chromatography monitoring mainly employs single-point sampling. However, due to the complex internal structure of transformers and uneven oil flow distribution, single-point sampling has monitoring blind spots. To address this, existing technologies have proposed multi-point oil chromatography monitoring schemes, such as setting up multiple sampling points at different locations within the transformer, or setting up multiple detection channels to achieve independent detection by each channel.

[0004] However, multi-point monitoring still faces technical shortcomings in practical applications. On the one hand, oil chromatography online monitoring devices operate in the complex environment of substations for extended periods, and are susceptible to data drift and distortion due to factors such as temperature, humidity, and electromagnetic interference. Frequent distortion of online monitoring data due to device malfunctions severely impacts the effectiveness and practicality of the monitoring system. On the other hand, existing technologies lack methods for layer-by-layer filtering of multi-point data, making it impossible to determine the reasonableness of data across multiple dimensions and levels. This results in distorted data mixing with genuine fault data in the diagnostic process, reducing the reliability of fault detection. Summary of the Invention

[0005] To address this issue, the present invention provides a transformer fault detection method based on multi-point oil chromatography data, which overcomes the problem in the prior art that the multi-point oil chromatography monitoring data is not considered for layer-by-layer validity identification, and that interfered data and real fault data are mixed in the fault detection process, resulting in poor reliability of multi-point oil chromatography data in transformer fault detection.

[0006] To achieve the above objectives, the present invention provides a transformer fault detection method based on multi-point oil chromatography data, comprising: Obtain oil chromatographic monitoring data from several oil chromatographic sampling points in the transformer; Based on the oil chromatography monitoring data, the spatial distribution deviation of the same gas component at each sampling point is determined, and the spatial distribution qualification is determined based on the spatial distribution deviation. In response to the spatial distribution being deemed acceptable, based on the oil chromatography monitoring data, the correlation deviation index of each sampling point is determined, and abnormal sampling points are removed from the oil chromatography sampling points based on the correlation deviation index to obtain a set of reliable sampling points. Based on the oil chromatography monitoring data, the time-series change rate of each point in the set of reliable points is determined, and the number of points in the set of reliable points whose time-series change rate is greater than or equal to a preset change rate threshold is counted. The synchronization jump rate is determined according to the number of points and the total number of points in the set of reliable points. The trusted point set is processed based on the synchronization jump rate, and the trusted point set is updated. In response to the number of trusted points in the updated trusted point set being greater than or equal to a preset number threshold, transformer fault detection is performed based on the oil chromatography monitoring data in the trusted point set.

[0007] By employing the above technical solution, this invention, through acquiring oil chromatographic monitoring data from various sampling points in the transformer, can determine the spatial distribution deviation of each sampling point and judge whether the spatial distribution is qualified. Only when the spatial distribution is qualified will subsequent analysis proceed, thus eliminating the possibility of overall data distortion before fault detection. By determining the correlation deviation index of each sampling point in response to a qualified spatial distribution and eliminating abnormal sampling points, data points affected by local interference can be isolated from the data source, preventing a single abnormal point from contaminating the entire fault detection process. By determining the temporal change rate of each sampling point and counting the number of points exceeding the threshold to determine the synchronous jump rate, data stability can be quantitatively evaluated in the temporal dimension, distinguishing local jumps. The system synchronizes changes with the global data. By processing and updating the set of trusted data points based on the synchronous jump rate, it can simultaneously identify multi-point synchronous anomaly signals during data filtering, providing data quality markers for subsequent fault detection. By responding to the number of trusted data points in the updated set meeting preset conditions, transformer fault detection is then performed based on trusted data. During the detection process, residual verification of the extracted fault features is performed using a preset physical model. This ensures that the fault detection results are based on data that has undergone multi-layered trust filtering and secondary verification by the physical model. This effectively prevents interfered data from entering the detection process mixed with real fault data, improving the reliability of multi-point oil chromatography data in transformer fault detection. Attached Figure Description

[0008] The present application will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0009] Figure 1 This is a flowchart of a transformer fault detection method based on multi-point oil chromatography data in an embodiment of the present invention; Figure 2 This is a flowchart illustrating the determination of the spatial distribution deviation of the same gas component at each sampling point in an embodiment of the present invention. Detailed Implementation

[0010] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0011] like Figure 1 As shown, the transformer fault detection method based on multi-point oil chromatography data in this embodiment of the invention includes: acquiring oil chromatography monitoring data from several oil chromatography sampling points in the transformer.

[0012] In this embodiment, multiple oil chromatography sampling points are set on the transformer, each located at a different height within the transformer tank. For example, for a 220kV forced oil circulation transformer, sampling points can be set at the top, middle, and bottom of the tank, for a total of three sampling points; for a 500kV large transformer, sampling points can be set at the top, upper-middle, lower-middle, and bottom of the tank, for a total of four sampling points. The spatial location of the sampling points should cover the main height range of the transformer tank to allow for subsequent spatial analysis of the gas concentration distribution.

[0013] In this embodiment, the online oil chromatography monitoring device corresponding to each sampling point samples and performs chromatographic analysis on the insulating oil according to a preset collection cycle, outputting the concentration values ​​of each gas component at each sampling point at the current collection time, i.e., oil chromatography monitoring data. The preset collection cycle can be set according to monitoring needs; for example, it can be set to once every 4 hours or once every 24 hours. The gas components involved in the oil chromatography monitoring data typically include hydrogen, methane, ethane, ethylene, acetylene, carbon monoxide, and carbon dioxide. Among them, acetylene is a key characteristic gas for judging discharge faults, and ethylene is a key characteristic gas for judging overheating faults.

[0014] In a specific example, assume that the transformer has three sampling points: the top sampling point P1, with spatial coordinates z=2.5m; the middle sampling point P2, with spatial coordinates z=1.2m; and the bottom sampling point P3, with spatial coordinates z=0.3m. At a certain sampling time t, the oil chromatography monitoring data obtained are as follows: at the top sampling point P1, the hydrogen concentration is 45 μL / L, methane concentration is 12 μL / L, ethylene concentration is 8 μL / L, acetylene concentration is 0.5 μL / L, and carbon monoxide concentration is 320 μL / L; at the middle sampling point P2, the hydrogen concentration is 38 μL / L, methane concentration is 10 μL / L, ethylene concentration is 6 μL / L, acetylene concentration is 0.3 μL / L, and carbon monoxide concentration is 280 μL / L; at the bottom sampling point P3, the hydrogen concentration is 30 μL / L, methane concentration is 8 μL / L, ethylene concentration is 5 μL / L, acetylene concentration is 0.2 μL / L, and carbon monoxide concentration is 250 μL / L.

[0015] like Figure 2 As shown, based on oil chromatography monitoring data, the spatial distribution deviation of the same gas component at each sampling point is determined, and the spatial distribution qualification is determined based on the spatial distribution deviation.

[0016] In this embodiment, spatial distribution deviation is used to characterize the degree of difference between the actual concentration distribution and the expected concentration distribution at each sampling point. Under normal operating conditions, due to the physical laws of oil circulation and gas diffusion, the concentrations of the same gas component at each sampling point exhibit a relatively stable spatial order and ratio relationship. Spatial distribution deviation reflects the amount of deviation of the current data from this stable relationship.

[0017] In this embodiment, the expected concentration distribution is obtained through statistical analysis of historical chromatographic data from each sampling point under normal operating conditions of the transformer. Specifically, long-term monitoring data of the concentration of each characteristic gas at each sampling point under normal operating conditions are recorded, such as normal operating data from the past 6 months to 1 year, forming the normal range and relative ranking relationship of the concentration at each point, which serves as a benchmark for judging the rationality of the spatial distribution. Normal operating conditions refer to a transformer load rate of 20%-100% of rated capacity, top oil temperature of 40℃-85℃, and all gas concentrations below industry-recognized values ​​(hydrogen ≤150μL / L, acetylene ≤5μL / L, total hydrocarbons ≤150μL / L), with the cooling system operating normally and all monitoring devices performing normal self-tests.

[0018] In a specific example, the expected concentration distribution pattern obtained based on historical data statistics is as follows: For hydrogen, due to its small molecular weight and low density, it tends to float and accumulate at the top of the oil tank after dissolving in oil. Therefore, the hydrogen concentration at the top sampling point P1 is usually higher than that at the middle sampling point P2, and the hydrogen concentration at the middle sampling point P2 is usually higher than that at the bottom sampling point P3. That is, it satisfies the ordering relationship of C(P1) > C(P2) > C(P3), and the concentration ratio of each point fluctuates within the normal range. For example, the normal range of C(P1) / C(P3) is 1.2-2.0.

[0019] In this embodiment, the spatial distribution deviation can be calculated as follows: for each gas component, calculate the normalized root mean square error between the actual concentration value and the expected concentration value at each sampling point. The specific calculation formula is: , Where d represents the spatial distribution deviation; This is the i-th sampling point; This represents the current actual concentration value at the i-th sampling point; Let N be the expected concentration value at the i-th sampling point. The expected concentration value represents the background concentration baseline value of the gas component at the i-th sampling point under normal, fault-free operation of the transformer. Specifically, the expected concentration value is obtained in the following way: K consecutive collection periods of confirmed fault-free operation data from the transformer's historical operation data are selected, with K ≥ 30 preset; or normal operation data within the most recent 6 months are selected, and invalid data during shutdown and maintenance are removed. The arithmetic mean of the gas component concentration at the i-th sampling point within the K collection periods is calculated. N is the total number of sampling points. This is the mean of the expected concentration values ​​for all sampling points.

[0020] In this embodiment, the process of determining the spatial distribution qualification based on the spatial distribution deviation includes: if the spatial distribution deviation is greater than or equal to a preset deviation threshold, the spatial distribution qualification is determined to be unqualified, the fault detection is terminated and a data anomaly alarm is output; if the spatial distribution deviation is less than the preset deviation threshold, the spatial distribution qualification is determined to be qualified and subsequent steps are continued.

[0021] In a specific example, the preset deviation threshold can be set to 0.3. If the calculated spatial distribution deviation of hydrogen concentration at each sampling point at the current acquisition time is 0.15, which is less than the preset deviation threshold of 0.3, then the spatial distribution is deemed acceptable, and subsequent steps are executed. If the calculated spatial distribution deviation is 0.45, which is greater than the preset deviation threshold of 0.3, it indicates that the data at each sampling point does not conform to the normal physical distribution pattern, and there may be a systemic fault in the monitoring device, such as oil circuit blockage causing synchronous distortion of data at each point, or column aging causing synchronous shift of data at all points. In this case, fault detection is terminated and a data anomaly alarm is output to prompt maintenance personnel to check the monitoring device.

[0022] In response to the spatial distribution being deemed acceptable, the correlation deviation index of each sampling point is determined based on oil chromatography monitoring data. Abnormal sampling points are then removed from the oil chromatography sampling points based on the correlation deviation index, resulting in a set of reliable sampling points.

[0023] In this embodiment, the correlation deviation index is used to characterize the deviation of the correlation between a sampling point and other sampling points in terms of gas concentration changes relative to the correlation under normal operating conditions. Under normal operating conditions, since all sampling points are located in the same oil circuit system, the gas concentration changes at each point exhibit a certain statistical correlation. When a sampling point is affected by local electromagnetic interference or sensor malfunction, the data changes at that sampling point will deviate from the normal correlation pattern with other points, manifested as an increase in the correlation deviation index.

[0024] In this embodiment, the process of determining the correlation deviation index of each sampling point includes: constructing a correlation matrix between each sampling point, wherein the elements in the correlation matrix are the Spearman rank correlation coefficients between each sampling point; determining the mean correlation degree between each sampling point and other sampling points based on the correlation matrix; and calculating the absolute value of the difference between the mean correlation degree and the baseline correlation degree under normal working conditions as the correlation deviation index.

[0025] Specifically, the Spearman rank correlation coefficient is a nonparametric statistic used to measure the degree of monotonic association between two variables, with a value range of [-1, 1]. It is calculated by converting the observed values ​​of both variables into ranks (i.e., sorted indices), and then calculating the Pearson correlation coefficient between the two sets of ranks. Compared to the Pearson correlation coefficient, the Spearman rank correlation coefficient is less sensitive to outliers, does not require the data to follow a normal distribution, and is more suitable for evaluating the association between sampling points in oil chromatography monitoring data.

[0026] Specifically, the data used to construct the correlation matrix are time series of the concentrations of each gas component at each sampling point within the most recent collection periods, with the collection period set to 30 periods. The baseline value of the correlation under normal operating conditions is obtained based on historical data statistics during the initial commissioning of the transformer or during normal operation after the most recent overhaul. For example, the baseline value of the correlation can be calculated using data that meets the normal operating conditions within the first 3 months after commissioning.

[0027] In a specific example, suppose the transformer has four sampling points P1, P2, P3, and P4. Based on the concentration data of each gas component from the most recent 30 sampling periods, a 4×4 correlation matrix R is constructed. The Spearman rank correlation coefficients between P1 and P2 are 0.92, between P1 and P3 0.88, and between P1 and P4 0.85; between P2 and P3 0.90, and between P2 and P4 0.87; and between P3 and P4 0.91. The correlation coefficient between each sampling point and itself is 1.00.

[0028] The mean correlation of P1 is (0.92 + 0.88 + 0.85) / 3 = 0.883. If the baseline correlation of P1 under normal operating conditions is 0.90, then the correlation deviation index of P1 is |0.883 - 0.90| = 0.017.

[0029] In this embodiment, the process of removing abnormal sampling points from oil chromatography sampling points based on the correlation deviation index includes: if the correlation deviation index of a sampling point is greater than or equal to a preset deviation index threshold, the sampling point is determined to be an abnormal sampling point and removed; if the correlation deviation index is less than the preset deviation index threshold, the sampling point is retained. After removing all abnormal sampling points, the remaining sampling points constitute a set of reliable points.

[0030] In a specific example, the preset deviation index threshold can be set to 0.15. If the correlation deviation index of P4 is 0.22, which is greater than the preset deviation index threshold of 0.15, then P4 is determined to be an abnormal sampling point, possibly affected by local electromagnetic interference or sensor aging, and P4 is removed from the oil chromatography sampling points. After removal, the set of reliable points is {P1, P2, P3}.

[0031] Optionally, if the correlation deviation index of all sampling points is less than a preset deviation index threshold, then all sampling points are retained. For example, if the correlation deviation index of P1 is 0.017, P2 is 0.025, P3 is 0.030, and P4 is 0.040, all less than 0.15, then the set of reliable points is {P1, P2, P3, P4}.

[0032] Based on oil chromatography monitoring data, the time-series change rate of each point in the set of reliable points is determined, and the number of points in the set of reliable points whose time-series change rate is greater than or equal to the preset change rate threshold is counted. The synchronous jump rate is determined based on the number of points and the total number of points in the set of reliable points.

[0033] In this embodiment, the temporal change rate is used to characterize the relative rate of change of gas concentration at a sampling point between two adjacent sampling times, evaluating the stability of the data from a time dimension. The synchronous jump rate is the ratio of the number of points with a temporal change rate greater than or equal to a preset change rate threshold to the total number of points in the set of reliable points, used to distinguish between two situations: multiple points experiencing simultaneous concentration abrupt changes and only a few points experiencing abrupt changes.

[0034] In this embodiment, the process of determining the time-series change rate of each sampling point includes: obtaining the gas concentration values ​​of each sampling point at the current time and the previous time; calculating the ratio of the difference between the gas concentration values ​​at the current time and the previous time to the gas concentration value at the previous time, as the time-series change rate. The specific calculation formula is as follows: , Among them, R(P) i P represents the time-series change rate at the i-th sampling point; i For the i-th sampling point; C(P) i C(P) represents the gas concentration value at the i-th sampling point at the current time t; i , t 1) For the i-th sampling point at the previous time t The gas concentration value of 1.

[0035] It should be noted that the gas concentration value C(P) at the current moment... i , t 1) When the value is zero or below the minimum detection limit of the instrument, the time-series change rate R(P) is... i Set it to zero to avoid the denominator being zero or abnormally amplifying the effects of small fluctuations.

[0036] Optionally, the time-series change rate can be calculated for a single gas component, or it can be calculated separately for multiple gas components and the maximum value is taken. In this embodiment, it is preferable to calculate the time-series change rate separately for the two key fault characteristic gases, acetylene and ethylene, and take the larger value of the two as the time-series change rate at that point.

[0037] In a specific example, suppose the set of reliable points is {P1, P2, P3}, and the acetylene concentrations at each point at the current time t are: P1 = 2.5 μL / L, P2 = 2.0 μL / L, P3 = 1.8 μL / L; and the acetylene concentrations at each point at the previous time t-1 are: P1 = 0.5 μL / L, P2 = 0.4 μL / L, P3 = 1.7 μL / L. Then the time-series change rates at each point are: The time-series rate of change of P1 = |2.5 - 0.5| / 0.5 = 4.0, or 400%; The time-series rate of change of P2 = |2.0 - 0.4| / 0.4 = 4.0, or 400%; The time-series rate of change of P3 = |1.8 - 1.7| / 1.7 ≈ 0.059, or 5.9%; In this embodiment, the preset change rate threshold can be determined based on the gas composition. For example, for acetylene, the preset change rate threshold can be set to 1.0, i.e., 100%; for ethylene, the preset change rate threshold can be set to 0.8, i.e., 80%.

[0038] In the example above, the preset change rate threshold is 1.0. Therefore, the time-series change rates of P1 and P2 (4.0) are greater than the preset change rate threshold, while the time-series change rate of P3 (0.059) is less than the preset change rate threshold. Thus, the number of points with a time-series change rate greater than or equal to the preset change rate threshold is 2, the total number of points in the reliable point set is 3, and the synchronization jump rate = 2 / 3 ≈ 0.667.

[0039] The set of trusted points is processed based on the synchronous jump rate, and the set of trusted points is updated.

[0040] In this embodiment, the process of processing the set of trusted points based on the synchronization jump rate includes: if the synchronization jump rate is less than a preset synchronization jump rate threshold, then points with a temporal change rate greater than or equal to the preset change rate threshold are marked as temporally unstable points and removed, and the set of trusted points is updated; if the synchronization jump rate is greater than or equal to the preset synchronization jump rate threshold, then all points in the set of trusted points are retained and no points are removed.

[0041] In a specific example, the preset synchronous jump rate threshold can be set to 0.5. The synchronous jump rate is 0.667, which is greater than the preset synchronous jump rate threshold of 0.5. At this time, multiple points P1 and P2 simultaneously experience concentration abrupt changes. Since the characteristic gas generated by the internal fault of the transformer will diffuse to each sampling point with the oil flow, the synchronous concentration change of multiple sampling points is a typical manifestation of a real fault. Therefore, all points {P1, P2, P3} in the set of reliable points are retained.

[0042] In another example, if only the timing change rate of P1 is 4.0, which is greater than the preset change rate threshold, while the timing change rates of P2 (0.08) and P3 (0.059) are both less than the preset change rate threshold, then the number of points with timing change rates exceeding the threshold is 1, and the synchronization jump rate = 1 / 3 ≈ 0.333, which is less than the preset synchronization jump rate threshold of 0.5. In this case, only a few points experience abrupt changes, while other points do not respond synchronously. Since the gas generated by a real fault inside the transformer will diffuse to various points with the oil flow, if it were a real fault, synchronous changes at multiple points should be observed; a sudden change at only a single point is more likely due to local electromagnetic interference, poor contact, or random noise affecting the sensor at that point. Therefore, P1 is marked as a timing unstable point and removed, and the updated set of reliable points is {P2, P3}.

[0043] In response to the number of trusted points in the updated trusted point set being greater than or equal to a preset threshold, transformer fault detection is performed based on the oil chromatography monitoring data in the trusted point set.

[0044] In this embodiment, the preset number threshold can be determined based on the total number of sampling points. For example, if the total number of sampling points is 4, the preset number threshold can be set to 2 or 3. The principle for setting the preset number threshold is: at least 2 points are required for spatial gradient analysis, and it is preferable to retain 3 or more points to improve spatial resolution. If the number of reliable points in the updated set of reliable points is less than the preset number threshold, an insufficient data alarm is output, prompting the maintenance personnel that the currently available data is insufficient to support reliable fault detection, and suggesting manual re-inspection or waiting for the data in the next collection cycle.

[0045] In this embodiment, the process of transformer fault detection based on oil chromatography monitoring data in the set of trusted locations includes: extracting fault features based on oil chromatography monitoring data in the set of trusted locations; substituting the fault features into a preset transformer oil flow-diffusion model for residual verification; and outputting fault detection results based on the residual verification results.

[0046] Specifically, the fault characteristics include spatial gradient characteristics and temporal rate characteristics. Spatial gradient characteristics are determined by the concentration gradient vector based on the concentration distribution of the same gas component at each sampling point in the set of reliable locations. The spatial gradient direction and magnitude of the concentration field are determined by least-squares fitting, using the spatial coordinates of each sampling point as the independent variable and the concentration of each gas component as the dependent variable. Temporal rate characteristics are determined by the gas production rate based on oil chromatography monitoring data from each sampling point in the set of reliable locations over multiple consecutive time periods. Specifically, linear fitting is performed on the concentration data of each gas component from the most recent several acquisition periods. The acquisition period is selected as 5 periods, and the slope of the fitted straight line is the gas production rate, with units of μL / (L·period).

[0047] For example, if the set of reliable points is {P1 (top, z=2.5m), P2 (middle, z=1.2m), P3 (bottom, z=0.3m)}, and the current acetylene concentrations are C(P1)=3.2μL / L, C(P2)=1.8μL / L, and C(P3)=0.5μL / L respectively, then the concentration gradient along the vertical direction is: dC / dz≈(3.2-0.5) / (2.5-0.3)≈1.23μL / (L·m), and the gradient direction is from bottom to top, that is, the concentration increases with height, indicating that the fault source may be located at the top of the transformer. If the acetylene concentration at point P1 is 0.5 μL / L, 1.2 μL / L, 2.0 μL / L, 2.8 μL / L, and 3.2 μL / L in the last 5 collection cycles, the gas production rate is 0.70 μL / (L·cycle) obtained by least squares linear fitting.

[0048] Specifically, the preset transformer oil flow-diffusion model is a physical model based on the convection-diffusion equation. It is used to describe the concentration distribution of characteristic gases generated by internal transformer faults, which propagate from the fault source to various sampling points under the combined effects of oil flow transport and molecular diffusion. The physical basis of this model is the coupling of Fick's second law and the convection transport equation, and it is applied to the physical consistency verification of multi-point oil chromatography monitoring data.

[0049] Specifically, the process of constructing the pre-defined transformer oil flow-diffusion model is as follows: First, establish the three-dimensional convection-diffusion equation for dissolved gases in the transformer oil tank: , Where C is the dissolved gas concentration in the oil (μL / L), t is time (s), v is the oil flow velocity vector (m / s), and D is the molecular diffusion coefficient of the gas in the transformer oil (m). 2 / s), S(x,y,z,t) is the gas production source term (μL / (L·s)). For gradient operators, 2 This is the Laplace operator. The physical meaning of this equation is: the rate of change of gas concentration at a certain point in the oil over time is equal to the sum of the concentration changes caused by oil convection transport, molecular diffusion, and gas production from the fault source.

[0050] Then, the oil flow velocity vector v is determined according to the transformer cooling method: For natural oil circulation transformers, the oil flow is generated by thermal convection driven by temperature difference. Hot oil rises along the tank wall after being heated from the winding surface, enters the radiator at the top for cooling, and then descends back to the bottom of the tank, forming a natural circulation loop. The typical oil flow velocity is 0.01-0.05 m / s. For forced oil circulation transformers, the oil flow is driven by an oil pump. Oil is pumped out from the bottom of the tank, cooled by the radiator, and then reinjected from the top or middle of the tank. The typical oil flow velocity is 0.1-0.5 m / s, and the specific value is calculated based on the oil circuit structure and rated flow rate of the oil pump in the transformer design drawings. In this embodiment, to reduce computational complexity, it is assumed that the oil flow in the tank is a uniform velocity field, and the above typical velocity is used as the value of the overall velocity vector v. The molecular diffusion coefficient D is determined based on the gas type and oil temperature. In mineral insulating oil at an oil temperature of 60°C, the molecular diffusion coefficient for hydrogen is approximately 2.5 × 10⁻⁶ m / s. -9 m 2 / s, methane is approximately 1.8 × 10⁻⁶. -9 m 2 / s, acetylene is approximately 1.5 × 10⁻⁶. -9 m 2 / s, ethylene is approximately 1.6 × 10⁻⁶. -9 m 2 / s, carbon monoxide is approximately 1.7 × 10⁻⁶. -9 m 2 / s; the diffusion coefficient varies with oil temperature according to the Arrhenius relation: , in, The activation energy is the diffusion energy, typically around 15-25 kJ / mol; R is the gas constant (8.314 J / (mol·K)); and T is the absolute temperature (K). The gas-producing source term S(x, y, z, t) is used as an unknown to be inverted during the model validation phase. Assuming the fault source is a point source located at coordinates (x0, y0, z0) and the gas production rate is q (μL / s), then: , in, The Dirac function is used. The geometric boundary conditions establish the computational domain based on the actual internal dimensions of the transformer tank. For example, for a 220kV transformer, the internal dimensions of the tank are approximately 3.5m long × 2.0m wide × 3.0m high, and the tank walls are set as flux-free boundaries. C / (n=0, where n is the wall normal), the oil surface is set as a free surface boundary. In this embodiment, the free surface boundary adopts a flux-free boundary condition, i.e. C / n=0, where n is the normal direction of the oil surface. The initial conditions are set as the background concentration field of each gas component under normal operating conditions. Specifically, the long-term average concentration of each sampling point during normal operating conditions is used as the node value. The initial concentration value of each grid node in the computational domain is determined by spatial interpolation (such as linear interpolation or least squares fitting), thus reflecting the spatial distribution gradient under normal operating conditions. It should be noted that in the actual numerical calculation, the point source is approximated by the uniform gas production within the grid cell where the source point is located; let the equivalent oil volume of the grid cell where the source point is located be V. cell The unit is L, and the volume is determined by the grid step size Δx. Δy Δz is obtained by conversion, i.e., V cell =Δx×Δy×Δz×10 3 (Because of 1m) 3 =1000L). Then the volumetric source term intensity S within this mesh cell is... cell =q / V cell That is, within this grid cell, S=S cell The volumetric source term intensity of other grid cells is zero. This ensures that the point source expression maintains dimensional consistency with the gas-producing source term S in the convection-diffusion equation.

[0051] Next, the computational domain is spatially divided into a three-dimensional uniform grid with a preset grid size of 0.1m × 0.1m × 0.1m. For an oil tank of 3.5m × 2.0m × 3.0m, the total number of grid nodes is approximately 21,000. The preset time step is Δt = 60s, and the total simulation duration is determined based on the acquisition cycle (e.g., if the acquisition cycle is 4 hours, the total simulation duration is 14,400s, totaling 240 time steps). The convection term is discretized using an upwind scheme, the diffusion term using a central difference scheme, and the time progression uses the implicit Euler method. The model output is the curve C showing the change in the concentration of each gas component at each sampling point over time. model (P i ,t).

[0052] Specifically, the model's input data includes: transformer structural parameters (tank internal dimensions, spatial coordinates of each sampling point, oil circuit structure), operating parameters (current top-level oil temperature, cooling method, current load rate), gas property parameters (diffusion coefficients of each characteristic gas at the current oil temperature), initial conditions (background concentration C0 of each gas component), and parameters to be inverted (fault source location and gas generation rate). The model's output data includes: the model-predicted concentration C of each gas component at each sampling point location. model (P i , t), and the residual values ​​between the model predictions and the actual observations.

[0053] Specifically, the extracted fault features are used as constraints, and the fault source region and gas production rate q are retrieved using a least-squares optimization algorithm. The fault source location is represented by spatial coordinates (x0, y0, z0), which are estimated values ​​obtained from data retrieved using finite sampling points. The actual fault region needs to be determined by considering the transformer structural dimensions to minimize the residual between the model prediction and the actual observation. The residual is calculated as follows: , Where i is the sampling point index, L is the total number of points in the set of trusted points, j is the time step index, and M is the total number of time steps participating in the verification. M is set to 5. To predict concentration values ​​for the model, These are the actual observed concentration values.

[0054] In this embodiment, the preset residual threshold is determined comprehensively based on the concentration level of the gas component and the measurement accuracy of the detection instrument. For example, for trace characteristic gases such as acetylene, their concentration under normal fault-free operating conditions is usually close to 0 μL / L (or lower than 0.5 μL / L). Considering the measurement error of the online monitoring device, the preset residual threshold is set to 0.2 μL / L. For characteristic gases with relatively high concentrations such as hydrogen and methane, their background concentration under normal operating conditions usually fluctuates in the range of 10-50 μL / L, and the preset residual threshold is set to 5 μL / L. For macroscopic gases such as carbon monoxide, their normal concentration can reach several hundred μL / L, and the preset residual threshold is set to 20 μL / L.

[0055] If the residual is less than a preset residual threshold, the fault characteristic is deemed reliable, and the fault detection result is output. The fault detection result includes the fault type and fault location.

[0056] Specifically, the fault type is determined based on the concentration ratios of each gas component and the content of characteristic gases in the multi-point oil chromatography fusion data. First, based on the gas concentration data after weighted fusion of concentrations from each reliable sampling point, three sets of characteristic gas ratios—C2H2 / C2H4, CH4 / H2, and C2H4 / C2H6—are calculated. Each set of ratios is assigned a corresponding code value according to a preset coding range, and the fault type is determined by the combination of the three sets of codes.

[0057] In this embodiment, the weighted fusion adopts an equal-weighted arithmetic average, that is, the concentration data of each reliable sampling point participates in the calculation with the same weight; in an alternative embodiment, the weight can also be determined based on the data reliability of each sampling point or the distance of its spatial location from the fault source, and sampling points with higher reliability or closer to the fault source are given higher weights.

[0058] For example, in one embodiment of the present invention, the coding intervals for each group of gas ratios are divided as follows: For the C2H2 / C2H4 ratio, a value of 0 is assigned when it is less than 0.1, a value of 1 is assigned when it is greater than or equal to 0.1 and less than 1, a value of 1 is assigned when it is greater than or equal to 1 and less than 3, and a value of 2 is assigned when it is greater than or equal to 3. For the CH4 / H2 ratio, a value of 1 is assigned when it is less than 0.1, a value of 0 is assigned when it is greater than or equal to 0.1 and less than 1, and a value of 2 is assigned when it is greater than or equal to 1. For the C2H4 / C2H6 ratio, a value of 0 is assigned when it is less than 1, a value of 1 is assigned when it is greater than or equal to 1 and less than 3, and a value of 2 is assigned when it is greater than or equal to 3.

[0059] Based on the combination of the three sets of codes mentioned above, the corresponding fault type judgment rules are as follows: When the code combination is (0, 0, 1), it is judged as a low temperature overheating fault, corresponding to a temperature below 150℃; when the code combination is (0, 2, 0), it is judged as a low temperature overheating fault, corresponding to a temperature of 150℃-300℃; when the code combination is (0, 2, 1), it is judged as a medium temperature overheating fault, corresponding to a temperature of 300℃-700℃; when the code combination is (0, X, 2), where X is 0, 1, or 2, it is judged as a high temperature overheating fault, corresponding to a temperature above 700℃; when the code combination is (0, 1, 0), it is judged as a partial discharge fault; when the code combination is (2, X, Y), where X and Y are 0, 1, or 2, it is judged as a low energy discharge fault (spark discharge); when the code combination is (1, X, Y), where X and Y are 0, 1, or 2, it is judged as a high energy discharge fault (arc discharge).

[0060] Based on this, auxiliary verification is carried out by combining the content of characteristic gases: if the acetylene content is abnormally high, it indicates a high-energy discharge fault; if the ethylene content is abnormally high, it indicates an overheating fault; if the carbon monoxide and carbon dioxide content are significantly high, while the hydrocarbon gas production rate does not show a significant increase, it indicates overheating decomposition of solid insulation materials.

[0061] The fault location is determined based on the inversion analysis of multi-point concentration gradient vectors. Specifically, using the spatial coordinates and concentration data of each sampling point in the set of reliable points as constraints, the range of the fault source is estimated through least squares optimization. Taking the bottom of the tank as the origin of the coordinate system, and combining the typical dimensional characteristics of the transformer body structure: if the estimated height of the fault source obtained by inversion is greater than or equal to 2.0m, the fault is determined to be located in the upper region of the transformer, corresponding to the upper part of the winding or the lead joint; if the estimated height is greater than or equal to 0.8m and less than 2.0m, the fault is determined to be located in the middle region of the transformer, corresponding to the main part of the winding; if z0 is less than 0.8m, the fault is determined to be located in the lower region of the transformer, corresponding to the bottom of the core or the lower end of the winding.

[0062] If the residual is greater than or equal to the preset residual threshold, the fault characteristic is determined to be unreliable, and a prompt suggesting manual re-inspection is output.

[0063] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A transformer fault detection method based on multi-point oil chromatography data, characterized in that, include: Obtain oil chromatographic monitoring data from several oil chromatographic sampling points in the transformer; Based on the oil chromatography monitoring data, the spatial distribution deviation of the same gas component at each sampling point is determined, and the spatial distribution qualification is determined based on the spatial distribution deviation. In response to the spatial distribution being deemed acceptable, based on the oil chromatography monitoring data, the correlation deviation index of each sampling point is determined, and abnormal sampling points are removed from the oil chromatography sampling points based on the correlation deviation index to obtain a set of reliable sampling points. Based on the oil chromatography monitoring data, the time-series change rate of each point in the set of reliable points is determined, and the number of points in the set of reliable points whose time-series change rate is greater than or equal to a preset change rate threshold is counted. The synchronization jump rate is determined according to the number of points and the total number of points in the set of reliable points. The trusted point set is processed based on the synchronization jump rate, and the trusted point set is updated. In response to the number of trusted points in the updated trusted point set being greater than or equal to a preset number threshold, transformer fault detection is performed based on the oil chromatography monitoring data in the trusted point set.

2. The transformer fault detection method based on multi-point oil chromatography data according to claim 1, characterized in that, Determining the spatial distribution qualification based on the spatial distribution deviation includes: If the spatial distribution deviation is less than a preset deviation threshold, then the spatial distribution is determined to be qualified. If the spatial distribution deviation is greater than or equal to a preset deviation threshold, the spatial distribution is determined to be unqualified, the fault detection is terminated, and a data anomaly alarm is output.

3. The transformer fault detection method based on multi-point oil chromatography data according to claim 2, characterized in that, Determine the spatial distribution deviation of the same gas component at each sampling point, including: Based on historical gas concentration data under normal operating conditions at each sampling point, the expected concentration distribution at each sampling point is established. The difference between the actual concentration distribution at each current sampling point and the expected concentration distribution is determined as the spatial distribution deviation.

4. The transformer fault detection method based on multi-point oil chromatography data according to claim 1, characterized in that, Abnormal sampling points are removed from the oil chromatography sampling points based on the correlation deviation index, including: If the correlation deviation index is greater than or equal to the preset deviation index threshold, the sampling point is determined to be an abnormal sampling point and is removed.

5. The transformer fault detection method based on multi-point oil chromatography data according to claim 4, characterized in that, Determine the correlation deviation index for each sampling point, including: Construct a correlation matrix between each sampling point, wherein the elements in the correlation matrix are the Spearman rank correlation coefficients between each sampling point; Based on the correlation matrix, the mean correlation degree between each sampling point and other sampling points is determined; The absolute value of the difference between the mean correlation degree and the baseline correlation degree of the corresponding sampling point under normal operating conditions is calculated and used as the correlation deviation index.

6. The transformer fault detection method based on multi-point oil chromatography data according to claim 1, characterized in that, Processing the set of trusted points based on the aforementioned synchronization jump rate includes: If the synchronization jump rate is less than the preset synchronization jump rate threshold, then the points with a timing change rate greater than or equal to the preset change rate threshold are marked as timing unstable points and removed. If the synchronization jump rate is greater than or equal to the preset synchronization jump rate threshold, then all points in the trusted point set are retained.

7. The transformer fault detection method based on multi-point oil chromatography data according to claim 6, characterized in that, The synchronization jump rate is the ratio of the number of points with a temporal change rate greater than or equal to a preset change rate threshold to the total number of points in the trusted point set.

8. The transformer fault detection method based on multi-point oil chromatography data according to claim 1, characterized in that, The process of determining the time-series rate of change for each sampling point includes: Obtain the gas concentration values ​​at each sampling point at the current time and the previous time. The ratio of the absolute value of the difference between the gas concentration values ​​at the current time and the previous time to the gas concentration value at the previous time is calculated as the time-series change rate.

9. The transformer fault detection method based on multi-point oil chromatography data according to claim 1, characterized in that, Transformer fault detection based on oil chromatography monitoring data from the aforementioned set of reliable locations includes: Fault features are extracted based on oil chromatography monitoring data from the set of reliable locations. These fault features include spatial gradient features and temporal rate features. The fault characteristics were substituted into a preset transformer oil flow-diffusion model for residual verification. If the residual is less than the preset residual threshold, the fault feature is determined to be reliable, and the fault detection result is output. If the residual is greater than or equal to the preset residual threshold, the fault characteristic is determined to be unreliable, and a prompt suggesting manual re-inspection is output. The spatial gradient feature is a concentration gradient distribution determined based on the concentration distribution of the same gas component at each sampling point in the set of reliable points; The time-series rate characteristic is the trend of gas production rate change determined based on oil chromatography monitoring data of each sampling point in the set of reliable locations over multiple consecutive time periods. The fault detection results include at least one of the following: transformer fault type and fault location.

10. The transformer fault detection method based on multi-point oil chromatography data according to claim 1, characterized in that, Also includes: If the number of trusted points in the updated trusted point set is less than the preset threshold, an insufficient data alarm will be output.

Images