A voltage transformer fault on-line diagnosis method

Abstract

The application discloses a kind of voltage transformer fault online diagnosis methods, this method includes: the voltage effective value, current effective value and temperature value normalization to the acquisition, obtain multidimensional state parameters;Three-dimensional phase space coordinate system is constructed, and state trajectory point is generated;Setting Poincare section, intersection is obtained and is identified periodicity characteristic according to time interval;Based on intersection position and periodicity characteristic, by support vector machine and historical fault mode library matching, judge fault state and mark;According to the marked classification fault type generates diagnosis report.The application is analyzed by the normalization phase space of multiple parameters and Poincare section, captures the periodicity anomaly of multiple parameter dynamic coupling, combines support vector machine intelligent diagnosis, realizes the online sensitive detection of voltage transformer latent fault, reduces misjudgment, reduces artificial experience dependence.

Description

Technical Field

[0001] This invention relates to the field of voltage transformer technology, and more specifically to an online fault diagnosis method for voltage transformers. Background Technology

[0002] Voltage transformers are key equipment in power systems, serving as devices for energy metering, relay protection, and control. Their operational reliability directly impacts power grid safety. Common voltage transformer faults include inter-turn short circuits, insulation degradation, and ferroresonance. Failure to detect and address these faults promptly can escalate into serious accidents such as insulation breakdown and explosions. Currently, fault diagnosis of voltage transformers primarily relies on two methods: periodic outage preventative testing and online monitoring. Periodic testing cannot reflect the dynamic state of the equipment under actual operating conditions, and outages disrupt power supply continuity. Online monitoring, through the installation of sensors to collect operating parameters in real time, offers greater timeliness.

[0003] Current fault detection methods largely rely on single electrical quantity thresholds for judgment, ignoring the coupling relationship between parameters such as current and temperature. This makes them prone to false alarms or missed alarms when load fluctuates or the environment changes. Offline periodic monitoring only acquires static data and cannot track the dynamic development of faults; faults that develop between two monitoring intervals may miss the optimal handling time. Some online monitoring systems only display numerical values ​​or provide simple alarms, without in-depth analysis of the time-series relationships of multiple parameters or extraction of periodic features, leading to misjudgments of normal operating condition transitions as faults. Fault type identification relies on manual comparison of waveforms or experience tables, which is inefficient, error-prone, and inconsistent in judgments in large-scale equipment scenarios. These limitations reduce the response speed to complex time-varying faults and increase operation and maintenance costs and risks. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and propose an online fault diagnosis method for voltage transformers.

[0005] The technical solution provided by this invention to solve the above problems is: an online fault diagnosis method for voltage transformers, the method comprising the following steps: S1: Collect the effective values ​​of voltage, current, and temperature through sensors, and perform normalization calculations on the effective values ​​of voltage, current, and temperature to obtain multidimensional state parameters; S2: Construct a three-dimensional phase space coordinate system based on multi-dimensional state parameters, and calculate the state trajectory points corresponding to each sampling time. S3: Based on the state trajectory points, set multiple Poincaré sections, obtain the intersection positions of the state trajectory points and the Poincaré sections, and identify periodic features by comparing the time intervals between adjacent intersection positions; S4: Based on the intersection location and periodic features, the support vector machine is invoked to calculate the similarity between the current intersection location and the fault intersection locations in the pre-stored historical fault mode library, determine whether the current intersection location is a fault state, and generate a fault diagnosis label; S5: Based on the fault diagnosis tags, classify and filter the fault types, generate a diagnostic report, and output the diagnostic results to the display terminal.

[0006] Preferably, the multidimensional state parameters include the processed normalized voltage, the effective value of the characteristic current, and the environmentally corrected temperature, and the state trajectory points include the calculated instantaneous spatial coordinates, the associated motion direction vector, and the geometric radius of curvature.

[0007] Preferably, the intersection location includes extracted cross-sectional penetration coordinates, local scatter distribution density, and mapped one-dimensional discrete sequence, and the periodic features include identified resonant frequencies, bifurcation limit cycle periods, and quantified chaotic measure indices.

[0008] Preferably, the fault diagnosis markers include the determined fault type, the matched confidence score, and the quantified state offset degree.

[0009] Preferably, the diagnostic report includes the identified abnormal fault classification, the predicted fault evolution trend, and recommended operation and maintenance handling suggestions.

[0010] Preferably, S1 specifically includes: The system collects the effective values ​​of voltage, current, and temperature. It reads the raw signal from the sensor and converts it into three values: effective voltage, effective current, and temperature. It calculates the maximum value of the three values, divides each value by the maximum value to obtain three normalized values ​​between 0 and 1, and combines the three normalized values ​​into a normalized state parameter value. Based on the normalized state parameter values, it is determined whether the three normalized components are all within the 0 to 1 interval. If the interval is within the range, a communication connection is established, and the normalized state parameter values ​​are sent to the phase space modeling module to generate normalized multidimensional state parameters.

[0011] Preferably, S2 specifically includes: Based on multidimensional state parameters, the correlation between each pair of parameters is calculated, and the three dimensions with the lowest correlation are selected as coordinate axes. The coordinate axis scale range is set according to the extreme values ​​of the selected dimensions at all sampling times, and a set of spatial axis parameter values ​​is generated. Using the set of spatial axis parameter values ​​as a reference, the corresponding multidimensional parameter values ​​are obtained for each sampling time. The parameter values ​​of each axis direction are normalized based on the length of the extreme value interval of that axis. The projection values ​​of each axis are combined into three-dimensional coordinates and recorded as the state points at the corresponding time. All state points at all times are merged to obtain the state trajectory points.

[0012] Preferably, S3 specifically includes: Multiple Poincaré sections are set for the acquisition of trajectory points. The extreme values ​​of each coordinate of the trajectory points are calculated. The position intervals of the equally spaced sections are divided based on the difference of extreme values. The normal vector and offset of each section are calculated to generate the coefficient set of the section equation. Call the coefficient set of the cross-sectional equation, calculate the distance between the coordinates of the trajectory point and the symbol of the cross-sectional plane at each sampling time, filter the adjacent trajectory point pairs with changes in the symbol distance, and interpolate to calculate the intersection point coordinate set at the crossing time. The time interval between adjacent intersection points is calculated based on the intersection point coordinate set. The difference between adjacent time intervals is calculated, and consecutive intervals with a difference less than the time interval threshold are merged into the same period to generate the intersection point position and periodic feature value.

[0013] Preferably, S4 specifically includes: The system retrieves the RMS voltage, RMS current, and temperature values ​​of historical fault intersections from a historical fault mode database as input features. An improved K-means clustering algorithm is used to aggregate the feature parameters of each intersection. Through iterative optimization, a set of cluster centroids for multiple fault mode clusters and statistical parameters of distance distribution within each cluster are determined. These cluster centroids and distance distribution parameters are then passed to the distance and threshold analysis module. Based on the cluster centroids, the feature vectors constructed from the RMS voltage, RMS current, and temperature values ​​at the current intersection are used to calculate Euclidean distances, obtaining the distances between the current intersection and each cluster centroid and generating a distance matrix. Dynamic thresholds are set for the distances based on the distance distribution parameters, and the distance matrix and threshold labels are passed to the time-series feature judgment module. Based on the distance matrix, threshold labels, and periodic features, a dynamic time warping model is used to reconstruct the time-series features of the current intersection's position over time, extracting a dynamic trend vector. Multi-dimensional discrimination logic is then used to comprehensively determine whether the current intersection is in a fault state, generating a fault diagnosis label. The above scheme improves the K-means clustering algorithm to perform unsupervised aggregation of voltage, current, and temperature features at historical fault intersection locations, achieving automatic division of fault mode clusters. It generates cluster center point sets and distance distribution statistics without manual annotation, enhancing the system's adaptive learning capability to historical data distribution. Simultaneously, it calculates the distance values ​​between the current intersection point and each cluster center point based on Euclidean distance and generates a distance matrix. Furthermore, it uses distance distribution statistics to set dynamic thresholds, allowing the thresholds to automatically adjust according to historical data distribution characteristics, improving the sensitivity and reliability of anomaly detection. Finally, it uses a dynamic time warping model to reconstruct the time series of the current intersection location, extracting dynamic trend vectors. Combining this with the distance matrix and threshold labels for multi-dimensional discrimination, it can capture the temporal patterns of fault evolution, compensating for the insufficient response of single spatial distance judgment to gradual faults. This enhances the diagnostic method's ability to identify fault modes and its real-time response efficiency under complex operating conditions.

[0014] Preferably, S5 specifically includes: Obtain fault diagnosis markers, read the fault feature values ​​carried by each marker, compare the feature values ​​with the boundary values ​​in the five preset type threshold intervals one by one, classify the markers into the corresponding types according to the intervals in which the values ​​are located, accumulate the number of markers belonging to the same type interval, and then calculate the quotient of each accumulated value and the total number of markers to obtain the fault type classification value. Based on the fault type classification value, the cumulative number of tags for each type is decomposed and its percentage of the total number of tags is calculated. The percentage value is compared with the boundary value of the matching interval in the report template library. The template number corresponding to the interval where the percentage is located is selected, and all selected template numbers are merged to form a sequence to generate a template matching code. Based on the template matching code, the data frame index bit corresponding to each number in the sequence is parsed, the data segment under each index bit is obtained by calling the data frame mapping table of the display terminal, and all data segments are concatenated in numerical order to form a complete display frame, thus obtaining the diagnostic result value.

[0015] Compared with existing technologies, the advantages of this invention are: by normalizing voltage, current, and temperature values ​​into multi-dimensional state parameters, the interference of dimensional differences on analysis is eliminated. A three-dimensional phase space coordinate system is constructed based on these multi-dimensional parameters, and the state trajectory points at each moment are calculated, allowing the parameter coupling relationship to be continuously presented in space. A Poincaré cross section is set to obtain the intersection points of the trajectory and the cross section; the time intervals between adjacent intersection points are compared to identify periodic fluctuation characteristics and capture the normal operating mode and abnormal deviations of the equipment. Support vector machines are used to match the current intersection point position with a historical fault mode library to automatically determine the fault state and generate labels. Based on the labels, fault types are classified and filtered, and a diagnostic report is output, achieving latent fault identification under multi-parameter fusion, improving detection sensitivity and accuracy, reducing false positives and false negatives, and reducing reliance on human experience. Attached Figure Description

[0016] The accompanying drawings, which are provided to further illustrate the invention and constitute a part of this invention, are illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention.

[0017] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0018] The following will describe in detail the implementation of the present invention with reference to the accompanying drawings and embodiments, so that the process of how the present invention uses technical means to solve technical problems and achieve technical effects can be fully understood and implemented accordingly.

[0019] In the description of this invention, it should be noted that the directional terms such as "center", "lateral", "longitudinal", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", and "counterclockwise" indicate the orientation and positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. They should not be construed as limiting the specific protection scope of this invention.

[0020] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features. Thus, the use of "first" and "second" to define a feature may explicitly or implicitly include one or more of that feature, and in the description of this invention, "a number" means two or more, unless otherwise explicitly specified.

[0021] In this invention, unless otherwise explicitly specified and limited, the terms "assembly," "connection," and "joining" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can also refer to a mechanical connection; they can refer to a direct connection or a connection through an intermediate medium; or they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0022] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0023] It should also be understood that the terminology used in this specification of embodiments of the invention is for the purpose of describing particular embodiments only and is not intended to limit the embodiments of the invention. As used in this specification of embodiments of the invention and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0024] As shown in the accompanying drawings, a method for online fault diagnosis of a voltage transformer is provided, the method comprising the following steps: S1: Collect the effective values ​​of voltage, current, and temperature through sensors, and perform normalization calculations on the effective values ​​of voltage, current, and temperature to obtain multidimensional state parameters; S2: Construct a three-dimensional phase space coordinate system based on multi-dimensional state parameters, and calculate the state trajectory points corresponding to each sampling time. S3: Based on the state trajectory points, set multiple Poincaré sections, obtain the intersection positions of the state trajectory points and the Poincaré sections, and identify periodic features by comparing the time intervals between adjacent intersection positions; S4: Based on the intersection location and periodic features, the support vector machine is invoked to calculate the similarity between the current intersection location and the fault intersection locations in the pre-stored historical fault mode library, determine whether the current intersection location is a fault state, and generate a fault diagnosis label; S5: Based on the fault diagnosis tags, classify and filter the fault types, generate a diagnostic report, and output the diagnostic results to the display terminal.

[0025] The multidimensional state parameters include the processed normalized voltage, the effective value of the characteristic current, and the environmentally corrected temperature; the state trajectory points include the calculated instantaneous spatial coordinates, the associated motion direction vector, and the geometric radius of curvature; the intersection points include the extracted cross-sectional penetration coordinates, the local scatter distribution density, and the mapped one-dimensional discrete sequence; the periodic features include the identified resonant frequency, the limit cycle period of the bifurcation, and the quantified chaotic measure index; the fault diagnosis markers include the determined fault type, the matched confidence score, and the quantified state offset degree; and the diagnostic report includes the determined abnormal fault classification, the predicted fault evolution trend, and the recommended operation and maintenance handling suggestions.

[0026] Chaos metrics: physical quantities used to quantitatively describe the complexity and deterministic characteristics of chaotic behavior in nonlinear dynamic systems. Since the system trajectory of a voltage transformer will enter a non-periodic chaotic state when nonlinear faults such as ferroresonance occur, known metrics such as the maximum Lyapunov exponent, correlation dimension, or information entropy are used as benchmarks to quantify the severity of the system entering a chaotic state.

[0027] For example, S1 specifically includes: The data acquisition card is connected to the secondary side terminal of the voltage transformer. The sampled values ​​of the three-phase voltage channels are 57.3V for phase A, 57.5V for phase B, and 57.4V for phase C. The average value of the three-phase voltage RMS, 57.4V, is taken as the RMS voltage value. The effective values ​​of the three-phase currents on the secondary side read from the current channel are 0.152A, 0.148A, and 0.150A, respectively. The average value of 0.150A is taken as the effective value of the current. The temperature sensor, placed close to the transformer housing, reads a temperature value of 31.2℃, which is taken as the temperature value. The three values ​​57.4, 0.150, and 31.2 are stored in a temporary array. The three elements in this array are compared sequentially. In the first comparison, 57.4 is compared with 0.150; 57.4 is larger and is retained. Then, 57.4 is compared with 31.2; 57.4 is larger and is therefore determined to be the maximum value. The normalization formula is ,in The values ​​range from 1 to 3, corresponding to the subscripts of the three original physical quantities of the sensor: voltage, current, and temperature. Indicates the first The dimensionless values ​​of the original physical quantities after normalization are then subjected to the first division operation. The second division operation The third division operation The three quotients, 1.000, 0.00261, and 0.544, all fall within the range of 0 to 1. These three numbers are combined into an array of length 3 as the normalized state parameter values. The normalized result was compared one by one with the lower bound of 0 and the upper bound of 1. 1.000 was within the closed interval, 0.00261 was within the interval, and 0.544 was within the interval. This result shows that all normalized results are valid and can form a multidimensional state parameter vector. .

[0028] Extract the normalized state parameter value array Each of the three components is checked individually to see if it satisfies the interval condition of being greater than or equal to 0 and less than or equal to 1. The first component, 1.000, is true within the closed interval; the second component, 0.00261, is true within the interval; and the third component, 0.544, is also true within the interval. After all three components pass the interval check, an RS485 communication connection is established with the phase space modeling module. The three components are packaged into a data frame in sequence, with a frame format start code of 0xAA, and data segments of 1.000, 0.00261, and 0.544 respectively, with an additional check code of 0x1F. The frame is sent to the phase space modeling module via the RS485 bus. After the module receives and verifies the data, it parses out the three normalized values ​​and writes them into register addresses 0x01, 0x02, and 0x03 respectively, combining them into a normalized multidimensional state parameter vector. .

[0029] For example, S2 specifically includes: Extracting the normalized multidimensional state parameter vector A three-dimensional array set is constructed from the normalized parameters of the current transformer at 100 consecutive sampling times, and the normalized voltage value array of the first 10 sampling times is extracted. and the normalized current value array Calculate the average voltage and average current Used in the Pearson correlation coefficient formula ,in and Representing voltage and current respectively Each sample value, The first term in the numerator represents the total number of samples. Item 2 Item 3 Item 4 Item 5 Item 6 Item 7 Item 8 Item 9 Item 10 The sum of the numerators is The first part of the denominator Calculate the square of each deviation, the first term Item 2 Item 3 Item 4 Item 5 Item 6 Item 7 Item 8 Item 9 Item 10 Summing gives 0.0000289, and taking the square root gives 0.005376. The second part of the denominator... Calculate the square of each current deviation, the first term. Item 2 Item 3 Item 4 Item 5 Item 6 Item 7 Item 8 Item 9 Item 10 Summing gives 0.000000003, taking the square root gives 0.0000548, substituting into the formula gives... The absolute value of the correlation coefficient, 0.578, is moderate. Similar to the correlation coefficients of 0.32 for voltage and temperature and 0.09 for current and temperature, the three-dimensional coordinate axes are constructed by selecting the current and temperature with the lowest absolute values ​​and adding voltage. The minimum value of the current axis is 0.00200 and the maximum value is 0.00300 at all sampling times, the minimum value of the temperature axis is 0.500 and the maximum value is 0.600, and the minimum value of the voltage axis is 0.950 and the maximum value is 1.050. The scale range of the current axis is set to 0.00200 to 0.00300, the range of the temperature axis is 0.500 to 0.600, and the range of the voltage axis is 0.950 to 1.050, generating a set of spatial axis parameter values.

[0030] Call the spatial axis parameter value set, for the multidimensional parameters at the first sampling time. For the voltage axis parameter value of 1.000, a normalization calculation is performed based on the extreme value range length of 0.100 for this axis. Subtracting the minimum value of 0.950 from 1.000 yields 0.050, which is then divided by 0.100 to obtain 0.500. For the current axis parameter value of 0.00261, subtracting the minimum value of 0.00200 yields 0.00061, which is then divided by the range of 0.00100 to obtain 0.610. For the temperature axis parameter value of 0.544, subtracting the minimum value of 0.500 yields 0.044, which is then divided by the range of 0.100 to obtain 0.440. Combining the three-axis projection values ​​of 0.500, 0.610, and 0.440 gives the three-dimensional coordinate points. This is recorded as the state point at time 1. This operation is repeated for time points 2 to 100 to obtain 100 state points, which are then merged into a state trajectory point sequence for storage.

[0031] For example, S3 specifically includes: The coordinates of all 100 points in the state trajectory point sequence were collected. The ranges were found as follows: voltage coordinate (maximum 0.650, minimum 0.350, range 0.300); current coordinate (maximum 0.700, minimum 0.300, range 0.400); and temperature coordinate (maximum 0.600, minimum 0.200, range 0.400). The maximum range of 0.400 was selected as the baseline. Each segment was set to a length of 0.080, dividing the area into equally spaced cross-sectional positions with a starting point of 0.200 and an ending point of 0.600, resulting in six cross-sectional positions: 0.200, 0.280, 0.360, 0.440, 0.520, and 0.600. The normal vector for each cross-section was determined. The offset is equal to the cross-sectional position value, generating 6 cross-sectional equations. , By sequentially taking the above position values, the normal vector and offset are combined into a set of coefficients for the section equation. Each coefficient contains three normal vector components and one offset. .

[0032] Call the coefficient set of the section equations, for the section Calculate the first trajectory point Sign distance to the cross section, point coordinates The value 0.440 minus the section position 0.200 equals 0.240, which is positive. (Point 2) Subtracting 0.200 from the value 0.439 gives 0.239, which is positive. Continue this calculation until the 49th point is found. Subtracting 0.200 from 0.201 gives 0.001, which is positive. (Point 50) Subtracting 0.200 from the value 0.199 yields -0.001, which is negative. The sign changes from positive to negative. Points 49 and 50 are considered a crossing pair, and a linear interpolation formula is used. ,in and These are coordinate components of two trajectory points before and after crossing. The positive sign distance from the previous point to the cross section. The negative sign distance from the next point to the cross section. For absolute values, interpolation calculations are performed on the voltage components: Interpolation of current components: Interpolation of temperature components: Obtain the coordinates of the intersection point. Repeat this interpolation operation for all 6 sections to obtain a total set of 15 intersection points.

[0033] Based on the intersection point coordinate set, the intersection points within each section are arranged in chronological order of sampling time. In section 0.200, the time interval between adjacent intersection points is 1 sampling interval between points 49 and 50; in section 0.280, the intervals are 3 and 2; in section 0.360, the intervals are 5, 3, and 2; in section 0.440, the intervals are 4, 2, and 3; in section 0.520, the intervals are 6, 1, and 4; and in section 0.600, the intervals are 2 and 5. The difference between adjacent time intervals is calculated. In section 0.200, there is only one interval with no difference; in section 0.280, 3 minus 2 equals 1; in section 0.360, 5 minus 3 equals 2 and 3 minus 2 equals 1; and in section 0.4... Subtracting 2 from 4 in section 40 yields 2, and subtracting 3 from 2 yields an absolute value of 1. Subtracting 1 from 6 in section 0.520 yields 5, and subtracting 4 from 1 yields an absolute value of 3. Subtracting 5 from 2 in section 0.600 yields an absolute value of 3. Setting the time interval threshold to 2, consecutive intervals with a difference less than or equal to 2 are merged. In section 0.280, a difference of 1 less than 2 is directly assigned to the same cycle. In section 0.360, differences of 2 and 1 both less than 2 are merged into one cycle. In section 0.440, differences of 2 and 1 both less than 2 are merged. In section 0.520, differences of 5 and 3 both greater than 2 are retained as two separate cycles. In section 0.600, a difference of 3 greater than 2 is retained as a separate cycle. The indices of the start and end intersection points of the cycles within each section are obtained, and an array of intersection positions and cycle feature values ​​is generated.

[0034] For example, S4 specifically includes: Two historical fault records were retrieved from the historical fault mode database. The first record showed a voltage of 0.5205V, a current of 0.4495A, and a temperature of 0.200℃ at the intersection point; the second record showed a voltage of 0.5300V, a current of 0.4400A, and a temperature of 0.205℃. These two sets of data were used as the initial cluster centers. Euclidean distance was calculated for the feature vectors (voltage, current, and temperature) of the current 15 intersection points using the distance formula... ,in , , The normalized coordinates of the current intersection point. , , Given the coordinates of a cluster center, take the intersection of the newly collected data. With the first cluster center Calculate the voltage difference Squared Current difference Squared temperature difference Squared Sum of three squares square root Distance from the center of the second cluster Calculate voltage difference Squared Current difference Squared temperature difference Squared Summation Extraction The minimum distance of 0.0015 corresponds to the first cluster being assigned to this cluster. This distance calculation and comparison is performed on all 15 intersection points. The first cluster has 10 intersection points, and the second cluster has 5. The mean of each cluster is recalculated. The mean voltage of the new center of the first cluster is 0.5210, the mean current is 0.4490, and the mean temperature is 0.201. The mean voltage of the new center of the second cluster is 0.5310, the mean current is 0.4410, and the mean temperature is 0.206. After another iteration, the assignment of each intersection point remains unchanged, and the final set of cluster center points and the distance distribution statistics within each cluster are obtained. The mean distance from each point to the center in the first cluster is 0.0012, and the standard deviation is 0.0002. The mean distance from the center to each point in the second cluster is 0.0018, and the standard deviation is 0.0003.

[0035] Based on the center of the first cluster Second cluster center For the current newly collected intersection points Construct feature vectors and calculate the Euclidean distance to the center of the first cluster. Distance to the center of the second cluster Generate distance matrix The dynamic threshold is set based on the average distance of the first cluster plus twice the standard deviation. The second cluster If 0.0015 is less than 0.0016 and does not exceed the threshold, and 0.01238 is greater than 0.0024 and exceeds the threshold, the distance matrix and threshold label are passed to the time series feature judgment module.

[0036] Receiver distance matrix In the second cluster of over-threshold markers and periodic features, the period length of the intersection point is 4 sampling intervals. The time-series feature judgment module calls dynamic time warping to warp the current sequence of 4 consecutive intersection points with the historical ferroresonant fault sequence. The voltage at the intersection point of the current sequence... Historical sequence The Euclidean distances between each pair of points in the normalized path are calculated as 0.0015, 0.0005, 0.0011, and 0.0015, respectively, with a cumulative normalized distance of 0.0046. A preset dynamic threshold of 0.0050 is set, and 0.0046 is less than 0.0050. Combined with the distance matrix, the distance of the first cluster (0.0015) is less than the threshold of 0.0016, indicating normal membership, while the distance of the second cluster (0.01238) is greater than the threshold of 0.0024, indicating strong offset. The comprehensive discrimination logic sets the fault determination condition as at least one distance exceeding the threshold and the normalized distance being less than 0.005. Both conditions are met: the distance of the second cluster exceeds the threshold and the normalized distance (0.0046) is less than 0.005. Therefore, a fault state is determined, and a fault diagnosis label type number F02 is generated with a confidence score of 0.93 and a state offset of 0.01238.

[0037] For example, S5 specifically includes: Retrieve all 15 fault diagnosis markers. Each marker carries fault characteristic values: voltage, current, and temperature. Read the characteristic values ​​of the first marker: voltage 0.5220, current 0.4485, temperature 0.202. Preset five threshold ranges: Type 1: voltage range 0.52 to 0.53, current range 0.44 to 0.45, temperature range 0.20 to 0.21; Type 2: voltage range 0.53 to 0.54, current range 0.43 to 0.44, temperature range 0.21 to 0.22; Type 3... The first category has voltages of 0.54 to 0.55, currents of 0.42 to 0.43, and temperatures of 0.22 to 0.23. The second category has voltages of 0.51 to 0.52, currents of 0.45 to 0.46, and temperatures of 0.19 to 0.20. The third category consists of all other combinations. Currently, the voltage of 0.5220 falls within the 0.52 to 0.53 range, the current of 0.4485 falls within the 0.44 to 0.45 range, and the temperature of 0.202 falls within the 0.20 to 0.21 range. All three conditions are met, classifying it as the first category. The second label... Voltage 0.5315 falls within the range of 0.53 to 0.54, current 0.4410 falls outside the range of 0.43 to 0.44 (but within the range of 0.44 to 0.45), and temperature 0.207 falls outside the range of 0.21 to 0.22 (but within the range of 0.20 to 0.21). These are classified into Category 5. A comparison is made of all 15 marks: 5 in Category 1, 3 in Category 2, 1 in Category 3, 2 in Category 4, and 4 in Category 5. The classification ratio formula is used. ,in To be classified as the first The number of flags for each fault type. Take 1 to 5. To mark the total number, For the first The proportion of the class, the first class Category 2 The third category Category 4 Category 5 Obtain the fault type classification value array .

[0038] Based on fault type classification value array The cumulative number of tags for each type is decomposed into 5, 3, 1, 2, and 4, and the percentages of these tags relative to the total number of tags (15) are calculated to be 33.3%, 20.0%, 6.7%, 13.3%, and 26.7%, respectively. Matching intervals are read from the report template library. When the percentage is greater than 30%, template T1 is matched; greater than 20% and less than or equal to 30%, T2 is matched; greater than 10% and less than or equal to 20%, T3 is matched; and less than or equal to 10%, T4 is matched. Specifically, the first type (33.3% and greater than 30%) matches T1; the second type (20.0% and greater than 20% and less than or equal to 30%) matches T2; the third type (6.7% and less than or equal to 10%) matches T4; the fourth type (13.3% and greater than 10% and less than or equal to 20%) matches T3; and the fifth type (26.7% and greater than 20% and less than or equal to 30%) matches T2. The template number sequence is [T1, T2, T4, T3, T2]. The sequences are merged to generate the template matching code "T1T2T4T3T2".

[0039] Based on the template matching code "T1T2T4T3T2", the first T1 in the parsed sequence corresponds to data frame index 1, the second T2 corresponds to index 2, the third T4 corresponds to index 4, the fourth T3 corresponds to index 3, and the fifth T2 corresponds to index 2. The data frame mapping table of the display terminal is called. Index 1 corresponds to the data segment "Ferromagnetic Resonance Warning", index 2 corresponds to "Current Over-Limit Warning", index 4 corresponds to "Temperature Normal", index 3 corresponds to "Voltage Fluctuation Instruction", and index 2 corresponds to "Current Over-Limit Warning". The data segments are concatenated in numerical order to form "Ferromagnetic Resonance Warning, Current Over-Limit Warning, Temperature Normal, Voltage Fluctuation Instruction, Current Over-Limit Warning", and the complete display frame is obtained and output to the display terminal.

[0040] The above description only illustrates the preferred embodiments of the present invention and should not be construed as limiting the scope of the claims. The present invention is not limited to the above embodiments, and variations in its specific structure are permitted. All modifications made within the scope of the independent claims of this invention are also within the scope of protection of this invention.

Claims

1. A method for online fault diagnosis of voltage transformers, characterized in that, The method includes the following steps: S1: Collect the effective values ​​of voltage, current, and temperature through sensors, and perform normalization calculations on the effective values ​​of voltage, current, and temperature to obtain multidimensional state parameters; S2: Construct a three-dimensional phase space coordinate system based on multi-dimensional state parameters, and calculate the state trajectory points corresponding to each sampling time. S3: Based on the state trajectory points, set multiple Poincaré sections, obtain the intersection positions of the state trajectory points and the Poincaré sections, and identify periodic features by comparing the time intervals between adjacent intersection positions; S4: Based on the intersection location and periodic features, the support vector machine is invoked to calculate the similarity between the current intersection location and the fault intersection locations in the pre-stored historical fault mode library, determine whether the current intersection location is a fault state, and generate a fault diagnosis label; S5: Based on the fault diagnosis tags, the fault types are classified and filtered, a diagnostic report is generated, and the diagnostic results are output to the display terminal.

2. The online fault diagnosis method for voltage transformers according to claim 1, characterized in that, The multidimensional state parameters include the processed normalized voltage, the effective value of the characteristic current, and the environmentally corrected temperature. The state trajectory points include the calculated instantaneous spatial coordinates, the associated motion direction vector, and the geometric radius of curvature.

3. The online fault diagnosis method for voltage transformers according to claim 1, characterized in that, The intersection location includes the extracted cross-sectional penetration coordinates, local scatter distribution density, and mapped one-dimensional discrete sequence; the periodic features include the identified resonant frequency, the limiting cycle period of the bifurcation, and the quantified chaotic measure index.

4. The online fault diagnosis method for voltage transformers according to claim 1, characterized in that, The fault diagnosis markers include the determined fault type, the matched confidence score, and the quantified state offset degree.

5. The online fault diagnosis method for voltage transformers according to claim 1, characterized in that, The diagnostic report includes the identified abnormal fault classification, the predicted fault evolution trend, and recommended operation and maintenance handling suggestions.

6. The online fault diagnosis method for voltage transformers according to claim 1, characterized in that, S1 specifically includes: The system collects the effective values ​​of voltage, current, and temperature. It reads the raw signal from the sensor and converts it into three values: effective voltage, effective current, and temperature. It calculates the maximum value of the three values, divides each value by the maximum value to obtain three normalized values ​​between 0 and 1, and combines the three normalized values ​​into a normalized state parameter value. Based on the normalized state parameter values, it is determined whether the three normalized components are all within the 0 to 1 interval. If the interval is within the range, a communication connection is established, and the normalized state parameter values ​​are sent to the phase space modeling module to generate normalized multidimensional state parameters.

7. The online fault diagnosis method for voltage transformers according to claim 1, characterized in that, S2 specifically includes: Based on multidimensional state parameters, the correlation between each pair of parameters is calculated, and the three dimensions with the lowest correlation are selected as coordinate axes. The coordinate axis scale range is set according to the extreme values ​​of the selected dimensions at all sampling times, and a set of spatial axis parameter values ​​is generated. Using the set of spatial axis parameter values ​​as a reference, the corresponding multidimensional parameter values ​​are obtained for each sampling time. The parameter values ​​of each axis direction are normalized based on the length of the extreme value interval of that axis. The projection values ​​of each axis are combined into three-dimensional coordinates and recorded as the state points at the corresponding time. All state points at all times are merged to obtain the state trajectory points.

8. The online fault diagnosis method for voltage transformers according to claim 1, characterized in that, S3 specifically includes: Multiple Poincaré sections are set for the acquisition of trajectory points. The extreme values ​​of each coordinate of the trajectory points are calculated. The position intervals of the equally spaced sections are divided based on the difference of extreme values. The normal vector and offset of each section are calculated to generate the coefficient set of the section equation. Call the coefficient set of the cross-sectional equation, calculate the distance between the coordinates of the trajectory point and the symbol of the cross-sectional plane at each sampling time, filter the adjacent trajectory point pairs with changes in the symbol distance, and interpolate to calculate the intersection point coordinate set at the crossing time. The time interval between adjacent intersection points is calculated based on the intersection point coordinate set. The difference between adjacent time intervals is calculated, and consecutive intervals with a difference less than the time interval threshold are merged into the same period to generate the intersection point position and periodic feature value.

9. The online fault diagnosis method for voltage transformers according to claim 1, characterized in that, S4 specifically includes: The system retrieves the RMS voltage, RMS current, and temperature values ​​of historical fault intersections from a historical fault mode database as input features. An improved K-means clustering algorithm is used to aggregate the feature parameters of each intersection. Through iterative optimization, a set of cluster centroids for multiple fault mode clusters and statistical parameters of distance distribution within each cluster are determined. These cluster centroids and distance distribution parameters are then passed to the distance and threshold analysis module. Based on the cluster centroids, the feature vectors constructed from the RMS voltage, RMS current, and temperature values ​​of the current intersection are used to calculate Euclidean distances, obtaining the distances between the current intersection and each cluster centroid and generating a distance matrix. Dynamic thresholds are set for these distances based on the distance distribution parameters. Based on the distance matrix, threshold labels, and periodic features, a dynamic time warping model is used to reconstruct the temporal features of the current intersection's position over time, extracting a dynamic trend vector. Multidimensional discrimination logic is then used to comprehensively determine whether the current intersection is in a fault state, generating a fault diagnosis label.

10. The online fault diagnosis method for a voltage transformer according to claim 1, characterized in that, S5 specifically includes: Obtain fault diagnosis markers, read the fault feature values ​​carried by each marker, compare the feature values ​​with the boundary values ​​in the five preset type threshold intervals one by one, classify the markers into the corresponding types according to the intervals in which the values ​​are located, accumulate the number of markers belonging to the same type interval, and then calculate the quotient of each accumulated value and the total number of markers to obtain the fault type classification value. Based on the fault type classification value, the cumulative number of tags for each type is decomposed and its percentage of the total number of tags is calculated. The percentage value is compared with the boundary value of the matching interval in the report template library. The template number corresponding to the interval where the percentage is located is selected, and all selected template numbers are merged to form a sequence to generate a template matching code. Based on the template matching code, the data frame index bit corresponding to each number in the sequence is parsed, the data segment under each index bit is obtained by calling the data frame mapping table of the display terminal, and all data segments are concatenated in numerical order to form a complete display frame, thus obtaining the diagnostic result value.

Images