A transformer bushing defect monitoring method and system based on multi-dimensional data fusion

Abstract

The application discloses a kind of transformer bushing defect monitoring method and system based on multidimensional data fusion, the pressure data sequence and temperature data sequence of multiple monitoring points inside transformer bushing are collected, and multidimensional data matrix is established based on sliding window mechanism;According to multidimensional data matrix, identify pressure field reconstruction event, and obtain the response time difference matrix of temperature change relative to pressure change, to carry out timing correction to temperature data sequence;Dynamic correlation model between oil level change and pressure, temperature is established using weighted fusion, and the weight distribution scheme of pressure, temperature is determined;According to dynamic correlation model and weight distribution scheme, multidimensional data fusion is carried out to real-time collected pressure data sequence and temperature data sequence, and transformer bushing defect monitoring is carried out according to obtained multidimensional feature vector, and monitoring result is obtained.The application can improve the accuracy of bushing defect identification and the reliability of bushing defect monitoring.

Description

Technical Field

[0001] This invention relates to the field of transformer defect monitoring technology, and in particular to a method and system for monitoring transformer bushing defects based on multi-dimensional data fusion. Background Technology

[0002] As a core piece of equipment in the power system, the safe and stable operation of power transformers directly affects the reliability of the entire power system. Bushings, as key components of power transformers, bear the important functions of high-voltage output and insulation protection. Real-time monitoring of their internal status is of decisive significance for preventing equipment failures and ensuring power grid security.

[0003] Traditional casing monitoring technology relies primarily on independent measurements of single parameters, lacking in-depth analysis of the coupling relationships between multidimensional parameters. For example, traditional monitoring methods often treat parameters such as pressure and temperature within the casing as independent indicators, ignoring the combined impact of oil level changes on each parameter. This leads to insufficient monitoring accuracy and a high risk of misjudgment. Furthermore, after obtaining multidimensional monitoring data, existing data processing algorithms mostly employ static fusion strategies, which cannot adapt to the dynamic changes in parameters caused by oil level fluctuations. For instance, abnormal fluctuations in the casing oil level trigger dynamic changes in the internal pressure field. This dynamic change process exhibits significant spatiotemporal inhomogeneity, and the dynamic changes in the pressure field further affect the heat conduction path within the casing, causing significant temporal differences in the responses of temperature measuring points at different locations. Ultimately, this results in temporal misalignment of multidimensional monitoring data. This data temporal misalignment directly affects the accurate identification of oil level-related defect features by the fusion algorithm. Moreover, when the algorithm cannot properly handle this temporal inconsistency, it generates a large number of missed detections or false alarms, severely reducing the reliability of defect monitoring. Summary of the Invention

[0004] The purpose of this invention is to provide a transformer bushing defect monitoring method and system based on multi-dimensional data fusion, which can effectively avoid misjudgment, missed detection and false alarm, achieve high-precision monitoring, and thus improve the accuracy of bushing defect identification and the reliability of bushing defect monitoring.

[0005] To achieve the above objectives, embodiments of the present invention provide a method for monitoring transformer bushing defects based on multi-dimensional data fusion, comprising:

[0006] According to the preset acquisition frequency, pressure data sequences and temperature data sequences of multiple monitoring points inside the transformer bushing are acquired, and a multi-dimensional data matrix containing monitoring point location, acquisition time point, pressure and temperature is established based on the sliding window mechanism.

[0007] The pressure change rate of each monitoring point at each collection time point is obtained based on the multidimensional data matrix, and when the pressure field reconstruction event is triggered based on the pressure change rate, the starting monitoring point and starting time point of the pressure field reconstruction are obtained.

[0008] Based on the starting time point and the multidimensional data matrix, the time delay coefficient of temperature change relative to pressure change at each monitoring point is obtained, and a response time difference matrix including the monitoring point location and the time delay coefficient is established.

[0009] Based on the starting monitoring point and the response time difference matrix, the temperature data sequence in the multidimensional data matrix is ​​time-series corrected to obtain a corrected data matrix.

[0010] Based on the starting monitoring point and the correction data matrix, a dynamic correlation model is established between the oil level change inside the transformer bushing and the pressure and temperature, and the weight allocation scheme of pressure and temperature in the dynamic correlation model is determined.

[0011] Based on the dynamic correlation model and the weight allocation scheme, multidimensional data fusion is performed on the real-time collected pressure data sequence and temperature data sequence to establish a multidimensional feature vector containing pressure, temperature and oil level changes. Based on the multidimensional feature vector, transformer bushing defects are monitored to obtain monitoring results. The monitoring results include the defect type and its severity.

[0012] To achieve the above objectives, embodiments of the present invention also provide a transformer bushing defect monitoring system based on multi-dimensional data fusion, comprising:

[0013] The multidimensional data matrix building module is used to collect pressure and temperature data sequences from multiple monitoring points inside the transformer bushing according to a preset acquisition frequency, and to build a multidimensional data matrix containing monitoring point locations, acquisition time points, pressure, and temperature based on a sliding window mechanism.

[0014] The pressure field reconstruction identification module is used to obtain the pressure change rate of each monitoring point at each collection time point according to the multi-dimensional data matrix, and when the pressure field reconstruction event is triggered according to the pressure change rate, it obtains the starting monitoring point and the starting time point of the pressure field reconstruction.

[0015] The response time difference matrix establishment module is used to obtain the time delay coefficient of temperature change relative to pressure change at each monitoring point based on the starting time point and the multidimensional data matrix, and to establish a response time difference matrix that includes the monitoring point location and the time delay coefficient.

[0016] A temperature sequence time-series correction module is used to perform time-series correction on the temperature data sequence in the multidimensional data matrix based on the starting monitoring point and the response time difference matrix to obtain a correction data matrix.

[0017] The correlation model and weight acquisition module is used to establish a dynamic correlation model between the oil level change and pressure and temperature inside the transformer bushing based on the starting monitoring point and the correction data matrix, and to determine the weight allocation scheme of pressure and temperature in the dynamic correlation model.

[0018] The transformer bushing defect monitoring module is used to perform multi-dimensional data fusion on the real-time collected pressure data sequence and temperature data sequence according to the dynamic correlation model and the weight allocation scheme, establish a multi-dimensional feature vector containing pressure, temperature and oil level changes, and perform transformer bushing defect monitoring based on the multi-dimensional feature vector to obtain monitoring results; wherein, the monitoring results include the defect type and its severity.

[0019] Compared with existing technologies, this invention provides a method and system for monitoring transformer bushing defects based on multidimensional data fusion. First, pressure and temperature data sequences from multiple monitoring points inside the transformer bushing are collected according to a preset acquisition frequency. A multidimensional data matrix containing monitoring point locations, acquisition times, pressure, and temperature is established based on a sliding window mechanism. Then, the pressure change rate at each monitoring point at each acquisition time point is obtained based on the multidimensional data matrix. When a pressure field reconstruction event is triggered based on the pressure change rate, the starting monitoring point and starting time point for pressure field reconstruction are obtained. Next, the time delay coefficient of temperature change relative to pressure change at each monitoring point is obtained based on the starting time point and the multidimensional data matrix. A multidimensional data matrix containing monitoring point locations is then established. The invention first establishes a response time difference matrix with a time delay coefficient. Then, based on the initial monitoring point and the response time difference matrix, it performs time-series correction on the temperature data sequence in the multidimensional data matrix to obtain a corrected data matrix. Next, based on the initial monitoring point and the corrected data matrix, it establishes a dynamic correlation model between oil level changes and pressure and temperature inside the transformer bushing, and determines the weight allocation scheme for pressure and temperature in the dynamic correlation model. Finally, according to the dynamic correlation model and the weight allocation scheme, it performs multidimensional data fusion on the real-time collected pressure data sequence and temperature data sequence to establish a multidimensional feature vector containing pressure, temperature, and oil level changes. Based on the multidimensional feature vector, it performs transformer bushing defect monitoring to obtain monitoring results, including the defect type and its severity. This embodiment of the invention can effectively avoid false judgments, missed detections, and false alarms, achieving high-precision monitoring, thereby improving the accuracy of bushing defect identification and the reliability of bushing defect monitoring. Attached Figure Description

[0020] Figure 1This is a flowchart of a transformer bushing defect monitoring method based on multi-dimensional data fusion provided in an embodiment of the present invention;

[0021] Figure 2 This is a structural block diagram of a transformer bushing defect monitoring system based on multi-dimensional data fusion, provided by an embodiment of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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 skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] This invention provides a method for monitoring transformer bushing defects based on multi-dimensional data fusion. (See also...) Figure 1 The diagram shown is a flowchart of a transformer bushing defect monitoring method based on multi-dimensional data fusion according to an embodiment of the present invention. The method includes steps S11 to S16:

[0024] Step S11: Collect pressure and temperature data sequences from multiple monitoring points inside the transformer bushing according to the preset acquisition frequency, and establish a multi-dimensional data matrix containing monitoring point locations, acquisition time points, pressure, and temperature based on the sliding window mechanism.

[0025] In one optional embodiment, the step of acquiring pressure and temperature data sequences from multiple monitoring points inside the transformer bushing according to a preset acquisition frequency, and establishing a multi-dimensional data matrix containing monitoring point locations, acquisition time points, pressure, and temperature based on a sliding window mechanism, specifically includes:

[0026] According to the preset acquisition frequency, pressure and temperature data of multiple monitoring points inside the transformer bushing are acquired, and the corresponding monitoring point locations and acquisition time points are recorded to obtain the original pressure data sequence and the original temperature data sequence.

[0027] The original pressure data sequence and the original temperature data sequence are verified to obtain valid pressure data sequence and valid temperature data sequence that pass the verification; wherein, the verification includes the removal of abnormal data;

[0028] The effective pressure data sequence and the effective temperature data sequence are sorted and organized according to the monitoring point location and the acquisition time point, and an initial data matrix containing the monitoring point location, acquisition time point, pressure and temperature is established based on the sliding window mechanism.

[0029] The pressure and temperature data of each monitoring point in the initial data matrix at the same acquisition time point are aligned to obtain a multi-dimensional data matrix containing the monitoring point location, acquisition time point, pressure, and temperature.

[0030] In this embodiment, pressure data (e.g., pressure data can be collected from multiple monitoring points inside the transformer bushing using a pre-set sampling frequency) and temperature data (e.g., temperature data can be collected from each monitoring point) are collected. The location of each monitoring point (e.g., spatial coordinates) and the sampling time for each data collection are recorded, resulting in original pressure and temperature data sequences containing the monitoring point location and sampling time. Then, the collected original pressure and temperature data sequences are validated to obtain valid pressure and temperature data sequences. Data validation includes, but is not limited to, removing abnormal data. For example, the change in pressure / temperature values ​​between adjacent sampling time points in the original pressure / temperature data sequence is calculated. If the change in pressure / temperature values ​​exceeds a certain threshold, it is marked as abnormal data and removed from the original pressure / temperature data sequence. Next, the valid pressure / temperature data sequences are validated... Based on the sequence and effective temperature data sequence, the data is sorted and organized according to the monitoring point location and collection time point. An initial data matrix containing monitoring point location, collection time point, pressure, and temperature is established based on a sliding window mechanism. For example, the matrix only stores data from the most recent 24 or 72 hours. When new data arrives, the oldest data in the matrix is ​​directly replaced by the new data, i.e., the old data is deleted, thus keeping the matrix size constant. The matrix size is limited by storage capacity and computational efficiency. Finally, based on the initial data matrix, the pressure and temperature data of each monitoring point at the same collection time point are aligned. For example, due to the abnormal data removal process described above, some monitoring points may have missing pressure / temperature data at certain collection time points. For these missing data, a linear interpolation method can be used to calculate the missing values ​​based on the pressure / temperature values ​​of the corresponding monitoring point at the corresponding preceding and following collection time points, and the data is filled in at the corresponding missing positions. After all missing data is filled in, a multi-dimensional data matrix containing monitoring point location, collection time point, pressure, and temperature is obtained.

[0031] The multidimensional data matrix (initial data matrix) is constructed using a three-dimensional storage method. The first dimension of the multidimensional data matrix represents the location of the monitoring point, the second dimension represents the time point of acquisition, and the third dimension contains two parameters: pressure value and temperature value. This matrix structure facilitates the rapid retrieval of data from a specific monitoring point at any time, and also facilitates horizontal multi-monitoring point comparative analysis and vertical time-series trend analysis. Through this systematic data organization method, accurate recording and comprehensive monitoring of the operating status of transformer bushings are achieved.

[0032] It should be noted that the arrangement of multiple monitoring points inside the transformer bushing needs to take into account the structural characteristics of the bushing and the monitoring requirements. In one possible implementation, a monitoring point can be set every 5 meters along the bushing axis, and a pressure sensor and a temperature sensor can be installed at each monitoring point. The pressure sensor can be a piezoresistive sensor, which works by measuring the pressure value by sensing the change in resistance of the strain gauge caused by the pressure of the medium inside the bushing. The temperature sensor is a platinum resistance thermometer, which works by using the characteristics of platinum resistance changing with temperature to obtain an accurate temperature reading.

[0033] It should be noted that the sampling frequency setting directly affects the timeliness and storage capacity of the data. For transformer bushings with rapid pressure changes, the sampling frequency can be set to once every 1 minute, while for relatively stable transformer bushings, the sampling frequency can be reduced to once every 10 minutes. Furthermore, each time data is collected, the current Unix timestamp (i.e., the sampling time point) can be automatically recorded, accurate to the millisecond level, to ensure the accuracy of the time stamp of the data collection.

[0034] It should be noted that the data verification process is crucial to ensuring data quality. Data validity can be determined by calculating the changes in data between adjacent time points. For example, if a monitoring point collects a pressure value of 10 MPa at a certain time point, and one minute later, the pressure value suddenly becomes 50 MPa at the next time point, the pressure change between adjacent time points under normal circumstances is no more than 0.5 MPa per minute. In this case, the 50 MPa pressure data will be marked as abnormal and discarded. Temperature data is verified using a similar method, and the normal temperature change is usually no more than 2°C per minute. This verification mechanism can effectively prevent erroneous data caused by sensor failure or signal interference from entering the subsequent analysis process.

[0035] It should be noted that linear interpolation plays a crucial role in handling missing data. When a monitoring point lacks data at a specific collection time, the nearest valid data before and after that collection time can be used for interpolation calculation. For example, assuming the pressure value at monitoring point A is 12 MPa at 10:00, pressure data is missing at 10:10, and the pressure value at 10:20 is 13 MPa, then the interpolation result at 10:10 is 12.5 MPa. This processing method ensures the continuity of data and provides a complete data foundation for subsequent trend analysis and defect identification.

[0036] Step S12: Obtain the pressure change rate of each monitoring point at each collection time point according to the multidimensional data matrix, and when the pressure field reconstruction event is triggered based on the pressure change rate, obtain the starting monitoring point and starting time point of the pressure field reconstruction.

[0037] In one optional embodiment, the step of obtaining the pressure change rate of each monitoring point at each acquisition time point based on the multidimensional data matrix, and obtaining the starting monitoring point and starting time point of pressure field reconstruction when determining that a pressure field reconstruction event is triggered based on the pressure change rate, specifically includes:

[0038] The pressure change rate of each monitoring point at each collection time point is obtained based on the multidimensional data matrix.

[0039] When the rate of pressure change at any monitoring point at any collection time exceeds the preset pressure fluctuation threshold, the corresponding collection time point is marked as an abnormal time point, and the corresponding monitoring point is marked as an abnormal monitoring point.

[0040] When any abnormal monitoring point has multiple consecutive abnormal time points, a pressure field reconstruction event is triggered, and the corresponding abnormal monitoring point is taken as the starting monitoring point for pressure field reconstruction, and the first abnormal time point of the starting monitoring point is taken as the starting time point for pressure field reconstruction.

[0041] In conjunction with the above embodiments, this embodiment, in its specific implementation, can obtain the pressure data sequence of each monitoring point at each collection time point from the multi-dimensional data matrix. For each monitoring point, the pressure change rate between adjacent collection time points is calculated. For example, the pressure value of the later collection time point can be subtracted from the pressure value of the previous collection time point, and then divided by the time interval between adjacent collection time points to obtain the pressure change rate at the corresponding collection time point, thereby obtaining the pressure change rate of each monitoring point at each collection time point. Next, the pressure change rate of each monitoring point at each collection time point is compared one by one with a preset pressure fluctuation threshold. If it is determined that the pressure change rate of a certain monitoring point at a certain collection time point exceeds the preset pressure fluctuation threshold, then the collection time point is marked as an abnormal time point, and the monitoring point is marked as an abnormal monitoring point. Further, if an abnormal monitoring point has multiple consecutive abnormal time points, for example, if an abnormal monitoring point has three consecutive abnormal time points, then it is determined that the abnormal monitoring point has triggered a pressure field reconstruction event, and the abnormal monitoring point is taken as the starting monitoring point for pressure field reconstruction, and the first abnormal time point appearing at the starting monitoring point is taken as the starting time point for pressure field reconstruction.

[0042] Furthermore, in this embodiment of the invention, within a preset time window after the starting time point (e.g., within 30 minutes; changes outside this time window are not considered to be affected by the same pressure field reconstruction event), it can be determined whether there are abnormal time points at other monitoring points within a preset range of the starting monitoring point (e.g., a distance range of 10 meters above and below the starting monitoring point). If so, it is determined that other monitoring points are also affected by pressure field reconstruction. At this time, the influence range of pressure field reconstruction can be determined based on the spatial coordinates of all other monitoring points affected by the same pressure field reconstruction event. For example, if abnormal pressure change rate is detected at monitoring points from a depth of 1000 meters to 1050 meters, the influence range is 50 meters. This quantitative assessment of the influence range provides accurate spatial positioning information for subsequent casing integrity evaluation and repair measures formulation, which helps to take timely and targeted countermeasures.

[0043] Understandably, when comparing the pressure change rate of each monitoring point at each collection time point with the preset pressure fluctuation threshold, if it is determined that the pressure change rate of all monitoring points at all collection time points does not exceed the preset pressure fluctuation threshold, then it is determined that the transformer bushing is currently operating normally and there is no defect. Subsequent processing steps can be omitted, but it is still necessary to periodically monitor the subsequent data changes.

[0044] It should be noted that when extracting pressure values ​​from the multidimensional data matrix, each monitoring point corresponds to a time series of pressure data. Assuming that the pressure value at a certain monitoring point is 15 MPa at 10:00 and 15.8 MPa at 10:05, with a time interval of 5 minutes, the pressure change rate is calculated as 0.8 MPa divided by 5 minutes, which gives 0.16 MPa / min. This point-by-point calculation method can accurately reflect the dynamic change characteristics of pressure.

[0045] It should be noted that the determination of pressure field reconstruction events adopts a continuous verification mechanism. An abnormal pressure change rate at a single acquisition time point may be caused by measurement error or instantaneous disturbance, while the occurrence of abnormal pressure change rates at multiple consecutive acquisition time points indicates that a fundamental change in the pressure field has indeed occurred. For example, if a monitoring point detects a pressure change rate of 0.35 MPa / min at 10:15, followed by pressure change rates of 0.32 MPa / min, 0.38 MPa / min, and 0.31 MPa / min at 10:20, 10:25, and 10:30 respectively, with the pressure change rates exceeding the threshold of 0.3 MPa / min for four consecutive acquisition time points, this continuous abnormal pressure change rate confirms the occurrence of a pressure field reconstruction event, and 10:15 can be determined as the starting time point of the pressure field reconstruction.

[0046] It should be noted that the spatial propagation characteristics of pressure field reconstruction reflect the changing laws of fluid dynamics inside the casing. When a pressure field reconstruction event occurs at a certain monitoring point, the pressure wave will propagate along the casing to adjacent areas. Assuming that monitoring point A first detects an abnormal pressure rate of change at a depth of 1000 meters, its adjacent monitoring point B at a depth of 1005 meters will also show a similar abnormal pressure rate of change at a later time. By tracking this propagation process, the expansion path of pressure field reconstruction can be accurately grasped.

[0047] Step S13: Based on the starting time point and the multidimensional data matrix, obtain the time delay coefficient of temperature change relative to pressure change at each monitoring point, and establish a response time difference matrix that includes the monitoring point location and the time delay coefficient.

[0048] In one optional embodiment, the step of obtaining the time delay coefficient of temperature change relative to pressure change at each monitoring point based on the starting time point and the multidimensional data matrix, and establishing a response time difference matrix including the monitoring point location and the time delay coefficient, specifically includes:

[0049] Extract the pressure data sequence and temperature data sequence of each monitoring point within each preset time period before and after the starting time point from the multidimensional data matrix;

[0050] For the same monitoring point, the extracted temperature data sequence is gradually shifted relative to the pressure data sequence on the time axis with a preset time step. The correlation coefficient between the temperature data sequence and the pressure data sequence after each shift is calculated based on the Pearson correlation coefficient. The total shift time corresponding to the maximum value of the correlation coefficient is used as the time delay coefficient of the temperature change relative to the pressure change at the corresponding monitoring point.

[0051] A response time difference matrix containing the location of each monitoring point and its time delay coefficient is established based on the time delay coefficient of all monitoring points.

[0052] In conjunction with the above embodiments, this embodiment, in its specific implementation, can extract the pressure data sequence and temperature data sequence of each monitoring point within a preset time period before and after the starting time point from the multidimensional data matrix, based on the starting time point of the pressure field reconstruction. Then, for the same monitoring point, the extracted temperature data sequence is progressively shifted relative to the extracted pressure data sequence on the time axis with a preset time step. Each shift on the time axis adds a preset time step to each acquisition time point of the temperature data sequence, while the temperature value remains unchanged. For example, suppose the temperature data sequence of a certain monitoring point extracted within the preset time period before and after the starting time point is [(t1, T1), (t2, T2)]. [t1+Δt, T1], [t2+Δt, T2], [t3+Δt, T3] are shifted once on the time axis with a preset time step of Δt. The shifted temperature data sequence becomes [(t1+Δt, T1), (t2+Δt, T2), (t3+Δt, T3)]. After each shift, the correlation coefficient between the shifted temperature data sequence and the pressure data sequence at the same monitoring point can be calculated based on the Pearson correlation coefficient. The total shift time corresponding to the maximum value of the correlation coefficient during the entire shift process is taken as the time delay coefficient of the temperature change relative to the pressure change at the same monitoring point. Finally, based on the time delay coefficients of all monitoring points, a response time difference matrix containing the monitoring point location and its corresponding time delay coefficient is established.

[0053] In this matrix, rows represent the locations of monitoring points, and columns represent the corresponding time delay coefficients. Assuming 10 monitoring points are arranged inside the casing, the response time difference matrix is ​​a 10×1 column vector. The first element corresponds to the time delay coefficient of monitoring point 1, the second element corresponds to the time delay coefficient of monitoring point 2, and so on. Through this matrix form, it is possible to quickly identify which monitoring points have abnormal response lags (this can be determined by judging whether the time delay coefficient of the monitoring point is greater than the preset time delay threshold; if so, the corresponding monitoring point is determined to have a response lag), and which monitoring points have response times that meet expectations. This structured data organization method not only facilitates storage and retrieval, but more importantly, it provides quantitative data support for subsequent casing integrity assessment and maintenance decisions.

[0054] It should be noted that the range of data sequence extraction directly affects the accuracy of the analysis results. Assuming the pressure field reconstruction event occurs at 14:30, data from the hour between 14:00 and 15:00 can be extracted for analysis. For example, the pressure data sequence might show a sharp increase at 14:30, from 12 MPa to 18 MPa, while the temperature data sequences at monitoring points at different depths will only begin to respond to this pressure change at different times. This is because when the pressure field is reconstructed, the pressure change is transmitted to different locations through the fluid medium within the casing, while temperature changes often lag behind pressure changes. Accurate identification of this lag is crucial for understanding the thermodynamic processes inside the casing. Therefore, it is necessary to perform time shifting on the time axis for the temperature data sequence at each monitoring point, gradually adjusting the shift from 0 seconds to 300 seconds. After each shift, the correlation coefficient between the temperature and pressure sequences is calculated to determine the time delay coefficient between them. The calculation process of the time delay coefficient reflects the physical characteristics of heat conduction.

[0055] For example, suppose monitoring point A, which is closer to the pressure change source (i.e., the initial monitoring point), has a temperature data sequence whose correlation coefficient with the pressure data sequence of monitoring point A reaches its maximum value of 0.92 after shifting for 15 seconds, indicating that the time delay coefficient of monitoring point A is 15 seconds. On the other hand, monitoring point B, which is farther away from the pressure change source (i.e., the initial monitoring point), needs to shift its temperature data sequence for 45 seconds to achieve a correlation coefficient of 0.88 with the pressure data sequence of monitoring point B, indicating that the time delay coefficient of monitoring point B is 45 seconds. This shows that it takes longer for heat to be transferred to monitoring point B, and this difference reflects the heat conduction path and heat transfer efficiency inside the casing.

[0056] Step S14: Perform time-series correction on the temperature data sequence in the multidimensional data matrix based on the starting monitoring point and the response time difference matrix to obtain the correction data matrix.

[0057] In one optional embodiment, the step of performing time-series correction on the temperature data sequence in the multidimensional data matrix based on the starting monitoring point and the response time difference matrix to obtain a corrected data matrix specifically includes:

[0058] The pressure data sequence of the starting monitoring point is extracted from the multidimensional data matrix as a reference sequence, and the temperature data sequence of each monitoring point is extracted as the temperature sequence to be corrected.

[0059] Based on the time delay coefficient of each monitoring point in the response time difference matrix, the temperature sequence to be corrected at the same monitoring point is shifted and corrected by time axis to obtain the intermediate temperature sequence of each monitoring point.

[0060] A dynamic time warping algorithm is used to calculate the optimal matching relationship between the intermediate temperature sequence of each monitoring point and the reference sequence at each acquisition time point.

[0061] Based on the optimal matching relationship corresponding to each monitoring point, the intermediate temperature sequence of the same monitoring point is time axis aligned and corrected to obtain the corrected temperature sequence of each monitoring point.

[0062] The multidimensional data matrix is ​​updated based on the corrected temperature sequence of each monitoring point to obtain the corrected data matrix.

[0063] In conjunction with the above embodiments, this embodiment, in its specific implementation, can extract the pressure data sequence of the starting monitoring point for pressure field reconstruction from the multidimensional data matrix as the reference sequence, and extract the temperature data sequence of each monitoring point from the multidimensional data matrix as the temperature sequence to be corrected; then, according to the time delay coefficient of each monitoring point in the response time difference matrix, a time offset is set for the temperature sequence to be corrected at the same monitoring point, and the corresponding temperature sequence to be corrected is time-axis shifted and corrected according to the time offset, thereby obtaining the intermediate temperature sequence of each monitoring point; then, the DTW (Dynamic Time Warping) algorithm is used to calculate the difference between the intermediate temperature sequence and the reference sequence at each monitoring point. The optimal matching relationship for data acquisition time points can be determined by, for example, using the DTW algorithm to calculate the Euclidean distance between the baseline sequence and the intermediate temperature sequence at each monitoring point at each acquisition time point, and then using a dynamic programming algorithm to progressively accumulate the minimum cumulative distance from the start of the sequence to each position. The optimal matching relationship between the baseline sequence and the intermediate temperature sequence at each acquisition time point is determined by backtracking the path of the minimum cumulative distance. Next, based on the optimal matching relationship corresponding to each monitoring point, time axis alignment correction is performed on the intermediate temperature sequence of the same monitoring point. That is, a new acquisition time point can be assigned to each temperature data point in the intermediate temperature sequence based on the optimal matching relationship (for example, assuming the optimal matching relationship shows temperature point t). j Corresponding pressure point p i Then t jChange the timestamp to p i The data is collected using timestamps, and the temperature data of each monitoring point is recombined with the pressure data at the corresponding time according to the newly allocated collection time points. This aligns the temperature and pressure data, which originally had a time delay, on the time axis, thus obtaining the corrected temperature sequence for each monitoring point. Finally, the temperature data sequence of the corresponding monitoring point in the multidimensional data matrix is ​​updated based on the corrected temperature sequence of each monitoring point, thus obtaining the corrected data matrix.

[0064] It should be noted that the time offset is set based on the precise information provided by the response time difference matrix. Assuming that the time delay coefficient of monitoring point A is 30 seconds and the time delay coefficient of monitoring point B is 45 seconds, the temperature data sequences of these two monitoring points can be shifted forward by the corresponding number of seconds. This initial time offset lays the foundation for subsequent fine alignment. In other words, the DTW algorithm processes the temperature data sequence after the initial adjustment of the time delay coefficient, rather than the temperature data sequence extracted from the multidimensional data matrix. This logic of "macroscopic alignment first, then microscopic optimization" ensures efficiency and improves the accuracy of time sequence correction, making the data of different monitoring points closer to the real physical correspondence in the time dimension.

[0065] It should be noted that the DTW algorithm plays a crucial role in processing the time-series alignment of casing monitoring data. This algorithm can handle the nonlinear time delay problem between different data sequences. In the casing monitoring scenario, pressure changes, as the main physical driving factor, will cause subsequent temperature changes. However, this change is not a simple fixed time delay, but exhibits complex time-varying characteristics depending on the fluid state and heat transfer conditions inside the casing. Therefore, the DTW algorithm can be used to correct the time axis of the time-series data of each monitoring parameter and adjust the time alignment benchmark of different monitoring parameters. Among them, the calculation of Euclidean distance is the core of the DTW algorithm. For example, if the pressure value at a certain time point is 15 MPa, and the temperature sequence values ​​at different time points are 80℃, 82℃, and 85℃, the algorithm will calculate the distance between the pressure value and each temperature value. This distance is not a physical distance, but a measure reflecting the degree of difference between the two parameter values. Through normalization processing, the two parameters with different dimensions, pressure and temperature, can be compared on the same scale.

[0066] It should be noted that the calculation of the cumulative distance embodies the idea of ​​dynamic programming. Starting from the starting point of the two sequences, the algorithm calculates the minimum cumulative distance to each position step by step. If the current position is the i-th point of the pressure sequence and the j-th point of the temperature sequence, its cumulative distance is equal to the distance of the current point plus the minimum cumulative distance from the three possible predecessor positions to the current position. This calculation method ensures that the found path is globally optimal. The backtracking process of the optimal matching path determines the final time alignment scheme. Starting from the end point of the cumulative distance matrix, the algorithm backtracks to the starting point along the direction of the fastest decrease in cumulative distance. Each point on the backtracking path represents a matching pair of the pressure sequence and the temperature sequence. For example, the 10th time point of the pressure sequence may match the 13th time point of the temperature sequence, indicating that there is a dynamic delay of 3 time units.

[0067] It should be noted that the corrected data matrix obtained after time-series correction achieves true physical event alignment, which can accurately reflect the causal relationship between pressure changes and temperature response. When the pressure changes abruptly at a certain moment, the temperature data of each monitoring point, after time correction, can reflect the response characteristics at the corresponding moment. This alignment not only improves the accuracy of data analysis, but more importantly, it provides a reliable data foundation for subsequent casing condition assessment and defect identification, enabling the defect identification process to more accurately identify potential problems in casing operation.

[0068] Step S15: Based on the starting monitoring point and the correction data matrix, establish a dynamic correlation model between the oil level change and pressure and temperature inside the transformer bushing, and determine the weight allocation scheme of pressure and temperature in the dynamic correlation model.

[0069] In one optional embodiment, the step of establishing a dynamic correlation model between oil level changes and pressure and temperature inside the transformer bushing based on the starting monitoring point and the correction data matrix, and determining the weight allocation scheme for pressure and temperature in the dynamic correlation model, specifically includes:

[0070] Based on the pressure data sequence and corrected temperature sequence of each monitoring point in the correction data matrix, the oil level change sequence inside the transformer bushing is calculated using fluid mechanics principles.

[0071] Based on the Pearson correlation coefficient, the correlation coefficients between the oil level change sequence and the pressure data sequence at the starting monitoring point, and the corrected temperature sequence at each monitoring point, were calculated respectively.

[0072] Each calculated correlation coefficient is compared with a preset correlation threshold. When any correlation coefficient is less than the correlation threshold, the weight of the pressure data sequence or corrected temperature sequence associated with the corresponding correlation coefficient is adjusted to the ratio of the corresponding correlation coefficient to the correlation threshold. Otherwise, the initial weight remains unchanged to obtain the weight allocation scheme of the pressure data sequence of the starting monitoring point and the corrected temperature sequence of each monitoring point.

[0073] According to the weighting scheme, the pressure data sequence of the starting monitoring point and the corrected temperature sequence of each monitoring point are weighted and fused to obtain a comprehensive parameter sequence.

[0074] The least squares method is used to establish a linear correspondence between the oil level change sequence and the comprehensive parameter sequence, so as to obtain a dynamic correlation model between the oil level change and pressure and temperature inside the transformer bushing.

[0075] In conjunction with the above embodiments, this embodiment, in its specific implementation, can, based on the pressure data sequence and corrected temperature sequence of each monitoring point in the correction data matrix, use fluid dynamics principles to calculate the oil level change sequence inside the transformer bushing (using the combined effect of pressure and temperature changes to calculate the oil level change, or it can be directly measured by an additional oil level sensor); then, based on the Pearson correlation coefficient, calculate the correlation coefficient between the oil level change sequence and the pressure data sequence of the initial monitoring point reconstructed by the pressure field, and calculate the correlation coefficient between the oil level change sequence and the corrected temperature sequence of each monitoring point. For example, by statistically calculating the covariance of the pressure data sequence and the oil level change sequence of the initial monitoring point and dividing it by the product of their standard deviations, the correlation coefficient between the pressure data sequence and the oil level change sequence of the initial monitoring point can be obtained. By statistically calculating the covariance of the corrected temperature sequence and the oil level change sequence of a certain monitoring point and dividing it by the product of their standard deviations, the correlation coefficient between the corrected temperature sequence and the oil level change sequence of that monitoring point can be obtained. Next, each calculated correlation coefficient is compared with a preset correlation threshold. When a correlation coefficient is less than the preset correlation threshold, the pressure data sequence or... The weight of the corrected temperature sequence is adjusted to the ratio of the correlation coefficient to the preset correlation threshold. Otherwise, the weight of the pressure data sequence or the corrected temperature sequence associated with the correlation coefficient is kept unchanged from the initial weight (the pressure data sequence of the starting monitoring point and the corrected temperature sequence of each monitoring point are given initial weights, for example, the initial weights are all 1), so as to obtain the weight allocation scheme of the pressure data sequence of the starting monitoring point and the corrected temperature sequence of each monitoring point, that is, to obtain the optimal fusion weight of the pressure data sequence of the starting monitoring point and the optimal fusion weight of the corrected temperature sequence of each monitoring point. Then, according to the obtained weight allocation scheme, the pressure data sequence of the starting monitoring point and the corrected temperature sequence of each monitoring point are weighted and fused to obtain the comprehensive parameter sequence containing pressure and temperature. Finally, the least squares method is used to establish a linear correspondence between the oil level change sequence and the comprehensive parameter sequence to obtain the dynamic correlation model between the oil level change and pressure and temperature inside the transformer bushing. For example, the expression of the dynamic correlation model is: H = a × comprehensive parameter sequence + b, where a and b are regression coefficients, which can be determined by minimizing the sum of squared errors between the predicted value and the actual value using the least squares method.

[0076] It's important to note that calculating the correlation coefficient is fundamental to establishing the relationship between oil level changes and pressure and temperature. Specifically, the correlation coefficient reflects the strength of the linear relationship between two variables. In casing monitoring scenarios, oil level changes are often closely related to pressure changes. When the pressure inside the casing increases, the fluid is compressed or pushed, causing the oil level to rise; conversely, when the pressure decreases, the oil level falls. This physical relationship can be quantified by calculating the product of the covariance and the standard deviation. The correlation coefficient calculation requires a synchronized time data series. Assuming that within a certain time period, the oil level rises from 2.5 meters to 3.2 meters, while the pressure increases from 10 MPa to 12 MPa, calculating the covariance of these two series reveals a highly consistent trend. A positive covariance indicates that they change in the same direction; a larger value indicates a stronger correlation. Dividing by the product of their respective standard deviations and normalizing the result yields a correlation coefficient between -1 and 1, facilitating comparisons between different parameters.

[0077] It should be noted that the setting of the correlation threshold reflects the criteria for judging the importance of parameters. Different parameters have different degrees of correlation with oil level. Pressure parameters usually have a direct physical causal relationship with oil level changes, and the correlation coefficient often reaches above 0.8. The influence of temperature parameters is relatively indirect, mainly affecting oil level by changing fluid viscosity and density, and the correlation coefficient may be around 0.5. Preferably, the correlation threshold can be set to 0.6. Parameters with correlation coefficients higher than this threshold are considered to have a significant impact, while parameters with correlation coefficients lower than this threshold need to have their weight reduced in the dynamic correlation model. The weight adjustment mechanism ensures the accuracy and stability of the dynamic correlation model.

[0078] For example, if the correlation coefficient of the corrected temperature sequence at a certain monitoring point is 0.45, which is lower than the correlation threshold of 0.6, its weight will be adjusted to 0.45 divided by 0.6, which is 0.75. This means that the corrected temperature sequence at this monitoring point only contributes 75% of the original value in the final comprehensive parameter calculation. If the correlation coefficient of the pressure data sequence at the starting monitoring point is 0.85, which is higher than the correlation threshold of 0.6, its weight will remain at 1, fully reflecting its dominant role in oil level changes.

[0079] It should be noted that the implementation of the weighted fusion algorithm reflects the idea of ​​the comprehensive influence of multiple parameters. For example, suppose that at a certain moment, the pressure value is 11 MPa with a weight of 1; temperature value 1 is 82℃ with a weight of 0.75; and temperature value 2 is 85℃ with a weight of 0.9. The weighted fusion parameter value is calculated as: 11 × 1 + 82 × 0.75 + 85 × 0.9. The resulting comprehensive parameter value reflects the weighted influence of multiple parameters. This approach retains the role of the main influencing factors while appropriately considering the contribution of secondary factors.

[0080] It should be noted that the dynamic correlation model is established by assuming a linear relationship. The optimal regression coefficient can be determined by minimizing the deviation between the actual oil level and the model prediction. The advantage of this method is that the model is simple in form, has high computational efficiency, and has clear physical meaning. The model established in this way can accurately predict the oil level change under given pressure and temperature conditions, providing reliable technical support for real-time monitoring and early warning of abnormalities in casing operation.

[0081] Step S16: Based on the dynamic correlation model and the weight allocation scheme, perform multi-dimensional data fusion on the real-time collected pressure data sequence and temperature data sequence to establish a multi-dimensional feature vector containing pressure, temperature and oil level changes, and perform transformer bushing defect monitoring based on the multi-dimensional feature vector to obtain monitoring results; wherein, the monitoring results include defect type and its severity.

[0082] In one optional embodiment, the step of performing multi-dimensional data fusion on the real-time collected pressure data sequence and temperature data sequence according to the dynamic correlation model and the weight allocation scheme to establish a multi-dimensional feature vector containing pressure, temperature and oil level changes, and performing transformer bushing defect monitoring based on the multi-dimensional feature vector to obtain monitoring results, specifically includes:

[0083] According to the weighting scheme, the pressure data sequence of the starting monitoring point and the corrected temperature sequence of each monitoring point are weighted to obtain the weighted pressure characteristics of the starting monitoring point and the weighted temperature characteristics of each monitoring point.

[0084] The weighted pressure characteristics of the starting monitoring point and the weighted temperature characteristics of each monitoring point are input into the dynamic correlation model to obtain the oil level change characteristics;

[0085] A standard feature vector is established based on the weighted pressure characteristics of the starting monitoring point, the weighted temperature characteristics of each monitoring point, and the oil level change characteristics;

[0086] Using the structure of the standard feature vector as a standard, multi-dimensional data fusion is performed on the real-time collected pressure data sequence and temperature data sequence to establish a multi-dimensional feature vector containing pressure, temperature and oil level changes;

[0087] The multidimensional feature vector is input into the trained defect recognition model to obtain the confidence level of the multidimensional feature vector being recognized as each defect category;

[0088] Based on the confidence level of the multidimensional feature vectors identified as each defect category, it is determined whether there is a defect type corresponding to the defect category in the transformer bushing. If there is, the severity of the defect type is determined according to the corresponding confidence level, so as to obtain the monitoring results based on the existing defect types and their severity.

[0089] In conjunction with the above embodiments, in specific implementation, this embodiment can, according to the obtained weight allocation scheme (including the optimal fusion weight of the pressure data sequence of the starting monitoring point and the optimal fusion weight of the corrected temperature sequence of each monitoring point), perform weighted processing on the pressure data sequence of the starting monitoring point and the corrected temperature sequence of each monitoring point respectively, to obtain the weighted pressure feature of the starting monitoring point (obtained by multiplying the pressure data sequence of the starting monitoring point by its optimal fusion weight) and the weighted temperature feature of each monitoring point (calculated by multiplying the corrected temperature sequence of each monitoring point by its corresponding optimal fusion weight); then, the obtained starting... The weighted pressure characteristics of the initial monitoring point and the weighted temperature characteristics of each monitoring point are input into the dynamic correlation model to obtain the corresponding oil level change characteristics. Then, based on the weighted pressure characteristics of the initial monitoring point, the weighted temperature characteristics of each monitoring point, and the oil level change characteristics, a standard feature vector is established. For example, this standard feature vector can be represented as a vector form of [weighted pressure value, weighted temperature value 1, weighted temperature value 2, ..., oil level change value]. Then, using the feature structure of the standard feature vector as a standard, multi-dimensional data fusion is performed on the real-time collected pressure data sequence and temperature data sequence (i.e., the real-time collected pressure data sequence and temperature data sequence are fused). The sequence execution steps S11~S14 correspond to the processing procedures, and then the multi-dimensional data fusion in this embodiment is performed according to the processing results to establish a multi-dimensional feature vector containing real-time pressure, real-time temperature, and real-time oil level changes, consistent with the feature structure of the standard feature vector. Next, the obtained multi-dimensional feature vector is input into the trained defect recognition model for recognition, and the confidence level of the multi-dimensional feature vector being identified as each defect category is obtained accordingly. Finally, based on the confidence level of the multi-dimensional feature vector being identified as each defect category, it is determined whether the transformer bushing has a defect type corresponding to the defect category. For example, the confidence level of a certain defect category is determined. If the confidence level is higher than the preset identification threshold, it is determined that the transformer bushing has a defect type corresponding to the defect category; otherwise, it is determined that the transformer bushing does not have a defect type corresponding to the defect category. Accordingly, if it is determined that the transformer bushing has a defect type corresponding to the defect category, the severity of the defect type corresponding to the defect category is determined according to the numerical range of the confidence level of the defect category, so as to obtain the monitoring results based on the existing defect type (e.g., normal state, leakage defect, valve failure, pressure abnormality, etc.) and its corresponding severity (e.g., slight leakage, moderate leakage, severe leakage, etc.).

[0090] It should be noted that the construction of multidimensional feature vectors embodies the core idea of ​​multidimensional data fusion. According to the weight allocation scheme determined earlier, if the optimal fusion weight of temperature value is 1 and the optimal fusion weight of temperature value is 0.8, then when constructing multidimensional feature vectors, pressure value directly uses the original value, while temperature value needs to be multiplied by a weight coefficient of 0.8. This weighted processing ensures that parameters that have a greater impact on oil level changes occupy a more important position in the feature space, thereby improving the accuracy of subsequent defect category classification.

[0091] It should be noted that, for the trained defect identification model, feature vectors under normal operating conditions and feature vectors when oil level anomalies occur can be extracted from historical monitoring data (the feature structure is consistent with that of standard feature vectors). Each feature vector is labeled with a corresponding category label to form a training dataset. The training dataset is then input into a support vector machine classification algorithm, which maps the feature vectors to a high-dimensional space through a kernel function. In the high-dimensional space, decision boundaries that can distinguish different categories are found, and the defect identification model is trained. This model calculates the confidence value of each category based on the distance between the feature vector and the decision boundary. The multi-dimensional feature vectors that are collected in real time and processed with weights are input into the trained defect identification model, and the model calculates the distance from the multi-dimensional feature vector to the decision boundary of each category and converts it into a confidence value.

[0092] It should be noted that the construction of the training dataset requires sufficient historical data support. Under normal operating conditions, the oil level remains stable within the range of 2.5 meters to 3.5 meters, corresponding to pressure values ​​between 10 MPa and 12 MPa, and temperatures between 80°C and 85°C. When an abnormal oil level occurs, such as a sudden drop below 2 meters, it is usually accompanied by an abnormal pressure drop below 8 MPa. These abnormal state data are labeled with corresponding defect types, such as "leakage defect" or "valve failure." The kernel function mapping of the support vector machine is the key to achieving nonlinear classification. In the original feature space, the data points of normal and defect states may overlap, making it difficult to separate them with simple linear boundaries. Through the path mapping of the support vector machine... The basis kernel function maps the original three-dimensional feature vectors to a higher-dimensional space, making the originally overlapping data points separable in the new space. This mapping preserves the relative relationships between data points while enhancing the distinguishability between different categories. The process of determining the decision boundary reflects the optimization idea of ​​the support vector machine. The algorithm finds the hyperplane that maximizes the interval between different categories. The data points closest to the decision boundary are called support vectors, which determine the location of the boundary. For example, the support vector for a normal state might be a data point with a pressure of 11.5 MPa and a temperature of 83°C, while the support vector for a leakage defect might be a data point with a pressure of 7.5 MPa and a temperature of 78°C. These key data points determine the precise location of the classification boundary.

[0093] It should be noted that the confidence score is calculated based on the distance from the feature vector to the decision boundary. Preferably, when a new feature vector is input into the model, the algorithm calculates the algebraic distance from the point to the decision boundary of each class. A positive distance indicates that the point is on the positive class side of the boundary, and a negative distance indicates that the point is on the negative class side. These distance values ​​can then be converted into a confidence score in the form of probabilities between 0 and 1 using the sigmoid function. For example, a confidence score of 0.9 means that the model has a 90% confidence that the sample belongs to a certain defect category.

[0094] It should be noted that the severity can be determined using a segmented mapping method. For example, for leakage defects, a confidence level between 0.7 and 0.8 is considered a minor leak, between 0.8 and 0.9 is a moderate leak, and above 0.9 is a severe leak. This classification method allows maintenance personnel to formulate corresponding handling strategies based on the severity of the defect. Minor defects can be repaired in a planned manner, while severe defects need to be dealt with immediately, thereby achieving refined management of casing operation risks.

[0095] In one alternative embodiment, the method further includes:

[0096] Based on the historical monitoring database, the defect evolution trend analysis is performed on the defect types contained in the monitoring results, and the corresponding defect early warning mechanism is triggered based on the analysis results and the monitoring results.

[0097] In one optional embodiment, the step of performing defect evolution trend analysis on the defect types included in the monitoring results based on a historical monitoring database, and triggering a corresponding defect early warning mechanism based on the analysis results and the monitoring results, specifically includes:

[0098] Using the defect types contained in the monitoring results as target defects, historical data corresponding to the same type of defects as the target defects are extracted from the historical monitoring database; wherein, the historical data includes each time point in time when the same type of defect was identified and its severity.

[0099] Based on the historical data, defect evolution trend analysis is performed to obtain the average evolution rate of the target defect;

[0100] The severity of the target defect contained in the monitoring results is taken as the current severity, and the deterioration time required for the target defect to reach the maximum severity from the current severity is predicted based on the average evolution rate.

[0101] The corresponding defect warning mechanism is triggered based on the deterioration time and the current severity; wherein, the defect warning mechanism is divided into different warning levels based on different deterioration times and different current severity.

[0102] In conjunction with the above embodiments, this embodiment, in its specific implementation, can use the defect type contained in the monitoring results as the target defect. Historical data corresponding to defects of the same type as the target defect (i.e., defects of the same type) can be extracted from the historical monitoring database. This historical data includes each time point in time when the same type of defect was identified and its severity, particularly the time point when the same type of defect was first discovered, its severity, and its evolution status at subsequent time points. Then, based on the extracted historical data, defect evolution trend analysis is performed on the same type of defect to obtain the average evolution rate of the target defect. For example, defects of the same type often have similar evolution patterns. Based on the extracted historical data, a linear regression method can be used, with time as the independent variable, to analyze the evolution trend of the same type of defect. The severity value of the defect is used as the dependent variable to fit the data, thereby obtaining the relationship between the severity value and time. The slope of the fitted line can be used as the average evolution rate of the same type of defect, which is equivalent to the average evolution rate of the target defect. Next, the severity of the target defect included in the monitoring results is taken as the current severity. Based on the average evolution rate of the target defect and the current severity, the deterioration time required for the target defect to reach the maximum severity from the current severity is predicted, thereby obtaining the defect deterioration time prediction value. Finally, based on the obtained defect deterioration time prediction value and the current severity of the target defect, a corresponding defect early warning mechanism is triggered. The defect early warning mechanism is divided into different early warning levels based on different deterioration times and different current severity.

[0103] For example, the expression for the relationship between severity values ​​and time obtained by fitting a linear regression method is: y = kx + c, where x represents time, y represents the severity value, k represents the slope, and c represents the intercept. The values ​​of the slope k and the intercept c can be obtained using the least squares method. For instance, if the fitted slope for a leakage defect is 8 units per month, this means that the severity increases by an average of 8 units per month.

[0104] It should be noted that the severity value can be expressed as a percentage. For example, a severity value of 0-30 indicates a minor defect, 30-70 indicates a moderate defect, and 70-100 indicates a severe defect. A severity value of 100 represents the highest severity, meaning that the defect has developed to a critical state that requires immediate attention. This quantification method makes the severity of different types of defects comparable, which is convenient for unified management and analysis.

[0105] It should be noted that the prediction of the deterioration time is calculated based on the current state and the rate of evolution. For example, assuming that the current severity of a defect is 45, which is a moderate defect, and the average rate of evolution obtained from historical monitoring data is 8 units per month, then the difference from 45 to 100 is 55. Dividing this by the average rate of evolution of 8 gives a deterioration time of about 7 months. This prediction indicates that if no intervention measures are taken, the defect will develop into its most severe state in 7 months.

[0106] It should be noted that the design of the warning levels in the defect warning mechanism reflects the refined requirements of risk management. For example, the warning levels of a complete defect warning mechanism are divided as follows:

[0107] Level 1 warning: The deterioration period is less than 3 months and the current severity level is moderate (30~70) or severe (70~100).

[0108] Level 2 warning: The deterioration period is between 3 and 6 months and the current severity is moderate (30-70) or severe (70-100).

[0109] Level 3 warning: The deterioration period is between 6 and 12 months and the current severity is moderate (30-70) or severe (70-100).

[0110] Level 4 warning: The deterioration period is >12 months and the current severity is moderate (30~70).

[0111] Monitoring only (can be understood as level zero warning): The current severity level is mild (0~30), and regular checks are performed without triggering a warning.

[0112] For example, when the deterioration period is less than 3 months and the defect is already of moderate severity, it indicates that the defect is deteriorating rapidly and the basic severity is high, requiring a Level 1 warning and immediate maintenance. When the deterioration period is between 3 and 6 months and the defect is already of moderate severity, a Level 2 warning should be triggered, and the defect can be included in the planned maintenance schedule.

[0113] It should be noted that this defect evolution trend analysis and early warning mechanism based on historical monitoring data has significant practical value. By accurately predicting the deterioration time of defects, maintenance personnel can rationally arrange maintenance plans and intervene before defects develop to a dangerous level. At the same time, the tiered early warning also avoids excessive investment of resources. Defects that evolve slowly can be monitored regularly, while defects that deteriorate rapidly can be responded to in a timely manner, thus achieving an organic combination of preventive maintenance and emergency response.

[0114] This invention provides a transformer bushing defect monitoring method based on multidimensional data fusion. It collects pressure and temperature data sequences from multiple monitoring points inside the transformer bushing, establishes a multidimensional data matrix based on a sliding window mechanism, identifies pressure field reconstruction events based on the multidimensional data matrix, and obtains the response time difference matrix of temperature changes relative to pressure changes to perform time-series correction on the temperature data sequence. Then, it uses weighted fusion to establish a dynamic correlation model between oil level changes and pressure and temperature, determines the weight allocation scheme for pressure and temperature, and finally performs multidimensional data fusion on the real-time collected pressure and temperature data sequences according to the dynamic correlation model and the weight allocation scheme. The system monitors transformer bushing defects based on the obtained multidimensional feature vectors, obtains monitoring results, analyzes the defect evolution trend to obtain the average evolution rate of the defects, and predicts the deterioration time required for the defects to reach the highest severity level from the current severity level based on the average evolution rate. Then, it triggers different levels of defect early warning mechanisms based on the deterioration time and the current severity of the defects, thereby effectively avoiding misjudgments, missed detections, and false alarms, achieving high-precision monitoring, and improving the accuracy and reliability of bushing defect identification and monitoring. At the same time, it can also accurately monitor oil level changes in the bushing, identify potential defects in a timely manner, effectively prevent safety accidents, and improve the reliability of oil well operation.

[0115] This invention also provides a transformer bushing defect monitoring system based on multi-dimensional data fusion, used to implement the transformer bushing defect monitoring method based on multi-dimensional data fusion described in any of the above embodiments. See [link to relevant documentation]. Figure 2 The diagram shown is a structural block diagram of a transformer bushing defect monitoring system based on multi-dimensional data fusion according to an embodiment of the present invention. The system includes:

[0116] The multidimensional data matrix establishment module 11 is used to collect pressure data sequences and temperature data sequences from multiple monitoring points inside the transformer bushing according to a preset acquisition frequency, and to establish a multidimensional data matrix containing monitoring point locations, acquisition time points, pressure, and temperature based on a sliding window mechanism.

[0117] The pressure field reconstruction identification module 12 is used to obtain the pressure change rate of each monitoring point at each collection time point according to the multi-dimensional data matrix, and when the pressure field reconstruction event is triggered according to the pressure change rate, obtain the starting monitoring point and starting time point of the pressure field reconstruction.

[0118] The response time difference matrix establishment module 13 is used to obtain the time delay coefficient of temperature change relative to pressure change at each monitoring point based on the starting time point and the multidimensional data matrix, and to establish a response time difference matrix including the monitoring point location and the time delay coefficient.

[0119] Temperature sequence time-series correction module 14 is used to perform time-series correction on the temperature data sequence in the multidimensional data matrix based on the starting monitoring point and the response time difference matrix to obtain a correction data matrix;

[0120] The correlation model and weight acquisition module 15 is used to establish a dynamic correlation model between the oil level change and pressure and temperature inside the transformer bushing based on the starting monitoring point and the correction data matrix, and to determine the weight allocation scheme of pressure and temperature in the dynamic correlation model.

[0121] The transformer bushing defect monitoring module 16 is used to perform multi-dimensional data fusion on the real-time collected pressure data sequence and temperature data sequence according to the dynamic correlation model and the weight allocation scheme, establish a multi-dimensional feature vector containing pressure, temperature and oil level changes, and perform transformer bushing defect monitoring according to the multi-dimensional feature vector to obtain monitoring results; wherein, the monitoring results include the defect type and its severity.

[0122] Preferably, the multidimensional data matrix establishment module 11 specifically includes:

[0123] The raw data sequence acquisition unit is used to acquire pressure and temperature data from multiple monitoring points inside the transformer bushing according to a preset acquisition frequency, and record the corresponding monitoring point location and acquisition time point to obtain the raw pressure data sequence and raw temperature data sequence.

[0124] The original data sequence verification unit is used to verify the original pressure data sequence and the original temperature data sequence to obtain a valid pressure data sequence and a valid temperature data sequence that pass the verification; wherein, the verification includes the removal of abnormal data;

[0125] The initial data matrix establishment unit is used to sort and organize the effective pressure data sequence and the effective temperature data sequence according to the monitoring point location and the acquisition time point, and establish an initial data matrix containing the monitoring point location, acquisition time point, pressure and temperature based on the sliding window mechanism.

[0126] The multidimensional data matrix establishment unit is used to perform data alignment processing on the pressure and temperature data of each monitoring point in the initial data matrix at the same acquisition time point to obtain a multidimensional data matrix containing the monitoring point location, acquisition time point, pressure, and temperature.

[0127] Preferably, the pressure field reconstruction and identification module 12 specifically includes:

[0128] The pressure change rate acquisition unit is used to acquire the pressure change rate of each monitoring point at each acquisition time point based on the multidimensional data matrix.

[0129] The abnormal time point and monitoring point identification unit is used to mark the corresponding collection time point as an abnormal time point and the corresponding monitoring point as an abnormal monitoring point when the pressure change rate of any monitoring point at any collection time point exceeds the preset pressure fluctuation threshold.

[0130] The pressure field reconstruction identification unit is used to determine that a pressure field reconstruction event is triggered when any abnormal monitoring point has multiple consecutive abnormal time points, and to take the corresponding abnormal monitoring point as the starting monitoring point of the pressure field reconstruction, and take the first abnormal time point of the starting monitoring point as the starting time point of the pressure field reconstruction.

[0131] Preferably, the response time difference matrix establishment module 13 specifically includes:

[0132] The first data sequence extraction unit is used to extract the pressure data sequence and temperature data sequence of each monitoring point within each preset time period before and after the starting time point from the multidimensional data matrix.

[0133] The time delay coefficient acquisition unit is used to progressively shift the extracted temperature data sequence relative to the pressure data sequence on the time axis with a preset time step for the same monitoring point, and calculate the correlation coefficient between the temperature data sequence and the pressure data sequence after each shift based on the Pearson correlation coefficient. The total shift time corresponding to the maximum value of the correlation coefficient is used as the time delay coefficient of the temperature change relative to the pressure change of the corresponding monitoring point.

[0134] The response time difference matrix building unit is used to build a response time difference matrix containing the location of the monitoring points and their time delay coefficients based on the time delay coefficients of all monitoring points.

[0135] Preferably, the temperature sequence timing correction module 14 specifically includes:

[0136] The second data sequence extraction unit is used to extract the pressure data sequence of the starting monitoring point from the multidimensional data matrix as a reference sequence, and to extract the temperature data sequence of each monitoring point as a temperature sequence to be corrected.

[0137] The first time-series correction unit is used to perform time-axis translation correction on the temperature sequence to be corrected of the same monitoring point according to the time delay coefficient of each monitoring point in the response time difference matrix, so as to obtain the intermediate temperature sequence of each monitoring point.

[0138] The optimal matching relationship acquisition unit is used to calculate the optimal matching relationship between the intermediate temperature sequence of each monitoring point and the reference sequence at each acquisition time point using a dynamic time warping algorithm.

[0139] The second time-series correction unit is used to perform time-axis alignment correction on the intermediate temperature sequence of the same monitoring point according to the optimal matching relationship corresponding to each monitoring point, so as to obtain the corrected temperature sequence of each monitoring point.

[0140] The multidimensional data matrix update unit is used to update the multidimensional data matrix according to the corrected temperature sequence of each monitoring point to obtain the corrected data matrix.

[0141] Preferably, the association model and weight acquisition module 15 specifically includes:

[0142] The oil level change sequence estimation unit is used to estimate the oil level change sequence inside the transformer bushing based on the pressure data sequence and the corrected temperature sequence of each monitoring point in the correction data matrix, using the principles of fluid mechanics.

[0143] The correlation coefficient calculation unit is used to calculate the correlation coefficient between the oil level change sequence and the pressure data sequence of the starting monitoring point, and the corrected temperature sequence of each monitoring point, based on the Pearson correlation coefficient.

[0144] The weight allocation scheme acquisition unit is used to compare each calculated correlation coefficient with a preset correlation threshold. When any correlation coefficient is less than the correlation threshold, the weight of the pressure data sequence or corrected temperature sequence associated with the corresponding correlation coefficient is adjusted to the ratio of the corresponding correlation coefficient to the correlation threshold. Otherwise, the initial weight remains unchanged to obtain the weight allocation scheme of the pressure data sequence of the starting monitoring point and the corrected temperature sequence of each monitoring point.

[0145] The comprehensive parameter sequence acquisition unit is used to perform weighted fusion processing on the pressure data sequence of the starting monitoring point and the corrected temperature sequence of each monitoring point according to the weight allocation scheme to obtain the comprehensive parameter sequence.

[0146] The dynamic correlation model acquisition unit is used to establish a linear correspondence between the oil level change sequence and the comprehensive parameter sequence using the least squares method, so as to obtain a dynamic correlation model between the oil level change and pressure and temperature inside the transformer bushing.

[0147] Preferably, the transformer bushing defect monitoring module 16 specifically includes:

[0148] The pressure and temperature feature acquisition unit is used to perform weighted processing on the pressure data sequence of the starting monitoring point and the corrected temperature sequence of each monitoring point according to the weight allocation scheme, so as to obtain the weighted pressure feature of the starting monitoring point and the weighted temperature feature of each monitoring point.

[0149] The oil level change feature acquisition unit is used to input the weighted pressure feature of the starting monitoring point and the weighted temperature feature of each monitoring point into the dynamic correlation model to obtain the oil level change feature;

[0150] The standard feature vector establishment unit is used to establish a standard feature vector based on the weighted pressure characteristics of the starting monitoring point, the weighted temperature characteristics of each monitoring point, and the oil level change characteristics;

[0151] The multidimensional feature vector establishment unit is used to perform multidimensional data fusion on the real-time collected pressure data sequence and temperature data sequence based on the structure of the standard feature vector, and to establish a multidimensional feature vector containing pressure, temperature and oil level changes.

[0152] The defect category identification unit is used to input the multidimensional feature vector into the trained defect identification model to obtain the confidence level of the multidimensional feature vector being identified as each defect category.

[0153] The defect monitoring result acquisition unit is used to determine whether there is a defect type corresponding to the defect type of the transformer bushing based on the confidence level of the multidimensional feature vector being identified as each defect category. If there is, the severity of the defect type is determined based on the corresponding confidence level, so as to obtain the monitoring result based on the existing defect type and its severity.

[0154] Preferably, the system further includes:

[0155] The defect evolution classification and early warning module is used to perform defect evolution trend analysis on the defect types contained in the monitoring results based on the historical monitoring database, and to trigger the corresponding defect early warning mechanism based on the analysis results and the monitoring results.

[0156] Preferably, the defect evolution classification and early warning module specifically includes:

[0157] The historical data extraction unit is used to extract historical data corresponding to the same type of defect as the target defect from the historical monitoring database, taking the defect type contained in the monitoring results as the target defect; wherein, the historical data includes each time point in time when the same type of defect was identified and its severity.

[0158] The defect evolution rate acquisition unit is used to perform defect evolution trend analysis based on the historical data to obtain the average evolution rate of the target defect;

[0159] A defect deterioration time prediction unit is used to predict the deterioration time required for the target defect to reach its maximum severity from the current severity, based on the average evolution rate, using the severity of the target defect contained in the monitoring results as the current severity.

[0160] A defect warning triggering unit is used to trigger a corresponding defect warning mechanism based on the deterioration time and the current severity; wherein, the defect warning mechanism is divided into different warning levels based on different deterioration times and different current severity.

[0161] It should be noted that the transformer bushing defect monitoring system based on multi-dimensional data fusion provided in this embodiment of the invention can realize all the processes of the transformer bushing defect monitoring method based on multi-dimensional data fusion described in any of the above embodiments. The functions and technical effects of each module and unit in the system are the same as those of the transformer bushing defect monitoring method based on multi-dimensional data fusion described in the above embodiments, and will not be repeated here.

[0162] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for monitoring transformer bushing defects based on multi-dimensional data fusion, characterized in that, include: According to the preset acquisition frequency, pressure data sequences and temperature data sequences of multiple monitoring points inside the transformer bushing are acquired, and a multi-dimensional data matrix containing monitoring point location, acquisition time point, pressure and temperature is established based on the sliding window mechanism. The pressure change rate of each monitoring point at each collection time point is obtained based on the multidimensional data matrix, and when the pressure field reconstruction event is triggered based on the pressure change rate, the starting monitoring point and starting time point of the pressure field reconstruction are obtained. Based on the starting time point and the multidimensional data matrix, the time delay coefficient of temperature change relative to pressure change at each monitoring point is obtained, and a response time difference matrix including the monitoring point location and the time delay coefficient is established. Based on the starting monitoring point and the response time difference matrix, the temperature data sequence in the multidimensional data matrix is ​​time-series corrected to obtain a corrected data matrix. Based on the starting monitoring point and the correction data matrix, a dynamic correlation model is established between the oil level change inside the transformer bushing and the pressure and temperature, and the weight allocation scheme of pressure and temperature in the dynamic correlation model is determined. Based on the dynamic correlation model and the weight allocation scheme, multidimensional data fusion is performed on the real-time collected pressure data sequence and temperature data sequence to establish a multidimensional feature vector containing pressure, temperature and oil level changes. Based on the multidimensional feature vector, transformer bushing defects are monitored to obtain monitoring results. The monitoring results include the defect type and its severity. The process of establishing a dynamic correlation model between oil level changes and pressure and temperature inside the transformer bushing based on the initial monitoring point and the correction data matrix, and determining the weight allocation scheme for pressure and temperature in the dynamic correlation model, specifically includes: Based on the pressure data sequence and corrected temperature sequence of each monitoring point in the correction data matrix, the oil level change sequence inside the transformer bushing is calculated using fluid mechanics principles. Based on the Pearson correlation coefficient, the correlation coefficients between the oil level change sequence and the pressure data sequence at the starting monitoring point, and the corrected temperature sequence at each monitoring point, were calculated respectively. Each calculated correlation coefficient is compared with a preset correlation threshold. When any correlation coefficient is less than the correlation threshold, the weight of the pressure data sequence or corrected temperature sequence associated with the corresponding correlation coefficient is adjusted to the ratio of the corresponding correlation coefficient to the correlation threshold. Otherwise, the initial weight remains unchanged to obtain the weight allocation scheme of the pressure data sequence of the starting monitoring point and the corrected temperature sequence of each monitoring point. According to the weighting scheme, the pressure data sequence of the starting monitoring point and the corrected temperature sequence of each monitoring point are weighted and fused to obtain a comprehensive parameter sequence. The least squares method is used to establish a linear correspondence between the oil level change sequence and the comprehensive parameter sequence, so as to obtain a dynamic correlation model between the oil level change and pressure and temperature inside the transformer bushing.

2. The transformer bushing defect monitoring method based on multi-dimensional data fusion as described in claim 1, characterized in that, The process involves acquiring pressure and temperature data sequences from multiple monitoring points inside the transformer bushing at a preset acquisition frequency, and establishing a multi-dimensional data matrix containing monitoring point locations, acquisition times, pressure, and temperature based on a sliding window mechanism. Specifically, this includes: According to the preset acquisition frequency, pressure and temperature data of multiple monitoring points inside the transformer bushing are acquired, and the corresponding monitoring point locations and acquisition time points are recorded to obtain the original pressure data sequence and the original temperature data sequence. The original pressure data sequence and the original temperature data sequence are verified to obtain valid pressure data sequence and valid temperature data sequence that pass the verification; wherein, the verification includes the removal of abnormal data; The effective pressure data sequence and the effective temperature data sequence are sorted and organized according to the monitoring point location and the acquisition time point, and an initial data matrix containing the monitoring point location, acquisition time point, pressure and temperature is established based on the sliding window mechanism. The pressure and temperature data of each monitoring point in the initial data matrix at the same acquisition time point are aligned to obtain a multi-dimensional data matrix containing the monitoring point location, acquisition time point, pressure, and temperature.

3. The transformer bushing defect monitoring method based on multi-dimensional data fusion as described in claim 1, characterized in that, The step of obtaining the pressure change rate of each monitoring point at each acquisition time point based on the multidimensional data matrix, and obtaining the starting monitoring point and starting time point of pressure field reconstruction when determining the triggering of a pressure field reconstruction event based on the pressure change rate, specifically includes: The pressure change rate of each monitoring point at each collection time point is obtained based on the multidimensional data matrix. When the rate of pressure change at any monitoring point at any collection time exceeds the preset pressure fluctuation threshold, the corresponding collection time point is marked as an abnormal time point, and the corresponding monitoring point is marked as an abnormal monitoring point. When any abnormal monitoring point has multiple consecutive abnormal time points, a pressure field reconstruction event is triggered, and the corresponding abnormal monitoring point is taken as the starting monitoring point for pressure field reconstruction, and the first abnormal time point of the starting monitoring point is taken as the starting time point for pressure field reconstruction.

4. The transformer bushing defect monitoring method based on multi-dimensional data fusion as described in claim 1, characterized in that, The process of obtaining the time delay coefficient of temperature change relative to pressure change at each monitoring point based on the starting time point and the multidimensional data matrix, and establishing a response time difference matrix including the monitoring point location and the time delay coefficient, specifically includes: Extract the pressure data sequence and temperature data sequence of each monitoring point within each preset time period before and after the starting time point from the multidimensional data matrix; For the same monitoring point, the extracted temperature data sequence is gradually shifted relative to the pressure data sequence on the time axis with a preset time step. The correlation coefficient between the temperature data sequence and the pressure data sequence after each shift is calculated based on the Pearson correlation coefficient. The total shift time corresponding to the maximum value of the correlation coefficient is used as the time delay coefficient of the temperature change relative to the pressure change at the corresponding monitoring point. A response time difference matrix containing the location of each monitoring point and its time delay coefficient is established based on the time delay coefficient of all monitoring points.

5. The transformer bushing defect monitoring method based on multi-dimensional data fusion as described in claim 1, characterized in that, The step of performing time-series correction on the temperature data sequence in the multidimensional data matrix based on the starting monitoring point and the response time difference matrix to obtain a corrected data matrix specifically includes: The pressure data sequence of the starting monitoring point is extracted from the multidimensional data matrix as a reference sequence, and the temperature data sequence of each monitoring point is extracted as the temperature sequence to be corrected. Based on the time delay coefficient of each monitoring point in the response time difference matrix, the temperature sequence to be corrected at the same monitoring point is shifted and corrected by time axis to obtain the intermediate temperature sequence of each monitoring point. A dynamic time warping algorithm is used to calculate the optimal matching relationship between the intermediate temperature sequence of each monitoring point and the reference sequence at each acquisition time point. Based on the optimal matching relationship corresponding to each monitoring point, the intermediate temperature sequence of the same monitoring point is time axis aligned and corrected to obtain the corrected temperature sequence of each monitoring point. The multidimensional data matrix is ​​updated based on the corrected temperature sequence of each monitoring point to obtain the corrected data matrix.

6. The transformer bushing defect monitoring method based on multi-dimensional data fusion as described in claim 1, characterized in that, The process involves fusing real-time collected pressure and temperature data sequences using the dynamic correlation model and weight allocation scheme to establish a multi-dimensional feature vector containing pressure, temperature, and oil level changes. Transformer bushing defects are then monitored based on this multi-dimensional feature vector to obtain monitoring results. Specifically, this includes: According to the weighting scheme, the pressure data sequence of the starting monitoring point and the corrected temperature sequence of each monitoring point are weighted to obtain the weighted pressure characteristics of the starting monitoring point and the weighted temperature characteristics of each monitoring point. The weighted pressure characteristics of the starting monitoring point and the weighted temperature characteristics of each monitoring point are input into the dynamic correlation model to obtain the oil level change characteristics; A standard feature vector is established based on the weighted pressure characteristics of the starting monitoring point, the weighted temperature characteristics of each monitoring point, and the oil level change characteristics; Using the structure of the standard feature vector as a standard, multi-dimensional data fusion is performed on the real-time collected pressure data sequence and temperature data sequence to establish a multi-dimensional feature vector containing pressure, temperature and oil level changes; The multidimensional feature vector is input into the trained defect recognition model to obtain the confidence level of the multidimensional feature vector being recognized as each defect category; Based on the confidence level of the multidimensional feature vectors identified as each defect category, it is determined whether there is a defect type corresponding to the defect category in the transformer bushing. If there is, the severity of the defect type is determined according to the corresponding confidence level, so as to obtain the monitoring results based on the existing defect types and their severity.

7. The transformer bushing defect monitoring method based on multi-dimensional data fusion as described in claim 1, characterized in that, The method further includes: Based on the historical monitoring database, the defect evolution trend analysis is performed on the defect types contained in the monitoring results, and the corresponding defect early warning mechanism is triggered based on the analysis results and the monitoring results.

8. The transformer bushing defect monitoring method based on multi-dimensional data fusion as described in claim 7, characterized in that, The step of performing defect evolution trend analysis on the defect types included in the monitoring results based on the historical monitoring database, and triggering a corresponding defect early warning mechanism based on the analysis results and the monitoring results, specifically includes: Using the defect types contained in the monitoring results as target defects, historical data corresponding to the same type of defects as the target defects are extracted from the historical monitoring database; wherein, the historical data includes each time point in time when the same type of defect was identified and its severity. Based on the historical data, defect evolution trend analysis is performed to obtain the average evolution rate of the target defect; The severity of the target defect contained in the monitoring results is taken as the current severity, and the deterioration time required for the target defect to reach the maximum severity from the current severity is predicted based on the average evolution rate. The corresponding defect warning mechanism is triggered based on the deterioration time and the current severity; wherein, the defect warning mechanism is divided into different warning levels based on different deterioration times and different current severity.

9. A transformer bushing defect monitoring system based on multi-dimensional data fusion, characterized in that, include: The multidimensional data matrix building module is used to collect pressure and temperature data sequences from multiple monitoring points inside the transformer bushing according to a preset acquisition frequency, and to build a multidimensional data matrix containing monitoring point locations, acquisition time points, pressure, and temperature based on a sliding window mechanism. The pressure field reconstruction identification module is used to obtain the pressure change rate of each monitoring point at each collection time point according to the multi-dimensional data matrix, and when the pressure field reconstruction event is triggered according to the pressure change rate, it obtains the starting monitoring point and the starting time point of the pressure field reconstruction. The response time difference matrix establishment module is used to obtain the time delay coefficient of temperature change relative to pressure change at each monitoring point based on the starting time point and the multidimensional data matrix, and to establish a response time difference matrix that includes the monitoring point location and the time delay coefficient. A temperature sequence time-series correction module is used to perform time-series correction on the temperature data sequence in the multidimensional data matrix based on the starting monitoring point and the response time difference matrix to obtain a correction data matrix. The correlation model and weight acquisition module is used to establish a dynamic correlation model between the oil level change and pressure and temperature inside the transformer bushing based on the starting monitoring point and the correction data matrix, and to determine the weight allocation scheme of pressure and temperature in the dynamic correlation model. The transformer bushing defect monitoring module is used to perform multi-dimensional data fusion on real-time collected pressure data sequences and temperature data sequences according to the dynamic correlation model and the weight allocation scheme, establish a multi-dimensional feature vector containing pressure, temperature and oil level changes, and perform transformer bushing defect monitoring based on the multi-dimensional feature vector to obtain monitoring results; wherein, the monitoring results include the defect type and its severity. The association model and weight acquisition module specifically include: The oil level change sequence estimation unit is used to estimate the oil level change sequence inside the transformer bushing based on the pressure data sequence and the corrected temperature sequence of each monitoring point in the correction data matrix, using the principles of fluid mechanics. The correlation coefficient calculation unit is used to calculate the correlation coefficient between the oil level change sequence and the pressure data sequence of the starting monitoring point, and the corrected temperature sequence of each monitoring point, based on the Pearson correlation coefficient. The weight allocation scheme acquisition unit is used to compare each calculated correlation coefficient with a preset correlation threshold. When any correlation coefficient is less than the correlation threshold, the weight of the pressure data sequence or corrected temperature sequence associated with the corresponding correlation coefficient is adjusted to the ratio of the corresponding correlation coefficient to the correlation threshold. Otherwise, the initial weight remains unchanged to obtain the weight allocation scheme of the pressure data sequence of the starting monitoring point and the corrected temperature sequence of each monitoring point. The comprehensive parameter sequence acquisition unit is used to perform weighted fusion processing on the pressure data sequence of the starting monitoring point and the corrected temperature sequence of each monitoring point according to the weight allocation scheme to obtain the comprehensive parameter sequence. The dynamic correlation model acquisition unit is used to establish a linear correspondence between the oil level change sequence and the comprehensive parameter sequence using the least squares method, so as to obtain a dynamic correlation model between the oil level change and pressure and temperature inside the transformer bushing.

Images