Bushing insulation condition detection method based on multi-parameter real-time monitoring

Abstract

This invention relates to the field of bushing inspection technology and discloses a method for bushing insulation condition detection based on real-time monitoring of multiple parameters. The method includes real-time acquisition of multi-dimensional monitoring data such as partial discharge pulse sequences, dielectric loss tangent fluctuation data, and surface leakage current density distribution in the bushing operating environment; extraction of insulation defect features from this dataset to generate an insulation defect distribution map containing surface discharge trajectories, internal air gap location markers, and carbonization channel density indices; the use of a dynamic threshold segmentation algorithm to divide high-risk and potential defect regions and determine boundary coordinates, thereby constructing an insulation condition assessment vector containing parameters such as the surface discharge intensity index; inputting this vector into a pre-trained insulation aging prediction model to output an estimated remaining insulation life of the bushing and a priority ranking of defect development. This method achieves comprehensive and accurate monitoring and prediction of bushing insulation condition, providing scientific support for bushing operation and maintenance decisions.

Description

Technical Field

[0001] This invention relates to the field of bushing inspection technology, specifically a bushing insulation status detection method based on real-time monitoring of multiple parameters. Background Technology

[0002] Bushings, as key components in power equipment, are widely used in core power equipment such as transformers and circuit breakers. Their insulation performance directly affects the stable operation of the power system. With the development of power systems towards higher voltage and larger capacity, bushings operate in complex environments, facing various factors such as electrical, thermal, and mechanical stresses. This can lead to gradual degradation of their insulation performance, resulting in defects such as surface discharge, internal air gap breakdown, and expansion of carbonized channels. If these defects are not detected and addressed in a timely manner, they may lead to bushing insulation failure, causing equipment malfunctions or even large-scale power outages, resulting in severe economic losses and social impact.

[0003] Currently, bushing insulation condition detection methods mainly rely on single-parameter monitoring, such as judging the insulation condition solely by monitoring partial discharge signals or dielectric loss tangent. These methods have significant limitations: they cannot comprehensively capture the complex characteristics of bushing insulation defects; partial discharge signals are susceptible to external electromagnetic interference, making it difficult to distinguish between actual defects and interference signals relying solely on this parameter; the dielectric loss tangent only reflects the overall insulation loss and cannot accurately locate local defects; and surface leakage current monitoring is mostly limited to qualitative analysis, making it difficult to quantify the severity of defects.

[0004] Current technologies for extracting defect features are relatively simple and lack systematic, graphical representation, making it difficult for maintenance personnel to intuitively grasp the distribution patterns and development trends of defects. In defect area segmentation, fixed threshold segmentation algorithms are commonly used. However, parameters such as temperature and humidity in the bushing operating environment are dynamically changing, and fixed thresholds cannot adapt to environmental fluctuations, easily leading to misjudgments or omissions of high-risk areas. Furthermore, existing assessment systems focus primarily on the current insulation condition, lacking effective prediction of remaining insulation life and failing to rationally prioritize the development of multiple defects. This results in a lack of foresight in maintenance work and difficulty in implementing targeted preventative maintenance. These problems make existing detection methods insufficient to meet the power system's need for accurate monitoring of bushing insulation condition, necessitating a more comprehensive and efficient detection technology to overcome the shortcomings of current technologies. Summary of the Invention

[0005] The purpose of this invention is to provide a bushing insulation status detection method based on real-time monitoring of multiple parameters, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a bushing insulation condition detection method based on real-time monitoring of multiple parameters, the method comprising:

[0007] A multi-dimensional monitoring data set is collected in real time from the bushing operating environment. The multi-dimensional monitoring data set includes partial discharge pulse sequence, dielectric loss tangent fluctuation data and surface leakage current density distribution.

[0008] The insulation defect feature extraction process is performed on the multi-dimensional monitoring data set to generate a bushing insulation defect distribution map. The insulation defect distribution map includes surface discharge trajectory, internal air gap location markers and carbonization channel density index.

[0009] A dynamic threshold segmentation algorithm is used to divide the insulation defect distribution map into regions to determine the boundary coordinates between high-risk defect regions and potential defect regions.

[0010] An insulation state evaluation vector is generated based on the boundary coordinates. The insulation state evaluation vector includes the surface discharge intensity index, the air gap volume ratio, and the carbonization channel expansion trend coefficient.

[0011] The insulation condition assessment vector is input into a pre-trained insulation aging prediction model, which outputs an estimate of the remaining insulation life of the bushing and a priority ranking of defect development.

[0012] Preferably, the insulation defect feature extraction process for the multi-dimensional monitoring data set includes:

[0013] The partial discharge pulse sequence is divided into discharge event subsets according to time windows, with each discharge event subset corresponding to a pulse cluster within a power frequency cycle; for each discharge event subset, the following processing is performed:

[0014] A dielectric loss phase distribution matrix is ​​constructed based on the fluctuation data of the dielectric loss tangent value, and the dielectric loss phase distribution matrix includes the loss angle offset under different voltage phases;

[0015] By combining the spatial mapping relationship between the surface leakage current density distribution and the subset of discharge events, a three-dimensional discharge energy density field is generated, which includes the axial energy gradient, radial energy diffusion rate and tangential energy concentration.

[0016] A time-domain superposition analysis was performed on the three-dimensional discharge energy density field of multiple consecutive power frequency cycles to calculate the radius of curvature of the surface discharge trajectory, the geometric center coordinates of the internal air gap location markers, and the spatial distribution weights of the carbonization channel density index.

[0017] Preferably, the generation of the three-dimensional discharge energy density field includes:

[0018] Based on the correlation between the loss angle offset and the amplitude of the partial discharge pulse in the dielectric loss phase distribution matrix, a dielectric loss-discharge energy conversion equation is established, and the first distribution function of the axial energy gradient is obtained by solving the equation.

[0019] Based on the attenuation characteristics of the surface leakage current density distribution in the radial direction of the bushing, a radial energy diffusion calculation model is constructed. The radial energy diffusion calculation model includes dynamic correction parameters for the material dielectric constant and ambient humidity.

[0020] Combining the spatiotemporal distribution characteristics of partial discharge pulse clusters with surface conductivity, an iterative calculation process for tangential energy concentration is established. The iterative calculation process includes a feedback adjustment mechanism for pulse interval and energy accumulation rate.

[0021] The outputs of the first distribution function, the radial energy diffusion calculation model, and the iterative calculation process of tangential energy concentration are fused into a spatial mesh to generate three-dimensional field data containing axial, radial, and tangential energy distributions.

[0022] Preferably, the process of dividing the insulation defect distribution map into regions using a dynamic threshold segmentation algorithm includes:

[0023] Extract the pixel intensity extreme points of the surface discharge trajectory in the insulation defect distribution map, and generate the first dynamic segmentation threshold based on the distribution density of the extreme points;

[0024] The second dynamic segmentation threshold is calculated based on the geometric feature parameters of the internal air gap location markers, wherein the geometric feature parameters include the ratio of the equivalent diameter of the air gap to its depth.

[0025] A third dynamic segmentation threshold is generated based on the spatial gradient change rate of the carbonization channel density index.

[0026] The first dynamic segmentation threshold, the second dynamic segmentation threshold, and the third dynamic segmentation threshold are weighted and fused to generate a composite dynamic segmentation threshold curve;

[0027] The composite dynamic segmentation threshold curve is used to perform binarization segmentation on the insulation defect distribution map, and the boundary coordinate set of high-risk defect area and potential defect area is output.

[0028] Preferably, the generation of the insulation state evaluation vector includes:

[0029] Extract the contour coordinates of high-risk defect regions from the boundary coordinate set, and calculate the integral mean of the discharge intensity index along the surface within the contour.

[0030] Based on the coordinates of the air gap location in the potential defect area, calculate the ratio of the air gap volume percentage to the total casing volume.

[0031] Based on the spatial distribution differences of carbonization channel density indices in the boundary coordinate set, the time derivative of the carbonization channel expansion trend coefficient is fitted.

[0032] The surface discharge intensity index, air gap volume ratio, and carbonization channel expansion trend coefficient are combined into a multi-dimensional vector according to preset weights.

[0033] Preferably, the processing of the pre-trained insulation aging prediction model includes:

[0034] The surface discharge intensity index is input into the first aging analysis layer, and the surface insulation degradation rate is determined by the discharge cumulative damage calculation module.

[0035] Input the air gap volume ratio into the second aging analysis layer, perform air gap electric field distortion analysis, and output the internal insulation strength decay curve.

[0036] Input the carbonization channel expansion trend coefficient into the third aging analysis layer, and calculate the channel expansion acceleration based on the carbonization path growth model;

[0037] By integrating the surface insulation degradation rate, internal insulation strength decay curve, and channel expansion acceleration, a comprehensive aging index is generated.

[0038] The remaining insulation life estimate is output based on the comparison results of the comprehensive aging index and the preset critical value, and the defect development priority is ranked according to the contribution of deterioration rate, decay curve and acceleration.

[0039] Preferably, the method further includes:

[0040] Real-time monitoring of ambient temperature and humidity changes, and calculation of the compensation coefficients for the phase distribution matrix of dielectric loss due to temperature and humidity.

[0041] The axial energy gradient distribution in the three-dimensional discharge energy density field is adjusted according to the compensation coefficient;

[0042] The dynamic correction parameters in the radial energy diffusion calculation model are corrected based on the rate of change of temperature and humidity.

[0043] The corrected three-dimensional discharge energy density field is re-input into the insulation aging prediction model to generate an estimated remaining insulation lifetime after environmental compensation.

[0044] Preferably, the compensation coefficients for calculating the effect of temperature and humidity on the phase distribution matrix of dielectric loss include:

[0045] Obtain the initial dielectric constant and temperature and humidity changes of the sleeve material under reference temperature and humidity, and establish a dielectric constant-temperature and humidity correlation function.

[0046] The offset compensation amount of the dielectric loss tangent is calculated based on the dielectric constant-temperature and humidity correlation function.

[0047] The offset compensation is superimposed on the dielectric loss phase distribution matrix during the generation process, and the compensated loss angle offset distribution data is output.

[0048] Preferably, the method further includes:

[0049] Collect the actual insulation resistance value and partial discharge quantity of the bushing within the preset detection cycle;

[0050] The deviation analysis between the actual insulation resistance value and the predicted degradation rate is performed to generate the first model correction coefficient;

[0051] The partial discharge quantity is compared with the predicted discharge intensity index in the time domain to generate the correction coefficients for the second model.

[0052] The weight parameters of the insulation aging prediction model are adjusted based on the correction coefficients of the first model and the second model to generate an optimized insulation aging prediction model.

[0053] Preferably, the method further includes:

[0054] A set of maintenance strategies is generated based on the priority of defect development. The set of maintenance strategies includes partial discharge suppression schemes, air gap filling processes, and carbonization channel blocking methods.

[0055] The execution timing of maintenance strategies is dynamically adjusted based on the remaining insulation life estimate, so that the implementation nodes of maintenance strategies are synchronized with the insulation life decay stage.

[0056] Compared with the prior art, the beneficial effects of the present invention are:

[0057] By integrating multi-dimensional monitoring data, the limitations of traditional single-parameter monitoring are overcome, enabling a more comprehensive reflection of the true state of bushing insulation. The coordinated acquisition of partial discharge pulse sequences, dielectric loss tangent fluctuation data, and surface leakage current density distribution covers key information on bushing insulation defects in terms of electrical signals, loss characteristics, and current distribution. It can simultaneously capture various defect characteristics such as surface discharge, internal air gaps, and carbonization channels, avoiding defect omissions or misjudgments caused by single-parameter monitoring. This provides a rich and reliable data foundation for subsequent feature extraction and condition assessment.

[0058] In the insulation defect feature extraction stage, the generated insulation defect distribution map presents the surface discharge trajectory, internal air gap location, and carbonization channel density in an intuitive form. This allows for clear identification of the spatial distribution and severity of defects, facilitating maintenance personnel to quickly grasp the core issues of bushing insulation. This map-based presentation transforms abstract monitoring data into concrete defect information, overcoming the drawbacks of fragmented and difficult-to-integrate information in traditional feature extraction methods. It lays a clear analytical foundation for subsequent area division and condition assessment.

[0059] The application of dynamic threshold segmentation algorithm overcomes the limitations of fixed thresholds in adapting to environmental changes. It adjusts the segmentation criteria based on fluctuations in the real-time operating environment of the bushing, accurately delineating the boundary coordinates between high-risk defect areas and potential defect areas. Compared to traditional fixed threshold algorithms, this method is better suited to the complexity and dynamism of the power equipment operating environment, effectively improving the accuracy of defect area segmentation, avoiding area judgment errors caused by changes in environmental parameters, and ensuring precise identification of high-risk areas.

[0060] The construction of the insulation condition assessment vector quantitatively integrates key indicators such as surface discharge intensity, air gap volume ratio, and carbonization channel expansion trend, forming a multi-dimensional condition assessment system that can comprehensively characterize the degree of bushing insulation degradation from different perspectives. This multi-indicator assessment method avoids the one-sidedness of single-indicator assessment, making the judgment of insulation condition more objective and comprehensive, and more accurately reflecting the overall performance and potential risks of bushing insulation.

[0061] The pre-trained insulation aging prediction model, by receiving insulation condition assessment vectors, can output remaining insulation lifetime estimates and defect development priority rankings, achieving full-process coverage from insulation condition monitoring to lifetime prediction. Trained on a large amount of historical data, this model possesses strong predictive accuracy and practicality, providing forward-looking decision-making support for operation and maintenance work. It helps maintenance personnel plan maintenance strategies in advance, rationally allocate maintenance resources, and take timely preventative measures against high-priority defects, reducing the probability of insulation failure leading to faults, extending bushing service life, and ensuring the stable and safe operation of the power system. Attached Figure Description

[0062] Figure 1 This is a diagram showing the insulation assessment status results;

[0063] Figure 2 A flowchart for the process of extracting features of insulation defects;

[0064] Figure 3 Flowchart for generating insulation condition assessment vectors;

[0065] Figure 4 This is a trend chart of the change in the insulation condition assessment vector. Detailed Implementation

[0066] 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.

[0067] Please see Figure 1 This invention provides a bushing insulation condition detection method based on multi-parameter real-time monitoring. The method includes: a multi-dimensional monitoring data set including partial discharge pulse sequences, dielectric loss tangent fluctuation data, and surface leakage current density distribution. These data are synchronously acquired through a high-precision sensor array to ensure real-time performance and integrity. The acquired data undergoes insulation defect feature extraction processing to generate a bushing insulation defect distribution map. This map includes surface discharge trajectories, internal air gap location markers, and carbonization channel density indices, thus intuitively reflecting the spatial distribution characteristics of insulation defects. Next, a dynamic threshold segmentation algorithm is used to divide the insulation defect distribution map into regions. The boundary coordinates between high-risk defect areas and potential defect areas are determined by calculating pixel intensity extreme points and geometric feature parameters. Based on the boundary coordinates, an insulation condition assessment vector is generated. This vector includes the surface discharge intensity index, air gap volume ratio, and carbonization channel expansion trend coefficient. These parameters quantify key indicators of insulation degradation. Finally, the insulation condition assessment vector is input into a pre-trained insulation aging prediction model. This model analyzes the correlation between vector components to output an estimated remaining insulation life of the bushing and a defect development priority ranking, providing data support for maintenance decisions.

[0068] Example 1: See Figure 2The multi-dimensional monitoring dataset includes partial discharge pulse sequences, dielectric loss tangent fluctuation data, and surface leakage current density distribution. The partial discharge pulse sequences are divided into discharge event subsets according to time windows. Each discharge event subset corresponds to a pulse cluster within a power frequency cycle. The partitioning process uses a fixed-duration sliding window algorithm, with the window width strictly synchronized with the power grid cycle to ensure complete discharge phase information within each window. For each discharge event subset, a dielectric loss phase distribution matrix is ​​constructed. The matrix generation is based on dielectric loss tangent fluctuation data. A phase analysis algorithm is used to extract the loss angle offset under different voltage phases. This algorithm uses phase-locked loop technology to track the voltage zero-crossing point, subdividing each power frequency cycle into multiple phase intervals and calculating the standard deviation of the loss tangent in each interval as the offset index. A three-dimensional discharge energy density field is generated by combining the spatial mapping relationship between the surface leakage current density distribution and the discharge event subsets. The spatial mapping relationship is established through coordinate transformation, dividing the bushing surface into grid cells. Each cell is associated with the corresponding leakage current measurement value and discharge pulse data. The three-dimensional discharge energy density field comprises three dimensions: axial energy gradient, radial energy diffusion rate, and tangential energy concentration. The axial energy gradient describes the energy distribution along the bushing length, the radial energy diffusion rate reflects the energy attenuation characteristics along the insulation layer thickness, and the tangential energy concentration characterizes the degree of energy concentration along the circumferential direction of the surface. To generate the three-dimensional discharge energy density field, a dielectric loss-discharge energy conversion equation is first established based on the correlation between the loss angle offset and the partial discharge pulse amplitude in the dielectric loss phase distribution matrix. The equation is in the form of a multiple regression model. The weighting coefficients of the offset and pulse amplitude are fitted using the least squares method, and the equation is solved to obtain the first distribution function of the axial energy gradient. This first distribution function takes the bushing axial position as the independent variable and outputs the energy gradient value at each point.

[0069] A radial energy diffusion calculation model is constructed based on the attenuation characteristics of the surface leakage current density distribution in the radial direction of the bushing. The attenuation characteristics are simulated using an exponential function. The model inputs include the material dielectric constant and dynamic correction parameters for ambient humidity. The dielectric constant is taken from a material database, and the ambient humidity is read from a real-time sensor. The dynamic correction parameters are updated every five minutes using an adaptive filtering algorithm. The radial energy diffusion calculation model solves for the radial propagation speed of energy and outputs a radial energy diffusion rate distribution map. An iterative calculation process for tangential energy concentration is established by combining the spatiotemporal distribution characteristics of partial discharge pulse clusters with surface conductivity. The spatiotemporal distribution characteristics are reconstructed using pulse timestamps and sensor position coordinates, and the surface conductivity is provided by the impedance measurement module. The iterative calculation process includes a feedback adjustment mechanism for pulse interval and energy accumulation rate. Each iteration adjusts the accumulation rate based on the previously calculated energy value. Feedback adjustment is triggered when the pulse interval is less than a threshold, and finally, a stable solution for tangential energy concentration is output. The outputs of the first distribution function, the radial energy diffusion calculation model, and the iterative calculation process of tangential energy concentration are spatially meshed. Mesh fusion employs a three-dimensional interpolation algorithm to unify data from different coordinate systems into a cylindrical coordinate system with the casing center as the origin, generating three-dimensional field data containing axial, radial, and tangential energy distributions. This data is stored as a multi-layer matrix structure. Temporal overlay analysis is performed on the three-dimensional discharge energy density field over multiple consecutive power frequency cycles. The overlay analysis uses a moving average method to calculate the average energy of each grid point over the time series, thereby deriving the radius of curvature of the surface discharge trajectory. The radius of curvature is calculated by fitting the geometric feature points of the discharge path. The geometric center coordinates of the internal air gap location markers are obtained by clustering algorithms to identify energy anomaly regions and then calculating the centroid. The spatial distribution weights of the carbonization channel density index are calculated based on the gradient change of the energy field, with the weight values ​​proportional to the gradient magnitude.

[0070] The update frequency of the dielectric loss phase distribution matrix is ​​synchronized with the subset division of discharge events. One frame of matrix data is generated for each power frequency cycle, and the matrix dimension is consistent with the number of phase intervals. Preprocessing of the partial discharge pulse sequence includes noise reduction and pulse classification. Noise reduction uses wavelet transform to filter out background interference, and pulse classification clusters pulses into different discharge types based on the amplitude-phase distribution map. The surface leakage current density distribution is measured using a ring electrode array. The electrode spacing is optimized according to the bushing size, and the density value is normalized by dividing the current value by the electrode area. Visualization of the three-dimensional discharge energy density field uses isosurface rendering technology, and the energy threshold is adjustable to highlight defect areas. The window length for time-domain overlay analysis is set to ten minutes, corresponding to approximately 300 power frequency cycles, ensuring statistical significance. The radius of curvature calculation of the surface discharge trajectory uses least squares fitting of the arc. The geometric center coordinates of the internal air gap location markers are accurate to the millimeter level. The spatial distribution weights of the carbonization channel density index are normalized to the 0-1 range for comparing the severity of different regions. The partitioning of discharge event subsets also considers pulse density factors. When the number of pulses per unit time exceeds a threshold, the time window is automatically narrowed to improve resolution. A temperature compensation coefficient is introduced to generate the dielectric loss phase distribution matrix, obtained by looking up the material's temperature characteristic curve. The axial energy gradient of the three-dimensional discharge energy density field is calculated using the central difference method. The radial energy diffusion rate model is verified by comparing simulated values ​​with measured leakage current distributions. The convergence condition for the iterative calculation of tangential energy concentration is set to an error of less than one percent for three consecutive iterations. The spatial grid fusion process checks data integrity, and missing values ​​are supplemented by interpolation of neighboring points. The results of the time-domain overlay analysis are stored in a time-series database, supporting backtracking queries of energy field changes over any time period. The radius of curvature of the surface discharge trajectory is compared with a preset safety range; an alarm is triggered when the range is exceeded. The geometric center coordinates of the internal air gap location marker are overlaid with the bushing design drawings. The spatial distribution weight of the carbonization channel density index is used to generate a thermal map to guide maintenance of key areas.

[0071] Partial discharge pulse sequences are acquired using high-frequency current transformers with a sampling rate of at least 10 MHz to ensure pulse waveform integrity. Fluctuations in the dielectric loss tangent are measured using a high-voltage bridge with an accuracy of 0.0001%. Surface leakage current density distribution is monitored using distributed fiber optic sensors with a spatial resolution better than one centimeter. The algorithm for partitioning discharge event subsets runs in real-time on an embedded processor, with processing latency controlled to the millisecond level. The construction of the dielectric loss phase distribution matrix utilizes parallel computation via a graphics processor to accelerate matrix operations. The generation cycle of the three-dimensional discharge energy density field is consistent with the power frequency cycle, updating one frame of data every 20 milliseconds. Time-domain overlay analysis is performed on a dedicated data analysis server with sufficient memory for large-scale matrix operations. The calculated radius of curvature of the surface discharge trajectory is automatically annotated on the bushing's three-dimensional model, and the geometric center coordinates of the internal air gap location are output in engineering coordinate format. The spatial distribution weights of the carbonization channel density index are integrated into the risk assessment report. The overall workflow for insulation defect feature extraction is designed as a pipeline architecture, with the output of each step directly serving as the input for the next step, reducing data transfer latency. Partial discharge pulse sequences, dielectric loss tangent fluctuation data, and surface leakage current density distribution are time-stamped and synchronized with microsecond-level accuracy. The three-dimensional discharge energy density field is stored using a layered compression format to balance data accuracy and storage space. Time-domain overlay analysis supports manual selection of time intervals for targeted analysis. The radius of curvature calculation for surface discharge trajectories is compatible with both straight-line and curved-line modes. The calculation of the geometric center coordinates of internal air gap location markers eliminates false defects caused by edge effects. The spatial distribution weight calculation of carbonization channel density indices incorporates a distance attenuation factor to reflect the proximity effect of channel expansion. The partitioning of discharge event subsets supports an overlapping window mode, with configurable overlap rates to improve temporal resolution. The generation of the dielectric loss phase distribution matrix includes outlier removal logic to prevent matrix distortion caused by measurement abrupt changes. The axial, radial, and tangential components of the three-dimensional discharge energy density field can be independently accessed for specialized analysis. Time-domain overlay analysis provides trend prediction capabilities, forecasting future changes in the energy field based on historical data.

[0072] Example 2: See Figure 3The insulation defect distribution map includes surface discharge trajectories, internal air gap location markers, and carbonization channel density indices. The map data is stored in a two-dimensional matrix format, with matrix element values ​​corresponding to the quantified intensity of defect features. After the dynamic threshold segmentation algorithm is activated, it extracts the pixel intensity extreme points of the surface discharge trajectories in the insulation defect distribution map. Extreme point detection uses a neighborhood comparison method, comparing intensity values ​​within the eight connected regions of each pixel to identify local maximum points. A first dynamic segmentation threshold is generated based on the extreme point distribution density, which is calculated as the number of extreme points per unit area. The threshold value is adjusted by the ratio of the density value to a preset benchmark value. A second dynamic segmentation threshold is calculated based on the geometric feature parameters of the internal air gap location markers. These geometric feature parameters include the air gap equivalent diameter to depth ratio. The equivalent diameter is calculated by treating the air gap area as an equivalent circle, and the depth ratio is determined by the ratio of the deepest point to the average depth in the three-dimensional air gap model. The second dynamic segmentation threshold is negatively correlated with the depth ratio; that is, the larger the depth ratio, the lower the threshold is set to capture potential risks. The third dynamic segmentation threshold is generated based on the spatial gradient change rate of the carbonization channel density index. The spatial gradient change rate is calculated by the Sobel operator to calculate the gradient magnitude of the carbonization density map, and the statistical average of the gradient magnitude is taken as the change rate index. The third dynamic segmentation threshold is positively correlated with the gradient change rate, and high gradient regions correspond to higher segmentation thresholds.

[0073] The first, second, and third dynamic segmentation thresholds are weighted and fused. The weighting coefficients are determined based on statistical analysis of historical defect data. The weight for surface discharge-related defects is set to 0.5, the weight for internal air gap defects is set to 0.3, and the weight for carbonization channel defects is set to 0.2, generating a composite dynamic segmentation threshold curve. The curve is a piecewise linear function, with the horizontal axis representing the defect feature intensity value and the vertical axis representing the segmentation threshold. The composite dynamic segmentation threshold curve is used to perform binarization segmentation of the insulation defect distribution map. Each pixel in the map is traversed. Points with intensity values ​​higher than the local threshold are classified as high-risk defect areas, and points with intensity values ​​lower than the local threshold but higher than the baseline threshold are classified as potential defect areas. The set of boundary coordinates between high-risk and potential defect areas is output. The boundary coordinates are extracted using an edge detection algorithm and stored as a polygon vertex sequence. When generating the insulation condition assessment vector, the contour coordinates of high-risk defect areas are extracted from the boundary coordinate set. These contour coordinates are obtained using a chain code tracking algorithm. The integral mean of the surface discharge intensity index within the contour is calculated. The integral calculation uses the Gaussian quadrature method to perform an area integral on the discharge intensity value within the contour region. The mean is calculated as the integral value divided by the area of ​​the contour region. Based on the air gap location marker coordinates of potential defect areas, the ratio of air gap volume percentage to the total bushing volume is calculated. The air gap location marker coordinates are mapped to actual spatial locations through 3D coordinate transformation. The volume of each air gap is obtained by accumulating the element volumes after finite element mesh generation. The air gap volume percentage is the ratio of the total air gap volume to the bushing insulation material volume.

[0074] The time derivative of the carbonization channel expansion trend coefficient is fitted based on the spatial distribution difference of the carbonization channel density index in the boundary coordinate set. The spatial distribution difference is calculated as the standard deviation of the density values ​​between high-risk and potential-risk areas. The time derivative is calculated by the density value difference over multiple consecutive detection cycles, and the central difference method is used to improve accuracy. The surface discharge intensity index, air gap volume ratio, and carbonization channel expansion trend coefficient are combined into a multi-dimensional vector according to preset weights. The weight allocation scheme is as follows: surface discharge intensity index weight 0.4, air gap volume ratio weight 0.35, and carbonization channel expansion trend coefficient weight 0.25. The vector format is a three-dimensional row vector, supporting matrix operations. The execution frequency of the dynamic threshold segmentation algorithm is set to once per minute, synchronized with the data acquisition cycle. The pixel resolution of the insulation defect distribution map is 1024×1024, corresponding to centimeter-level accuracy on the actual bushing surface. The extreme point distribution density is calculated using the sliding window method, with a window size of 50×50 pixels. The calculation of the air gap equivalent diameter to depth ratio requires three-dimensional point cloud data, which is acquired by a laser scanner. The spatial gradient rate of change of the carbonization channel density index is calculated using a 3×3 convolution kernel, with the kernel coefficients set according to a Gaussian distribution. A smoothing process is introduced into the generation of the composite dynamic segmentation threshold curve, employing a five-point cubic smoothing algorithm to eliminate threshold jumps.

[0075] The boundary coordinate set of high-risk defect areas and potential defect areas is stored in GeoJSON format, containing polygon ring information and attribute tables. The contour coordinate extraction algorithm uses area filtering to ignore noise areas smaller than 10 square pixels. The integral calculation of the surface discharge intensity index uses an adaptive step-size algorithm, automatically refining integration points in areas with drastic intensity changes. Air gap volume calculation uses tetrahedral mesh discretization, with the mesh size dynamically adjusted according to the air gap size. The time derivative calculation of the carbonization channel expansion trend coefficient requires data from at least five consecutive time points, with a five-minute interval. The normalization processing of the insulation state assessment vector maps each component to the 0-100 range, facilitating comparison of parameters with different dimensions. Vector data is updated to the database in real time, retaining one year of historical records for trend analysis. The parameters of the dynamic threshold segmentation algorithm support remote configuration, allowing sensitivity adjustment based on field conditions. Topological checking of the boundary coordinate set excludes self-intersecting polygons, ensuring the rationality of area division. The calculation of the surface discharge intensity index incorporates a temperature correction coefficient, obtained by looking up a table based on the material temperature rise characteristic curve. The calculation of the air gap volume ratio takes into account the difference in insulation thickness at different parts of the bushing, with thickness data imported from the design drawings. The fitting of the carbonization channel expansion trend coefficient uses quadratic polynomial regression, with a goodness of regression requirement of 0.9 or higher.

[0076] The application of the composite dynamic segmentation threshold curve employs a pixel-by-pixel comparison method, where each pixel is interpolated on the curve based on its spatial location to obtain the local threshold. High-risk defect areas are marked with a red semi-transparent overlay, while potential defect areas are marked with a yellow semi-transparent overlay, displayed in layers on the visualization interface. The insulation condition assessment vector is transmitted using a lightweight JSON format, including timestamps and device identifiers. An anomaly detection mechanism for vector data monitors abrupt changes in component values, triggering a review process when a change exceeds 20%. Performance optimization of the dynamic threshold segmentation algorithm utilizes multi-threading technology, processing images in blocks in parallel. The compressed storage of the boundary coordinate set employs a run-length encoding algorithm to reduce storage space usage. The integral calculation of the surface discharge intensity index skips signal-deficient regions, which are supplemented by interpolation of neighboring points. The calculation of the air gap volume ratio is periodically cross-validated with X-ray inspection results, triggering a calibration process when the error exceeds 5%. The calculation of the time derivative of the carbonization channel expansion trend coefficient uses a robust estimation method to eliminate gross error points. The persistent storage of the insulation condition assessment vector establishes a time-series database, supporting rapid retrieval by time range. The evaluation metrics for the dynamic threshold segmentation algorithm include segmentation accuracy and recall. Accuracy, tested with manually labeled samples, exceeds 95%. The spatial index of the boundary coordinate set uses an R-tree structure to accelerate region lookup. Visualization of the insulation condition assessment vector provides both radar charts and parallel coordinate plots for easy multi-parameter comparative analysis. The output of the region segmentation processing is in a standardized GIS layer format, which can be overlaid with the plant map. The interface for the insulation condition assessment vector supports RESTful API calls, facilitating integration with other systems. The dynamic threshold segmentation algorithm's log records detailed results for each processing step, used for fault diagnosis. Boundary coordinates of high-risk defect areas are automatically pushed to the inspection system, generating a list of key inspection areas. Monitoring data for potential defect areas are entered into a long-term trend tracking queue, generating periodic analysis reports. Alarm thresholds are set for the component values ​​of the insulation condition assessment vector, with threshold levels categorized into warning, alert, and critical.

[0077] See Figure 4 The figure, divided into upper and lower subplots, illustrates the changing trends of core indicators in the bushing insulation condition assessment vector over time. The upper subplot shows the changing trends of the surface discharge intensity index and the air gap volume ratio. The surface discharge intensity index generally shows an upward trend, reflecting the cumulative process of surface insulation degradation. The air gap volume ratio gradually increases, reflecting the development degree of internal air gap defects. The lower subplot shows the changes in the carbonization channel expansion trend coefficient, which continuously increases over time, indicating that the expansion rate of carbonization channels is accelerating. These indicators quantify the degradation state of bushing insulation from different dimensions, providing multi-dimensional status basis for subsequent insulation aging prediction and maintenance strategy formulation.

[0078] Example 3: The insulation condition assessment vector includes three components: surface discharge intensity index, air gap volume ratio, and carbonization channel expansion trend coefficient. The pre-trained insulation aging prediction model adopts a hierarchical processing architecture. The model structure includes three independent aging analysis layers and a comprehensive fusion module. Each aging analysis layer specifically processes data for one type of defect. The surface discharge intensity index is input into the first aging analysis layer. The first aging analysis layer contains a discharge cumulative damage calculation module. This module is built based on damage mechanics theory and converts time-varying discharge intensity data into the degradation rate of the material's insulation performance. The calculation of the discharge cumulative damage calculation module follows the following mathematical relationship: ;in: This indicates the rate of surface insulation degradation, expressed as a percentage of decay per day. The partial discharge intensity index represents the real-time collected data, and the unit is picocous. This is the material's reference discharge withstand strength, also measured in picocoos. It is a discharge aging acceleration factor, which is a dimensionless power-law exponent. It is a material-related aging rate constant, measured in units of percentage decay per day. This represents the aging activation energy of insulating materials, measured in electron volts. It is the Boltzmann constant; This is the absolute temperature of the insulating material, measured in Kelvin. The dimensions on both sides of the formula remain consistent: the degradation rate on the left is measured as a percentage per day, and the combined values ​​on the right are also measured as a percentage per day.

[0079] The air gap volume ratio is input into the second aging analysis layer, which performs air gap electric field distortion analysis. The analysis process calculates the non-uniformity of the electric field distribution around the air gap defect based on electrostatic field theory. The second aging analysis layer outputs the internal insulation strength decay curve. The curve is generated using a finite element numerical calculation method, obtaining the maximum electric field intensity at the air gap edge by solving the Poisson equation. The internal insulation strength decay curve uses the air gap volume ratio as the independent variable, outputting the decay trajectory of the insulation strength retention percentage. The carbonization channel expansion trend coefficient is input into the third aging analysis layer. The third aging analysis layer calculates the channel expansion acceleration based on a carbonization path growth model. The growth model describes the carbonization process as a fractal growth process, and the expansion acceleration calculation involves the time differential operation of the trend coefficient. A comprehensive aging index is generated by fusing the surface insulation degradation rate, the internal insulation strength decay curve, and the channel expansion acceleration. The fusion algorithm uses a weighted geometric average method, with weight coefficients obtained through training with historical aging data. The comparison between the comprehensive aging index and the preset critical value uses a multi-level threshold judgment logic, with thresholds set according to the insulation material life termination standard. Based on the comparison results, the remaining insulation life estimate is output. The life estimate calculation is based on the time integral operation of the aging index reaching the critical value, and the integration process takes into account the nonlinear characteristics of the aging rate. The defect development priority is sorted according to the contribution of deterioration rate, decay curve and acceleration. The contribution is calculated using the variance analysis method to analyze the sensitivity of each input parameter to the comprehensive aging index.

[0080] The training dataset for the pre-trained insulation aging prediction model comes from long-term operational monitoring records of multiple bushings, covering the entire aging cycle. Model parameter optimization employs an adaptive moment estimation algorithm, with the learning rate dynamically adjusted between 0.0001 and 0.01. Model validation uses hold-out validation, dividing the dataset into training and test sets in a 7:3 ratio. The online update mechanism for the insulation aging prediction model is set to incremental training weekly, with new data added to the training set after quality checks. Model inference is executed on an embedded edge computing device, whose computing power meets real-time requirements. The material parameters for the discharge cumulative damage calculation module are derived from standard accelerated aging tests, with test conditions conforming to relevant International Electrotechnical Commission (IEC) standards. Adaptive mesh refinement technology is used for the air gap electric field distortion analysis, automatically refining mesh nodes in regions with large electric field gradients. The fractal dimension of the carbonization path growth model is determined through scanning electron microscopy image analysis, achieving nanometer-level image resolution. The scaling transformation of the comprehensive aging index maps parameters of different dimensions to a standard normal distribution space, facilitating multi-parameter fusion processing. The threshold classification of the multi-level threshold judgment logic is based on the breakdown probability distribution of insulating materials and is divided into two levels: early warning threshold and alarm threshold.

[0081] The output format of the remaining insulation lifetime estimate includes point estimates and interval estimates, with the interval estimates generated using the Monte Carlo simulation method. The results of defect development priority ranking are displayed as a weighted score list, with scores calculated based on both current severity and development speed. Input data preprocessing for the insulation aging prediction model includes data cleaning and feature standardization; anomaly identification utilizes the isolated forest algorithm. A correlation mapping is established between the model output results and periodically offline test data, with the mapping error serving as the basis for model calibration. The real-time data stream processing of the discharge cumulative damage calculation module employs a sliding window mechanism, with the window size dynamically adjusted according to discharge activity characteristics. The material parameter database for air gap electric field distortion analysis includes temperature-dielectric constant characteristic curves, obtained through dielectric spectrum analysis. Monitoring data of the carbonization channel expansion trend coefficient is fused with thermal imaging data to improve the accuracy of trend prediction. Historical data storage for comprehensive aging indicators uses a time-series database, with data compression algorithms preserving key trend characteristics. Threshold parameters for the multi-level threshold judgment logic support remote configuration, with complete audit logs recording configuration changes. The remaining insulation lifetime estimate display interface supports multi-timescale displays, including short-term trend and long-term prediction views.

[0082] The algorithm for prioritizing defect development requires transparency by providing interpretable ranking criteria, with each ranking result accompanied by explanations of key influencing factors. The interface protocol of the insulation aging prediction model adopts industry-standard communication protocols, supporting seamless integration with existing monitoring systems. Status monitoring during model calculation includes processing latency and resource utilization monitoring, with alarms triggered by abnormal states. The cumulative period of the discharge cumulative damage calculation module is configurable to adapt to the differences in bushings at different voltage levels. Boundary condition settings for air gap electric field distortion analysis consider the electromagnetic interference effects of adjacent phases, employing a multi-conductor system model. The stochastic process simulation of the carbonization path growth model introduces the Markov chain Monte Carlo method to improve the model's ability to characterize uncertainties. The storage of the calculation results of the comprehensive aging index adopts a multi-copy mechanism to ensure data security and reliability. The judgment results of the multi-level threshold judgment logic generate detailed event records, including key intermediate variables of the judgment process. The prediction results of the remaining insulation life estimate generate standardized health status reports, with report formats conforming to industry standards. The update triggering conditions for the defect development priority ranking include both time-triggered and event-triggered modes, with adjustable event trigger sensitivity. The insulation aging prediction model employs containerization technology to isolate its operating environment, ensuring model independence. A model version rollback mechanism supports rapid restoration to historical stable versions. The damage accumulation algorithm in the discharge cumulative damage calculation module supports pause and reset functions, adapting to equipment downtime conditions. Parallel computation tasks for air gap electric field distortion analysis are distributed to multiple computing nodes, with automatic load balancing among nodes. The parameter calibration process for the carbonization path growth model provides convergence diagnostics to ensure the reliability of parameter estimates. A data quality check is added to the calculation process of the comprehensive aging index, triggering a data repair process for low-quality data. The judgment rules for the multi-level threshold judgment logic support graphical configuration, reducing configuration complexity. The output interface for remaining insulation lifetime estimation supports multiple data exchange protocols, including MQTT and OPCUA. The result push for defect development priority ranking supports multi-target address configuration, ensuring timely information delivery.

[0083] Example 4: Real-time monitoring of ambient temperature and humidity changes is achieved through temperature and humidity sensors installed near the bushing. The sensor's measurement range covers a temperature range of -40°C to 80°C and a relative humidity range of 10% to 98%. Calculating the compensation coefficients for the temperature and humidity on the dielectric loss phase distribution matrix requires obtaining the initial dielectric constant and temperature and humidity changes of the bushing material under reference temperature and humidity conditions. The reference temperature and humidity conditions are set at 20°C and 50% relative humidity, and the initial dielectric constant value is obtained from the material certification report. A polynomial regression model is used to establish the dielectric constant-temperature and humidity correlation function. The model parameters are determined through laboratory temperature and humidity cycling tests, simulating the on-site operating environment. The offset compensation amount is calculated based on the dielectric constant-temperature and humidity correlation function, and is the relative difference between the dielectric constant at the current temperature and humidity and the reference dielectric constant. The offset compensation amount is superimposed on the dielectric loss phase distribution matrix during its generation process. The superposition operation is performed at the matrix element level, with each phase interval applying an independent compensation coefficient. The compensated loss angle offset distribution data is output, maintaining the same two-dimensional array structure as the original matrix.

[0084] The axial energy gradient distribution in the three-dimensional discharge energy density field is adjusted based on the compensation coefficient, and the material parameters in the energy conversion equation are recalculated during the adjustment process. The dynamic correction parameters in the radial energy diffusion calculation model are corrected based on the temperature and humidity change rate, which is obtained by differentially analyzing data from the three most recent sampling points. The corrected three-dimensional discharge energy density field is then re-input into the insulation aging prediction model, which receives input data with environmental compensation labels. An estimated remaining insulation lifetime after environmental compensation is generated, with the environmental conditions labeled as reference information in the estimation results.

[0085] The temperature and humidity sensors are arranged according to the principle of equal spacing, with sensors installed in four directions around the bushing. The update cycle of the dielectric constant-temperature and humidity correlation function is set to once a year, taking into account material aging factors. The calculation of offset compensation incorporates a humidity influence factor, the value of which is determined based on the hygroscopic characteristics of the insulation material. The compensation calculation of the dielectric loss phase distribution matrix is ​​executed in parallel on the graphics processor to improve computational efficiency. The compensated loss angle offset distribution data is stored as a floating-point matrix, with precision retained to six decimal places. The adjustment algorithm for the axial energy gradient distribution adopts a point-by-point correction method, with the correction amount calculated independently for each grid point. The dynamic correction parameter update frequency of the radial energy diffusion calculation model is synchronized with the temperature and humidity acquisition, updating once per minute. Timestamp verification is added to the data stream re-entered into the insulation aging prediction model to ensure data time sequence consistency. The remaining insulation life estimate after environmental compensation is displayed on the monitoring interface, highlighting the degree of compensation impact. Quality control of temperature and humidity monitoring data includes sensor self-testing and data rationality checks, with abnormal data automatically removed. The establishment process of the dielectric constant-temperature and humidity correlation function includes a cross-validation step, with the verification mean square error less than 5%. The application range of the offset compensation amount is set with upper and lower limits to prevent over-compensation from causing data distortion. The compensation calculation log for the dielectric loss phase distribution matrix records the intermediate results of each compensation step. The compensation-reduced loss angle offset distribution data provides a before-and-after comparison view for easy analysis of the compensation effect. The adjustment amount of the axial energy gradient distribution is smoothed to avoid drastic changes in gradient values. The parameter correction of the radial energy diffusion calculation model considers the coupling effect of temperature and humidity, employing a multiple regression model. The data re-entered into the model is standardized to maintain the same distribution as the training data. A difference report is generated for the remaining insulation lifetime estimate after environmental compensation, comparing the numerical changes before and after compensation. Refer to Table 1 for the temperature and humidity correction coefficients of the bushing material's dielectric constant.

[0086] Table 1: Dielectric Constant and Temperature / Humidity Correction Coefficient for Sleeve Materials The temperature and humidity sensor is calibrated every three months, with calibration standards transferred to national benchmarks. The dielectric constant-temperature-humidity correlation function is established using least squares fitting, with a goodness of fit requirement of 0.95 or higher. The calculation of offset compensation incorporates the influence of temperature gradients, considering the temperature difference effect at different parts of the bushing. Timeout protection is implemented for the compensation calculation of the dielectric loss phase distribution matrix, triggering degradation processing upon timeout. The compensated loss angle offset distribution data is provided via an application programming interface for other systems. The algorithm for adjusting the axial energy gradient distribution is validated by comparing simulated and measured values. The parameter correction of the radial energy diffusion calculation model records version history and supports parameter rollback. Data re-entered into the insulation aging prediction model is marked with quality indicators, limiting the influence weight of low-quality data. An uncertainty analysis report is generated for the remaining insulation life estimate after environmental compensation, including confidence interval calculations. Time-division multiplexing technology is used to store temperature and humidity monitoring data, improving data transmission efficiency. Stepwise regression is used to select variables for the dielectric constant-temperature-humidity correlation function, eliminating insignificant variables. The application timing of offset compensation depends on the magnitude of temperature and humidity changes; compensation calculations are skipped when changes are small. The computational resources for compensating the dielectric loss phase distribution matrix are dynamically adjusted, allocating more computing resources during peak periods. Incremental backup is used for the backup strategy of the compensated loss angle offset distribution data to reduce storage space usage. Consistency checks are performed on the adjusted results of the axial energy gradient distribution to remove obvious outliers. Limits are set on the parameter correction amplitude of the radial energy diffusion calculation model to prevent parameter oscillations. Flow control is added to the data stream re-entered into the model to avoid model overload. A trend correlation is established between the environmentally compensated remaining insulation lifetime estimate and historical data to identify abnormal changes.

[0087] Fault diagnosis for temperature and humidity sensors includes communication interruption detection and measurement drift detection. The applicable range of the dielectric constant-temperature and humidity correlation function is clearly marked, and warning messages are displayed for data outside the range. The calculation of offset compensation supports a manual correction mode, allowing for expert intervention. The progress of the compensation calculation of the dielectric loss phase distribution matrix is ​​displayed in real time, improving operational transparency. The compensated loss angle offset distribution data provides data export functionality, with export formats including CSV and MAT files. The axial energy gradient distribution adjustment algorithm offers multiple adjustment strategies, which can be selected based on operating conditions. The parameter correction results of the radial energy diffusion calculation model generate a correction report, which includes the basis for correction and effect evaluation. Data re-entered into the model is accompanied by data source tracking information, facilitating problem tracing. The estimated remaining insulation lifetime after environmental compensation generates a trend prediction chart, showing the trend over a future period. The temperature and humidity monitoring data is displayed using a multi-dimensional view, including real-time curves and historical comparison views. The process of establishing the dielectric constant-temperature and humidity correlation function provides a visualization tool, which displays the fitted curve and residual distribution. The calculation process of offset compensation is logged in detail, including input and output parameter records. Error analysis is performed on the compensation calculation results of the dielectric loss phase distribution matrix, and the analysis report helps improve the compensation algorithm. Access permissions for the compensated loss angle offset distribution data are managed hierarchically, with different roles having different operating permissions. The adjustment values ​​for the axial energy gradient distribution are subject to statistical process control; exceeding limits triggers an investigation process. The parameter correction interface for the radial energy diffusion calculation model provides parameter influence analysis, displaying the sensitivity of each parameter to the results. Data re-entered into the model undergoes integrity verification; data failing verification is processed in an anomaly handling procedure.

[0088] Example 5: The actual insulation resistance and partial discharge of the bushing are collected within a preset detection period. The preset detection period is set according to the equipment voltage level and operating history, typically ranging from three to twelve months. The actual insulation resistance is measured using a megohmmeter at rated DC voltage. Measurement points include the main insulation between the bushing's center guide rod and the grounding flange, as well as the creepage distance on the bushing's surface. Partial discharge is detected using a measurement circuit composed of a coupling capacitor and a detection impedance. The pulse current signal is processed by a bandpass filter with a bandwidth of 100kHz to 1MHz. Deviation analysis is performed between the actual insulation resistance and the predicted degradation rate. The predicted degradation rate is derived from the output of the insulation aging prediction model. The deviation analysis calculates the percentage relative error between the measured and predicted values. A first model correction coefficient is generated, calculated using an error backpropagation algorithm. The coefficient's magnitude is positively correlated with the deviation amplitude. The partial discharge and the predicted discharge intensity index are compared in the time domain. The time domain comparison uses a time series alignment algorithm, comparing the measured discharge and the predicted value point-to-point at the same time scale. Generate the second model correction coefficients. The second model correction coefficients are calculated based on the root mean square error of the comparison results. The larger the error, the higher the value of the correction coefficient.

[0089] The weight parameters of the insulation aging prediction model are adjusted based on the correction coefficients of the first and second models. The weight parameter adjustment employs a gradient descent optimization method with a learning rate set to 0.001. An optimized insulation aging prediction model is generated and stored as a new version file, with the version number including a timestamp. A set of maintenance strategies is generated based on defect development priority, which is a key output of the model. The maintenance strategy set includes partial discharge suppression schemes, air gap filling processes, and carbonization channel blocking methods, each corresponding to a defect type and treatment standard. The execution sequence of the maintenance strategies is dynamically adjusted based on the remaining insulation life estimate, expressed in monthly units. The execution sequence of the maintenance strategies considers equipment downtime windows and resource allocation cycles.

[0090] The actual insulation resistance measurement process follows standard operating procedures, and the bushing surface is cleaned and dried before measurement. The partial discharge detection system is calibrated regularly, with a calibration cycle not exceeding six months, to ensure measurement accuracy. The deviation analysis algorithm includes data validity checks, eliminating abnormal measurement values ​​that are significantly outside the reasonable range. The generation of the first model correction coefficient introduces a sliding window mechanism, considering the deviation trend of the most recent three detection cycles. The time-domain comparison algorithm handles the time asynchrony problem, using cubic spline interpolation to align the time series. The calculation of the second model correction coefficient weights recent data, with recent data having a higher weight than earlier historical data. The weight parameter adjustment process retains a snapshot of the model before adjustment, facilitating rollback to previous versions when necessary. The optimized insulation aging prediction model is cross-validated before formal use, using an independent validation dataset. The maintenance strategy set is generated using rule engine technology, and the rule base contains handling plans for various defect scenarios. The partial discharge suppression scheme details the installation location and parameter setting requirements of the suppressor. The air gap filling process specifies the performance indicators and construction process standards of the injected materials. The carbonization channel blocking method describes the specific operational steps of physical blocking and chemical treatment. The dynamic adjustment algorithm for the execution sequence of maintenance strategies comprehensively considers equipment importance coefficients and maintenance cost factors. The storage of insulation resistance measurement data includes environmental temperature and humidity records, facilitating environmental compensation during subsequent data analysis. Partial discharge detection reports include typical discharge spectra and statistical characteristic parameters, providing a complete record of discharge activity. Deviation analysis results generate visual charts, displaying the curves of measured and predicted values ​​and their areas of difference. The calculation process of the first model correction coefficient records the contribution of each input variable; contribution analysis helps understand the sources of deviation. The time-domain comparison report highlights abrupt changes in discharge quantity; abrupt change analysis helps identify situations of rapid deterioration in insulation condition. An upper limit is set on the adjustment range of the second model correction coefficient to prevent excessive changes to model characteristics in a single correction. The convergence condition for weight parameter adjustment is set to five consecutive iterations where the correction amount is less than a threshold, ensuring model stability. Sensitivity testing is performed on the optimized model before deployment to test the model's output response to input changes. Version management of the maintenance strategy set is linked to the model version to ensure the matching of strategies and models. The evaluation criteria for the implementation effect of the partial discharge suppression scheme clearly define quantitative indicators, such as the percentage reduction in discharge quantity. The quality control points for the air gap filling process include key inspection steps to ensure treatment effectiveness. The effectiveness of the carbonization channel blocking method is verified using insulation resistance recovery rate as the evaluation index. The execution timing optimization algorithm takes seasonal factors into account to avoid scheduling outdoor maintenance work during severe weather.

[0091] The detection cycle is set considering the equipment's operating load, with the detection interval appropriately shortened during high-load periods. The insulation resistance measurement voltage is selected based on the equipment's rated voltage to avoid insulation damage caused by overvoltage. Anti-interference measures for the partial discharge detection system include hardware filtering and software digital signal processing technology. The deviation analysis report generates automatic conclusion statements, with statement templates containing a description of the deviation degree and improvement suggestions. The adaptive adjustment mechanism of the first model correction coefficients gradually reduces the correction magnitude based on the model's usage time. Time-domain comparative analysis provides various statistical quantities, including correlation coefficients, peak deviation, and phase difference. The generation of the second model correction coefficients incorporates an expert review process, inviting domain experts to participate in coefficient determination in complex cases. A historical record of weight parameter adjustments is established, documenting the reasons and effects of each adjustment. The optimized model provides both online and offline operating modes to adapt to different application scenarios. The maintenance strategy set update mechanism is synchronized with equipment technical upgrades, promptly incorporating new technologies and processes. The selection logic for partial discharge suppression schemes considers the applicability and cost-effectiveness of different suppression technologies. The selection database for air gap filling materials contains performance comparison data for various materials. The decision tree analysis of carbonization channel blocking methods covers channel processing schemes of different lengths and depths. The results of dynamic adjustments to the execution sequence are pushed to the equipment management information system, which automatically generates maintenance work orders.

[0092] The data acquisition equipment has an automatic diagnostic function, triggering an alarm and recording fault information when equipment malfunctions. The insulation resistance measurement process includes a self-test procedure; if the self-test fails, the measurement is stopped and the equipment is prompted for inspection. The data format of the partial discharge detection report conforms to international standards, facilitating data exchange and comparative analysis. The deviation analysis algorithm provides confidence interval calculation, with the interval width reflecting the reliability of the analysis results. The generation process of the first model correction coefficients undergoes uncertainty propagation analysis to analyze the impact of each input error on the coefficient accuracy. Time-domain comparative analysis supports multi-scale comparisons, including both macro-trend comparisons and micro-feature comparisons. The application scope of the second model correction coefficients is clearly defined, distinguishing between scenarios applicable to global and local adjustments. The optimization objective of weight parameter adjustment is set to minimize prediction errors while maintaining the model's generalization ability. The optimized model provides an application programming interface, supporting third-party systems to call model services. The generation logic of the maintenance strategy set incorporates cost constraints, optimizing resource investment while ensuring effectiveness. The implementation progress monitoring of the partial discharge suppression scheme sets key nodes, triggering early warnings for node delays. The on-site operation instructions for the air gap filling process include safety precautions and emergency response plans. The effectiveness tracking period for the carbonization channel blocking method is clearly defined, and the treatment effect is evaluated periodically during the tracking period. The execution timing optimization results take into account the parallel operation of equipment and coordinate the downtime of multiple devices to reduce system impact.

[0093] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0094] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for detecting the insulation condition of a casing based on multi-parameter real-time monitoring, characterized in that, The method comprises the following steps: Real-time acquisition of a multi-dimensional monitoring data set in a sleeve operating environment, the multi-dimensional monitoring data set including a partial discharge pulse sequence, dielectric loss tangent fluctuation data, and surface leakage current density distribution; Insulation defect feature extraction processing is performed on the multi-dimensional monitoring data set to generate a sleeve insulation defect distribution map, the insulation defect distribution map including a surface discharge track, an internal air gap position marker, and a carbonization channel density index; A dynamic threshold segmentation algorithm is used to perform region division processing on the insulation defect distribution map to determine the boundary coordinates of a high-risk defect region and a potential defect region; An insulation state evaluation vector is generated according to the boundary coordinates, the insulation state evaluation vector including a surface discharge intensity index, an air gap volume proportion rate, and a carbonization channel expansion trend coefficient; The insulation state evaluation vector is input into a pre-trained insulation aging prediction model to output a sleeve residual insulation life estimate and a defect development priority ranking; The insulation defect feature extraction processing on the multi-dimensional monitoring data set comprises: The partial discharge pulse sequence is divided into discharge event subsets according to time windows, and each discharge event subset corresponds to a pulse cluster in a power frequency cycle; for each discharge event subset, the following processing is performed: A dielectric loss phase distribution matrix is constructed according to the dielectric loss tangent fluctuation data, the dielectric loss phase distribution matrix including loss angle offsets at different voltage phases; A three-dimensional discharge energy density field is generated in combination with the spatial mapping relationship between the surface leakage current density distribution and the discharge event subsets, the three-dimensional discharge energy density field including an axial energy gradient, a radial energy diffusion rate, and a tangential energy accumulation degree; Time-domain superposition analysis is performed on the three-dimensional discharge energy density fields of a plurality of continuous power frequency cycles to calculate the curvature radius of the surface discharge track, the geometric center coordinates of the internal air gap position marker, and the spatial distribution weight of the carbonization channel density index; The generation of the three-dimensional discharge energy density field comprises: A dielectric loss-discharge energy conversion equation is established based on the correlation between the loss angle offsets in the dielectric loss phase distribution matrix and the partial discharge pulse amplitudes, and a first distribution function of the axial energy gradient is obtained by solving the equation; A radial energy diffusion calculation model is constructed according to the attenuation characteristics of the surface leakage current density distribution in the radial direction of the sleeve, the radial energy diffusion calculation model including dynamic correction parameters of material dielectric constant and environmental humidity; An iterative calculation process of the tangential energy accumulation degree is established in combination with the spatio-temporal distribution characteristics of the partial discharge pulse cluster and the surface conductivity, the iterative calculation process including a feedback adjustment mechanism of pulse interval and energy accumulation rate; The output results of the first distribution function, the radial energy diffusion calculation model, and the tangential energy accumulation degree iterative calculation process are fused in a spatial grid to generate three-dimensional field data including axial, radial, and tangential energy distribution.

2. The method for detecting the insulation condition of the casing based on multi-parameter real-time monitoring according to claim 1, characterized in that, The region division processing on the insulation defect distribution map using the dynamic threshold segmentation algorithm comprises: Pixel intensity extreme points of the surface discharge track in the insulation defect distribution map are extracted, and a first dynamic segmentation threshold is generated according to the distribution density of the extreme points; The second dynamic segmentation threshold is calculated based on a geometric feature parameter of an internal air gap position marker, and the geometric feature parameter includes an air gap equivalent diameter-depth ratio; A third dynamic segmentation threshold is generated according to a spatial gradient change rate of a carbonization channel density index; The first dynamic segmentation threshold, the second dynamic segmentation threshold and the third dynamic segmentation threshold are weighted and fused to generate a composite dynamic segmentation threshold curve; The composite dynamic segmentation threshold curve is used for binarization segmentation of an insulation defect distribution map, and a boundary coordinate set of a high-risk defect region and a potential defect region is output.

3. The method for detecting the insulation condition of the casing based on multi-parameter real-time monitoring according to claim 2, characterized in that, The generated insulation state evaluation vector includes: The profile coordinates of the high-risk defect region are extracted from the boundary coordinate set, and the integral mean of the surface discharge intensity index in the profile is calculated; According to the air gap position marker coordinates of the potential defect region, the proportion relationship between the air gap volume ratio and the total volume of the sleeve is calculated; Based on the spatial distribution difference of the carbonization channel density index in the boundary coordinate set, the time derivative of the carbonization channel expansion trend coefficient is fitted; The surface discharge intensity index, the air gap volume ratio and the carbonization channel expansion trend coefficient are combined into a multi-dimensional vector according to a preset weight.

4. The method for detecting the insulation condition of the casing based on multi-parameter real-time monitoring according to claim 3, characterized in that, The processing of the pre-trained insulation aging prediction model includes: The surface discharge intensity index is input into a first aging analysis layer, and the surface insulation deterioration rate is determined through a discharge cumulative damage calculation module; The air gap volume ratio is input into a second aging analysis layer, and the air gap electric field distortion analysis is performed to output an internal insulation strength decay curve; The carbonization channel expansion trend coefficient is input into a third aging analysis layer, and the channel expansion acceleration is calculated based on a carbonization path growth model; The surface insulation deterioration rate, the internal insulation strength decay curve and the channel expansion acceleration are fused to generate a comprehensive aging index; According to the comparison result of the comprehensive aging index and the preset critical value, the residual insulation life estimate value is output, and the defect development priority is sorted according to the contribution degrees of the deterioration rate, the decay curve and the acceleration.

5. The method for detecting the insulation condition of the casing based on multi-parameter real-time monitoring according to claim 4, characterized in that, Further includes: Real-time monitoring of environmental temperature and humidity change data, calculating the compensation coefficient of temperature and humidity on dielectric loss phase distribution matrix; According to the compensation coefficient, adjust the axial energy gradient distribution in the three-dimensional discharge energy density field; Based on the temperature and humidity change rate, correct the dynamic correction parameter in the radial energy diffusion calculation model; The corrected three-dimensional discharge energy density field is re-input into the insulation aging prediction model to generate the residual insulation life estimate value after environmental compensation.

6. The method for detecting the insulation condition of the casing based on multi-parameter real-time monitoring according to claim 5, characterized in that, The compensation coefficient of temperature and humidity on dielectric loss phase distribution matrix includes: Obtain the initial dielectric constant of the sleeve material under the reference temperature and humidity and the temperature and humidity change amount, and establish a dielectric constant-temperature and humidity correlation function; According to the dielectric constant-temperature and humidity correlation function, calculate the offset compensation amount of the dielectric loss tangent value; The offset compensation amount is superimposed in the generation process of the dielectric loss phase distribution matrix, and the compensated loss angle offset amount distribution data is output.

7. The method for detecting the insulation condition of the casing based on multi-parameter real-time monitoring according to claim 1, characterized in that, Further includes: Collect the actual insulation resistance value and the partial discharge amount of the sleeve within a preset detection period; Deviation analysis is performed on the actual insulation resistance value and the predicted deterioration rate to generate a first model correction coefficient; The partial discharge amount and the predicted discharge intensity index are compared in time domain to generate a second model correction coefficient; The weight parameters of the insulation aging prediction model are adjusted according to the first model correction coefficient and the second model correction coefficient, and an optimized insulation aging prediction model is generated.

8. The method for detecting the insulation condition of the casing based on multi-parameter real-time monitoring according to claim 7, characterized in that, Further comprising: A maintenance strategy set is generated according to the defect development priority ranking, and the maintenance strategy set comprises a partial discharge suppression scheme, a gas gap filling processing flow and a carbonization channel blocking method; The execution timing of the maintenance strategy is dynamically adjusted based on the residual insulation life estimation, so that the implementation node of the maintenance strategy is synchronized with the insulation life attenuation stage.

Images