Intelligent evaluation method for insulation state of high-voltage equipment

Abstract

The application discloses a high-voltage equipment insulation state intelligent evaluation method, relates to the technical field of electrical engineering, and comprises the following steps: collecting insulation state data and performing pretreatment, performing feature extraction and data fusion on the pretreated insulation state data, and generating an electric field disturbance feature set; adopting a Tikhonov regularization inverse problem algorithm, inversely deducing a micro-defect density field inside insulation material based on the electric field disturbance feature set, and fusing the electric field disturbance feature set and the micro-defect density field into an insulation feature data set through normalization processing; constructing an insulation evaluation model, inputting the insulation feature data set into the insulation evaluation model for regression prediction, and outputting an insulation state score; setting three-level early warning thresholds, comparing the insulation state score with the three-level early warning thresholds, judging the insulation state, and performing early warning notification. Through deep feature extraction and fusion of multi-source data, the evaluation of the insulation state of high-voltage equipment is realized.

Description

Technical Field

[0001] This invention relates to the field of electrical engineering technology, and in particular to an intelligent assessment method for the insulation status of high-voltage equipment. Background Technology

[0002] With the increasing complexity of high-voltage equipment capacity and operating environment, intelligent assessment methods for the insulation status of high-voltage equipment have gradually become an important research direction in power operation and maintenance. Existing technologies mainly involve quantitative analysis of equipment status through the measurement of insulation resistance and discharge signals, as well as electric field distribution characteristics. Simultaneously, data fusion methods are used to comprehensively evaluate multi-source information, providing a technical foundation for high-voltage equipment health assessment. Finally, through multi-source feature fusion and micro-defect modeling, insulation monitoring can evolve from single-parameter judgment to overall status assessment, providing data support for intelligent operation and maintenance.

[0003] However, existing technologies still have significant problems in practical applications. The lack of effective fusion and continuous judgment mechanisms for multi-source feature information makes it impossible to reflect the relationship between micro-defects within the insulation material and its overall state. In complex electromagnetic environments, the mapping between electric field disturbances and micro-defect distribution is biased, making it difficult for the generated insulation state score to accurately reflect the true state of the equipment, thus affecting the reliability of graded early warning and intelligent operation and maintenance. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an intelligent assessment method for the insulation status of high-voltage equipment to solve the problem that the relationship between internal micro-defects and the overall status of insulation materials cannot be reflected.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides an intelligent assessment method for the insulation status of high-voltage equipment, comprising, Insulation state data is collected and preprocessed. Feature extraction and data fusion are performed on the preprocessed insulation state data to generate an electric field disturbance feature set. The Tikhonov regularized inverse problem algorithm is used to infer the micro-defect density field inside the insulation material based on the electric field perturbation feature set. The electric field perturbation feature set and the micro-defect density field are then fused into an insulation feature dataset through normalization. Construct an insulation assessment model, input the insulation feature dataset into the insulation assessment model for regression prediction, and output an insulation status score; A first threshold, a second threshold, and a third threshold are preset, and the insulation status score is compared with the first threshold, the second threshold, and the third threshold to determine the insulation status and issue an early warning notification.

[0007] As a preferred embodiment of the intelligent evaluation method for insulation status of high-voltage equipment according to the present invention, the insulation status data includes insulation material data, discharge signal, insulation resistance value and electric field strength data; The data on the insulating material includes the internal electric field distribution and micro-defects of the insulating material.

[0008] As a preferred embodiment of the intelligent assessment method for insulation status of high-voltage equipment described in this invention, the preprocessing refers to noise reduction, normalization, and timestamp alignment.

[0009] As a preferred embodiment of the intelligent assessment method for the insulation state of high-voltage equipment described in this invention, the specific steps for extracting features from the preprocessed insulation state data and fusing them to generate an electric field disturbance feature set are as follows: Wavelet packet decomposition was used to extract electric field variation features from electric field intensity data. Seasonal trend decomposition (STL) is used to extract resistance fluctuation characteristics from insulation resistance values; The time-frequency characteristics of the discharge signal are obtained by using wavelet transform to perform time-frequency analysis. The feature splicing method is used to splice and fuse the electric field change features, resistance fluctuation features and time-frequency domain features of the discharge signal to generate an electric field disturbance feature set.

[0010] As a preferred embodiment of the intelligent assessment method for insulation status of high-voltage equipment described in this invention, the method of using the Tikhonov regularized inverse problem algorithm to infer the micro-defect density field inside the insulation material based on the electric field disturbance feature set refers to constructing a linear equation system of micro-defects using electromagnetic field theory and the finite difference method, substituting the electric field disturbance feature set into the linear equation system of micro-defects, and solving the linear equation system of micro-defects through the Tikhonov regularization term to obtain the micro-defect density field.

[0011] As a preferred embodiment of the intelligent assessment method for insulation status of high-voltage equipment described in this invention, the construction of a linear equation set for micro-defects using electromagnetic field theory and the finite difference method refers to defining the physical relationship between the internal electric field distribution of the insulation material and micro-defects based on electromagnetic field theory and the Poisson equation, and then discretizing the Poisson equation using the finite difference method to obtain a linear equation set for micro-defects.

[0012] As a preferred embodiment of the intelligent insulation status assessment method for high-voltage equipment described in this invention, the specific steps of constructing an insulation assessment model, inputting the insulation feature dataset into the insulation assessment model for regression prediction, and outputting an insulation status score are as follows: The gradient boosting decision tree is used as the basic framework for the insulation assessment model to construct the insulation assessment model. Each gradient boosting decision tree starts from the root node and selects a split point to split the insulation feature dataset according to the splitting criterion, thus obtaining child nodes; Set a splitting threshold, and repeat the splitting process for each child node until the number of child nodes is lower than the splitting threshold. Stop splitting then. During the splitting process, the child nodes are passed down to the leaf nodes along the splitting path of the gradient boosting decision tree. When a child node reaches a leaf node, the leaf node outputs a predicted insulation state value. An insulation status score is generated by weighted summation of multiple insulation status predictions.

[0013] As a preferred embodiment of the intelligent assessment method for the insulation status of high-voltage equipment according to the present invention, the specific steps of setting a three-level early warning threshold, comparing the insulation status score with the three-level early warning threshold, determining the insulation status, and issuing an early warning notification are as follows: A first threshold, a second threshold, and a third threshold are preset, wherein the first threshold < the second threshold < the third threshold; When the insulation status score is less than the first threshold, the insulation status is determined to be an emergency state, and a warning message is sent to the maintenance personnel via the email server. When the first threshold is less than or equal to the insulation status score and less than the second threshold, the insulation status is judged to be of medium risk, and the relevant maintenance personnel are notified by email to strengthen monitoring. When the second threshold is less than or equal to the insulation status score and less than the third threshold, the insulation status is determined to be a state of concern, and an early warning report is sent to the department head via SMS. When the insulation condition score is greater than or equal to the third threshold, the insulation condition is judged to be normal.

[0014] In a second aspect, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the intelligent assessment method for insulation status of high-voltage equipment as described in the first aspect of the present invention.

[0015] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the intelligent assessment method for insulation status of high-voltage equipment as described in the first aspect of the present invention.

[0016] The beneficial effects of this invention are as follows: By extracting and fusing deep features from multi-source data, this invention enables the assessment of the insulation status of high-voltage equipment. This not only improves the sensitivity to changes in insulation status but also captures the development dynamics of micro-defects within the insulation material. By fusing multiple insulation status data, potential insulation problems can be identified at an early stage, making maintenance work more targeted. This reduces the risk of failures caused by insulation failure, ensures the stable operation of power installations, provides technical support for assessing insulation status, helps reduce unexpected power outages and equipment damage, and improves the safety and operating efficiency of power installations. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart for an intelligent assessment method for the insulation status of high-voltage equipment.

[0019] Figure 2 This is a flowchart of the electric field disturbance feature set.

[0020] Figure 3 This is a flowchart of an insulation feature dataset.

[0021] Figure 4 A flowchart for constructing an insulation assessment model. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figures 1-4This embodiment provides an intelligent assessment method for the insulation status of high-voltage equipment, including the following steps: S1. Collect insulation state data and preprocess it. Extract features and fuse data from the preprocessed insulation state data to generate an electric field disturbance feature set.

[0026] Insulation status data includes insulation material data, discharge signals, insulation resistance values, and electric field strength data; By deploying Rydberg atomic field meters on the surface of insulating materials, such as cable joints and transformer windings, and using electromagnetically induced transparent EIT technology, the weak electric field caused by PD events can be detected, the displacement of atomic energy levels can be measured, and the discharge signal can be obtained. Discharge signal refers to the transient discharge signal that occurs in insulating materials due to micro-defects, reflecting the insulation state of the insulating material; Insulation resistance values ​​are collected using an insulation resistance meter; Insulation resistance value refers to the resistance value of an insulating material to prevent current from passing through it, and is used to reflect the ability of an insulating material to prevent leakage current. Using the field-milled electrostatic probe of Trek-Model-347, the electric field strength is detected by a charge displacement device on the rotating blade to obtain electric field strength data; For example, one field-ground electrostatic probe is installed per meter of cable, and four field-ground electrostatic probes are configured near the transformer winding to form a three-dimensional electric field measurement network. The calibration accuracy of the field-ground electrostatic probes is set to ±2%, supporting a dynamic range of 0.1-100kV / m, adapting to equipment of different voltage levels, and the field-ground electrostatic probes are optimized through electromagnetic simulation to ensure that the coverage of the field-ground electrostatic probes reaches more than 90% and reduce blind spots. Electric field strength data refers to the electric field strength inside and around the insulating material; Data on insulating materials includes the internal electric field distribution and micro-defects within the insulating material; Using a quantum magnetometer based on nitrogen-vacancy NV centers, the electric field changes caused by dielectric polarization of the insulating material are detected by laser excitation and microwave pulses, and the electric field distribution inside the insulating material is obtained. By measuring the change in dielectric constant caused by micro-defects using a capacitive sensor, the location and size information of the micro-defects can be obtained. Combine quantum magnetometers and capacitive sensors to fully cover areas with insulating materials (such as transformer oil paper insulation layers), and adjust the number of sensors (10 to 20) according to the size of the equipment. Preprocessing refers to noise reduction, normalization, and timestamp alignment; Apply moving average filtering to remove noise from insulation status data; Maximum-minimum normalization was used to normalize the electric field strength data, discharge signal, and insulation resistance value. The formula is as follows: Normalized value = (current value - minimum value) / (maximum value - minimum value); The insulation status data is timestamped using linear interpolation. For insulation status data with missing time points, linear interpolation is performed using the values ​​of the two time points before and after the missing time point. Wavelet packet decomposition was used to extract electric field variation features from electric field intensity data. Multi-level wavelet packet decomposition of electric field intensity data is performed using wavelet basis functions to capture local electric field distortion caused by micro-defects. Each level of wavelet packet decomposition decomposes the electric field intensity data into low-frequency components (approximation coefficients) and high-frequency components (detail coefficients). After each level of wavelet packet decomposition, the entropy of the electric field intensity data is used as the electric field change feature. For example, the Daubechies wavelet basis (db4) is used to perform 4-level decomposition of the electric field intensity data, and the energy of the high-frequency subband is selected as the electric field change feature. Seasonal trend decomposition (STL) is used to extract resistance fluctuation characteristics from insulation resistance values; The STL algorithm is used to decompose the insulation resistance value into trend component, seasonal component and residual component. The trend component reflects the long-term trend, the seasonal component captures the periodic fluctuations, and the residual component represents the irregular fluctuations. Finally, the statistical characteristics (such as variance) of the residuals are extracted from the seasonal component and used as the resistance fluctuation characteristics of the insulation resistance value. The time-frequency characteristics of the discharge signal are obtained by using wavelet transform to perform time-frequency analysis. Morlet wavelets are used to perform multi-scale decomposition of the discharge signal (such as 8-level decomposition). Each level of decomposition will divide the discharge signal into low-frequency and high-frequency components. After each layer of decomposition, a set of low-frequency signals and a set of high-frequency signals are obtained. The low-frequency signals are then decomposed again, and the decomposition process is repeated until there are 8 layers. Wavelet coefficients are obtained after each layer of decomposition. Extract time-frequency domain features from wavelet coefficients; The feature splicing method is used to splice and fuse the electric field change features, resistance fluctuation features and time-frequency domain features of the discharge signal to generate an electric field disturbance feature set.

[0027] S2. The Tikhonov regularized inverse problem algorithm is used to infer the micro-defect density field inside the insulation material based on the electric field disturbance feature set. The electric field disturbance feature set and the micro-defect density field are then fused into an insulation feature dataset through normalization.

[0028] The method of using the Tikhonov regularized inverse problem algorithm to infer the micro-defect density field inside the insulating material based on the electric field disturbance feature set refers to constructing a linear equation system of micro-defects using electromagnetic field theory and the finite difference method, substituting the electric field disturbance feature set into the linear equation system of micro-defects, and solving the linear equation system of micro-defects through the Tikhonov regularization term to obtain the micro-defect density field. By measuring changes in the electric field and combining them with the physical laws of electromagnetic field theory, the spatial distribution characteristics of micro-defects inside insulating materials can be inferred, providing data support for evaluating the insulation performance of insulating materials. When micro-defects exist within an insulating material, the dielectric constant of the insulating material changes, leading to electric field distortion. This distortion can be addressed by measuring the electric field when micro-defects are present within the insulating material. By comparing the electric field with and without micro-defects, the electric field perturbation is obtained. The electric field perturbation reflects the influence of micro-defects on the electric field strength and is the data basis for inferring the micro-defect density field. The formula for electric field disturbance is: = - ; in, Indicates electric field disturbance (unit: volts / meter). Electric field when there are no defects (unit: volts / meter). Electric field representing a micro-defect (unit: volts per meter); For example, when an insulating material is energized, an electric field is generated around the insulating material. Under normal circumstances, the insulating material of the insulating layer is uniform, and the electric field distribution is regular and stable. However, when there are micro-defects inside the insulating layer, the electric field will be disturbed near the micro-defects. Specifically, the electric field strength is increased or decreased. The electric field disturbance is closely related to the type, size and distribution of the micro-defects. Void-type micro-defects lead to an enhanced electric field, while impurity-type micro-defects alter the direction of the electric field. Based on electromagnetic field theory, the physical relationship between the internal electric field distribution and micro-defects of insulating materials is defined using the Poisson equation. The specific steps are as follows: When there are micro-defects inside the insulating material, the dielectric constant of the insulating material will change due to the presence of the micro-defects, causing the electric field distribution to deviate from the normal state. According to electromagnetic field theory, micro-defects can cause electric field distortion, which manifests as electric field perturbation in electric field strength. This electric field perturbation is related to the micro-defect density field, which represents the distribution density of micro-defects in insulating materials. The physical relationship between the internal electric field distribution and micro-defects of an insulating material is defined using the Poisson equation, and the formula is as follows: ; in, This indicates electric field disturbance (unit: volts per meter). Represents the second derivative. This represents the density of microdefects (unit: coulombs per cubic meter). Indicates dielectric constant (unit: farad / meter); According to the Poisson equation, the higher the density of micro-defects, the more obvious the electric field distortion. For example, void-type micro-defects will reduce the dielectric constant and cause the electric field to concentrate, while impurity-type micro-defects will cause additional charge effects, further complicating the electric field distribution. The Poisson equation is discretized using the finite difference method to obtain a system of linear equations for micro-defects. The finite difference method is a numerical method that transforms continuous physical relationships into discrete difference forms by dividing the insulating material region into a three-dimensional mesh. The specific discretization process involves establishing a uniform grid structure within the insulating region where the insulating material is located, dividing the region into multiple small cells, each representing the location of a micro-defect. The physical relationship is then approximated as a difference equation between the electric field and the micro-defect density, as shown in the formula. ; in, This represents the coefficient matrix based on mesh partitioning and dielectric constant distribution. Represents the electric field vector. Indicates the density of micro-defects; For example, a 1 mm grid size is often used for high-precision analysis. Secondly, the physical relationship between the electric field and the micro-defect density field is approximated as a system of linear equations using the finite difference method, generating the relationship between the coefficient matrix, the electric field vector, and the micro-defect density field. The coefficient matrix is ​​constructed based on the grid division and dielectric constant distribution, reflecting the physical connection between each grid point and its neighboring points. The non-uniformity of the dielectric constant distribution directly affects the structure of the coefficient matrix. The dielectric constant is obtained through a capacitance sensor. For example, the smaller the dielectric constant in a densely populated micro-defect region, the more likely it is to cause changes in the matrix elements. Substituting the electric field disturbance feature set into the linear equations of the micro-defects, we obtain the micro-defect density field. The electric field disturbance feature set needs to be converted into a discrete three-dimensional spatial distribution. The specific method is as follows: Spatial interpolation is performed on the electric field variation features obtained from wavelet packet decomposition, the resistance fluctuation features obtained from STL decomposition, and the time-frequency domain features of the discharge signal to obtain a three-dimensional electric field disturbance feature set. The electric field disturbance feature set is then normalized to eliminate the influence of different electric field disturbance feature sets in terms of their dimensions. Substituting the normalized electric field disturbance feature set into the linear equation system, since the micro-defect density is unknown, the Tikhonov regularized inverse problem algorithm is needed to solve the inverse problem. The micro-defect density field is obtained by solving the linear equations of the micro-defect system using the Tikhonov regularization term. When collecting insulation status data, noise will be present in the collected insulation status data due to sensor accuracy limitations and environmental electromagnetic interference. This noise will cause oscillations when solving the micro-defect density field. In addition, meshing will cause the column vectors of the coefficient matrix to be approximately linearly correlated. Directly solving the micro-defect density field will amplify the error, resulting in the calculated micro-defect density field having no physical meaning. For example, the micro-defect density field may have negative values, which does not match the actual micro-defect distribution. The Tikhonov regularization term transforms the ill-conditioned problem into an optimization problem through prior constraints, thereby suppressing the interference of noise on solving the micro-defect density field. Tikhonov regularization defines the objective function by balancing the minimization of data fitting error and the complexity of the solution, thus constraining the smoothness and physical rationality of the micro-defect density field. For example, the Tikhonov regularization term makes the micro-defect density values ​​of adjacent grid points close, avoiding drastic spatial jumps. The solution process usually adopts the conjugate gradient method to iteratively optimize the Tikhonov regularization objective function and generate a smooth micro-defect density field that conforms to physical laws. Compared with direct solution, Tikhonov regularization improves the robustness and reliability of the micro-defect density field and is suitable for real-world scenarios with high noise levels. The effect of micro-defects on the electric field is essentially achieved by changing the insulation properties of the insulating material itself. The originally uniform dielectric properties inside the insulating material will produce structural defects when micro-defects are present, thereby reducing the dielectric properties. This change in dielectric properties is positively correlated with the density of micro-defects. The denser the micro-defects, the lower the dielectric constant. Through the material structure relationship in electromagnetic field theory, the quantitative correlation between micro-defect density and dielectric constant is used as the basis of the linear equation system of micro-defects. By normalizing the data, the electric field disturbance feature set and the micro-defect density field are fused into an insulation feature dataset. The normalization method is Z-score standardization, which can eliminate scale differences between different physical units. By calculating the mean and standard deviation of the electric field perturbation feature set and the micro-defect density field, the electric field perturbation feature set and the micro-defect density field are scaled to a uniform range. After Z-score standardization, the electric field disturbance feature set and the micro-defect density case are directly combined into an insulation feature dataset; Since the electric field disturbance feature set and the micro-defect density field have different physical dimensions, direct fusion would lead to scale differences. By normalizing, the electric field disturbance feature set and the micro-defect density field are scaled to the same range, which is convenient for subsequent input into the insulation evaluation model.

[0029] S3. Construct an insulation assessment model, input the insulation feature dataset into the insulation assessment model for regression prediction, and output the insulation status score.

[0030] The gradient boosting decision tree is used as the basic framework for the insulation assessment model to construct the insulation assessment model. The insulation assessment model consists of multiple gradient boosting decision trees. The gradient boosting decision tree is constructed iteratively by building multiple decision trees to gradually obtain the true insulation state. Each decision tree is optimized based on the residual of the previous decision tree. For example, the average value of the insulation status score in the training set is used as the baseline prediction value of the insulation assessment model. Then, for each decision tree, the residual between the baseline prediction value and the true value of the current insulation assessment model is calculated. The residual represents the direction and magnitude of the error. The decision tree is fitted to the residuals to generate a new decision tree for predicting residual values. The predictions of the new decision tree are scaled by a learning rate (e.g., 0.1) and added to the predictions of the current insulation assessment model for updating. This iterative optimization process continues until a preset number of decision trees (e.g., 100 decision trees) is reached. The learning rate is used to control the contribution of each decision tree. A lower learning rate (e.g., 0.1) makes the insulation assessment model update smoother and reduces the risk of overfitting, but requires more decision trees to achieve convergence. Finally, the predictions of all decision trees are weighted and summed to generate an insulation status score. The weights are determined by the learning rate and the fitting effect of each decision tree. Meanwhile, the mean squared error (MSE) is used as the loss function to minimize the residuals. The insulation feature dataset was divided into a training set and a validation set (80%:20%) for training and evaluating the insulation assessment model. Input the training set into the insulation evaluation model; Each gradient boosting decision tree starts from the root node and, according to the splitting criterion, selects the splitting point of the insulation feature dataset to split the insulation feature dataset, thus obtaining child nodes; The splitting criterion refers to the selection of split points by minimizing the residual squared error in gradient boosting decision trees. Specifically, for the root node of the current decision tree, all features (such as electric field strength and micro-defect density field) and potential split points of the insulation feature dataset are traversed. The sum of squared residuals of all insulation feature datasets within the root node is calculated. Then, the split point of the insulation feature dataset is divided into left and right parts, and the sum of squared residuals of the left and right parts is recalculated. Finally, the loss reduction before and after splitting is compared by gain to determine whether to continue splitting until splitting stops. The splitting process of a decision tree refers to each decision tree starting from the root node. The root node contains the entire insulation feature dataset. Each row in the dataset represents a sample (such as the measurement of the insulation state of high-voltage equipment), and each column represents an insulation state feature (such as electric field strength and micro-defect density). The splitting criterion is used to reduce mean square error. The decision tree tries different splitting points for each insulation state feature (such as electric field change features and resistance fluctuation features), divides it into two groups according to the splitting points, calculates the number of residual sums of squares reduced, and finally selects the feature and splitting point that can reduce the residual sums of squares to the greatest extent as the splitting method of the current node. For example, when the insulation feature dataset has 1000 samples, and each sample has 10 insulation state features (electric field disturbance features and micro-defect density field), when the decision tree finds that the condition "micro-defect density is greater than 0.3" can divide the samples into two groups, one group with more micro-defects (poor insulation state) and the other group with fewer micro-defects (good insulation state), then this condition is the splitting criterion, and 0.3 is the splitting point. Once the splitting criterion is selected (e.g., micro-defect density greater than 0.3), the decision tree will split the insulation feature dataset into a left child node (all samples with micro-defect density ≤ 0.3) and a right child node (all samples with micro-defect density > 0.3). Set a splitting threshold, and repeat the splitting process for each child node until the number of child nodes is lower than the splitting threshold. Stop splitting then. During the splitting process, the child nodes are passed down to the leaf nodes along the splitting path of the gradient boosting decision tree. In this embodiment, the split threshold is a threshold that controls the minimum number of samples in the child nodes, which is used to control the growth of the decision tree and prevent the child nodes from being too small, which would lead to unstable predictions and overfitting. The splitting threshold is set based on the total number of samples in the insulation feature dataset and can be adjusted according to the actual equipment monitoring sample density.

[0031] For example, the insulation feature dataset contains 1000 sample data points. The splitting threshold can be set to 50-100 samples (i.e., the child nodes contain at least 5%-10% of the original samples) to ensure that the insulation state prediction values ​​output by the leaf nodes are statistically representative. For high-voltage equipment with different voltage levels and sensor densities, the threshold can be adjusted according to the acquisition frequency × monitoring period. For example, the splitting threshold can be increased when monitoring at high density and decreased when monitoring at low density.

[0032] Stopping splitting means that when the decision tree can no longer improve the micro-defect by continuing to split, the decision tree will stop splitting. The child node when splitting stops is the leaf node. The conditions for stopping splits are used to control the growth of the decision tree, prevent overfitting and waste of computational resources. The main conditions for stopping splits include the minimum number of samples, the maximum tree depth, and the minimum MSE reduction. By setting the minimum number of samples, the maximum tree depth, and the minimum MSE reduction, at each split, we check whether the number of samples in the child nodes is lower than the minimum number of samples, whether the tree depth has reached the maximum depth, and whether the minimum MSE reduction is lower than the split threshold. When any one of these conditions is met, the split stops. For example, the minimum number of samples is set to 10, that is, when the number of child node samples is less than 10, the splitting stops to avoid the prediction being unstable due to the nodes being too small. The maximum tree depth is set to 5 to limit the height of the tree to balance the complexity and generalization ability of the insulation assessment model. The minimum MSE reduction is set to 0.01 to ensure that each split can reduce the residual sum of squares and avoid invalid splitting. It should be noted that, in addition to the number of child nodes being lower than the splitting threshold, splitting should also be stopped when there are no more splitting points that can further reduce the loss function and when the number of splits reaches the maximum tree depth. When a child node reaches a leaf node, the leaf node outputs a predicted insulation state value. The optimization objective of gradient boosting decision tree is to generate the insulation state prediction value of leaf node, that is, to minimize the residual squared error. When the decision tree stops splitting, the child node becomes a leaf node, and the leaf node outputs the average value of the residuals of all samples in the current node as the insulation state prediction value. For example, if a leaf node contains 20 samples with residuals of 0.5, -0.3, and 0.2, the predicted insulation state value is the average residual (e.g., 0.1). An insulation status score is generated by weighted summation of multiple insulation status predictions. The insulation status score is obtained by weighted summation of the insulation status prediction values ​​of multiple decision trees. The insulation status prediction values ​​are continuous values ​​(e.g., -10 to 10). The continuous values ​​(e.g., -10 to 10) are mapped to the range of [0, 100] of the insulation status score through linear transformation. The linear transformation is to calculate the sum of the insulation state predictions of all decision trees to obtain the original predictions, and then scale the original values ​​to [0, 100] using a linear normalization formula; For example, if the original predicted value range is [min, max], the mapping formula is: (Predicted insulation condition value - min) / (max - min) × 100; If there is no clear range, the mapping boundary can be determined based on the training set score distribution (e.g., mean ± 3 standard deviations). In a physical sense, the score reflects the health status of the insulating material. For example, 100 indicates perfect insulation with no significant micro-defects and a uniform electric field distribution; 0 indicates severe defects and insulation material failure; and intermediate values ​​(e.g., 50) indicate moderate defects and a risk of discharge. The insulation assessment model is evaluated using a validation set. The parameters of the insulation assessment model are optimized using the mean square error loss function to obtain the mean square error. The parameters of the insulation assessment model are then adjusted based on the mean square error. The optimal split point for the insulation feature dataset is selected by using a splitting criterion that minimizes the squared residual error. This ensures that each gradient boosting decision tree can identify the data splitting method that is most helpful in reducing prediction error. This ensures that each gradient boosting decision tree can improve the prediction accuracy of insulation state during the construction process. The splitting criterion effectively guides the growth direction and depth of the tree structure by calculating and comparing the loss reduction before and after different split points. This allows the final gradient boosting decision tree to not only deeply mine the key features in the data, but also avoid overfitting. Subsampling (80% of samples), L2 regularization (penalizing large predicted values), and early stopping (stopping when the mean square error of the validation set fails to improve after 10 consecutive iterations) are used to control overfitting.

[0033] S4. Set a three-level warning threshold, compare the insulation status score with the three-level warning threshold, determine the insulation status and issue a warning notification.

[0034] Based on the insulation status score, a three-level early warning threshold is set; The three-level early warning thresholds are low-risk threshold, medium-risk threshold, and high-risk threshold; For example, the high-risk threshold is 40%, the medium-risk threshold is 80%, and the low-risk threshold is 90%. The insulation status score is compared with the three-level early warning threshold to determine the insulation status. When the insulation status score is less than the high-risk threshold, the insulation status is judged to be in an emergency state, and a warning message is sent to maintenance personnel via the email server. When the high-risk threshold is less than or equal to the insulation status score and less than the medium-risk threshold, the insulation status is determined to be of medium risk, and relevant maintenance personnel are notified via email to strengthen monitoring. When the medium risk threshold is less than or equal to the insulation status score and less than the low risk threshold, the insulation status is judged to be in a state of concern, and an early warning report is sent to the department head's mobile phone via SMS. When the insulation status score is greater than or equal to the low-risk threshold, the insulation status is judged to be normal. When the insulation status is in a state of concern, the department head should conduct a weekly inspection plan, checking the insulation status once a week after receiving the SMS message until the status is normal. The SMS message content is as follows: Equipment CBL-001, score 70, state of concern, time 2025-04-29 10:00, location substation A, please arrange weekly inspection. Once the insulation status is declared to be in an emergency, an emergency warning notification should be immediately sent to the maintenance personnel to verify the insulation status and relevant sensor data. After the maintenance personnel confirm that the insulation status is indeed abnormal, they should authorize and confirm the triggering of the shutdown device through the operation interface to shut down the high-voltage equipment to prevent damage to the high-voltage equipment. After shutdown, maintenance personnel should be dispatched to conduct on-site inspection and repair, and the time of confirmation and shutdown operation, the responsible person, and relevant equipment information should be recorded.

[0035] The weekly inspection includes using a portable electric field sensor to measure the electric field distribution of the equipment, verifying changes in the insulation condition score, conducting dielectric loss tests, assessing the aging of the insulation material, and checking whether the operating environment (such as temperature and humidity) exacerbates micro-defects. After repair, a dielectric strength tester is used to verify the repair effect. The insulation condition score must drop below 40 before operation can be resumed. Repair records are archived, including defect type, repair measures and test results, to provide data support for subsequent repairs. It should be noted that after the warning notification is sent, the delivery status should be verified. For warning notifications in non-emergency situations, if delivery fails (such as network interruption), they should be resent, with a maximum of 3 retries, each with an interval of 5 minutes. If they still fail after 3 retries, the person in charge should be contacted through backup channels (such as telephone notification). Emergency warnings should be sent immediately through multiple channels (SMS, telephone, and SCADA alarms) and trigger automatic shutdown devices, without relying on notification confirmation, to ensure that electrical equipment responds within a safe time window. To ensure security, all notifications are transmitted via TLS encryption for SMS and email to prevent data breaches.

[0036] This embodiment also provides a computer device applicable to the intelligent assessment method for the insulation status of high-voltage equipment, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the intelligent assessment method for the insulation status of high-voltage equipment as proposed in the above embodiment.

[0037] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0038] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the intelligent assessment method for the insulation status of high-voltage equipment as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0039] In summary, this invention achieves the assessment of the insulation status of high-voltage equipment through deep feature extraction and fusion of multi-source data. This not only improves the sensitivity to changes in insulation status but also captures the development dynamics of micro-defects within the insulation material. By fusing multiple insulation status data, potential insulation problems can be identified at an early stage, making maintenance work more targeted. This reduces the risk of failures due to insulation failure, ensures the stable operation of power installations, provides technical support for assessing insulation status, helps reduce unexpected power outages and equipment damage, and improves the safety and operational efficiency of power installations.

[0040] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for intelligent assessment of the insulation status of high-voltage equipment, characterized in that: include, Insulation status data is collected and preprocessed. The insulation status data includes insulation material data, discharge signal, insulation resistance value and electric field strength data. The insulation material data includes the internal electric field distribution and micro-defects of the insulation material. The preprocessing refers to noise reduction, normalization and timestamp alignment. Feature extraction and data fusion are performed on the preprocessed insulation state data to generate an electric field disturbance feature set. The specific steps for generating the electric field disturbance feature set are as follows: Wavelet packet decomposition was used to extract electric field variation features from electric field intensity data. Seasonal trend decomposition (STL) is used to extract resistance fluctuation characteristics from insulation resistance values; The time-frequency characteristics of the discharge signal are obtained by using wavelet transform to perform time-frequency analysis. The feature splicing method is used to splice and fuse the electric field change features, resistance fluctuation features and time-frequency domain features of the discharge signal to generate an electric field disturbance feature set; The Tikhonov regularized inverse problem algorithm is used to inversely deduce the micro-defect density field inside the insulating material based on the electric field disturbance feature set. After normalization, the electric field disturbance feature set and the micro-defect density field are merged into an insulation feature dataset. The Tikhonov regularized inverse problem algorithm is used to inversely deduce the micro-defect density field inside the insulating material based on the electric field disturbance feature set. This means that a linear equation system of micro-defects is constructed using electromagnetic field theory and the finite difference method. The electric field disturbance feature set is substituted into the linear equation system of micro-defects, and the Tikhonov regularization term is used to solve the linear equation system of micro-defects to obtain the micro-defect density field. Construct an insulation assessment model, input the insulation feature dataset into the insulation assessment model for regression prediction, and output an insulation status score; Three-level early warning thresholds are set, and the insulation status score is compared with the three-level early warning thresholds to determine the insulation status and issue an early warning notification.

2. The intelligent assessment method for insulation status of high-voltage equipment as described in claim 1, characterized in that: The construction of a linear equation system for micro-defects using electromagnetic field theory and the finite difference method refers to defining the physical relationship between the internal electric field distribution of insulating materials and micro-defects based on electromagnetic field theory and the Poisson equation, and then discretizing the Poisson equation using the finite difference method to obtain a linear equation system for micro-defects.

3. The intelligent assessment method for insulation status of high-voltage equipment as described in claim 2, characterized in that: The specific steps for constructing an insulation assessment model, inputting the insulation feature dataset into the model for regression prediction, and outputting an insulation status score are as follows: The gradient boosting decision tree is used as the basic framework for the insulation assessment model to construct the insulation assessment model. Each gradient boosting decision tree starts from the root node and selects a split point to split the insulation feature dataset according to the splitting criterion, thus obtaining child nodes; Set a splitting threshold, and repeat the splitting process for each child node until the number of child nodes is lower than the splitting threshold. Stop splitting then. During the splitting process, the child nodes are passed down to the leaf nodes along the splitting path of the gradient boosting decision tree. When a child node reaches a leaf node, the leaf node outputs a predicted insulation state value. An insulation status score is generated by weighted summation of multiple insulation status predictions.

4. The intelligent assessment method for insulation status of high-voltage equipment as described in claim 3, characterized in that: The process involves setting three warning thresholds, comparing the insulation status score with the three warning thresholds, determining the insulation status, and issuing a warning notification. The specific steps are as follows: Based on the insulation status score, a three-level early warning threshold is set; When the insulation status score is less than or equal to the low-risk threshold, the insulation status is judged to be normal. When the low-risk threshold < insulation status score ≤ medium-risk threshold, the insulation status is judged to be in a state of concern, and an early warning report is sent to the department head's email address via SMS. When the insulation condition score exceeds the high-risk threshold, the insulation condition is determined to be in an emergency state, and a warning message is sent to maintenance personnel via the email server.

5. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the intelligent evaluation method for insulation status of high-voltage equipment as described in any one of claims 1 to 4.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the intelligent evaluation method for the insulation status of high-voltage equipment as described in any one of claims 1 to 4.

Images