Voltage transformer state influence factor analysis method, system, equipment and medium

Abstract

The invention discloses a voltage transformer state influence factor analysis method, system and device and a medium, and belongs to the technical field of voltage fault analysis, and the method comprises the steps: collecting state parameters of a voltage transformer, carrying out the preprocessing of the state parameters, and constructing a nonlinear coupling relation between multidimensional feature tensor capture features; analyzing voltage fluctuation characteristics of the voltage transformer, and automatically correcting the insulation state evaluation model; and inputting abnormal feature vectors obtained by monitoring into a recursive attribution decision tree model, carrying out multi-dimensional abnormal clustering and causal analysis, generating a tracing report, and carrying out multi-dimensional evaluation and verification on an analysis result. According to the invention, continuous dynamic full-coverage monitoring of the equipment insulation health condition is realized, the accuracy and robustness of anomaly detection are improved, the phenomena of missing detection and false alarm are reduced, the scientificity of anomaly tracing analysis and the interpretation of early warning decision are improved, and the reliability of the system is improved. And the intelligent level of the monitoring and early warning system is improved by continuously adapting to new abnormal types and complex operation environments.

Description

Technical Field

[0001] This invention relates to the field of voltage fault analysis technology, specifically to a method, system, equipment, and medium for analyzing the influencing factors of voltage transformer status. Background Technology

[0002] With the increasing scale and intelligence of power systems, voltage transformers, as core components of power metering, play a crucial role in ensuring stable power supply. In recent years, a large number of high-voltage voltage transformers in the power grid have experienced insulation degradation and frequent faults due to complex environments, load fluctuations, and variable operating conditions. Traditional insulation testing and risk assessment methods, relying on periodic offline tests and empirical criteria, suffer from limitations such as limited data samples, assessment lag, and insensitivity to anomalies. They struggle to capture early, hidden degradation and complex anomalies caused by the coupling of multiple factors. Existing technologies have limited modeling capabilities for multi-source influencing factors such as load and environment, failing to comprehensively and dynamically reveal the insulation degradation mechanisms and early warning indicators under various complex operating conditions. This leads to strong reliance on and subjectivity in operation and maintenance decisions, resulting in missed detections.

[0003] Traditional voltage transformer condition monitoring relies on manual inspections and periodic tests, resulting in low data acquisition frequency and limited coverage, making it difficult to promptly detect early-stage problems and dynamic anomalies. Existing methods, which use empirical thresholds and single-physical-quantity monitoring, lack synergistic analysis and deep coupling modeling of multiple influencing factors of environmental load voltage. Furthermore, they lack sufficient understanding of insulation degradation mechanisms under complex operating conditions.

[0004] Anomaly detection is largely based on static rules, resulting in poor flexibility and high rates of missed detection and false positives when dealing with novel or complex insulation anomalies. Current technologies lack comprehensive anomaly tracing capabilities, failing to effectively identify dominant risk factors. Maintenance personnel often struggle to obtain clear anomaly evolution information, leading to insufficient support for decision-making. The difficulty in deeply integrating expert knowledge with data-driven models limits the models' adaptability and self-evolution, making it difficult to meet the demands of rapid power grid development and increasingly complex operating environments. In summary, current technologies exhibit significant shortcomings in real-time performance, comprehensiveness, intelligence, and interpretability, requiring further breakthroughs and improvements. Summary of the Invention

[0005] In view of the above-mentioned problems, the present invention is proposed.

[0006] Therefore, this invention aims to propose a method for monitoring insulation status and providing early warning of degradation that integrates real-time monitoring of anomalies from multiple data sources with the ability to identify the causal relationship of faults, thereby meeting the high precision and reliability requirements of modern power grids for the full life cycle management and risk control of voltage transformers.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a method for analyzing the influencing factors of voltage transformer status, comprising, The system collects the state parameters of the voltage transformer and preprocesses them to construct a multidimensional feature tensor to capture the nonlinear coupling relationship between features. It analyzes the voltage fluctuation characteristics of the voltage transformer, identifies extreme value anomaly areas in real time, evaluates the current electric field distribution of the transformer, and automatically corrects the insulation state evaluation model. The abnormal feature vectors obtained from the monitoring are input into the recursive attribution decision tree model to perform multidimensional anomaly clustering and causal analysis, generate a cause-tracing report, and perform multidimensional evaluation and verification of the analysis results.

[0008] As a preferred embodiment of the voltage transformer condition influencing factor analysis method of the present invention, the preprocessing includes: collecting environmental parameters at the voltage transformer operating site and performing improved aggregation processing to construct a correlation matrix to measure the influence intensity of environmental parameters on insulation condition in the past period. A weight self-adjustment mechanism is set up so that when any parameter is detected to be abnormal, the weight of the current parameter is adjusted, and all parameters are normalized and weighted to calculate the environmental health index.

[0009] As a preferred embodiment of the voltage transformer state influencing factor analysis method of the present invention, the step of constructing a multidimensional feature tensor to capture the nonlinear coupling relationship between features includes collecting and decomposing the real-time waveform data of the secondary current of the transformer. Synchronously extract associated features and assemble them into a multidimensional load feature sequence tensor.

[0010] As a preferred embodiment of the voltage transformer state influencing factor analysis method described in this invention, the construction of a multidimensional feature tensor to capture the nonlinear coupling relationship between features further includes using tensor decomposition to perform low-rank modeling of the multidimensional load feature sequence tensor. Minimize reconstruction error and capture nonlinear coupling relationships between features.

[0011] As a preferred embodiment of the voltage transformer condition influencing factor analysis method described in this invention, the automatically corrected insulation condition evaluation model includes using a high-precision sensor to measure the operating voltage of the voltage transformer. Full-time high-frequency continuous data acquisition is performed to obtain voltage time series sequences. The voltage time series sequences are then divided into intervals, and time windows are set. Extract the maximum value from each small segment. and minimum value The piecewise extreme value perturbation is obtained. ; Continuous detection based on sliding window method The statistical characteristics of the changes, combined with the distribution drift criterion. or ,in, and for The mean at time t and time t-1, and for The standard deviations at time t and time t-1 and This is the sensitivity threshold; For the detected voltage disturbance range, input the current voltage distribution. State of isolation from history Finite element simulation model is used to evaluate the electric field distribution in key parts of the current transformer. Where x is the spatial coordinate, the change in electric field is compared. Breakdown or aging threshold of insulating material The relationship between dynamic identification of critical risk zones in local insulation of equipment: when any local electric field peak value If the threshold is continuously exceeded, the weight is adjusted based on the extreme value perturbation interval under drift analysis, and the insulation state evaluation function is updated in real time. in, For the corrected insulation health, The intensity coefficient of the linkage between disturbance and distribution. This is a comprehensive response function based on statistics and physical fields; When voltage fluctuations are detected that lead to a deterioration in insulation condition, sensitivity to risks in the current time zone and location is increased.

[0012] The beneficial effects of the preferred technical solution in the embodiments of the present invention are as follows: Based on the operating voltage time-series data of the voltage transformer, the local electric field distribution caused by voltage disturbances, the insulation health status and its dynamic changes under the influence of external disturbances; the degradation risk of the insulation critical zone under small fluctuations is analyzed; a dynamic insulation health correction evaluation system with adaptive feedback capability is obtained, which provides early warning of insulation critical risks under small voltage extreme disturbances and local electric field anomalies, significantly improving the detection sensitivity and anomaly expression capability of early signs of insulation degradation in the transformer, and providing a powerful intelligent monitoring means for equipment operation safety.

[0013] As a preferred embodiment of the voltage transformer state influencing factor analysis method described in this invention, the step of performing multidimensional anomaly clustering and causal analysis includes aggregating features based on environmental health. Secondary load anomaly score and voltage extreme value disturbances and the corrected insulation state The results of multi-source monitoring together form a multidimensional anomaly feature vector. The input is fed into a recursive attribution decision tree model, with the root node set as the overall anomaly rating variable. The tree structure transforms various input features into branch judgments, with each node based on a feature threshold. Recursive partitioning: in, , , , , belong During the recursive attribution process, each branch node stores the causal path. The information was used to deduce the dominant factor sequence leading to the anomaly; the analysis mapped the anomalous time-series trajectories of all features into a causal directed graph. ,node Corresponding to the multi-source monitoring features extracted during modeling, the edges This is reflected in the anomaly identification and dynamic health assessment, based on the correlation indicators, using the maximum a posteriori probability method to infer the causal risk path and obtain the risk transmission chain that dominates insulation degradation; The model generates a cause-finding report for each anomaly, and the report is based on the decision-making path. The critical path visualization in the causal graph G shows the attribution of responsibility, the evolution of risk, and possible triggering scenarios.

[0014] The beneficial effects of the preferred technical solutions in the embodiments of the present invention are as follows: They analyze the characteristic data from environmental, load, and voltage anomaly monitoring, as well as the causal relationships and risk propagation paths in the dynamic evolution of the insulation state of the instrument transformer; they analyze and attribute the core risk sources that dominate insulation degradation and equipment anomalies; they form an automated, multi-dimensional, and traceable anomaly attribution and explanation system; they efficiently locate the dominant risk factors, accurately output clear anomaly tracing reports, and improve the efficiency of instrument transformer anomaly diagnosis, risk warning, and scientific decision-making.

[0015] As a preferred embodiment of the voltage transformer state influencing factor analysis method described in this invention, the multi-dimensional evaluation and verification of the analysis results includes: constructing an expert-model collaborative interaction and annotation platform; and the model's preliminary analysis outputting anomaly types. The results are automatically pushed to domain experts, who evaluate and label the output based on their experience and historical cases: the actual category is... and risk level Manual annotations are made on key causal paths or risk triggers, and interactive data is presented in tuples. Formal recurrent feedback is fed into the training queue; Employing a self-evolutionary algorithm that combines expert experience transfer and feedback reinforcement, all newly acquired expert-annotated samples are added to the training set during model training and updates, dynamically expanding the knowledge base. Model parameters are optimized by minimizing the comprehensive loss function. Iterative updates, the comprehensive loss function is defined as: in, The loss represents the difference between the model output and the expert-annotated categories. The consistency between the measurement model's judgment level and the expert's implicit risk perception. For model parameters The regularization term is L2 regularized. Adjust the weighted hyperparameters according to the relative importance of the loss in the overall objective.

[0016] The beneficial effects of the preferred technical solution in the embodiments of the present invention are as follows: it applies to all types of monitored anomalies, risk ratings, causal paths, expert knowledge rules, and model parameter spaces, focusing on the consistency and complementarity between model recognition accuracy and expert practical experience; it enables the model to integrate and learn from big data and expert knowledge, continuously improving the model's analytical capabilities, anomaly recognition, and risk classification accuracy; it quickly adapts to new or complex anomaly scenarios and automatically generates highly interpretable analysis reports.

[0017] Another objective of this invention is to provide a system for analyzing the factors affecting the state of voltage transformers.

[0018] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a voltage transformer state influencing factor analysis system, comprising: a data acquisition module, a correction module, and an analysis module; The acquisition module acquires the state parameters of the voltage transformer, preprocesses the state parameters, and constructs a multi-dimensional feature tensor to capture the nonlinear coupling relationship between features. The correction module analyzes the voltage fluctuation characteristics of the voltage transformer, identifies extreme value anomaly areas in real time, assesses the current electric field distribution of the transformer, and automatically corrects the insulation status evaluation model. The analysis module inputs the abnormal feature vectors obtained from monitoring into the recursive attribution decision tree model to perform multidimensional anomaly clustering and causal analysis, generate a tracing report, and conduct multidimensional evaluation and verification of the analysis results.

[0019] The present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, characterized in that the processor executes the computer program to implement the steps of the method for analyzing the influencing factors of voltage transformer status.

[0020] The present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that, when the computer program is executed by a processor, it implements the steps of the method for analyzing the influencing factors of voltage transformer status.

[0021] The beneficial effects of this invention are as follows: This invention improves the real-time monitoring of voltage transformer insulation status through online fusion acquisition of multi-source data; it achieves continuous, dynamic, and comprehensive monitoring of the equipment's insulation health status. It significantly enhances the ability to detect insulation degradation and potential anomalies in a timely manner, thus improving the ability to detect early-stage fault hazards.

[0022] This invention possesses adaptability to complex operating conditions and the ability to automatically identify multiple types of insulation anomalies, improving the accuracy and robustness of anomaly detection and reducing missed detections and false alarms. Through causal reasoning and anomaly evolution path analysis methods, it identifies dominant risk factors and their mechanisms of action, enhancing the scientific rigor of anomaly cause analysis and the interpretability of early warning decisions.

[0023] This invention introduces an expert-model interaction and self-evolutionary optimization mechanism to integrate expert experience with data models, enabling the system to continuously learn and optimize itself, and constantly adapt to new anomaly types and complex operating environments, thereby improving the intelligence level of the monitoring and early warning system. Attached Figure Description

[0024] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0025] Figure 1 The above is a flowchart of a method for analyzing the influencing factors of voltage transformer status, provided as an embodiment of the present invention.

[0026] Figure 2 This is a decision tree model diagram of a method for analyzing the influencing factors of voltage transformer status, provided in one embodiment of the present invention.

[0027] Figure 3 The present invention provides a simulation experiment flowchart for a method for analyzing the influencing factors of voltage transformer status, as an embodiment of the present invention.

[0028] Figure 4 The output waveform diagram is provided by a voltage transformer state influencing factor analysis method according to an embodiment of the present invention.

[0029] Figure 5 The diagram illustrates the influence of temperature and humidity on the output voltage of a voltage transformer, as provided in an embodiment of the present invention, for analyzing the factors affecting the state of a voltage transformer. Detailed Implementation

[0030] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0031] Example 1, referring to Figure 1 and Figure 2 This is one embodiment of the present invention, which provides a method for analyzing the influencing factors of voltage transformer status, including: S100: Collect the state parameters of the voltage transformer, preprocess the state parameters, and construct a multi-dimensional feature tensor to capture the nonlinear coupling relationship between features; S200: Analyzes voltage fluctuation characteristics of voltage transformers, identifies extreme value anomaly areas in real time, assesses the current electric field distribution of transformers, and automatically corrects the insulation status evaluation model. S300. Input the abnormal feature vectors obtained from monitoring into the recursive attribution decision tree model, perform multidimensional anomaly clustering and causal analysis, generate a tracing report, and perform multidimensional evaluation and verification of the analysis results. It should be noted that traditional voltage transformer condition monitoring relies on manual inspections and periodic tests, resulting in low data acquisition frequency and limited coverage, making it difficult to promptly detect early hidden dangers and dynamic anomalies during operation. Existing methods use empirical thresholds and single physical quantity monitoring, lacking collaborative analysis and deep coupling modeling of multiple sources of environmental load voltage influencing factors. Furthermore, they lack sufficient understanding of insulation degradation mechanisms under complex operating conditions.

[0032] Therefore, addressing the aforementioned problems, this invention, through steps S100-S300, proposes a system for intelligent assessment and anomaly early warning of voltage transformer insulation status. This system encompasses multi-source monitoring data acquisition, comprehensive environmental and load health aggregation, intelligent correction of voltage extreme disturbances and electric field distribution, anomaly causal attribution, and expert collaborative self-evolutionary optimization. Through high-frequency sensing, hierarchical data fusion, physical-statistical coupling modeling, causal decision analysis, and expert knowledge transfer, it achieves highly sensitive dynamic identification of early-stage minor anomalies in insulation degradation. This improves the accuracy, interpretability, and operational efficiency of intelligent diagnostics for power equipment, meeting the high standards required for insulation risk management in complex environments under smart grids.

[0033] Example 2, refer to Figure 1 and Figure 2 This is one embodiment of the present invention, which provides a method for analyzing the influencing factors of voltage transformer status, including: In this embodiment of the invention, step S100 involves acquiring the state parameters of the voltage transformer and preprocessing these parameters to construct a multidimensional feature tensor to capture the nonlinear coupling relationship between features. This includes the following steps S101-S102: S101. In the implementation of intelligent aggregation and weighted evolution modeling of environmental multi-factors, it is first necessary to synchronously and in real time collect temperature T, humidity H, ultraviolet intensity U, and air pollutant concentrations such as PM2.5 and sulfides at the operating site of the voltage transformer based on various sensor systems. and the volume fraction of the surrounding specific gases. Environmental data is used to form a multidimensional environmental observation vector. .

[0034] In view of the fact that the weights of various factors change significantly with time, space, equipment status, and environmental interactions during actual operation, this invention abandons the traditional static weighting method and proposes a hierarchical dynamic correlation aggregation algorithm: Based on historical insulation performance data (such as insulation resistance and leakage current) and the sequences of various environmental factors, a sliding correlation matrix within a time window is constructed. Used to measure the impact of the i-th environmental factor on the insulation status over the past w time period. The influence strength is defined by the weight of each factor. Its dynamic evolution follows the following: in, Forgetting factor, Control the relative importance of historical and current relevance.

[0035] In an embodiment of the present invention, S102, all parameters are normalized and weighted to calculate environmental health indicators, including the following steps A1-A2: A1. To adapt to the interaction and feedback between the environment and equipment, this invention sets up a weight self-adjustment mechanism. When a sudden abnormality is detected in a certain factor, such as a sudden increase in pollutants causing a sharp drop in insulation performance, the weight of the factor is automatically increased to reflect its actual impact. A2. Calculate environmental health indicators by normalizing and weighting each environmental factor using weights: in, The standardized value of the i-th factor enables dynamic aggregation of high-dimensional environmental states. This aggregation index not only tracks the changing trends of each factor in real time, but also fully captures risk signals through a weighted self-adjustment mechanism when new or complex environmental combinations occur, thereby greatly improving the intelligent monitoring level of internal insulation hazards and environmental risk sources of voltage transformers.

[0036] In an optional implementation, the normalized weighted aggregation in S102 can be a fixed-weighted aggregation. Based on historical data or expert experience, a fixed weight value is preset for each environmental factor. These weight values ​​reflect the relative importance of each factor to the insulation state of the voltage transformer and are not adjusted with time and environmental changes. The environmental factor data are standardized to eliminate the influence of dimensions, and then the standardized factor values ​​are directly weighted and summed using the fixed weights to obtain the environmental health index. However, this implementation cannot adapt to the dynamic changes of environmental factors.

[0037] In another optional implementation, the normalized weighted synthesis in S102 can also be a moving average weighted synthesis. Based on historical data, the moving average of the correlation between each environmental factor and the insulation state is calculated periodically as the weight. The weight is updated at a low frequency rather than evolving in real time. In real-time monitoring, the data of each environmental factor is first standardized, and then the standardized factor values ​​are weighted and summed using the latest updated moving average weight to obtain the environmental health index. However, this implementation cannot respond to sudden environmental events in real time.

[0038] In summary, various environmental parameters are collected from the operating site of the voltage transformer, including temperature, humidity, ultraviolet radiation, air pollutant concentration, and ambient gas composition. Addressing the shortcomings of traditional weighting methods that rely on blindly static weighting, a novel algorithm based on hierarchical dynamic correlation aggregation is proposed. This algorithm dynamically calculates the influence weight of each environmental factor by utilizing the coupled evolution relationship between historical environmental data and insulation status. Furthermore, a real-time self-adjustment mechanism is established based on the interaction and feedback between environmental and equipment status. This allows for the automatic identification of the risk manifestations of emerging environmental combinations, achieving high-dimensional environmental health factor aggregation and dynamic characteristic modeling, thereby improving the monitoring sensitivity of insulation risk sources.

[0039] In this embodiment of the invention, step S200 involves analyzing the voltage fluctuation characteristics of the voltage transformer, identifying extreme value anomaly areas in real time, evaluating the current electric field distribution of the transformer, and automatically correcting the insulation state evaluation model. This includes the following steps S201-S202: S201, Current on the secondary side of the transformer Real-time waveform data is acquired at a high sampling rate and decomposed into fundamental frequency components and several higher harmonic components, denoted as: in, and The first The amplitude and initial phase of the second harmonic. Where is the fundamental frequency, N is all the collected harmonic data, and k is the variable index; Synchronous extraction of power factor Total Harmonic Distortion and the proportion of typical nonlinear components in the current load. Multiple sets of associated features are assembled into a multidimensional load feature sequence tensor. ; To handle the nonlinear composite relationships between features, a multi-feature probability tensor decomposition method is adopted, which minimizes the following objective function: Where U, V, and W are the factor matrices to be trained. Represents the tensor outer product. Let be the coupling strength of the r-th feature group; U, V, and W are the factor matrices to be trained. , , Let R represent the r-th eigenvectors of the factor matrices U, V, and W respectively, where R is the total number of features, i.e., the rank. Represents the square of the Frobenius norm; In an embodiment of the present invention, the decomposition process includes the following steps B1-B2: B1. Through tensor decomposition, i.e. Tucker decomposition, training is performed to organize multi-source monitoring data (different types, such as electrical parameters, environmental parameters, and operating conditions) according to a unified tensor structure to construct a high-order data tensor. B2. Based on the objective function of minimizing the reconstruction error, the factor matrices U, V, and W are trained using an optimization algorithm (Stochastic Gradient Descent SGD). B3. During the training process, the elements of the factor matrix are repeatedly updated to ensure that the factor matrix decomposed can reconstruct the original multi-source data tensor as accurately as possible through tensor multiplication, reflecting the coupling and dynamic evolution relationship between various features, thereby extracting the most representative low-dimensional feature representation for anomaly identification and state assessment.

[0040] In an optional implementation, the decomposition process in S201 can be based on the feature extraction scheme of CP decomposition. CP decomposition decomposes the tensor into a linear combination of multiple rank-tensors, thereby extracting simple linear coupling relationships between features. By minimizing the objective function of reconstruction error, the Alternating Least Squares (ALS) method is used for optimization, and the elements of the factor matrix are iteratively updated to obtain low-dimensional feature representations from the decomposed factor matrix. However, the CP decomposition in this implementation assumes linear relationships between features, which cannot fully capture the complex nonlinear coupling characteristics of voltage transformers, resulting in decreased identification accuracy under highly dynamic load environments.

[0041] In another alternative implementation, the decomposition process in S201 can also be based on a feature extraction scheme of nonnegative tensor decomposition. The nonnegative tensor decomposition method is applied to decompose the tensor into a product of nonnegative factor matrices, emphasizing the additive combination relationship between features. The multiplicative update algorithm is used for optimization, and the elements of the factor matrix are iteratively adjusted to ensure that the reconstructed tensor approximates the original data. Low-dimensional features are extracted from the nonnegative factor matrix for analyzing the positive coupling effect between voltage transformer features. However, nonnegative constraints may not be able to handle negative correlation or negative value data in voltage fluctuation features.

[0042] It should be noted that, The coupling strength of the r-th group of features is a parameter that measures the correlation and synergy of multi-source features on the r-th dimension of the latent factor. Its data source comes from statistical analysis of various types of raw monitoring data and adaptive adjustment of the model during training; Specifically, The anomaly detection performance can be determined by observing the correlation of each feature across the factor dimensions after tensor decomposition and its contribution to the overall anomaly detection effect. In actual training, Initial values ​​are often obtained automatically by the model, or they can be set and constrained with the help of some prior knowledge or expert experience, so as to better reflect the actual physical or working condition coupling relationship between various types of data.

[0043] Tensor decomposition enables high-dimensional probabilistic mapping of load states, restores nonlinear responses, and reveals sensitive areas and vulnerabilities in transformer insulation under multi-feature coupling. Based on tensor decomposition results and anomaly response fusion criteria, a recursive enhanced anomaly detection function is defined. as follows: in To decompose the constructed prediction tensor, For characteristic variance, recursive factor Control the memory weight of historical outlier scores; whenever If the set dynamic threshold is exceeded, the system will determine it as a minor anomaly induced by the load and activate the risk-sensitive interval tracking mechanism. The continuously recursive anomaly score can provide early warning of potential threats to the insulation and measurement accuracy of the instrument transformer under complex working conditions such as load step changes and high-order harmonic surges, effectively realizing sensitive and intelligent identification of faults affected by hidden loads.

[0044] It should be noted that, The prediction tensor, constructed by decomposition, is obtained based on historical multi-source monitoring data through tensor decomposition and factor matrix reconstruction: By constructing a training set using various types of data collected in the past, the factor matrices (such as U, V, W) in the tensor decomposition model are trained to obtain a low-rank structure that can characterize the coupling relationship between features. When it is necessary to predict the data state at the current or future moment, the tensor at the target moment is reconstructed using the learned factor matrix and existing partial observation data to obtain... ; This predictive tensor reflects the normal evolution trend of multi-source features under the current historical conditions, providing a reference for subsequent anomaly detection and early warning.

[0045] It should also be noted that whenever (Usually refers to the actual observed tensor and the predicted tensor) When the residuals, abnormal measurements, or deviations between the values ​​exceed the set dynamic threshold, the system will issue a warning. The setting of dynamic thresholds is based on historical data statistical analysis, adaptive adjustment to changes in operating conditions, and model self-learning adjustment. Using historical residual data, the mean and standard deviation are statistically calculated, and thresholds are set based on experience or confidence intervals. The threshold parameters are updated in real time, enabling them to dynamically adjust according to changes in the environment and load conditions, thereby enhancing their adaptability to sudden anomalies and trend evolution.

[0046] S202. Using a high-precision sensor to measure the operating voltage of the voltage transformer. Full-time high-frequency continuous data acquisition is performed to obtain voltage time series sequences. The voltage time series sequences are then divided into intervals, and time windows are set. Extract the maximum value from each small segment. and minimum value The piecewise extreme value perturbation is obtained. ; Continuous detection based on sliding window method The statistical characteristics of the changes, combined with the distribution drift criterion. or ,in, and for The mean at time t and time t-1, and for The standard deviations at time t and time t-1 and This is the sensitivity threshold; it allows for the timely detection of abnormal voltage extreme value disturbances. For the detected voltage disturbance range, input the current voltage distribution. State of isolation from history Finite element simulation model is used to evaluate the electric field distribution in key parts of the current transformer. Where x is the spatial coordinate, the change in electric field is compared. Breakdown or aging threshold of insulating material The relationship between the device and the local insulation critical risk zone is dynamically identified. When any local electric field peak value If the threshold is continuously exceeded, the weight is adjusted based on the extreme value perturbation interval under drift analysis, and the insulation state evaluation function is updated in real time. in, For the corrected insulation health, The intensity coefficient of the linkage between disturbance and distribution. This is a comprehensive response function based on statistics and physical fields; Once a small voltage fluctuation is detected leading to a rapid deterioration in insulation condition, the system significantly enhances sensitivity to risks in that time zone and location. Continuous data feedback and iterative model corrections improve the ability to detect early signs of insulation degradation and provide accurate early warnings of critical insulation states under minor disturbances.

[0047] It should be noted that, The determination of the (disturbance and distribution linkage strength coefficient) depends on an in-depth analysis of the interaction between disturbances (load fluctuations, environmental changes) and the distribution of equipment insulation status in multi-source monitoring data: First, statistical analysis of historical monitoring data was conducted to explore the impact of external disturbances such as load rate and environmental humidity on changes in insulation health. The factor matrix obtained by tensor decomposition is used to quantify the contribution and coupling strength of each perturbation feature and health status factor in different dimensions. Based on statistical results and model output, combined with expert experience and actual on-site working conditions, Set appropriate initial values; With the continuous accumulation of data The parameters can be adaptively adjusted through the model to dynamically reflect the latest disturbance-response distribution, ensuring the accuracy and robustness of insulation health assessment.

[0048] It should also be noted that, The (comprehensive response function based on statistics and physical fields) is a comprehensive mathematical expression that incorporates statistical laws and the physical degradation mechanism of insulation in this invention. The determination process is as follows: Based on historical multi-source monitoring data, a response relationship model between disturbance characteristics and health indicators was constructed using correlation analysis, regression modeling and statistical methods. Based on the theoretical mechanism of typical physical processes of insulation aging and moisture absorption in voltage transformers, the variation law of key parameters such as dielectric loss factor and leakage current is incorporated into the model in the form of functions. The function is expressed as a weighted linear, nonlinear, or mechanism-driven composite mapping. By integrating the above statistical laws with physical mechanisms, the insulation health assessment has both data-driven adaptability and a physical interpretation basis, improving the model's ability to express and predict abnormal evolution under different working conditions.

[0049] In summary, this study analyzes multiple features of the secondary load's real-time waveform, including power factor, harmonic distortion, and the proportion of nonlinear load. A multi-feature tensor decomposition and anomaly response fusion algorithm is designed to map the complex influence of the secondary load on the transformer's performance into a high-dimensional probabilistic characteristic space, automatically extracting minor anomalies caused by load fluctuations. The algorithm incorporates a recursive enhancement mechanism for anomaly points, enabling the identification of potential impacts on the transformer's internal insulation and measurement accuracy under rapid load changes or sudden high-order harmonic occurrences, achieving sensitive prediction of faults induced by hidden loads.

[0050] In an embodiment of the present invention, in step S300, the abnormal feature vectors obtained from monitoring are input into a recursive attribution decision tree model to perform multidimensional anomaly clustering and causal analysis, generate a tracing report, and perform multidimensional evaluation and verification of the analysis results, including the following steps S301-S302: S301. In an embodiment of the present invention, multidimensional anomaly clustering and causal analysis are performed, including the following steps C1-C2: C1, which will be characterized by the aggregation of environmental health features Secondary load anomaly score and voltage extreme value disturbances and the corrected insulation state The results of multi-source monitoring together form a multidimensional anomaly feature vector. The input is fed into a recursive attribution decision tree model, with the root node set as the overall anomaly rating variable. The tree structure transforms various input features into branch judgments, with each node based on a feature threshold. Recursive partitioning: in, , , , , belong ; C2. During the recursive attribution process, each branch node stores the causal path. The information was used to deduce the dominant factor sequence that led to the anomaly; In an optional implementation, the anomaly clustering in S301 can be based on K-means clustering and association rule analysis to collect multidimensional anomaly feature vectors composed of environmental health indicators, load anomaly scores, voltage extreme value disturbances, and corrected insulation states. The feature vectors are grouped using the K-means clustering algorithm, the number of clusters is preset, and the vectors are divided to the nearest cluster center through iterative calculation. The feature mean distribution of each cluster is analyzed to identify the dominant risk factors and list the anomaly dominant factor sequence. However, this implementation is sensitive to the selection of the initial center and cannot capture complex nonlinear relationships.

[0051] In another alternative implementation, the anomaly clustering in S301 can also be based on hierarchical clustering and a decision list. Hierarchical clustering is used to calculate the similarity between feature vectors, which are then gradually merged to form a tree-like clustering structure. The data is divided into multiple clusters according to the cutting threshold of the tree structure, with each cluster corresponding to a risk category. A decision list is constructed for each cluster, and the risk type is determined based on the order threshold of the feature values. However, this is not suitable for real-time processing of large-scale data.

[0052] To implement the anomaly clustering attribution model, a causal path inversion algorithm is employed. After completing the preliminary steps of multi-source feature fusion modeling, dynamic anomaly detection, and health assessment, the algorithm maps the anomaly time-series trajectories of all features into a causal directed graph. ,node Corresponding to the multi-source monitoring features extracted during modeling, the edges This is reflected in the causal connections between risk factors based on anomaly identification and dynamic health assessment; Specifically, the correlation strength is quantified based on the feature coupling parameters obtained from the tensor decomposition above, the probability of abnormal co-occurrence, and the responsiveness of perturbation factors to changes in health status.

[0053] Based on correlation indicators, the maximum a posteriori probability method is used to automatically infer causal risk paths and accurately identify the risk transmission chain that dominates insulation degradation. This enables a structured expression of abnormal evolution mechanisms and quantitative identification of main risk sources, providing a solid theoretical and data foundation for subsequent equipment health management, risk warning, and intervention decisions.

[0054] In an embodiment of the present invention, generating a cause-of-fact report includes the following step D1: D1. The model generates a cause-of-fact report for each anomaly, and the report is based on the decision-making path. The critical path visualization in the cause-effect graph G shows the attribution of responsibility, the evolution of risks, and possible triggering scenarios, providing maintenance engineers with vivid and intuitive suggestions for tracing the root causes and handling them.

[0055] In an optional implementation, the attribution report generated in S301 can be based on the attribution report generation of the rule engine. A set of static rules is defined in advance based on expert experience and historical data to map the threshold combination of abnormal feature vectors to specific risk types. When a multidimensional abnormal feature vector is input, the rule engine matches the preset rules one by one and determines the abnormal category and dominant factor according to whether the feature value exceeds the threshold. For the matched rules, a text-based attribution report is automatically generated, which includes the attribution of responsibility, a brief analysis of risk evolution and possible triggering scenarios, and is output in the form of a structured list. However, it is only suitable for a single abnormal scenario.

[0056] In another optional implementation, the cause-finding report generated in S301 can also be based on clustering and association rules. Unsupervised clustering is performed on historical abnormal feature vectors to identify common abnormal pattern categories. Association rule learning is used to extract frequently co-occurring feature combinations from the clustering results to form causal association rules. When a new abnormal feature vector is input, its distance to each cluster center is calculated, and it is assigned to the nearest cluster. The cause-finding report is generated according to the association rules of that cluster, describing the dominant factor sequence and risk chain in text. However, the response to sudden new abnormal patterns is delayed.

[0057] By using automated decision tree node interpretation and path highlighting, the model can accurately classify various typical and complex anomalies and generate traceable cause reports, significantly improving the efficiency and scientific nature of on-site anomaly diagnosis and achieving closed-loop intelligent management and control of the insulation risk of instrument transformers throughout the entire process.

[0058] S302. Construct an expert-model collaborative interaction and annotation platform, and conduct preliminary analysis of the anomaly types output by the model. The results are automatically pushed to domain experts, who evaluate and label the output based on their experience and historical cases: the actual category is... (Including load-dominant risk and composite risk) and risk level: Manual annotations are made on key causal paths or risk triggers, and interactive data is presented in tuples. Formal recurrent feedback is fed into the training queue; The model employs a self-evolutionary algorithm that combines expert experience transfer and feedback reinforcement to continuously absorb newly acquired human experience and knowledge. The definition of the comprehensive loss function aims to achieve multiple collaborative optimizations of the model for automatic data feature discrimination and expert knowledge feedback.

[0059] During model training and updating, all newly acquired expert-annotated samples are added to the training set to dynamically expand the knowledge base. The model parameters are optimized by minimizing the comprehensive loss function. Iterative updates, the comprehensive loss function is defined as: in, The loss represents the difference between the model output and the expert-annotated categories. The consistency between the measurement model's judgment level and the expert's implicit risk perception. For model parameters The regularization term is L2 regularized. Adjust the weighted hyperparameters according to the relative importance of the loss in the overall objective.

[0060] The loss function consists of three parts: The first item ( This is used to measure the error between the model's automatic anomaly identification and the expert-labeled true categories, ensuring the continuous improvement of the model's basic discrimination ability; The second item ( This is to measure the degree of matching between the anomaly type output by the model and the risk level rated by the expert, so that the model can gradually learn the implicit rules of the expert's classification of risk severity. The third item ( ) represents the regularization penalty for model parameters, which limits model complexity and prevents overfitting.

[0061] Weighting coefficient Adjust the relative importance of the above three parts of the loss in the overall objective.

[0062] Furthermore, an experience transfer mechanism is introduced: for heuristic rules manually summarized by experts in typical complex scenarios, differentiable rule embedding operators (such as knowledge distillation or soft logic rule priority) are used to supplement the criteria in a timely manner. By combining data-driven and expert input as dual-track reinforcement, the model uses Bayesian model updates or stepwise regression based on historical learning performance to continuously optimize the parameter space and decision boundary, forming a "experience-model-feedback" three-element cycle, thereby achieving rapid adaptation and high-precision identification of new anomalies in any complex scenario; The model continuously evolves, combining the advantages of human knowledge and big data statistics, and continuously improves the accuracy, adaptability and robustness of anomaly identification and mechanism analysis, effectively meeting the needs of highly intelligent and interpretable decision support in the complex and ever-changing operating environment of smart grids.

[0063] Example 3, referring to Figures 3-5 This invention provides a method for analyzing factors affecting the state of a voltage transformer, and scientific demonstration is conducted through experiments to verify the beneficial effects of the invention.

[0064] Experimental objectives: To simulate the operating status of voltage transformers under the influence of various environmental factors; and to test the ability to monitor abnormal risks and trace root causes based on a hierarchical dynamic correlation aggregation algorithm.

[0065] Data collected: 1. Environmental data, collected hourly for 7 days, totaling 168 records; Temperature (T): 20~40°C; Humidity (H): 30%~95%; UV intensity (U): 100~900; PM2.5 concentration (P1): 10~250; SO2 concentration (P2): 0.005~0.07; Gas G1 (methane): 0%~1%; Gas G2 (hydrogen sulfide): 0%~0.1%; Table 2 Experimental Environment Data Collection

[0066] 2. Insulation condition data: Insulation resistance (MΩ): 5~100; Leakage current (μA): 0.5~10; Table 3 Insulation Data

[0067] 3. Mark the abnormal event segment: 60~72 hours: extreme high temperature + high humidity + high SO2, set as abnormal segment.

[0068] like Figure 3 The simulation flowchart shown illustrates the simulation analysis: Before performing live disconnection of the lead wire, the voltage transformer is in its initial state, at which point its main function is to measure or monitor the voltage level in the power grid. During this stage, the transformer's output signal is relatively stable, electromagnetic interference is low, and the equipment operates within its design parameter range, accurately reflecting the voltage state of the power grid. The voltage transformer's design enables it to provide reliable voltage measurements under normal operating conditions, providing crucial data for power system monitoring and protection.

[0069] The operating state of a voltage transformer changes significantly during operation, which affects its performance and measurement accuracy.

[0070] At the start of live disconnection of the drain line, the voltage transformer transitions from an initial steady state to an intermediate state. The transformer is affected by various factors such as electromagnetic interference, temperature changes, and mechanical vibration, causing fluctuations in the output signal. Electromagnetic interference originates from the redistribution of current during the disconnection operation, generating additional electromagnetic fields that interfere with the transformer's internal magnetic field.

[0071] Temperature changes can also cause changes in the resistance and capacitance parameters of the internal components of the transformer, affecting output stability.

[0072] like Figure 4 Mechanical vibration can also cause poor contact or structural deformation of the current transformer, further affecting its performance.

[0073] During the transition period, the measurement accuracy of the voltage transformer decreases, leading to increased output signal error. This is because the transformer is sensitive to environmental changes, such as variations in temperature and humidity.

[0074] like Figure 5 Poor contact or structural deformation due to mechanical vibration may also cause instability in the output signal of the current transformer during the transient phase, resulting in momentary fluctuations or continuous deviations.

[0075] Instrument transformers require a certain amount of time to adapt to new operating environments and conditions. They exhibit unstable characteristics until they stabilize in response to voltage changes in the power grid. This adjustment period is crucial to the transformer's performance, determining its ability to quickly adapt to changes in the operating environment and return to a stable state.

[0076] Instrument transformers can self-adjust or be adjusted via external control to adapt to new operating conditions. This includes adjusting internal circuit parameters, improving shielding measures, or enhancing structural stability. While the performance of the instrument transformer may decrease, adjustments and optimizations can minimize this impact and ensure that the transformer reaches a new stable state as quickly as possible.

[0077] After adjustment and adaptation, the voltage transformer will reach a new stable state. The transformer's output signal tends to stabilize, accurately reflecting the voltage state of the power grid. At this time, the transformer's performance parameters are slightly different from the initial state, but remain within an acceptable error range.

[0078] A voltage transformer in operation at a 110kV substation is equipped with the multi-source data real-time monitoring terminal of this invention to collect three types of data: electrical parameters, environmental parameters, and operating conditions. This invention's method is used to fuse and analyze these data, enabling early warning of insulation degradation anomalies.

[0079] Table 3. Collected multi-source monitoring data

[0080] Multi-source data collection and preliminary screening were carried out. Terminal devices collected various parameters at regular intervals. Outliers and missing values ​​were removed through data consistency and validity checks, and data tables were generated as shown in Table 3.

[0081] Data fusion and dynamic trend analysis are performed using the data fusion algorithm of this invention to jointly analyze electrical, environmental, and operating condition data, with a focus on the dynamic changes in dielectric loss factor, load rate, and ambient humidity.

[0082] Intelligent anomaly identification and early warning are performed by introducing a set threshold and a self-learning model (e.g., the dielectric loss factor early warning threshold is 0.70%). The system detects that the dielectric loss factor collected at 10:50 and 11:00 is continuously exceeding the standard, and the load and humidity are both showing an upward trend. The model judges that an "insulation degradation early warning" signal has been issued.

[0083] Cause attribution and handling recommendations: System tracing analysis revealed that the anomaly was accompanied by an increase in load rate and humidity, indicating that the main influencing factors were environment and operating conditions. It is recommended that maintenance personnel focus on checking the terminal sealing condition and insulation moisture.

[0084] Expert interaction confirms and inspections are optimized, and technical experts are notified for manual review and confirmation. The system guides on-site drying of the transformers and inspection of sealing points. At the same time, this type of abnormal data is fed back to the system to optimize model parameters. Through this embodiment, the system achieves timely and accurate early warning of insulation degradation, effectively guides operation and maintenance, realizes intelligent and closed-loop abnormal management, and significantly improves equipment operation safety and operation and maintenance efficiency.

[0085] Example 4 is an embodiment of the present invention, and the above is a schematic scheme of a method for analyzing the influencing factors of voltage transformer condition. It should be noted that the technical solution of a voltage transformer condition influencing factor analysis system and the technical solution of the voltage transformer condition influencing factor analysis method described above belong to the same concept. Details not described in detail in the technical solution of the voltage transformer condition influencing factor analysis system in this embodiment can be found in the description of the technical solution of the voltage transformer condition influencing factor analysis method described above.

[0086] This embodiment provides a voltage transformer state influencing factor analysis system, including: a data acquisition module, a correction module, and an analysis module; The acquisition module acquires the state parameters of the voltage transformer, preprocesses the state parameters, and constructs a multi-dimensional feature tensor to capture the nonlinear coupling relationship between features. The correction module analyzes the voltage fluctuation characteristics of the voltage transformer, identifies extreme value anomaly areas in real time, assesses the current electric field distribution of the transformer, and automatically corrects the insulation status evaluation model. The analysis module inputs the abnormal feature vectors obtained from monitoring into the recursive attribution decision tree model to perform multidimensional anomaly clustering and causal analysis, generate a tracing report, and conduct multidimensional evaluation and verification of the analysis results.

[0087] This embodiment also provides an electronic device applicable to a method for analyzing the influencing factors of voltage transformer status, comprising: 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 method for analyzing the influencing factors of voltage transformer status as proposed in the above embodiment.

[0088] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements a method for analyzing the influencing factors of voltage transformer status as proposed in the above embodiments.

[0089] The storage medium proposed in this embodiment and the method for analyzing the influencing factors of voltage transformer status proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0090] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0091] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not 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 voltage transformer state influencing factor analysis method, characterized by: The method comprises the following steps: Collecting state parameters of the voltage transformer and preprocessing the state parameters to construct a multi-dimensional feature tensor to capture the nonlinear coupling relationship between features; Analyzing voltage fluctuation characteristics of the voltage transformer, identifying extreme abnormal areas in real time, evaluating the electric field distribution of the current transformer, and automatically correcting the insulation state evaluation model; Inputting the abnormal feature vector obtained by monitoring into a recursive causal decision tree model to perform multi-dimensional abnormal clustering and causal analysis, generating a traceability report, and performing multi-dimensional evaluation and verification on the analysis results.

2. The voltage transformer condition influencing factor analysis method of claim 1, wherein: The preprocessing includes collecting environmental parameters of the voltage transformer operating site and performing improved aggregation processing to construct a correlation matrix to measure the influence intensity of environmental parameters on the insulation state in the past period; A weight self-adjusting mechanism is set to adjust the current parameter weight when any parameter is abnormal, and all parameters are normalized and weighted to calculate the environmental health index.

3. The voltage transformer condition influencing factor analysis method of claim 2, wherein: The method further comprises the following steps: Synchronously extracting associated features to assemble a multi-dimensional load feature sequence tensor.

4. The voltage transformer condition influencing factor analysis method of claim 3, wherein: The method further comprises the following steps: Using tensor decomposition to perform low-rank modeling on the multi-dimensional load feature sequence tensor; 5. The voltage transformer condition influencing factor analysis method of claim 4, wherein: The automatic correction insulation state evaluation model comprises the following steps: using a high-precision sensor to detect the operating voltage of the voltage transformer Collecting continuously at high frequency in the whole period to obtain a voltage time sequence, dividing the voltage time sequence into intervals, and setting a time window Extracting the maximum value on each small section And the minimum value Obtaining a segmented extreme disturbance ; Continuous detection of statistical property changes based on a sliding window approach in combination with a distribution drift criterion or wherein and are the mean values at time t and time t-1, and are the standard deviations at time t and time t-1, and are sensitivity thresholds;​ For the detected voltage disturbance interval, input the current voltage distribution and historical insulation state , evaluate the current mutual inductor key site electric field distribution by using finite element simulation model , wherein x is the spatial coordinate, by comparing the electric field change amount with the relationship of insulation material breakdown or aging threshold , dynamically identify the local insulation critical risk area of the device: when any local electric field peak continues to exceed the threshold, combine the extreme disturbance interval under the drift analysis to correct the weight, and update the insulation state evaluation function in real time: wherein, is the corrected insulation health, is the disturbance and distribution linkage intensity coefficient, is the comprehensive response function based on statistics and physical field; Minimizing reconstruction error and capturing nonlinear coupling relationships between features.

6. The voltage transformer condition influencing factor analysis method of claim 5, wherein: The multidimensional anomaly clustering and causal analysis includes clustering features based on environmental health. Secondary load anomaly score and voltage extreme disturbances and the corrected insulation state The results of multi-source monitoring together form a multidimensional anomaly feature vector. The input is fed into a recursive attribution decision tree model, with the root node set as the overall anomaly rating variable. The tree structure transforms various input features into branch judgments, with each node based on a feature threshold. Recursive partitioning: wherein, , , , , belongs to ; each branch node in the recursive attribution process saves the information of the causal path , and the leading factor sequence causing the anomaly is reversed; the analysis maps all the abnormal time series trajectories of the characteristics into a causal directed graph , and the nodes correspond to the multi-source monitoring characteristics extracted during modeling, and the edges reflect the risk transmission chain of the leading insulation degradation based on the correlation index, the maximum posterior probability method for inferring the causal risk path, and the acquisition of the risk transmission chain of the leading insulation degradation. The model generates a traceability report for each anomaly, reporting the decision path and the key paths in the causal graph G visualize the responsibility attribution, risk evolution, and possible triggering scenarios.

7. The voltage transformer condition influencing factor analysis method of claim 6, wherein: The multi-dimensional evaluation and verification of the analysis result comprises constructing an interactive and labeling platform of expert-model cooperation, and the model preliminarily analyzes the output abnormal type The automatic pushing is performed to a field expert, the expert evaluates and labels the output result based on own experience and historical cases, the real category is and the risk level is , and manual annotation is made on a key causal path or a risk trigger factor, and the interactive data is fed back in the form of a tuple into a training queue. In the model training and updating process, all newly obtained expert labeled samples are supplemented into the training set, the knowledge base is dynamically expanded, and the model parameters are minimized by minimizing the comprehensive loss function Iterative update, the comprehensive loss function is defined as: wherein, a difference loss between the model output and the expert-labeled class, a consistency loss measuring agreement between the model decision class and the expert implicit risk perception, a regularization term for the model parameters using L2 regularization, weight hyperparameters respectively adjusting the relative importance of the losses in the overall objective.

8. A voltage transformer state influencing factor analysis system applying the voltage transformer state influencing factor analysis method according to any one of claims 1 to 7, characterized by, When it is detected that voltage fluctuation leads to a deterioration trend of the insulation state, the sensitivity to the current time zone and part risk is improved. The method comprises the following steps: A collection module collects state parameters of the voltage transformer and preprocesses the state parameters to construct a multi-dimensional feature tensor to capture the nonlinear coupling relationship between features; A correction module analyzes voltage fluctuation characteristics of the voltage transformer, identifies extreme abnormal areas in real time, evaluates the electric field distribution of the current transformer, and automatically corrects the insulation state evaluation model; 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-8 when the computer program is executed by the processor. An analysis module inputs the abnormal feature vector obtained by monitoring into a recursive causal decision tree model to perform multi-dimensional abnormal clustering and causal analysis, generates a traceability report, and performs multi-dimensional evaluation and verification on the analysis results.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The processor executes the computer program to realize the steps of the voltage transformer state influencing factor analysis method according to any one of claims 1 to 7. The computer program is executed by the processor to realize the steps of the voltage transformer state influencing factor analysis method according to any one of claims 1 to 7.

Images