Power transformer insulation characteristic data analysis and performance prediction method and system

Abstract

The invention belongs to the technical field of power transformer insulation design and verification, and particularly relates to a power transformer insulation characteristic data analysis and performance prediction method and system, and the method comprises the steps: constructing a structured sample set containing multi-dimensional design parameters and lightning impulse response tags based on a three-phase differential prototype; identifying key variables in the structured sample set and extracting physical consistency constraints; and inputting the key variable and the physical consistency constraint into a preset XGBoost prediction model, and outputting a prediction value and a confidence interval. According to the method, the influence of multi-dimensional parameters on the insulation impact indexes is disassembled into clear main effects and interaction effects, and a traceable mechanism map is formed. Compared with a traditional method which depends on single-point experience judgment or post trial and error, the method has the advantages that standardized data evidences and explainable output are provided, and designers can conveniently and quickly position key constraints, formulate optimization priorities and directly backfill results to design specification and parameter boundaries.

Description

Technical Field

[0001] This invention belongs to the field of power transformer insulation design and verification technology, specifically relating to a method and system for power transformer insulation characteristic data analysis and performance prediction. Background Technology

[0002] The insulation impulse performance of power transformers is typically assessed using 1.2 / 50 μs full-wave and sheared-wave tests according to IEC 60076-3 and GB / T 1094.3 to obtain indicators such as U50%, BIL, and end overvoltage ratio. Equivalent circuits or field-circuit coupling models are then used to evaluate end field strength and wavefront propagation. While this paradigm is mature in engineering, it has significant limitations in the design phase: constrained by the "three-phase consistency" requirement and limited prototype manufacturing time, a single test cannot cover multiple variables such as axial / radial / end gaps, shielding and equalizing ring dimensions, cardboard thickness, dielectric constant and dielectric loss, moisture content, and temperature. Measurement links are primarily based on voltage divider and current channels, with sparse measurement points and high time synchronization requirements. Inconsistent data caliber across batches makes it difficult to form a structured sample set that can be directly used for modeling, resulting in high evaluation costs, long cycles, and difficulties in comparing parallelization schemes.

[0003] More importantly, the parameter mechanisms of lightning impulse response are complex and unclear. Under high-frequency transients, the combined effects of winding distributed capacitance, lead parasitics, three-dimensional edge fields, and the dispersion / nonlinearity of the oil-paper dielectric mean that the influence of geometric clearance and dielectric parameters on U50% / BIL is not a simple monotonic relationship. There are strong correlations and significant interactions between variables such as end shield size and axial clearance, dielectric loss and water content, and lead path and field strength gradient. Traditional single-factor or empirical fitting methods are insufficient to identify main effects and coupling effects, and it is difficult to provide verifiable mechanistic explanations. At the same time, engineering practice lacks an integrated analysis and prediction system: one that can extract features from design parameter tables in a standardized manner and align them with experimental / simulation labels, stably predict U50% / BIL and end overvoltage ratio under small sample conditions, and simultaneously provide variable importance, interaction relationships, and uncertainty quantification to support design decisions. Under the premise of complying with IEC / GB lightning impulse test specifications, existing methods are difficult to conduct parallel evaluation and reliable prediction of multi-dimensional insulation design parameters of power transformers under controllable cost and time constraints. Specific problems include: low sample acquisition efficiency and inconsistent data caliber across batches, making it difficult to form structured data that can be used for modeling; strong coupling and nonlinearity exist among variables such as geometric gap, shielding size, dielectric constant / dielectric loss, paper thickness and moisture content under lightning impact, making it difficult for traditional single-factor or empirical fitting to explain the mechanism of parameter action and interaction effects; data-driven methods are prone to overfitting under small sample conditions, lack uncertainty and interpretability, resulting in insufficient predictive stability for key indicators such as U50% / BIL and end overvoltage ratio; and the lack of integrated analysis and prediction tools for the design stage makes it impossible to provide unified and traceable quantitative basis for scheme comparison and engineering decision-making. Summary of the Invention

[0004] The purpose of this invention is to provide To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for analyzing insulation characteristic data and predicting performance of power transformers, comprising: Based on a three-phase differentiated prototype, a structured sample set containing multi-dimensional design parameters and lightning impulse response labels was constructed. Identify key variables in the structured sample set and extract physical consistency constraints; Input the key variables and physical consistency constraints into the preset XGBoost prediction model, and output the predicted values ​​and confidence intervals.

[0005] Preferably, the construction of a structured sample set containing multi-dimensional design parameters and lightning impulse response labels based on a three-phase differentiated prototype includes: Based on the design parameter table of the three-phase differentiated prototype, the independent variables are read, and the field validity is checked, the unit is normalized, and the missing / abnormal annotation is completed to obtain multi-dimensional design parameters; By importing standard lightning impulse test or verified simulation output labels into the multidimensional design parameters, a structured sample set is obtained by constructing dimensionless geometric ratios, equivalent parameters of the medium, and phase coding characteristics, and by using Grubbs / IQR rules to remove outliers and aligning the time and unit aperture within the same impulse waveform / amplitude window.

[0006] Preferably, the independent variables include: the insulation distance between the winding and the end, dielectric parameters, the geometry of the shield and equalizing ring, the winding and lead path and grounding arrangement, the impulse voltage level and waveform, and the environment and temperature; The standard lightning impulse test or verified simulation output labels include: U50% / BIL, end-transmitted overvoltage ratio, wavefront steepness tolerance threshold, end maximum field strength, or a breakdown / non-breakdown binary label.

[0007] Preferably, the step of identifying key variables in the structured sample set and extracting physical consistency constraints includes: Significance tests, global sensitivity analysis, principal component analysis, and multifactor ANOVA were performed on the structured sample set to identify key variables and their interaction effects affecting U50%, BIL, and end overvoltage ratio; and monotonicity and boundary constraints reflecting engineering physical laws were extracted as physical consistency constraints.

[0008] Preferably, the step of inputting key variables and physical consistency constraints into a preset XGBoost prediction model and outputting predicted values ​​and confidence intervals includes: Key variables are transformed into standardized, dimensionless feature vectors; physical consistency constraints are formalized into constraint conditions. The XGBoost prediction model is based on feature vectors to obtain predicted values ​​and confidence intervals; Based on the constraints, the predicted values ​​and confidence intervals are checked for physical priors and normative boundaries. If a deviation from the trend or a violation of the boundary occurs, it is automatically marked and blocked; otherwise, the predicted values ​​and confidence intervals are output.

[0009] Preferably, the preset XGBoost prediction model is obtained through the following steps: Obtain a structured sample set, identify key variables in the structured sample set, and extract physical consistency constraints; Key variables are transformed into standardized, dimensionless feature vectors; physical consistency constraints are formalized into constraints usable for modeling. The feature vectors are partitioned using a hierarchical k-fold cross-validation method to obtain k sets of combined samples consisting of training and test sets; The XGBoost base model is input with k sets of combined samples, and the constraint condition mapping is used as the model constraint. The global hyperparameter optimization algorithm is used to fine-tune the XGBoost base model, output the optimal hyperparameters, and construct the XGBoost prediction model based on the optimal hyperparameters.

[0010] In a second aspect, the present invention provides a system for analyzing the insulation characteristics and predicting the performance of power transformers, characterized in that it comprises: The acquisition unit is used to construct a structured sample set containing multi-dimensional design parameters and lightning impulse response labels based on a three-phase differentiated prototype; The data analysis unit is used to identify key variables in structured sample sets and extract physical consistency constraints. The performance prediction unit is used to input key variables and physical consistency constraints into a preset XGBoost prediction model and output predicted values ​​and confidence intervals.

[0011] Preferably, in the performance prediction unit, the key variables and physical consistency constraints are input into a preset XGBoost prediction model, and the output predicted values ​​and confidence intervals include: Key variables are transformed into standardized, dimensionless feature vectors; physical consistency constraints are formalized into constraint conditions. The XGBoost prediction model is based on feature vectors to obtain predicted values ​​and confidence intervals; Based on the constraints, the predicted values ​​and confidence intervals are checked for physical priors and normative boundaries. If a deviation from the trend or a violation of the boundary occurs, it is automatically marked and blocked; otherwise, the predicted values ​​and confidence intervals are output.

[0012] In a third aspect, the present invention provides an electronic device, characterized in that it includes a processor and a memory, the processor being configured to execute a computer program stored in the memory to implement the power transformer insulation characteristic data analysis and performance prediction method as described in any one of claims 1 to 6.

[0013] In a fourth aspect, the present invention provides a computer-readable storage medium, characterized in that the computer-readable storage medium stores at least one instruction, which, when executed by a processor, implements the data analysis and performance prediction method for insulation characteristics of power transformers as described in any one of claims 1 to 6.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: (I) Achieving Dynamic Decoupling and Mechanism Identification of Design Variables and Lightning Impulse Response. Through a series of methods including "significance test—global sensitivity—principal component analysis—multi-factor variance analysis," this invention decomposes the influence of multi-dimensional parameters such as axial, radial, and end gaps, shielding and equalizing ring dimensions, cardboard thickness, dielectric constant and dielectric loss, moisture content and temperature, as well as impact amplitude and waveform, on insulation impact indicators into clear main effects and interaction effects, forming a traceable mechanism map. This "dynamic decoupling" capability not only identifies which variables are more sensitive in the wavefront stage and which have a stronger impact in the wave tail stage, but also presents the importance and interaction of variables using a unified standard, providing engineering-oriented, rule-based conclusions (e.g., "increasing critical gaps more effectively improves withstand capability," "shielding and gaps need to be optimized synergistically," "dielectric loss and moisture content have a linkage risk"). Compared with traditional methods relying on single-point experience judgments or post-event trial and error, this invention provides standardized data evidence and interpretable output, facilitating designers to quickly locate key constraints, formulate optimization priorities, and directly backfill the results into design specifications and parameter boundaries.

[0015] (II) Achieving rapid assessment of lightning impact characteristics with reduced time and computational cost, supporting large-scale parallel optimization design. Based on the physical priors and constraints provided by numerical analysis, this invention employs a gradient boosting tree model for prediction, combined with cross-validation and uncertainty assessment, achieving engineering-grade accuracy highly consistent with bench tests and high-fidelity simulations. After training, the evaluation of a single scheme requires only a very short time, and batch calculations of thousands of design schemes can be performed on conventional workstations, significantly accelerating the process compared to traditional three-dimensional field-path coupling simulations and significantly reducing the manpower required for mesh modeling, parameter scanning, and post-processing. As a result, design teams can complete a wider range of scheme coverage and sensitivity exploration within the same project cycle, transforming "small-scale, high-cost verification" into "broad-coverage, low-cost screening," requiring only secondary fine simulation and experimental confirmation for boundary or innovative schemes. This capability directly supports parallelized insulation optimization workflows, shortens design iteration cycles, reduces prototype and testing resource consumption, and serves engineering decision-making and compliance review with interpretable reports, visual dashboards, and standardized evaluation indicators. Attached Figure Description

[0016] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention; Figure 2 This is a schematic diagram of numerical analysis according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the XGBoost model construction process according to an embodiment of the present invention; Figure 4 These are the hyperparameters of the XGBoost model in this embodiment of the invention; Figure 5 The model weight factor output by the XGBoost model in this embodiment of the invention; Figure 6 This is a flowchart of the heuristic algorithm for optimizing the XGBoost model according to an embodiment of the present invention. Figure 7 This is a system structure block diagram according to an embodiment of the present invention; Figure 8 This is a structural block diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0017] The present invention will now be described in detail with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0018] The following detailed description is exemplary and intended to provide further detailed explanation of the invention. Unless otherwise specified, all technical terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terminology used in this invention is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention.

[0019] See Figure 1 This application discloses a method for analyzing insulation characteristic data and predicting performance of power transformers, including: S1: Based on the three-phase differentiated prototype, construct a structured sample set containing multi-dimensional design parameters and lightning impulse response labels; S2: Identify key variables in the structured sample set and extract physical consistency constraints; S3: Input the key variables and physical consistency constraints into the preset XGBoost prediction model, and output the predicted values ​​and confidence intervals.

[0020] In some embodiments, the construction of a structured sample set containing multidimensional design parameters and lightning impulse response labels based on a three-phase differentiated prototype includes: Based on the design parameter table of the three-phase differentiated prototype, the independent variables are read, and the field validity is checked, the unit is normalized, and the missing / abnormal annotation is completed to obtain multi-dimensional design parameters; By importing standard lightning impulse test or verified simulation output labels into the multidimensional design parameters, a structured sample set is obtained by constructing dimensionless geometric ratios, equivalent parameters of the medium, and phase coding characteristics, and by using Grubbs / IQR rules to remove outliers and aligning the time and unit aperture within the same impulse waveform / amplitude window.

[0021] In some embodiments, the independent variables include: the distance between the winding and the end insulation, dielectric parameters, shielding and equalizing ring geometry, winding and lead path and grounding arrangement, impulse voltage level and waveform, and environment and temperature. The standard lightning impulse test or verified simulation output labels include: U50% / BIL, end-transmitted overvoltage ratio, wavefront steepness tolerance threshold, end maximum field strength, or a breakdown / non-breakdown binary label.

[0022] In some embodiments, identifying key variables in the structured sample set and extracting physical consistency constraints includes: Significance tests, global sensitivity analysis, principal component analysis, and multifactor ANOVA were performed on the structured sample set to identify key variables and their interaction effects affecting U50%, BIL, and end overvoltage ratio; and monotonicity and boundary constraints reflecting engineering physical laws were extracted as physical consistency constraints.

[0023] In some embodiments, the step of inputting key variables and physical consistency constraints into a preset XGBoost prediction model and outputting predicted values ​​and confidence intervals includes: Key variables are transformed into standardized, dimensionless feature vectors; physical consistency constraints are formalized into constraint conditions. The XGBoost prediction model is based on feature vectors to obtain predicted values ​​and confidence intervals; Based on the constraints, the predicted values ​​and confidence intervals are checked for physical priors and normative boundaries. If a deviation from the trend or a violation of the boundary occurs, it is automatically marked and blocked; otherwise, the predicted values ​​and confidence intervals are output.

[0024] In some embodiments, the preset XGBoost prediction model is obtained through the following steps: Obtain a structured sample set, identify key variables in the structured sample set, and extract physical consistency constraints; Key variables are transformed into standardized, dimensionless feature vectors; physical consistency constraints are formalized into constraints usable for modeling. The feature vectors are partitioned using a hierarchical k-fold cross-validation method to obtain k sets of combined samples consisting of training and test sets; The XGBoost base model is input with k sets of combined samples, and the constraint condition mapping is used as the model constraint. The global hyperparameter optimization algorithm is used to fine-tune the XGBoost base model, output the optimal hyperparameters, and construct the XGBoost prediction model based on the optimal hyperparameters.

[0025] In some embodiments, the purpose of this invention is to establish an integrated "numerical analysis-modeling and prediction" technology for power transformers under lightning impact conditions, enabling unified data processing and performance prediction capabilities under multidimensional insulation design parameters and limited sample conditions. To this end, this invention uses statistical numerical analysis as a precursor, quantitatively identifying dominant variables and key interactions, mitigating multicollinearity, and clarifying parameter mechanisms through significance testing (P-value), sensitivity analysis, principal component analysis (PCA), and analysis of variance (ANOVA). Based on this, a prediction model centered on XGBoost is constructed, combined with reasonable hyperparameter global optimization and physical constraints, to improve the prediction accuracy and stability of indicators such as U50% / BIL and end overvoltage ratio under small sample scenarios, and outputs variable importance and uncertainty ranges, providing interpretable quantitative basis for design decisions. Through the above methods, this invention aims to overcome the problems of low sample acquisition efficiency, insufficient mechanism identification, and unstable prediction generalization in existing technologies, achieving parallel, low-cost, and rapid evaluation and scheme comparison for different combinations of structural parameters.

[0026] This invention proposes an integrated "numerical analysis-predictive modeling" technical solution for power transformers under lightning impact conditions. Based on multi-source samples formed from differentiated three-phase prototypes, it first uses statistical numerical methods to reveal the parameter action mechanism and solidify physical consistency constraints. Then, a gradient boosting tree model is used to complete performance prediction and uncertainty assessment, forming an engineering system that can directly serve insulation design and scheme comparison. The system consists of a data interface and verification module, a numerical analysis module, a feature engineering and physical constraint module, a machine learning modeling and hyperparameter optimization module, an uncertainty and interpretability module, and a result visualization and reporting module. It is deployed on a computing platform with data storage and computing power (CPU ≥ 8 cores, memory ≥ 32 GB, preferably equipped with a GPU to accelerate parallel evaluation) and is version-managed through a message bus and model repository. The following is a two-part description.

[0027] Numerical analysis methods (sample construction and mechanism identification) were employed. First, independent variables were retrieved from the design parameter table of the three-phase differentiated prototype, and field validity verification, unit normalization, and missing / abnormal labeling were completed. Table fields included at least: winding and end insulation distances (axial, radial, end, lead-shield-equalizing ring gap), dielectric parameters, shield and equalizing ring geometry (inner and outer diameters, width, relative position), winding and lead paths and grounding arrangements, impulse voltage levels and waveforms, environmental conditions, and temperature. Simultaneously, corresponding test / simulation labels were imported, including U50% / BIL, end-transmitted overvoltage ratio, wavefront steepness tolerance threshold, and maximum end field strength or breakdown / non-breakdown binary labels. To eliminate scale differences and collinearity, dimensionless geometric ratios, dielectric equivalent parameters, and phase coding characteristics were constructed. Grubbs / IQR rules were used to remove outliers, and time alignment and unit caliber within the same impulse waveform / amplitude window were used to ensure consistency across batches.

[0028] Subsequently, significance testing and variance decomposition were used to identify main effects and interaction effects: Within the framework of generalized linear regression, p-value tests were performed on key variables and interaction terms (significance threshold α=0.05, reporting adjusted p and effect size), and multivariate ANOVA was used to quantify the contribution of variables to the target indicator. For scenarios with nonlinearity and significant interactions, global sensitivity analysis was used to characterize the influence intensity and uncertainty range of variables on U50% / BIL, K_tr, etc. To alleviate collinearity and compress dimensionality, principal component analysis (PCA) was performed based on the covariance matrix, retaining principal components with a cumulative contribution rate ≥95%, and extracting orthogonal feature subspaces. Combined with impact physics, consistency verification was performed on the monotonicity and boundary patterns of "increased geometric gap → no decrease in tolerance, increased dielectric loss → no improvement in tolerance, increased water content → increased dielectric loss," outputting the causal relationship, significance ranking, and interaction spectrum of "variable-indicator," and solidifying the monotonicity and acceptable value range as physical priors and constraints for subsequent modeling. Through the above steps, a structured dataset and its interpretable meta-information (significance, sensitivity, PCA loading, ANOVA effect size) are formed after standardization, denoising, and mechanism annotation.

[0029] (II) Machine Learning Modeling for Performance Prediction Analysis (Model Training, Calibration, and Evaluation). Based on the constraints and features of the numerical analysis output, a prediction model with gradient boosting trees as its core is constructed, modeling both regression targets (U50% / BIL, K_tr, S_crit, E_max) and optional classification targets (breakdown / non-breakdown). Data partitioning employs hierarchical k-fold cross-validation (k=5–10, hierarchically divided by waveform / amplitude / phase). Feature scaling and encoding consistency transformation are performed on the training set, and monotonicity priors are mapped to model constraints (monotonically increasing for critical gaps and monotonically decreasing for tan δ / water content) to suppress overfitting in small samples and physics-incorrect predictions. The hyperparameter set employs global optimization search, preferentially using particle swarm optimization / quantum behavior particle swarm optimization (QPSO) or Bayesian optimization for adaptive exploration. The fitness function is taken as RMSE / MAE or AUC (classification) from cross-validation. An early stopping strategy and robust evaluation through repeated cross-validation are built-in to output the optimal result. After the final model training, error and uncertainty assessments are performed: for regression tasks, bootstrap (B≥1000) and error propagation are used to provide 95% confidence intervals, or quantile regression trees are used to provide upper and lower confidence bands; for classification tasks, reliability plots and isothermal calibration are used to ensure consistency of probability outputs. To ensure engineering interpretability, SHAP values ​​and partial dependency curves (PDP / ICE) are calculated to provide variable contributions, interaction sensitivity, and design suggestion intervals. The consistency between "model interpretation—numerical analysis conclusions—physical priors" is compared, and if any deviation from physical trends is found, feature / constraint rollback and retraining are triggered. The system outputs include: predicted values ​​and confidence intervals of target indicators, importance ranking and interactive heatmaps of key variables, radar charts comparing the current design with the "safety margin" boundary, and parallel evaluation reports for batch solutions; it also generates a traceable run-manifest (data source, field definition, model version, superparticipant random seed, calibration record), which can be called with one click in the design process via table / interface, enabling fast and low-cost parallel prediction and solution comparison for different combinations of structural parameters.

[0030] Numerical analysis-driven proxy modeling application scenarios: Based on three-phase differentiated prototypes and standard lightning impulse data, an insulation impulse performance prediction model is established to replace some high-cost tests and high-fidelity simulations, enabling rapid evaluation of indicators such as U50% / BIL, end overvoltage ratio, and wavefront steepness tolerance, serving the insulation design and scheme comparison of power transformers.

[0031] Innovative Methodological Design: A two-layer framework of "pre-statistical analysis + learning prediction" is constructed. The first layer employs p-value tests, global sensitivity analysis, principal component analysis, and multivariate analysis of variance to identify main effects and interactions, mitigate collinearity, and solidify physical monotonicity and boundary constraints. The second layer uses XGBoost as the core for modeling, combined with global hyperparameter optimization (such as QPSO / Bayesian) and hierarchical cross-validation to improve generalization, stability, and interpretability under small sample conditions.

[0032] Rapid prediction capability coupled with insulation structure parameters: After inputting design variables such as axial / radial / end gap, shield and equalizing ring size, paper thickness, dielectric constant and dielectric loss, moisture content and temperature, impulse voltage and waveform, the system can output predicted values ​​such as U50% / BIL and end overvoltage ratio, as well as uncertainty ranges, and generate variable importance and interaction graphs, supporting large-scale parallel evaluation and batch optimization, significantly reducing computation and test time costs.

[0033] Example This embodiment aims to overcome the problems of traditional lightning impulse assessment, such as reliance on numerous bench tests or high-fidelity simulations, long modeling cycles, and difficulty in parallel assessment. It proposes a method and system for predicting insulation characteristics using a combination of numerical analysis and XGBoost modeling. This method utilizes limited samples from differentiated three-phase prototypes and standard impulse tests / verified simulations. First, it clarifies the parameter action mechanism through statistical numerical analysis, and then uses machine learning to achieve rapid and high-precision performance prediction, providing engineering design with low-cost and scalable insulation impulse assessment capabilities.

[0034] (1) Data construction stage Data source: Based on the full-wave and shear wave conditions of lightning impulse in IEC 60076-3 and GB / T 1094.3, parallel data acquisition was performed on three-phase differentiated prototypes; optional field-path coupling simulation verified by actual measurements was introduced to supplement the sample coverage.

[0035] Input variables: Read from “Table 1 Design Parameters” and standardize the parameters, including at least axial / radial / end gaps, lead and shield / equalizing ring geometry, cardboard thickness, dielectric constant and dielectric loss, moisture content and temperature, impulse voltage level and waveform, grounding and lead path, etc.

[0036] Table 1 Design Parameters

[0037] Output labels: U50%, BIL, end overvoltage ratio, wavefront steepness tolerance, end maximum field strength or breakdown / non-breakdown.

[0038] (2) Numerical analysis stage, see Figure 2 Significance test: Under a unified regression framework, p-value tests are performed on key variables and interaction terms to identify significant and marginally significant factors.

[0039] Sensitivity analysis: Use global sensitivity (such as Sobol or equivalent methods) or Morris semi-global method to quantify the contribution and interaction strength of each variable to the index; if necessary, output "dynamic sensitivity" by wavefront / wavetail time window.

[0040] Principal component analysis: This method addresses multicollinearity by extracting orthogonal principal components with high cumulative contribution rates, thus reducing dimensionality for subsequent modeling.

[0041] Multifactor analysis of variance: quantifies the effect size of main effects and interaction effects, forming a traceable "variable-response" explanatory map.

[0042] Physical prior solidification: Engineering principles such as "increasing critical gaps does not reduce tolerance" and "increasing dielectric loss or water content does not increase tolerance" are solidified into monotonicity and boundary constraints for subsequent modeling.

[0043] (3) Machine learning modeling stage, see Figure 3 The above processes S1, S2 and S3 are heuristic processes, which involve hyperparameter tuning, meaning that parameter optimization is required to achieve optimal model performance. Figure 4 The diagram shows the hyperparameters that need to be optimized during the model heuristic process.

[0044] Model selection: The XGBoost regression / classifier is used as the core, and is used for regression indicators (U50%, BIL, end overvoltage ratio, maximum field strength, etc.) and binary classification tasks (breakdown or not).

[0045] Training strategies: hierarchical k-fold cross-validation and early stopping; imposing monotonicity or boundary constraints on key geometric and medium variables; using repeated cross-validation to improve robustness for small sample scenarios.

[0046] Results output: Output error metrics and stability assessments on the validation set, and generate variable importance and interaction visualizations (such as feature importance ranking, partial dependency / interaction heatmaps).

[0047] (4) Hyperparameter optimization stage, see Figure 6 Search methods: Global methods such as quantum behavior particle swarm optimization or Bayesian optimization are used to optimize learning rate, tree depth, subsampling ratio, minimum leaf weight, column sampling rate, regularization coefficient, etc.

[0048] Objectives and constraints: The objective is to minimize the cross-validation error, while constraining the physical consistency of the predictions (keeping the monotonic default rate as low as possible).

[0049] Robust evaluation: Multiple repetitions and a fixed random seed are used to record the search trajectory and the optimal parameter set, which are then incorporated into the model version repository.

[0050] (5) Prediction and evaluation stage Rapid prediction: For any new combination of structural parameters, the system quickly outputs predicted values ​​such as U50%, BIL, and end overvoltage ratio, generating compliance and safety margin prompts.

[0051] Batch evaluation: Supports parallel computing and sorting of thousands to tens of thousands of schemes for scheme comparison and design optimization, and performs high-fidelity simulation or bench verification only on boundary schemes.

[0052] Consistency check: Physical prior and standard boundary checks are performed on the prediction results, and any deviations from the trend or out-of-bounds errors are automatically marked and blocked.

[0053] The above embodiments provide a complete technical path from sample construction, numerical analysis, model training, hyperparameter optimization to rapid prediction. Without changing the "three-phase consistency" principle of engineering products, samples can be efficiently obtained through three-phase differentiated prototypes, and rapid evaluation and parallel optimization of lightning impact performance for different combinations of design variables can be achieved with less time and computational cost.

[0054] In summary, this invention relates to a data analysis algorithm for the insulation characteristics of power transformers and its performance prediction system, belonging to the technical field of insulation performance evaluation and intelligent modeling of power equipment. The method includes the following steps: First, multidimensional variables with standardized calibers are read from the prototype design parameter table, covering at least axial / radial / end gaps, shielding and equalizing ring dimensions, cardboard thickness, dielectric constant and dielectric loss, moisture content and temperature, impulse voltage level and waveform, etc.; simultaneously, the data is integrated with standard lightning impulse tests and / or verified simulation data (such as U50%, BIL, end overvoltage ratio, wavefront steepness tolerance, maximum field strength, or breakdown status) to construct a structured sample set. Second, numerical analysis is conducted: significance testing (P-value), global sensitivity analysis, principal component analysis, and multifactor ANOVA are used to identify dominant variables and key interactions, mitigate multicollinearity, and establish physical priors and monotonicity constraints such as "increased gap does not decrease tolerance, increased dielectric loss does not increase tolerance." Then, a performance prediction model based on XGBoost is established, and hyperparameter global optimization is performed by combining quantum behavior particle swarm optimization or Bayesian search. Hierarchical cross-validation and early stopping strategies are used to improve generalization stability under small sample conditions. Furthermore, uncertainty assessment and interpretability analysis are performed on the model output, generating reports on variable importance, interaction relationships, and design sensitivity. Finally, new design parameters are input into the trained model to quickly obtain predicted values ​​and confidence intervals for indicators such as U50%, BIL, and end overvoltage ratio, enabling batch screening and optimization of multiple schemes. Correspondingly, the system consists of a data interface and governance module, a numerical analysis module, a feature engineering and physical constraint module, a model training and hyperparameter optimization module, a prediction and uncertainty assessment module, and a result visualization and reporting module. This method significantly reduces the overhead of bench testing and high-fidelity simulation while maintaining engineering-grade accuracy, supports large-scale parallel evaluation under high-dimensional multivariable conditions, and is suitable for insulation verification and optimization decisions in the transformer design stage.

[0055] like Figure 7 As shown, based on the same inventive concept as the above embodiments, the present invention also provides a power transformer insulation characteristic data analysis and performance prediction system, comprising: The acquisition unit is used to construct a structured sample set containing multi-dimensional design parameters and lightning impulse response labels based on a three-phase differentiated prototype; The data analysis unit is used to identify key variables in structured sample sets and extract physical consistency constraints. The performance prediction unit is used to input key variables and physical consistency constraints into a preset XGBoost prediction model and output predicted values ​​and confidence intervals.

[0056] In some embodiments, the performance prediction unit inputs key variables and physical consistency constraints into a preset XGBoost prediction model, and outputs predicted values ​​and confidence intervals, including: Key variables are transformed into standardized, dimensionless feature vectors; physical consistency constraints are formalized into constraint conditions. The XGBoost prediction model is based on feature vectors to obtain predicted values ​​and confidence intervals; Based on constraints, physical prior and normative boundary checks are performed on the predicted values ​​and confidence intervals. If a deviation from the trend or exceeding the boundary occurs, it is automatically marked and blocked; otherwise, the predicted value and confidence interval are output. Example 3 like Figure 8 As shown, the present invention also provides an electronic device 100 for implementing a method for analyzing the insulation characteristics of power transformers and predicting their performance; The electronic device 100 includes a memory 101, at least one processor 102, a computer program 103 stored in the memory 101 and executable on at least one processor 102, and at least one communication bus 104.

[0057] The memory 101 can be used to store the computer program 103. The processor 102 implements the steps of the power transformer insulation characteristic data analysis and performance prediction method by running or executing the computer program stored in the memory 101 and calling the data stored in the memory 101.

[0058] The memory 101 may primarily include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created based on the use of the electronic device 100 (such as audio data), etc. In addition, the memory 101 may include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other non-volatile solid-state storage device.

[0059] At least one processor 102 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Processor 102 may be a microprocessor or any conventional processor. Processor 102 is the control center of electronic device 100, connecting various parts of electronic device 100 via various interfaces and lines.

[0060] The memory 101 in the electronic device 100 stores multiple instructions to implement a method for analyzing the insulation characteristics of a power transformer and predicting its performance. The processor 102 can execute multiple instructions to achieve the following: Based on a three-phase differentiated prototype, a structured sample set containing multi-dimensional design parameters and lightning impulse response labels was constructed. Identify key variables in the structured sample set and extract physical consistency constraints; Input the key variables and physical consistency constraints into the preset XGBoost prediction model, and output the predicted values ​​and confidence intervals.

[0061] If the modules / units integrated in the electronic device 100 are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, and read-only memory (ROM).

[0062] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0063] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0064] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0065] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0066] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0067] Finally, 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 the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for analyzing insulation characteristic data and predicting performance of power transformers, characterized in that, include: Based on a three-phase differentiated prototype, a structured sample set containing multi-dimensional design parameters and lightning impulse response labels was constructed. Identify key variables in the structured sample set and extract physical consistency constraints; Input the key variables and physical consistency constraints into the preset XGBoost prediction model, and output the predicted values ​​and confidence intervals.

2. The method for analyzing insulation characteristics and predicting performance of power transformers according to claim 1, characterized in that, The structured sample set, constructed based on the three-phase differentiated prototype and containing multi-dimensional design parameters and lightning impulse response labels, includes: Based on the design parameter table of the three-phase differentiated prototype, the independent variables are read, and the field validity is checked, the unit is normalized, and the missing / abnormal annotation is completed to obtain multi-dimensional design parameters; By importing multidimensional design parameters into standard lightning impulse test or verified simulation output labels, and constructing dimensionless geometric ratios, equivalent parameters of the medium, and phase coding characteristics, and using Grubbs / IQR rules to remove outliers, and aligning the time and unit aperture within the same impulse waveform / amplitude window, a structured sample set is obtained.

3. The method for analyzing insulation characteristics and predicting performance of power transformers according to claim 2, characterized in that, The independent variables include: the distance between the winding and the end insulation, dielectric parameters, shielding and equalizing ring geometry, winding and lead path and grounding arrangement, impulse voltage level and waveform, and environment and temperature; The standard lightning impulse test or verified simulation output labels include: U50% / BIL, end-transmitted overvoltage ratio, wavefront steepness tolerance threshold, end maximum field strength, or a breakdown / non-breakdown binary label.

4. The method for analyzing insulation characteristics and predicting performance of power transformers according to claim 1, characterized in that, The process of identifying key variables in the structured sample set and extracting physical consistency constraints includes: Significance tests, global sensitivity analysis, principal component analysis, and multifactor ANOVA were performed on the structured sample set to identify key variables and their interaction effects affecting U50%, BIL, and end overvoltage ratio; and monotonicity and boundary constraints reflecting engineering physical laws were extracted as physical consistency constraints.

5. The method for analyzing insulation characteristics and predicting performance of power transformers according to claim 1, characterized in that, The process of inputting key variables and physical consistency constraints into a preset XGBoost prediction model and outputting predicted values ​​and confidence intervals includes: Key variables are transformed into standardized, dimensionless feature vectors; physical consistency constraints are formalized into constraint conditions. The XGBoost prediction model is based on feature vectors to obtain predicted values ​​and confidence intervals; Based on the constraints, the predicted values ​​and confidence intervals are checked for physical priors and normative boundaries. If a deviation from the trend or a violation of the boundary occurs, it is automatically marked and blocked; otherwise, the predicted values ​​and confidence intervals are output.

6. The method for analyzing insulation characteristics and predicting performance of power transformers according to claim 1, characterized in that, The preset XGBoost prediction model is obtained through the following steps: Obtain a structured sample set, identify key variables in the structured sample set, and extract physical consistency constraints; Key variables are transformed into standardized, dimensionless feature vectors; physical consistency constraints are formalized into constraints usable for modeling. The feature vectors are partitioned using a hierarchical k-fold cross-validation method to obtain k sets of combined samples consisting of training and test sets; The XGBoost base model is input with k sets of combined samples, and the constraint condition mapping is used as the model constraint. The global hyperparameter optimization algorithm is used to fine-tune the XGBoost base model, output the optimal hyperparameters, and construct the XGBoost prediction model based on the optimal hyperparameters.

7. A system for analyzing the insulation characteristics and predicting the performance of power transformers, characterized in that, include: The acquisition unit is used to construct a structured sample set containing multi-dimensional design parameters and lightning impulse response labels based on a three-phase differentiated prototype; The data analysis unit is used to identify key variables in structured sample sets and extract physical consistency constraints. The performance prediction unit is used to input key variables and physical consistency constraints into a preset XGBoost prediction model and output predicted values ​​and confidence intervals.

8. The power transformer insulation characteristic data analysis and performance prediction system according to claim 7, characterized in that, In the performance prediction unit, key variables and physical consistency constraints are input into a preset XGBoost prediction model, and the output predicted values ​​and confidence intervals include: Key variables are transformed into standardized, dimensionless feature vectors; physical consistency constraints are formalized into constraint conditions. The XGBoost prediction model is based on feature vectors to obtain predicted values ​​and confidence intervals; Based on the constraints, the predicted values ​​and confidence intervals are checked for physical priors and normative boundaries. If a deviation from the trend or a violation of the boundary occurs, it is automatically marked and blocked; otherwise, the predicted values ​​and confidence intervals are output.

9. An electronic device, characterized in that, It includes a processor and a memory, the processor being used to execute a computer program stored in the memory to implement the power transformer insulation characteristic data analysis and performance prediction method as described in any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one instruction, which, when executed by a processor, implements the power transformer insulation characteristic data analysis and performance prediction method as described in any one of claims 1 to 6.

Images