CN-121997764-A - Intelligent prediction and optimization method and system for electromagnetic parameters of transformer

CN121997764ACN 121997764 ACN121997764 ACN 121997764ACN-121997764-A

Abstract

The invention belongs to the technical field of transformers, and particularly relates to an intelligent prediction and optimization method and system for electromagnetic parameters of a transformer. The method aims at solving the problems that the theory and actual measurement are out of alignment caused by dynamic change of the process coefficient, the precision and the safety of a single model cannot be considered, and the efficiency is reduced due to the fact that design iteration depends on manual trial and error. And finally, verifying a predicted result through comparison of a predicted value and an actual measured value, triggering an alarm and model updating mechanism if the result is not ideal, improving the design primary success rate, and effectively reducing the research and development cost and period.

Inventors

HUANG YAFEN
HU SHENGCHUN
TAN YONG
TIAN HAO

Assignees

新华都特种电气股份有限公司

Dates

Publication Date: 20260508
Application Date: 20260130

Claims (10)

1. An intelligent prediction and optimization method for electromagnetic parameters of a transformer is characterized by comprising the following steps: Obtaining design parameters and corresponding test parameters of a plurality of historical transformer prototypes to form a historical sample set; based on the historical sample set, respectively training to obtain a first type of prediction model suitable for sample space internal prediction and a second type of prediction model suitable for sample space external prediction; The method comprises the steps of obtaining design parameters of a transformer to be predicted, judging the position of the design parameters in a feature space, carrying out electromagnetic parameter prediction by adopting a first type prediction model if the design parameters are positioned in a sample space determined by a historical sample set to obtain a predicted value; Calculating the deviation between the predicted value and the measured value of the transformer to be predicted, judging whether the deviation is smaller than a preset deviation, outputting the predicted value of the transformer to be predicted if the deviation is smaller than the preset deviation, triggering an alarm if the deviation is not smaller than the preset deviation, adding a current sample into the historical sample set, and updating a corresponding prediction model.
2. The method according to claim 1, wherein the step of training to obtain a first type of prediction model applicable to intra-sample-space prediction and a second type of prediction model applicable to extra-sample-space prediction, respectively, based on the historical sample set, comprises: preprocessing the historical sample set, wherein the preprocessing comprises encoding the design parameters and category variables in the test parameters; dividing the preprocessed historical sample set into a training set and a testing set; training and parameter optimization are carried out on the models in the first type candidate model pool and the second type candidate model pool on the training set, so that a plurality of first type candidate models and a plurality of second type candidate models are obtained; evaluating performance of a plurality of said first class candidate models and a plurality of said second class candidate models on said test set; And selecting a model with optimal performance from a plurality of first type candidate models as the first type prediction model based on an evaluation result, selecting a model with optimal performance from a plurality of second type candidate models as the second type prediction model, and respectively carrying out serialization storage on the first type prediction model and the second type prediction model.
3. The method of claim 2, wherein the step of evaluating the performance of the plurality of first-type candidate models and the plurality of second-type candidate models over the test set comprises: inputting design parameters in any group of test set into each candidate model for calculation to obtain predicted values of a plurality of electromagnetic parameters predicted by a plurality of candidate models; scoring all predicted values based on a preset model performance evaluation criterion to obtain comprehensive scoring indexes; and screening out the model with high comprehensive scoring index as the model with optimal performance.
4. A method according to claim 3, wherein the step of scoring all predicted values of the candidate model on the test set based on a preset model performance evaluation criterion to obtain a comprehensive scoring index comprises: based on the predicted values of a plurality of electromagnetic parameters, respectively calculating the RMSE root mean square error, the MAPE average absolute percentage error and the R2 decision coefficient of each predicted value and the test parameter corresponding to the design parameter; And carrying out fusion analysis on the RMSE root mean square error, the MAPE average absolute percentage error and the R2 decision coefficient corresponding to each predicted value to obtain a comprehensive scoring index.
5. The method according to claim 1, characterized in that it comprises, simultaneously with the step of outputting the predicted value of the transformer to be predicted, the following parallel steps: Calculating at least one error evaluation index of the current prediction, and storing the error evaluation index in association with the complete information of the current prediction, wherein the complete information at least comprises a model version identifier triggering the current prediction, a prediction time stamp and an associated unique product identifier; And carrying out confidence enhancement on the generated prediction model based on the successful prediction result.
6. The method of claim 2, wherein the step of determining the position of the substrate comprises, The step of triggering the alarm comprises the steps of generating alarm information and directionally pushing the alarm information to a design engineer and/or an algorithm engineer, wherein the alarm information comprises a result obtained by analyzing a feature set with high contribution degree to the target electromagnetic parameter prediction and/or a matching result of a history similar case, and/or The step of coding the category variables in the design parameters and the test parameters comprises the steps of identifying the category variables in the design parameters and the test parameters of a plurality of historical transformer prototypes, and performing numerical coding conversion on the identified category variables to generate numerical characteristics which can be processed by a machine learning model, wherein the category variables comprise at least one of product identification and specification parameters, working condition environment parameters, structural material parameters and design target parameters.
7. The method of claim 1, wherein the step of determining the position of the substrate comprises, The first type of prediction model is set as a machine learning model based on a tree model, wherein the machine learning model based on the tree model comprises a gradient lifting decision tree model and a random forest model, the gradient lifting decision tree model comprises a XGBoost model or a LightGBM model, and/or The second type of prediction model is set as a prediction model based on regression analysis, and the regression model comprises a linear regression model, a polynomial regression model, a Gaussian process regression model and a quantile regression model.
8. An intelligent prediction and optimization system for electromagnetic parameters of a transformer, characterized in that it can execute the intelligent prediction and optimization method for electromagnetic parameters of a transformer according to any one of claims 1 to 7, said system comprising: the data acquisition module is configured to acquire design parameters of a plurality of historical transformer prototypes, corresponding test parameters and design parameters of transformers to be predicted; The training module is configured to respectively train and obtain a first type of prediction model suitable for the prediction in a sample space and a second type of prediction model suitable for the prediction out of the sample space based on the historical sample set, and add samples corresponding to the deviation between a predicted value and the measured value in the prediction process more than or equal to a preset deviation into the historical sample set and update the corresponding prediction models; The prediction module is configured to judge the position of the design parameter in a feature space, and if the design parameter is positioned in a sample space determined by a historical sample set, the first type prediction model is adopted for electromagnetic parameter prediction to obtain a predicted value; The response and self-optimization module is configured to calculate the deviation between the predicted value and the actual measured value of the transformer to be predicted, judge whether the deviation is smaller than a preset deviation, output the predicted value of the transformer to be predicted if the deviation is smaller than the preset deviation, and trigger an alarm if the deviation is not smaller than the preset deviation.
9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor implements the intelligent prediction and optimization method of the electromagnetic parameters of the transformer according to any one of claims 1 to 7 and is applied to the intelligent prediction and optimization system of the electromagnetic parameters of the transformer according to claim 8 when executing the computer program.
10. A computer readable storage medium storing a computer program, wherein the computer program when executed by a processor implements the intelligent prediction and optimization method of the electromagnetic parameters of the transformer according to any one of claims 1 to 7 and is applied to the intelligent prediction and optimization system of the electromagnetic parameters of the transformer according to claim 8.

Description

Intelligent prediction and optimization method and system for electromagnetic parameters of transformer Technical Field The invention belongs to the technical field of transformers, and particularly relates to an intelligent prediction and optimization method and system for electromagnetic parameters of a transformer. Background The transformer is a core device of the power system, and the design accuracy of electromagnetic parameters such as no-load loss, no-load current, load impedance and the like directly determines the energy efficiency level, manufacturing cost and operation reliability of the device. Traditional electromagnetic parameter design relies on classical electromagnetic theory models such as Maxwell's equations, finite element analysis and the like, influences of real factors such as material characteristics, manufacturing tolerances and the like on theoretical calculation are compensated by introducing empirical process coefficients, and finally, the design is verified and corrected by manufacturing a prototype and performing actual measurement. However, with the continuous progress of modern material technology such as high-performance silicon steel sheet and the optimization of manufacturing process such as high-precision assembly fixture, the material performance and process conditions tend to be dynamically changed, which makes the fixed "process coefficient" set based on experience have a large difference from the actual production condition, and a large deviation is generated between the theoretical value and the actual measured value, so that the product performance, cost control and research and development efficiency of the transformer are restricted. Currently, the method for reducing the deviation between the theoretical value and the actual value comprises manual parameter adjustment and static fitting based on historical data. Specifically, the manual parameter adjustment depends on experience of a senior engineer, each process coefficient is manually and repeatedly adjusted by comparing actual measurement data of a prototype with theoretical calculation values, and the product reaches the standard after multiple iterations of 'design-trial-test'. The method is highly dependent on personal experience, human errors and inefficiency exist in the parameter adjusting process, and the problems that the product performance is difficult to guarantee, the research and development period is long and the trial production cost is high are caused. And based on a static fitting mode of historical data, a fixed empirical formula or correction coefficient is fitted through simple regression analysis by utilizing test data accumulated in the past to replace part of manual experience. However, this method builds a "static" mathematical model, which fails rapidly once the production conditions change, such as changing material lots, upgrading the production line, or pushing out new product lines, and cannot adapt to the dynamic evolution of process parameters. In view of this, the present invention has been made. Disclosure of Invention The invention aims to solve the problems that the theory and actual measurement are out of alignment caused by dynamic change of process coefficients, the precision and safety of a single model cannot be considered, and the efficiency is reduced due to the fact that design iteration depends on manual trial and error. In order to achieve the above purpose, the invention provides an intelligent prediction and optimization method for electromagnetic parameters of a transformer, comprising the following steps: Obtaining design parameters and corresponding test parameters of a plurality of historical transformer prototypes to form a historical sample set; based on the historical sample set, respectively training to obtain a first type of prediction model suitable for sample space internal prediction and a second type of prediction model suitable for sample space external prediction; The method comprises the steps of obtaining design parameters of a transformer to be predicted, judging the position of the design parameters in a feature space, carrying out electromagnetic parameter prediction by adopting a first type prediction model if the design parameters are positioned in a sample space determined by a historical sample set to obtain a predicted value; Calculating the deviation between the predicted value and the measured value of the transformer to be predicted, judging whether the deviation is smaller than a preset deviation, outputting the predicted value of the transformer to be predicted if the deviation is smaller than the preset deviation, triggering an alarm if the deviation is not smaller than the preset deviation, adding a current sample into the historical sample set, and updating a corresponding prediction model. The method comprises the steps of obtaining a first type of prediction model suitable for sample space internal prediction and a second type o