Machine Learning-Based Design Method for Transformer Insulation Structure

Abstract

The present invention relates to a design method for the insulation structure of a transformer based on machine learning, including collecting the parameters of previous design schemes and the simulation data of the electric field distribution of the transformer; establishing the feature engineering for the transformer insulation design; constructing and training a prediction model for the parameters of the pre-transformer insulation structure; applying the trained model to the simulation of a new design scheme; and determining the design condition parameters of the transformer. The design method for the insulation structure of a transformer based on machine learning of the present invention analyzes a large amount of historical design data and simulation results, automatically constructs complex mapping relationships, thereby improving the efficiency and accuracy of the design of the transformer insulation structure, optimizing the insulation performance, and realizing automatic optimization and feedback adjustment to solve the limitations of traditional design methods in dealing with complex electric field distributions and multi-factor influences, and ensuring that the transformer design meets the high standards and high requirements of modern power systems.

Description

Technical Field

[0001] The present invention relates to the technical field of transformers, and more specifically, to a design method for the insulation structure of a transformer based on machine learning. Background Art

[0002] As a crucial device in the power system, the reliability of a transformer is directly related to the safe and stable operation of the entire power system. The insulation structure design of a transformer is one of the key links to ensure its safe operation. Insulation problems are one of the main causes of transformer failures. Especially during long-term operation and under complex working conditions, defects in the insulation structure often lead to serious safety accidents, such as insulation aging, partial discharge, and even transformer explosion. Therefore, ensuring the reasonable design of the transformer insulation structure can not only extend the service life of the equipment but also prevent potential fault hazards and ensure the safe operation of the power system. With the continuous development of the power system, the operating environment of transformers has become increasingly complex, and the requirements for their insulation structure design have also increased. Traditional design methods are difficult to fully meet the high standards and requirements of modern power equipment.

[0003] Traditional design methods for transformer insulation structures mainly rely on standard parameters in design manuals and the experience of designers. However, this method has significant limitations. First, the standard parameters in design manuals are usually formulated based on typical scenarios and historical data, and it is difficult to cover all possible actual application scenarios. Especially when facing complex and non-standard working conditions, the design results often fail to achieve the expected effects. Second, although the experience of designers plays an important role in the design process, the accumulation of experience takes time. New designers are difficult to master sufficient design skills in the short term, and even experienced designers may make judgment errors when facing complex design tasks with the interaction of multiple factors, resulting in inaccurate design results. In addition, traditional design methods are less efficient in dealing with complex electric field distributions and the interaction of multiple design parameters, and are prone to affecting the stability and reliability of the design due to errors in manual calculations or empirical judgments. These problems make it difficult for traditional methods to quickly and accurately respond to the changing operating conditions and higher design requirements in modern power systems.

[0004] As a powerful data analysis and modeling tool, machine learning demonstrates great potential in the design of transformer insulation structures. Machine learning can automatically establish complex mapping relationships by analyzing a large amount of historical design data and simulation results, thereby more accurately predicting the optimal design parameters of the insulation structure. Compared with traditional methods, the design method based on machine learning has significant advantages. First of all, machine learning can greatly improve the design efficiency, reduce human errors in the design process, and ensure the stability and reliability of the design results. Secondly, when facing complex multi-factor influences, machine learning models can better capture and process the relationships between these factors, thereby optimizing the design of the insulation structure. In addition, machine learning can also achieve automatic optimization and feedback adjustment of the design process. When the design results do not meet expectations, the model can quickly make adjustments to ensure that the final design meets all requirements. To sum up, the design method of transformer insulation structure based on machine learning can not only cope with the complexity of modern power systems, but also has extremely strong application prospects, which is an important development direction in the future transformer design field.

[0005] The design method of a 35kV distribution three-winding transformer proposed in Patent CN 103310963B includes determining the 0.4kV side capacity, transformer body design, material selection, operating loss calculation, and economic analysis. This method is applicable to 35kV substations in remote areas. By optimizing the design and selectively equipping with on-load tap-changers, the cost and losses are reduced to achieve economic operation. Although Patent CN 103310963B proposes a design method for a 35kV distribution three-winding transformer applicable to remote areas, it relies on traditional design processes and manual standards, fails to address the special design requirements under complex operating conditions, and insufficiently considers the mutual influences among multiple factors, resulting in insufficient flexibility and accuracy in the design.

[0006] The optimization design method of a transformer main insulation structure proposed in Patent CN 117150791B involves determining the objective function, setting the insulation structure parameters as decision variables, and obtaining the constraint conditions. By solving the objective function, the objective solution set is obtained and comprehensively evaluated, and finally the optimized parameters of the main insulation structure are determined. This method makes up for the deficiencies of existing design methods and improves the reliability of the main insulation structure. Although Patent CN 117150791B optimizes the transformer main insulation structure, it still relies on traditional optimization methods and does not introduce intelligent algorithms, resulting in low efficiency in dealing with complex design parameters and electric field distributions, and being unable to automatically adjust the optimization results, and the design process may take a long time.

[0007] A design method of a 35kV distribution three-winding transformer proposed in Patent CN 103310963B includes: determining the capacity of the 0.4kV side according to the substation layout, setting the transformer capacity and conducting the main body design, selecting appropriate materials, calculating the operating losses, and performing economic analysis. This method is aimed at substations in remote areas. By configuring a 35 / 10 / 0.4kV three-winding transformer and optimizing the capacity of the 0.4kV side, the cost and losses are reduced. At the same time, considering the installation of an on-load tap-changer, economic operation is achieved. Patent CN103310963B also relies on traditional design methods and manual standard parameters, and does not provide an effective solution for the complexity of the insulation structure, making it difficult to cope with the complex working conditions and high-standard design requirements in modern power systems, which may lead to inaccurate and unstable design results.

[0008] A design method of a distribution transformer proposed in Patent CN106298215B includes configuring a design scheme, determining the operating environment requirements, setting the manufacturing cost and performance indicators, and comparing the life cycle cost (LCC). By establishing a cost model and sensitivity analysis, the increase in the life cycle cost is restricted, and the technical economy is improved. Although Patent CN 106298215B improves the economy of transformer design through the life cycle cost model, its design method is mainly based on traditional sensitivity analysis and does not use modern data analysis technology to improve the design accuracy and optimization efficiency. Especially when facing multi-dimensional design factors, it may be difficult to achieve the optimal solution.

[0009] A design method of a power transformer based on fast multi-objective optimization proposed in Patent CN115310353B includes creating an optimization model and using an improved NSGA-II algorithm for solution. The improvement is reflected in combining artificial solutions with random solutions when initializing the population, so as to facilitate algorithm convergence, reduce the time consumption, avoid local optima, and at the same time maintain a wide distribution of the solution space, meeting the requirements of optimization effects and fast calculations. Although Patent CN 115310353B uses an improved NSGA-II algorithm for multi-objective optimization, this method still relies on some artificial solutions during initialization, which may lead to limited convergence efficiency of the algorithm. In addition, when facing more complex design scenarios, the adaptability and adjustment ability of the algorithm may be insufficient, and there is still a risk of falling into local optimal solutions. Summary of the Invention

[0010] Aiming at the deficiencies in the above-mentioned existing technologies, the present invention provides a machine learning-based transformer insulation structure design method that improves the efficiency and accuracy of transformer insulation structure design, optimizes the insulation performance, and realizes automatic optimization and feedback adjustment by analyzing a large amount of historical design data and simulation results and constructing complex mapping relationships.

[0011] The technical solution adopted by the present invention is:

[0012] A machine learning-based design method for transformer insulation structure, comprising:

[0013] Step S1, collecting parameters of previous design schemes and simulation data of transformer electric field distribution;

[0014] Step S2, establishing a feature engineering for transformer insulation design;

[0015] Step S3, constructing and training a prediction model for pre-transformer insulation structure parameters;

[0016] Step S4, applying the trained model to the simulation of a new design scheme;

[0017] Step S5, determining whether the electric field distortion rate and insulation size meet the requirements. If they meet the requirements, go to Step S6; if they do not meet the requirements, go back to Step S4 for iterative calculation;

[0018] Step S6, determining the transformer design condition parameters.

[0019] Preferably, in Step S1, collecting parameters of previous design schemes and simulation data of transformer electric field distribution includes:

[0020] Collecting transformer design data in existing design schemes;

[0021] With the help of a finite element analysis numerical simulation tool, performing electric field distribution simulation on the existing design scheme to obtain data of electric field intensity, electric field distortion rate, and insulation thickness of the transformer under actual working conditions;

[0022] The electric field distortion rate (D) is defined as:

[0023]

[0024] Where: E max and E min are respectively the maximum and minimum values of the electric field intensity; E avg is the average value of the electric field intensity.

[0025] Preferably, in Step S2, feature engineering is to convert the original design data and simulation results into features that can be effectively understood and processed by a machine learning model, and extract key parameters highly relevant to the performance of the transformer insulation structure;

[0026] By describing the complex relationship between electric field distribution, material properties, and working environment, feature engineering can improve the prediction ability and generalization performance of the model, thereby optimizing the insulation design, enhancing the application effect of the model in new schemes, and ensuring that the transformer meets the requirements of efficient and safe operation of modern power systems;

[0027] The process of establishing feature engineering includes the following steps:

[0028] First, collect multi-dimensional data related to transformer insulation design, such as material parameters, electric field distribution, environmental conditions, and historical design results.

[0029] Clean, normalize, and perform feature screening on the multi-dimensional data to ensure data quality and consistency.

[0030] Combined with electromagnetic field theory and simulation analysis, extract key features that can reflect insulation performance, such as local electric field strength and thermal field coupling characteristics.

[0031] Through dimensionality reduction techniques and feature combination, construct a feature set suitable for machine learning models to support efficient training and optimization of the models.

[0032] Feature selection is carried out by analyzing parameters such as voltage level, material dielectric constant, winding spacing, oil-paper thickness, and electric field strength, which directly affect the insulation performance and electric field distribution of the transformer.

[0033] In order to capture more complex design influencing factors, the interaction relationships between features are also constructed. After these features are selected, they can effectively improve the prediction ability and generalization ability of the model, enabling it to maintain efficient performance in different design tasks.

[0034] Feature engineering not only helps the model better understand the input data but also reduces the dimension of the data, improving the efficiency and accuracy of model training.

[0035] Preferably, in step S3, construct a pre-transformer insulation structure parameter prediction model and training method, and construct a machine learning model for predicting transformer insulation structure parameters.

[0036] Constructing a machine learning model for predicting transformer insulation structure parameters depends on high-quality data and appropriate algorithm selection.

[0037] First, collect and preprocess the data, including parameter normalization and outlier removal, to ensure the quality of the model input data.

[0038] Then, establish a feature set, combine key features of transformer insulation design, such as material properties, electric field strength, and environmental temperature, and train the model with the training data.

[0039] Finally, apply the trained model to the simulation data, verify its prediction accuracy and applicability in the new design scheme, and continuously iterate and optimize the model.

[0040] The machine learning model based on the random forest algorithm is trained using a large amount of historical data and simulation data to establish the mapping relationship between input features and target outputs, and gradually optimize the prediction results.

[0041] The steps of processing data by the random forest algorithm are as follows:

[0042] By constructing multiple decision trees and integrating their prediction results, the robustness and accuracy of the model are improved; when processing transformer insulation design data, the dataset is first divided into a training set and a test set, and the features are standardized; in the training stage, each tree randomly extracts some samples and features from the dataset for modeling to ensure the diversity of the model. Then, the outputs of all trees are integrated into the final prediction result by voting or averaging; the random forest algorithm has good generalization ability, can handle high-dimensional data and non-linear relationships, and has strong fault tolerance for missing data and noise, which is very suitable for the complex data modeling requirements in transformer insulation design;

[0043] The model is trained using the collected historical data and simulation data, and the model learns the mapping relationship between the input features and the target output to gradually optimize the prediction result;

[0044] Using the collected historical data and simulation data of previous transformer designs, first, the input features (such as material parameters, electric field distribution, and environmental conditions) and the target outputs (such as insulation performance indicators and electric field distribution results) in the data are labeled and corresponding to form training samples;

[0045] During the training process, the model continuously iterates to learn the non-linear mapping relationship between the input features and the target output, gradually reducing the prediction error;

[0046] After each iteration, the model adjusts the parameters to reduce the error and optimize its prediction ability. Finally, the prediction result is the evaluation of the insulation structure performance in the new transformer design scheme, such as the local electric field intensity distribution, insulation breakdown voltage, and thermal field stability, to ensure that the new scheme meets the expected insulation design requirements;

[0047] The performance of the model is continuously evaluated during the training process, and cross-validation technology is used to ensure that the model does not overfit.

[0048] Preferably, in step S4, the trained model is applied to the simulation of the new design scheme;

[0049] The trained model is directly applied to the new transformer design;

[0050] After the designer inputs the new design conditions, the model will output the corresponding insulation structure parameters according to the trained mapping relationship;

[0051] These prediction results need to be verified for their rationality through simulation tools to ensure that the design scheme can achieve the expected performance in actual applications;

[0052] Simulation verification evaluates whether the predicted insulation structure of the model meets the design specifications by calculating key indicators such as the electric field distribution and electric field distortion rate of the new design scheme.

[0053] The steps of simulation verification are as follows:

[0054] In simulation verification, by inputting the insulation structure parameters predicted by the model, the electric field distribution and key performance indicators of the new design scheme, such as the local electric field strength and electric field distortion rate, are calculated in the simulation tool. During the simulation process, the results are compared with the design specifications to ensure that all indicators meet the requirements. If the simulation results show that the electric field distortion rate in some areas is too high or the electric field distribution is uneven, the input parameters need to be adjusted and the simulation is repeated until the predicted results meet the expected performance in the simulation verification, thereby ensuring the reliability and safety of the insulation structure design.

[0055] Preferably, in step S5, it is determined whether the electric field distortion rate, insulation size, and electric field uniformity meet the requirements;

[0056] After completing the simulation verification, it is necessary to evaluate the design scheme predicted by the model to determine whether it meets the predetermined design requirements;

[0057] The key indicators include the electric field distortion rate, insulation thickness, and electric field uniformity;

[0058] If the simulation results show that these indicators are all within the range of the design specifications, the design scheme can be considered feasible;

[0059] If the simulation results do not meet the requirements, a feedback mechanism is initiated to analyze the differences between the model prediction results and the actual simulation, and the model or feature selection is adjusted based on these differences.

[0060] The beneficial effects of the present invention compared with the prior art:

[0061] The method for designing a transformer insulation structure based on machine learning according to the present invention automatically constructs a complex mapping relationship by analyzing a large amount of historical design data and simulation results, thereby improving the efficiency and accuracy of the transformer insulation structure design, optimizing the insulation performance, and realizing automatic optimization and feedback adjustment to solve the limitations of traditional design methods in dealing with complex electric field distributions and multi-factor influences, and ensuring that the transformer design meets the high standards and high requirements of modern power systems.

[0062] Compared with the traditional methods that rely on manual standards and designers' experience in the prior art, by introducing machine learning technology, the present invention can automatically process complex design tasks, significantly reducing the time for manual calculation and simulation verification. This automated design process significantly improves the design efficiency and greatly shortens the design cycle.

[0063] The present invention uses a large amount of historical data and simulation data for model training, and can more accurately predict the optimal insulation structure parameters. Compared with the possible empirical errors and manual calculation deviations in traditional methods, the design results provided by the present invention have higher accuracy and reliability, and can effectively cope with complex electric field distributions and multi-factor interactions.

[0064] Existing technologies usually rely on standard parameters under typical scenarios and are difficult to cope with special or complex operating conditions. Through the analysis of multi-dimensional data by the machine learning model of the present invention, the mutual relationships between different design factors can be better captured and processed, making the design scheme more flexible and capable of adapting to a wider range of working conditions and usage environments.

[0065] When the design parameters of existing technologies do not meet the requirements, manual adjustment usually needs to be carried out again, resulting in low efficiency. The present invention introduces an automatic optimization and feedback adjustment mechanism. When the insulation structure parameters predicted by the model do not meet the design requirements, feedback adjustment can be quickly carried out to automatically optimize the design scheme and ensure that the design meets all standards and requirements.

[0066] Existing technologies rely on manual design and simulation, and are prone to errors due to the experience and judgment of designers. The present invention reduces the interference of human factors through a data-driven machine learning model, making the design process more objective and stable, and ensuring the consistency and reliability of the design results. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 is a schematic flow chart of a method for designing a transformer insulation structure based on machine learning;

[0068] Figure 2 is a schematic flow chart of a method for collecting parameters of previous design schemes and simulation data of transformer electric field distribution;

[0069] Figure 3 is a schematic flow chart of a method for constructing and training a pre-transformer insulation structure parameter prediction model. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0070] The present invention will be described in detail below with reference to the accompanying drawings and embodiments:

[0071] Attached Figure 1 It can be seen that a method for designing a transformer insulation structure based on machine learning

[0072] Step S1, collect parameters of previous design schemes and simulation data of transformer electric field distribution;

[0073] Step S2, establish a transformer insulation design feature engineering;

[0074] Step S3, construct and train a pre-transformer insulation structure parameter prediction model;

[0075] Step S4: Apply the trained model to the simulation of the new design solution;

[0076] Step S5: Determine whether the electric field distortion rate and the insulation size meet the requirements. If they meet the requirements ("yes"), go to Step S6; if they do not meet the requirements ("no"), go back to Step S4 to perform iterative calculations again;

[0077] Step S6: Determine the transformer design condition parameters.

[0078] As Figure 2 shown, in transformer design, historical design data and simulation data are the basis for constructing a machine learning model.

[0079] Preferably, Step S1: Collect the parameters of previous design solutions and the simulation data of the transformer electric field distribution, including:

[0080] Systematically collect key design parameters, namely: the transformer design data in the existing design solutions; these data include voltage level, current magnitude, geometric structure, and insulation material properties, and all these data are associated with the design and performance of the transformer;

[0081] With the help of a finite element analysis numerical simulation tool, systematically collect key design parameters and perform electric field distribution simulation on the existing design solutions, that is, obtain the electric field strength, electric field distortion rate, and insulation thickness data of the transformer under actual working conditions;

[0082] Denoise and standardize the simulation data. The simulation data can provide detailed electric field distribution information that is difficult to obtain by traditional design methods, thus more comprehensively reflecting the influence of design parameters on the performance of the transformer. To ensure the quality of the training data of the model, the collected data is denoised, standardized, and the deviated data is removed to ensure the reliability of the data and the accuracy of the model prediction.

[0083] The electric field distortion rate (D) is defined as:

[0084]

[0085] In the formula: E max and E min are respectively the maximum and minimum values of the electric field strength; E avg is the average value of the electric field strength.

[0086] Preferably, Step S2: Establish the transformer insulation design feature engineering, including:

[0087] Feature engineering is to convert the original data into a set of features that can be effectively utilized by machine learning models. For transformer insulation design, it is crucial to extract key features from the original design parameters and simulation data.

[0088] Feature engineering is to transform the original design data and simulation results into features that can be effectively understood and processed by machine learning models, and extract key parameters that are highly relevant to the performance of the transformer insulation structure;

[0089] By describing the complex relationships among the electric field distribution, material properties, and working environment, feature engineering can improve the prediction ability and generalization performance of the model, thereby optimizing the insulation design, enhancing the application effect of the model in new schemes, and ensuring that the transformer meets the requirements of efficient and safe operation of modern power systems;

[0090] The process of establishing feature engineering includes the following steps:

[0091] First, collect multi-dimensional data related to transformer insulation design, such as material parameters, electric field distribution, environmental conditions, and historical design results;

[0092] Clean, normalize, and perform feature screening on the multi-dimensional data to ensure data quality and consistency;

[0093] By removing outliers in the design schemes and simulation data and filling in missing values, the integrity and accuracy of the data are guaranteed. Then, normalize data with different dimensions, such as converting parameters like electric field strength and insulation thickness to a unified scale range, so as to ensure that during model training, the influence of each parameter on the prediction result remains consistent. Finally, remove factors with low correlation to the transformer insulation performance through feature screening, and retain key feature data such as material parameters and electric field distribution to optimize the input parameters of the model.

[0094] Combined with electromagnetic field theory and simulation analysis, extract key features that can reflect insulation performance, such as local electric field strength and thermal field coupling characteristics;

[0095] Through finite element analysis, simulate the electric field distortion rate, maximum and minimum electric field strengths, average electric field strength of the transformer under working conditions, and the thermal stability of materials under different working environments, and extract local electric field strength, insulation thickness, and thermal field coupling characteristics, etc. as key features.

[0096] Through dimensionality reduction techniques and feature combination, construct a feature set suitable for machine learning models to support the efficient training and optimization of the model;

[0097] Using dimensionality reduction techniques, the most representative features are extracted from high-dimensional electric field simulation data to reduce feature redundancy. Then, through feature combination, key features such as electric field intensity, thermal field coupling characteristics, and material parameters are combined to form new composite features, enabling these features to more accurately reflect the insulation performance.

[0098] Feature selection analyzes parameters such as voltage level, material dielectric constant, winding spacing, oil-paper thickness, and electric field intensity parameters, which directly affect the insulation performance and electric field distribution of transformers.

[0099] Feature selection analyzes parameters such as voltage level, material dielectric constant, winding spacing, oil-paper thickness, and electric field intensity, identifies and screens out the key features that have the most impact on the insulation performance and electric field distribution of transformers, thus simplifying the dataset. Based on feature selection, "feature combination" appropriately combines these selected important features to generate new composite features, which can more comprehensively and effectively capture the complex relationships between parameters, thereby improving the prediction performance and accuracy of the model.

[0100] To capture more complex design influencing factors, the interaction relationships between features are also constructed. After these features are selected, they can effectively improve the prediction ability and generalization ability of the model, enabling it to maintain efficient performance in different design tasks.

[0101] The construction of the interaction relationships between features is achieved by identifying and calculating the interactions between different features and their comprehensive impact on the insulation performance of transformers.

[0102] By analyzing features such as voltage level, material dielectric constant, winding spacing, oil-paper thickness, and electric field intensity, it can be found how the combination of these features jointly affects the electric field distribution.

[0103] This process usually uses technical methods such as polynomial expansion or interaction term models based on statistics and machine learning to capture complex non-linear relationships.

[0104] Feature selection is to screen out the key features that are most critical to the model prediction effect through feature importance evaluation (such as feature importance scoring based on statistical indicators, machine learning algorithms, etc.), ensuring that the model maintains efficient performance in different design tasks. This method of feature selection and construction of interaction relationships can improve the prediction ability and generalization ability of the model.

[0105] Feature engineering not only helps the model better understand the input data, but also reduces the dimensionality of the data, improving the efficiency and accuracy of model training.

[0106] Such as Figure 3As shown, the random forest algorithm is selected as the machine learning model for predicting the insulation structure of the transformer; the collected historical design data and simulation data are used as the input data of the model; the final prediction model is formed, the performance of the model is evaluated, and the model parameters are adjusted; the final prediction model is formed.

[0107] Preferably, in step S3, a prediction model and training method for the parameters of the pre-transformer insulation structure are constructed.

[0108] Construct a machine learning model for predicting the parameters of the transformer insulation structure.

[0109] Constructing a machine learning model for predicting the parameters of the transformer insulation structure depends on high-quality data and appropriate algorithm selection.

[0110] First, collect and preprocess the data, including parameter normalization and outlier removal, to ensure the quality of the model input data.

[0111] Then, establish a feature set, combine the key features of the transformer insulation design, such as material properties, electric field strength, and ambient temperature, and train the model with the training data.

[0112] Finally, apply the trained model to the simulation data, verify its prediction accuracy and applicability in the new design scheme, and continuously iterate and optimize the model.

[0113] Based on the machine learning model of the random forest algorithm, a large amount of historical data and simulation data are used for training to establish the mapping relationship between the input features and the target output, and gradually optimize the prediction results.

[0114] The steps for the random forest algorithm to process data are as follows:

[0115] By constructing multiple decision trees and integrating their prediction results, the robustness and accuracy of the model are improved; when processing the transformer insulation design data, first divide the data set into a training set and a test set, and standardize the features; in the training stage, each tree randomly extracts some samples and features from the data set for modeling to ensure the diversity of the model. Then, through voting or averaging, the outputs of all trees are integrated into the final prediction result; the random forest algorithm has good generalization ability, can handle high-dimensional data and non-linear relationships, and has strong fault tolerance for missing data and noise, which is very suitable for the complex data modeling requirements in transformer insulation design.

[0116] Use the collected historical data and simulation data to train the model, and the model learns the mapping relationship between the input features and the target output to gradually optimize the prediction results.

[0117] Using the historical data and simulation data of previous transformer designs collected, first, the input features (such as material parameters, electric field distribution, and environmental conditions) in the data are labeled and corresponded with the target outputs (such as insulation performance indicators and electric field distribution results) to form training samples;

[0118] During the training process, the model continuously iterates to learn the non-linear mapping relationship between the input features and the target outputs, gradually reducing the prediction error;

[0119] After each iteration, the model adjusts the parameters to reduce the error and optimize its prediction ability. Finally, the prediction result is an evaluation of the insulation structure performance in the new transformer design scheme, such as the local electric field intensity distribution, insulation breakdown voltage, and thermal field stability, to ensure that the new scheme meets the expected insulation design requirements;

[0120] The performance of the model is continuously evaluated during the training process by using cross-validation techniques to ensure that the model does not overfit.

[0121] Preferably, in step S4: Apply the trained model to the simulation of the new design scheme;

[0122] The trained model is directly applied to the new transformer design;

[0123] After the designer inputs the new design conditions, the model will output the corresponding insulation structure parameters according to the trained mapping relationship;

[0124] These prediction results need to be verified for their rationality through simulation tools to ensure that the design scheme can achieve the expected performance in actual applications;

[0125] The simulation verification evaluates whether the insulation structure predicted by the model meets the design specifications by calculating key indicators such as the electric field distribution and electric field distortion rate of the new design scheme.

[0126] The steps of the simulation verification are as follows:

[0127] The simulation verification calculates the electric field distribution and key performance indicators of the new design scheme, such as the local electric field intensity and electric field distortion rate, by inputting the insulation structure parameters predicted by the model; during the simulation process, the results are compared with the design specifications to ensure that all indicators meet the requirements; if the simulation results show that the electric field distortion rate in some areas is too high or the electric field distribution is uneven, the input parameters need to be adjusted and the simulation is repeated until the prediction results meet the expected performance in the simulation verification, thereby ensuring the reliability and safety of the insulation structure design.

[0128] Preferably, in step S5, determine whether the electric field distortion rate and insulation size meet the requirements;

[0129] After completing the simulation verification, it is necessary to evaluate the design scheme predicted by the model to determine whether it meets the predetermined design requirements;

[0130] The key indicators include the electric field distortion rate, insulation thickness, and electric field uniformity;

[0131] If the simulation results show that these indicators are within the design specifications, the design scheme can be considered feasible;

[0132] If the simulation results do not meet the requirements, a feedback mechanism is initiated to analyze the differences between the model prediction results and the actual simulation, and the model or feature selection is adjusted based on these differences.

[0133] When the simulation results do not meet the requirements, the design is optimized by adjusting multiple key parameters. First, adjust the thickness or number of layers of the insulation layer to improve the overall insulation strength and breakdown voltage. Second, affect the electric field distribution by changing the dielectric constant and other properties (such as dielectric loss) of the material, thereby improving the electric field distortion rate. Modify the boundary conditions or apply different voltage waveforms to make the electric field more uniform. Finally, consider optimizing the mesh density and time step of the model to improve the accuracy and stability of the simulation.

[0134] This feedback process will continuously optimize the model to enable it to more accurately predict the design results, ultimately ensuring the reliability and effectiveness of the design scheme.

[0135] Step S6, determine the design condition parameters of the transformer;

[0136] When the simulation verification results meet all the design requirements, the final design scheme of the transformer insulation structure can be determined;

[0137] The design parameters predicted by the model have passed the simulation verification, proving that it can perform excellently in practical applications. The output of the design scheme includes specific parameters such as the thickness of the insulation material and the spacing between windings, which will be directly used in production and manufacturing.

[0138] The core innovation of the present invention lies in introducing machine learning technology into the design process of the transformer insulation structure. By analyzing a large amount of historical design data and simulation results, a complex mapping relationship is automatically constructed, thereby improving the design efficiency and accuracy and avoiding the errors caused by experience and manual calculation in traditional design methods.

[0139] The present invention establishes a transformer insulation design feature engineering, extracts and optimizes key features, such as voltage level, material dielectric constant, winding spacing, oil-paper thickness, and electric field strength, etc. These features are carefully selected and interactively constructed to enhance the prediction ability of the machine learning model and enable it to maintain efficient performance in different design tasks.

[0140] The present invention proposes a machine learning model based on the random forest algorithm and uses a large amount of historical data and simulation data for training to establish a mapping relationship between input features and target outputs, and gradually optimize the prediction results.

[0141] The present invention proposes a set of simulation verification and feedback optimization mechanisms. The design scheme predicted by the model is verified through a simulation tool to judge its rationality and effectiveness. If the design result does not meet the expectations, the present invention will activate the feedback mechanism, analyze the differences between the model prediction and the actual simulation, and make adjustments to the model or feature selection to ensure the reliability of the final design scheme.

[0142] The automatic optimization and feedback adjustment mechanism in the present invention enables timely adjustment and optimization when the insulation structure parameters predicted by the model do not meet the design requirements, thereby ensuring the efficiency and reliability of the design scheme and solving the limitations of traditional design methods in the face of complex electric field distributions and multi-factor influences.

[0143] The method for designing the transformer insulation structure based on machine learning in the present invention automatically constructs a complex mapping relationship by analyzing a large amount of historical design data and simulation results, thereby improving the efficiency and accuracy of the transformer insulation structure design, optimizing the insulation performance, and realizing automatic optimization and feedback adjustment to solve the limitations of traditional design methods in dealing with complex electric field distributions and multi-factor influences, and ensuring that the transformer design meets the high standards and requirements of modern power systems.

[0144] Compared with the traditional method that relies on manual standards and designers' experience in the prior art, the present invention can automatically process complex design tasks by introducing machine learning technology, greatly reducing the time for manual calculation and simulation verification. This automated design process significantly improves the design efficiency and greatly shortens the design cycle.

[0145] The present invention uses a large amount of historical data and simulation data for model training, and can more accurately predict the optimal insulation structure parameters. Compared with the possible empirical errors and manual calculation deviations in traditional methods, the design results provided by the present invention have higher accuracy and reliability, and can effectively cope with complex electric field distributions and multi-factor interactions.

[0146] The prior art usually relies on standard parameters under typical scenarios and is difficult to cope with special or complex operating conditions. Through the analysis of multi-dimensional data by the machine learning model in the present invention, the mutual relationships between different design factors can be better captured and processed, making the design scheme more flexible and capable of adapting to a wider range of working conditions and usage environments.

[0147] When the design parameters of the prior art do not meet the requirements, manual adjustment usually needs to be carried out again, resulting in low efficiency. The present invention introduces an automatic optimization and feedback adjustment mechanism. When the insulation structure parameters predicted by the model do not meet the design requirements, it can quickly perform feedback adjustment, automatically optimize the design scheme, and ensure that the design meets all standards and requirements.

[0148] The prior art relies on manual design and simulation, which is prone to errors due to the experience and judgment of designers. The present invention reduces the interference of human factors through a data-driven machine learning model, making the design process more objective and stable, and ensuring the consistency and reliability of the design results.

[0149] In some embodiments, the machine learning-based transformer insulation structure design method according to the present invention can be understood as a dynamic combination of some implementation details, especially when it comes to algorithms, calculation methods, or software systems or devices. Key variables in the system or method can be set, adjusted, or optimized according to the details of the construction method, so as to achieve those expected technical effects. Some details of the machine learning-based transformer insulation structure design method of the present invention can be understood as including parameters in a definable model, especially in those embodiments such as machine learning. The details of the present invention can be used to describe how to select and adjust the hyperparameters of a neural network. In some embodiments of the present invention, those details can be defined as key parameters in an algorithm, or configuration parameters between various components of a certain system, and the overall performance of the system or method is optimized through these parameters. In some embodiments of the present invention, the method steps or details or their permutations and combinations of the present invention can be described by modules or devices. For example, the technical details or features of the nth device or the nth module implementing the nth step. In some method embodiments of the method of the present invention, the result of the last step obtained by permutation and combination can be output to the manufacturing device or equipment of the corresponding transformer or transformer component to achieve automatic production, or at the same time, the result is fed back to the starting step to further form a more optimal structure parameter construction. After multiple iterations, the result of the last step is output to the manufacturing device or equipment of the corresponding transformer or transformer component to achieve more optimal automatic production.

[0150] As described above, it is only a preferred embodiment of the present invention, and there is no limitation on the structure of the present invention in any form. Any simple modification, equivalent change, and modification made to the above embodiments based on the technical essence of the present invention all fall within the scope of the technical solution of the present invention.

Claims

1. A design method for the insulation structure of a transformer based on machine learning, characterized in that, Including: Step S1: Collect the parameters of previous design schemes and the simulation data of the transformer electric field distribution; Step S2: Establish the feature engineering of the transformer insulation design; Step S3: Construct and train the prediction model of the pre-transformer insulation structure parameters; Step S4: Apply the trained model to the simulation of the new design scheme; Step S5: Determine whether the electric field distortion rate and the insulation size meet the requirements. If they meet the requirements, go to Step S6; if they do not meet the requirements, go back to Step S4 for iterative calculation; Step S6: Determine the design condition parameters of the transformer; Step S3: Construct and train the prediction model of the pre-transformer insulation structure parameters, including the steps of: Collect and preprocess the data, including parameter normalization and outlier removal; Establish the feature set by combining the key features of the transformer insulation design; And train the model with the training data; Apply the trained model to the simulation data to verify its prediction accuracy and applicability in the new design scheme, and continuously iterate and optimize the model; Based on the machine learning model of the random forest algorithm, use a large amount of historical data and simulation data for training, establish the mapping relationship between the input features and the target output, and gradually optimize the prediction results; The steps of processing data by the random forest algorithm are as follows: By constructing multiple decision trees and integrating their prediction results, when processing the transformer insulation design data, first divide the data set into a training set and a test set, and perform standardization processing on the features; In the training stage, each tree randomly extracts some samples and features from the data set for modeling; Integrate the output of all trees into the final prediction result by voting or taking the average; Use the collected historical data and simulation data to train the model. The model learns the mapping relationship between the input features and the target output to gradually optimize the prediction results; Use the collected historical data and simulation data of previous transformer designs to label and correspond the input features and the target output in the data to form training samples; During the training process, the model continuously iterates to learn the non-linear mapping relationship between the input features and the target output, gradually reducing the prediction error; After each iteration, the model adjusts the parameters to reduce the error and optimize its prediction ability; finally, the prediction result is the evaluation of the insulation structure performance in the new transformer design scheme; The performance of the model is continuously evaluated during the training process, and cross-validation is used to ensure that the model does not overfit.

2. The method for designing the transformer insulation structure based on machine learning according to claim 1, wherein: Step S1: Collect the parameters of previous design schemes and the simulation data of the transformer electric field distribution, including the steps of: Collect the transformer design data in the existing design schemes; With the help of the finite element analysis numerical simulation tool, perform the electric field distribution simulation on the existing design scheme to obtain the electric field intensity, electric field distortion rate, and insulation thickness data of the transformer under actual working conditions; The electric field distortion rate D is defined as: where: E max and E min are the maximum and minimum values of the electric field strength respectively; E avg is the average value of the electric field strength.

3. The method for designing the transformer insulation structure based on machine learning according to claim 1, wherein: Step S2: Establish the feature engineering of the transformer insulation design, including the steps of: Collect multi-dimensional data related to transformer insulation design; Clean, normalize, and perform feature screening on the multi-dimensional data; Combined with electromagnetic field theory and simulation analysis, extract key features that can reflect insulation performance; Through dimensionality reduction techniques and feature combination, construct a feature set suitable for machine learning models to support the efficient training and optimization of the models; Determine feature selection by analyzing parameters such as voltage level, material dielectric constant, winding spacing, oil-paper thickness, and electric field strength.

4. The machine learning-based transformer insulation structure design method according to claim 1, characterized in that: Step S4, apply the trained model to the simulation of the new design scheme, including the steps: The trained model is directly applied to the new transformer design; After inputting the new design conditions, the model will output the corresponding insulation structure parameters according to the trained mapping relationship; The rationality of these prediction results is verified through simulation tools; The simulation verification evaluates whether the insulation structure predicted by the model meets the design specifications by calculating key indicators such as the electric field distribution and electric field distortion rate of the new design scheme; The steps of the simulation verification are as follows: The simulation verification calculates the electric field distribution and key performance indicators of the new design scheme in the simulation tool by inputting the insulation structure parameters predicted by the model; During the simulation process, compare the results with the design specifications to ensure that all indicators meet the requirements; if the simulation results show that the electric field distortion rate in some areas is too high or the electric field distribution is uneven, adjust the input parameters and re-simulate until the prediction results meet the expected performance in the simulation verification.

5. The machine learning-based transformer insulation structure design method according to claim 1, characterized in that: Step S5, determine whether the electric field distortion rate, insulation size, and electric field uniformity meet the requirements, including the steps: After completing the simulation verification, evaluate the design scheme predicted by the model to determine whether it meets the predetermined design requirements; The key indicators include the electric field distortion rate, insulation thickness, and electric field uniformity; If the simulation results show that the indicators are within the range of the design specifications, the design scheme is considered feasible; If the simulation results do not meet the requirements, start the feedback mechanism, analyze the differences between the model prediction results and the actual simulation, and adjust the model or feature selection according to these differences.

Images