A pouring optimization method and system for a transformer winding

Abstract

The present application relates to the technical field of transformers, and provides a method and system for optimizing the casting of transformer windings, including: constructing a three-dimensional model of the transformer winding and obtaining a casting parameter space; randomly generating first casting parameters within the parameter space, combining the model for bubble prediction and insulation layer thickness distribution to obtain bubble parameters and thickness distribution; performing distortion recognition on the insulation layer distribution to obtain a distortion degree, and calculating a casting fitness in combination with the bubble parameters; configuring a casting adjustment step size according to the bubble parameters and the distortion degree, adjusting the first casting parameters, continuously optimizing the parameters, and finally obtaining the optimal casting parameters for winding casting. The present application solves the technical problems of bubble generation and uneven insulation layer thickness during the casting process of transformer windings, and realizes the technical effect of avoiding the generation of bubbles and improving the casting quality and performance of transformer windings by dynamically adjusting the casting process parameters.

Description

Technical Field

[0001] This application relates to the technical field of transformers, and particularly to a method and system for optimizing the casting of transformer windings. Background Art

[0002] A transformer is an important device in the power system, and the quality of its winding directly affects the working performance and reliability of the transformer. In the traditional manufacturing process of transformer windings, the casting process of the winding has a crucial impact on the insulation performance and mechanical strength of the transformer. Currently, the casting technology of transformer windings still faces many challenges, especially in terms of bubble control and the uniformity of the insulation layer thickness. The presence of bubbles may lead to non-uniform insulation layers, thereby affecting the thermal stability and electrical performance of the transformer; while the non-uniformity of the insulation layer thickness may lead to a decrease in the mechanical strength of the winding, and even cause failures under high-voltage working conditions. Traditional casting processes mostly rely on manual experience and the setting of quantitative parameters, but lack effective means for real-time monitoring and adjustment of the casting process. Traditional casting methods cannot optimize and adjust for each batch of windings, making it difficult to meet the requirements of more efficient and precise production. In addition, with the continuous complexity of the application environment of transformers, it is required that the windings maintain an efficient and stable working state within a wider temperature and voltage range, and traditional processes and equipment are already difficult to meet these requirements. Summary of the Invention

[0003] This application provides a method and system for optimizing the casting of transformer windings, aiming to solve the technical problems of bubble generation and non-uniform insulation layer thickness during the casting process of transformer windings.

[0004] In view of the above problems, this application provides a method and system for optimizing the casting of transformer windings.

[0005] In the first aspect disclosed in this application, a method for optimizing the casting of a transformer winding is provided. The method includes: constructing a three-dimensional model of the transformer winding to be cast, and obtaining a casting parameter space; randomly generating a first set of casting parameters within the casting parameter space, and combining the three-dimensional model of the winding to predict bubbles after casting and generate the insulation layer thickness distribution, obtaining a first set of bubble parameters and a first insulation layer thickness distribution; performing distortion recognition and analysis on the first insulation layer thickness distribution to obtain a first distortion degree, and calculating a first casting fitness in combination with the first set of bubble parameters; configuring a casting adjustment step size according to the first set of bubble parameters and the first distortion degree, adjusting the first set of casting parameters, and continuing to optimize the casting parameters to obtain the optimal casting parameters for casting the transformer winding.

[0006] Another aspect disclosed in this application provides a casting optimization system for a transformer winding. The system includes: constructing a three-dimensional model of the transformer winding to be cast and obtaining a casting parameter space; randomly generating a first set of casting parameters within the casting parameter space, and combining with the three-dimensional model of the winding to predict the bubbles after casting and generate the insulation layer thickness distribution, obtaining a first set of bubble parameters and a first insulation layer thickness distribution; performing distortion recognition and analysis on the first insulation layer thickness distribution to obtain a first distortion degree, and calculating a first casting fitness in combination with the first set of bubble parameters; configuring a casting adjustment step size according to the first set of bubble parameters and the first distortion degree, adjusting the first set of casting parameters, and continuing to optimize the casting parameters to obtain the optimal casting parameters for casting the transformer winding.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0008] The above-mentioned method for optimizing the casting of a transformer winding first constructs a three-dimensional model of the transformer winding to obtain the relevant casting parameter space, providing a basis for subsequent optimization; subsequently, randomly generates preliminary casting parameters within the casting parameter space and conducts simulations in combination with the three-dimensional model of the winding to predict the bubble distribution and the change in the insulation layer thickness after casting, providing preliminary data for the next optimization; then, uses the distortion recognition and analysis of the first insulation layer thickness distribution to obtain a first distortion degree, identifying uneven regions or potential structural problems in the insulation layer distribution, and then combines the casting fitness calculated from the first set of bubble parameters to help further evaluate the advantages and disadvantages of the casting scheme during the optimization process, avoiding insulation layer defects or structural instability caused by improper casting; finally, according to the comprehensive analysis of the first set of bubble parameters and the distortion degree, adjusts the step size of the casting parameters, gradually optimizes the casting scheme, and finally obtains the optimal casting parameters through repeated adjustment and optimization, thereby ensuring the quality of the transformer winding casting. This optimization process can significantly reduce problems caused by inappropriate parameters, such as excessive bubbles or uneven insulation layers, improve the stability and reliability of the casting process, and thus ensure the long-term performance and safety of the transformer winding.

[0009] The above description is only an overview of the technical solutions of this application. In order to be able to understand the technical means of this application more clearly, it can be implemented according to the content of the description. And in order to make the above and other purposes, features, and advantages of this application more obvious and understandable, the following specifically illustrates the embodiments of this application. Brief Description of the Drawings

[0010] To more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the attached drawings required for description in the embodiments. Obviously, the attached drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other attached drawings can be obtained based on these attached drawings.

[0011] Figure 1 It is a schematic flowchart of a casting optimization method for a transformer winding in an embodiment.

[0012] Figure 2 It is an architecture diagram of a casting optimization system for a transformer winding in an embodiment.

[0013] Explanation of reference numerals: Parameter space acquisition module 1, parameter distribution acquisition module 2, parameter distribution acquisition module 3, transformer winding casting module 4. Detailed implementation manners

[0014] The embodiments of the present application provide a casting optimization method and system for a transformer winding to solve the technical problems of bubble generation and uneven insulation layer thickness during the casting process of the transformer winding.

[0015] The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the attached drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0016] It should be noted that the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or server including a series of steps or units does not necessarily have to be limited to those clearly listed steps or units, but may include other steps or modules not clearly listed or inherent to these processes, methods, products or devices.

[0017] Embodiment 1, as Figure 1 shown, the present application provides a casting optimization method for a transformer winding, and the method includes:

[0018] Construct a three-dimensional model of the transformer winding to be cast and obtain the casting parameter space.

[0019] In the embodiment of the present application, the system terminal first needs to construct a three-dimensional model of the transformer winding to be cast. This process first collects the detailed geometric shape of the winding, including the arrangement and shape of the wires on the winding surface. Since the wires on the winding usually have uneven surfaces, the complexity of this structure requires that the casting process can accurately adapt to these changing surfaces. By establishing a three-dimensional model of the winding, the spatial distribution and morphological characteristics of each wire in the winding can be accurately described, providing accurate basic data for subsequent casting optimization. Based on this three-dimensional model, the system terminal will obtain the casting parameter space, which is a space containing the parameter of casting speed. The casting speed directly affects the injection process and flow state of the resin. In the case of an irregular winding surface, the casting speed needs to be precisely controlled to ensure that the resin can smoothly fill all the voids, avoiding problems such as air bubbles or insufficient resin infiltration.

[0020] Further, the present application provides a method for constructing a three-dimensional model of a transformer winding to be cast and obtaining a casting parameter space, including:

[0021] Performing surface topography acquisition on the transformer winding to be cast, and constructing a three-dimensional model of the transformer winding; obtaining a casting speed adjustment space for casting the transformer winding, and constructing it as a casting parameter space.

[0022] Preferably, the system terminal uses a high-precision three-dimensional scanner or other surface acquisition devices to scan the transformer winding to be cast. In this process, the geometric shape of the winding surface can be obtained through laser scanning or optical measurement technology. The scanning device can capture the shape, size, and unevenness of the winding wires. Through scanning, the obtained surface data is converted into point cloud data, which represents the spatial coordinate information of each point on the winding surface. Subsequently, these point cloud data are input into 3D modeling software (such as AutoCAD, SolidWorks, etc.). The 3D modeling software obtains a complete three-dimensional model of the winding through fitting and processing of the point cloud data. This model can accurately reflect the size, shape, and surface roughness of the winding, providing basic data for the simulation and optimization of the subsequent casting process. After the three-dimensional model of the winding is completed, it is necessary to construct a pouring speed adjustment space for the casting process. This space involves the key parameter for controlling the resin flow, namely the pouring speed. The pouring speed affects the flow behavior of the resin, which in turn affects whether the resin can uniformly penetrate into all parts of the winding. On the uneven surface of the winding wires, the flow speed of the resin needs to be precisely controlled to avoid uneven casting results. Therefore, by analyzing the geometric shape of different regions in the three-dimensional model of the winding and the physical characteristics of the resin flow during the casting process, a pouring speed adjustment space is constructed. This space contains the influencing factors of different pouring speeds on the resin flow and distribution. The construction of the pouring speed adjustment space needs to be combined with physical simulation software to simulate the resin flow at different speeds and generate a parameter space that can be adjusted in real time for subsequent optimization control. Finally, by using the constructed pouring speed adjustment space as the pouring parameter space, the construction of this pouring parameter space is the basis of the entire optimization, laying a solid theoretical and data foundation for the automatic adjustment and optimization of parameters during the optimization process.

[0023] Randomly generate a first pouring parameter within the pouring parameter space, and combine it with the three-dimensional model of the winding to predict the bubbles after pouring and generate the insulation layer thickness distribution, obtaining the first bubble parameter and the first insulation layer thickness distribution.

[0024] In one embodiment, the system terminal randomly generates a pouring speed as the first pouring parameter from within the pouring parameter space. Randomly generating the first pouring parameter is to explore different possibilities within the pouring parameter space. During the pouring process, the resin will flow and cover the surface of the winding. The unevenness and irregularities on the surface of the winding (such as the unevenness of the wires) will affect the infiltration effect of the resin, resulting in differences in the thickness of the insulating layer at different positions. In addition, changes in the pouring speed will also affect the fluidity of the resin, possibly leading to incomplete resin filling in some areas, thus generating bubbles or uneven insulating layer thickness. To accurately analyze this process, the system terminal, through the established bubble prediction path, predicts the number of bubbles that may be generated during the pouring process based on the first pouring parameter and the three-dimensional model of the winding, obtaining the first bubble parameter. The generation of these bubbles is usually caused by poor resin flow or air entrainment. These bubbles will affect the density and uniformity of the insulating layer. At the same time, based on the first pouring parameter and the three-dimensional model of the winding, the generation of the insulating layer thickness distribution after pouring will also be carried out. This step simulates the infiltration situation of the resin at different positions through the insulating layer thickness generation path, generating a predicted insulating layer thickness distribution as the first insulating layer thickness distribution. This first insulating layer thickness distribution can accurately reflect the thickness differences of the insulating layer at different positions, especially in the areas where the surface of the winding is uneven. Through these analysis processes, the first bubble parameter and the first insulating layer thickness distribution can be obtained. These data will help the system terminal understand the potential problems during the pouring process, and then provide a basis for optimizing the pouring parameters, ultimately achieving the goal of improving the quality of the transformer winding.

[0025] Further, the present application provides for randomly generating a first pouring parameter within the pouring parameter space, combining the three-dimensional model of the winding, performing bubble prediction and insulating layer thickness distribution generation after pouring, and obtaining the first bubble parameter and the first insulating layer thickness distribution, including:

[0026] Randomly generating a first pouring parameter within the pouring parameter space; inputting the first pouring parameter and the three-dimensional model of the winding into a pre-constructed bubble prediction path to predict and obtain the first bubble parameter, wherein the bubble prediction path is trained and constructed using a sample pouring parameter, a sample three-dimensional model set of windings, and a sample bubble parameter set, and each sample bubble parameter includes a sample bubble number; generating the insulating layer thickness distribution after pouring according to the three-dimensional model of the winding and the first pouring parameter to obtain the first insulating layer thickness distribution.

[0027] Preferably, the system terminal takes the first pouring parameter and the constructed three-dimensional winding model as input data and inputs them into a pre-constructed bubble prediction path. The bubble prediction path is established using machine learning and is specifically used to predict the bubble parameters generated under different pouring conditions. This path is trained by combining a set of sample pouring parameters (including sample pouring speeds), a set of sample three-dimensional winding models (including sample winding structures and surface morphologies of different types), and a set of sample bubble parameters (including the number of bubbles generated under different sample pouring conditions). Taking the 3D convolutional neural network (3D-CNN) in machine learning as an example, the system terminal constructs a bubble prediction path based on the 3D convolutional neural network, including an input layer, a 3D convolutional layer, a pooling layer, a fully connected layer, and an output layer, etc. Then, through methods such as voxelization and point cloud interpolation, the set of sample three-dimensional winding models is converted into a set of sample winding voxel data, which together with the set of sample pouring parameters and the set of sample bubble parameters form a training set, a validation set, and a test set. Subsequently, the convolutional kernels of the 3D convolutional layer and the weights of the fully connected layer are set using the random initialization method, and then the training set is input into the initialized bubble prediction path for forward propagation. The data is passed layer by layer through the input layer, the 3D convolutional layer, the pooling layer, the fully connected layer, etc., to generate predicted bubble parameters. After that, the loss value between the predicted bubble parameters and the sample bubble parameters is quantified by the mean squared error (MSE) loss function. The gradient of the loss with respect to each layer's weight is calculated layer by layer through the backpropagation algorithm, and the Adam optimizer is used to optimize the path parameters, adjusting the weights to minimize the value of the loss function and ensuring that the bubble prediction path gradually converges. This process will perform multiple training epochs. After each epoch, the performance of the bubble prediction path is tested using the validation set, the MSE on the validation set is calculated, and the accuracy of the bubble prediction path in the bubble number prediction task is evaluated. If the loss on the validation set does not decrease significantly, the hyperparameters such as the learning rate and batch size can be adjusted to further optimize the bubble prediction path. After passing the validation, the test set is used to test the current bubble prediction path. If the bubble prediction path can achieve the expected accuracy on the test set, the training process is completed, and the current bubble prediction path is output. Otherwise, the path structure is further adjusted or data augmentation techniques are used to improve the training effect. During use, this bubble prediction path can predict the number of bubbles generated during the actual pouring process based on the input three-dimensional winding model and the first pouring parameter, thereby obtaining the first bubble parameter. In addition, the system terminal analyzes the insulation layer thickness distribution after pouring the three-dimensional winding model and the first pouring parameter through an insulation layer thickness generation path constructed based on a generative adversarial network, and generates the first insulation layer thickness distribution according to the analysis results to reflect the distribution of the resin on the winding, thereby ensuring the uniformity and quality of the insulation layer.

[0028] Further, the present application provides for generating the thickness distribution of the insulation layer after casting according to the three-dimensional winding model and the first casting parameters to obtain the first insulation layer thickness distribution, including:

[0029] Collecting a set of sample casting parameters and a set of sample three-dimensional winding models according to the transformer winding casting detection data within the historical time, and collecting the insulation layer thicknesses at multiple positions of the transformer winding after casting with different sample casting parameters and sample three-dimensional winding models to obtain a set of sample insulation layer thickness distributions; constructing a generation path for the insulation layer thickness based on a generative adversarial network, wherein the generation path for the insulation layer thickness includes a generator and a discriminator; using the set of sample casting parameters, the set of sample three-dimensional winding models, and the set of sample insulation layer thickness distributions as supervised training data to train the generator and the discriminator until convergence to obtain the generation path for the insulation layer thickness; and inputting the three-dimensional winding model and the first casting parameters into the generation path for the insulation layer thickness to generate and obtain the first insulation layer thickness distribution.

[0030] Optionally, based on the casting detection data of the transformer winding within the historical time, the system terminal collects and obtains a sample casting parameter set and a sample winding three-dimensional model set, and also collects the insulation layer thickness data at multiple positions of the transformer winding after casting with different sample casting parameters and winding three-dimensional models, forming a sample insulation layer thickness distribution set. These three sets will provide supervised data for the subsequent training of the generative adversarial network (GAN). Subsequently, an insulation layer thickness generation path is constructed based on the generative adversarial network (GAN). This generation path includes a generator and a discriminator. The task of the generator is to generate a predicted insulation layer thickness distribution based on the input casting parameters and winding three-dimensional model. Its goal is to deceive the discriminator so that it cannot distinguish between real and generated thickness data. The discriminator is used to judge whether the generated thickness distribution is real. Its goal is to distinguish between the generated fake data and the real data, so that the thickness distribution generated by the generator is close enough to the actual situation. The generator and the discriminator are continuously optimized through adversarial training until the generator can generate an insulation layer thickness distribution highly consistent with the actual data. During the training process, the system terminal starts from the prepared sample data and uses the sample casting parameter set, the sample winding three-dimensional model set, and the sample insulation layer thickness distribution set as supervised training data. These data provide input information for the generator and help the generator learn the mapping relationship from the casting conditions and winding model to the insulation layer thickness distribution. At the beginning of the training process, the discriminator is trained first. In each training batch, the discriminator receives the real insulation layer thickness distribution data and the fake data generated by the generator. The discriminator outputs a probability value indicating the likelihood that the data is real data by judging these two types of data. In this process, the discriminator uses the cross-entropy loss function to measure the accuracy of its judgment and calculates the gradient through backpropagation to update the weights. Then the generator is trained. The input of the generator is the winding three-dimensional model and the casting parameters, and the output is the generated insulation layer thickness distribution. The generator does not directly calculate the error between the generated data and the real data, but adjusts the generated result through the feedback of the discriminator. Specifically, the generator updates the weights by maximizing the realness score of the discriminator for the data it generates, so that the generated thickness distribution is more real.The loss function of the generator is calculated based on the output of the discriminator. The generator is continuously optimized, and ultimately the generated thickness distribution becomes closer and closer to the actual data. This training process is carried out alternately. Each time during training, the discriminator is first optimized, and the generator is trained through the feedback of the discriminator. Through the judgment of the discriminator, the generator gradually adjusts the generation process to produce more realistic results. In this adversarial process, the generator and the discriminator play against each other. The discriminator attempts to distinguish between real and fake data, while the generator attempts to make the discriminator unable to distinguish the fake data it generates from the real data. The adversarial optimization of the generator and the discriminator will continue until the thickness distribution generated by the generator is realistic enough and the discriminator is also difficult to make an accurate judgment. The training process will last for multiple epochs. At the end of each epoch, the performance of the generator and the discriminator is evaluated using the validation set. Through the evaluation, it is judged whether the thickness distribution generated by the generator gradually approaches the real data to ensure the training effect of the generator and the discriminator. As the training progresses, the generator will be continuously optimized to generate a more and more realistic thickness distribution, while the discriminator will gradually improve its ability to distinguish between real and fake data. When the generator and the discriminator reach an adversarial balance, the training process is completed. After the training is over, the generator can accurately predict the corresponding insulation layer thickness distribution according to the new pouring parameters and the three-dimensional model of the winding. At this time, the system terminal inputs the three-dimensional model of the winding to be predicted and the first pouring parameter into the trained generator to generate the first insulation layer thickness distribution. This prediction result provides a basis for the subsequent pouring optimization, helps to adjust the key parameters during the pouring process, and thus ensures that the insulation layer thickness of the transformer winding is uniform and meets the quality requirements.

[0031] Perform distortion identification and analysis on the first insulation layer thickness distribution to obtain the first distortion degree, and calculate the first pouring fitness in combination with the first bubble parameter.

[0032] In one embodiment, after generating the first insulating layer thickness distribution, the next step is to perform distortion identification analysis on this distribution to ensure that the generated insulating layer thickness meets the expected quality standards. Since there are uneven thicknesses on the winding surface, it may cause the insulating layer to be too thin or too thick in some areas. Therefore, the system terminal needs to calculate the distortion degree to measure the non-uniformity of the insulating layer thickness. The calculation of the distortion degree is based on the thickness distribution of the insulating layer surface itself. By analyzing the depth difference between the concave and convex regions of the transformer winding and the thickness difference of the first insulating layer thickness distribution, the degree of non-uniformity is quantified, thereby obtaining the first distortion degree. This first distortion degree is used to describe the non-uniformity of the insulating layer. The larger the distortion degree, the more uneven the thickness of the insulating layer, which may affect the performance and safety of the transformer. Subsequently, the calculated first distortion degree is combined with the first bubble parameter, and the first casting fitness calculated through the fitness calculation formula is used to reflect the quality of the current casting process. The higher the fitness, the better the quality of the insulating layer, and the smaller the impact of bubbles and thickness distortion on the quality. Otherwise, the parameters of the casting process need to be adjusted to improve the casting effect and ensure that the final transformer winding meets the ideal quality standards.

[0033] Furthermore, the present application provides a method for performing distortion identification analysis on the first insulating layer thickness distribution to obtain a first distortion degree, and calculating a first casting fitness in combination with the first bubble parameter, including:

[0034] According to the three-dimensional winding model, extract the maximum depth difference between the concave and convex regions of the transformer winding; according to the first insulating layer thickness distribution, calculate the difference between the minimum insulating layer thickness and the maximum insulating layer thickness to obtain the first insulating layer thickness difference; calculate the ratio of the first insulating layer thickness difference to the maximum depth difference as the first distortion degree; calculate the first casting fitness of the first casting parameter according to the first distortion degree and the first bubble parameter.

[0035] Optionally, the system terminal extracts the characteristics of the winding surface morphology based on the three-dimensional winding model, especially the depression depth of the depression area and the protrusion height of the protrusion area. By adding the maximum protrusion height and the maximum depression depth, the maximum depth difference of the transformer winding is obtained. This maximum depth difference reflects the unevenness of the winding surface geometry. Subsequently, based on the first insulation layer thickness distribution, the minimum insulation layer thickness and the maximum insulation layer thickness are extracted, and the difference between the maximum insulation layer thickness and the minimum insulation layer thickness is calculated to obtain the first insulation layer thickness difference. The purpose of this step is to evaluate the thickness variation range of the insulation layer, that is, to identify the difference between the thinnest and thickest areas in the insulation layer. This difference indicates the uniformity of the insulation layer. The larger the difference, the higher the degree of non-uniformity of the insulation layer. After that, the first insulation layer thickness difference is compared with the maximum depth difference, and their ratio is calculated as the first distortion degree. The larger the first distortion degree, the greater the unevenness on the winding surface and the greater the thickness difference of the insulation layer, indicating that the pouring process may be relatively uneven. A lower distortion degree means that the unevenness of the winding surface is smaller, and at the same time, the thickness of the insulation layer is more uniform. Finally, the first distortion degree and the first bubble parameter are input into the fitness calculation formula to calculate the first pouring fitness. The higher the fitness, the more uniform the pouring process, the fewer bubbles, and the smaller the thickness difference of the insulation layer. On the contrary, it indicates that the pouring quality is poor and the parameters need to be optimized. Through this series of calculation steps, the fitness of the pouring process can be finally obtained, providing a basis for adjusting and optimizing the pouring parameters to ensure the uniformity and quality of the insulation layer of the transformer winding.

[0036] Furthermore, the present application provides a first pouring fitness for calculating the first pouring parameters according to the first distortion degree and the first bubble parameter, as follows: ; where JZF is the pouring fitness, is the bubble weight, is the distortion weight, The sum of and is 1, is the bubble parameter,

[0037] Optionally, the system terminal inputs the first distortion degree and the first bubble parameter into the fitness calculation formula. The specific content of the fitness calculation formula is as follows: ; where JZF is the pouring fitness, which is used to comprehensively evaluate the effect of the pouring process, is the bubble weight, indicating the proportion of the bubble parameter in the fitness evaluation, is the distortion weight, indicating the proportion of the distortion degree in the fitness evaluation, The sum of and is the bubble parameter, representing the number of bubbles during the actual casting process. is the threshold value of the number of bubbles for the transformer winding casting (determined based on business requirements). K is the thickness variance of the first insulation layer thickness distribution, reflecting the uniformity of the thickness distribution. The smaller the variance, the more uniform the distribution. J is the distortion degree, reflecting the relationship between the winding geometric shape and the non-uniformity of the insulation layer thickness. Through this formula, the first casting fitness comprehensively considers the two main factors of bubble generation and thickness distribution during the casting process, thus providing a quantitative basis for optimizing the casting process parameters.

[0038] Configure the casting adjustment step size according to the first bubble parameter and the first distortion degree, adjust the first casting parameter, continue to optimize the casting parameter, obtain the optimal casting parameter, and perform the casting of the transformer winding.

[0039] In one embodiment, during the optimization of the casting parameters, according to the first bubble parameter and the first distortion degree, configure an appropriate casting adjustment step size to iteratively adjust the current casting parameters. The configuration of the adjustment step size aims to balance the optimization speed and accuracy, ensure that the parameter adjustment can gradually approach the optimal solution, and at the same time avoid oscillations caused by too large a step size or slow convergence caused by too small a step size. Specifically, calculate the adjustment step size for each round of optimization according to the bubble parameter and the distortion degree of the thickness distribution. This process comprehensively considers the generation of bubbles and the non-uniformity of the insulation layer thickness distribution during the current casting process. If the bubble parameter and the distortion degree deviate far from the target range, the configured adjustment step size will be larger to quickly correct the parameters. When the bubble parameter and the distortion degree gradually approach the target value, the adjustment step size will gradually decrease to improve the optimization accuracy. During the adjustment process, each optimization is based on the current casting parameters, combined with the configured adjustment step size, to iteratively update the casting parameters. Through multiple rounds of iteration, continuously optimize the casting parameters to make them gradually approach the parameter combination that can achieve the best casting effect. The optimization goal is to make the first bubble parameter and the first distortion degree reach the expected values, thereby obtaining the best casting fitness. When the optimization converges, the finally determined casting parameters are the optimal casting parameters. By performing the casting of the transformer winding with the determined optimal casting parameters, the generation of bubbles can be minimized, ensuring the thickness uniformity and overall quality of the insulation layer, thus meeting the quality requirements of actual production.

[0040] Furthermore, the present application provides configuring the casting adjustment step size according to the first bubble parameter and the first distortion degree, adjusting the first casting parameter, continuing to optimize the casting parameter, obtaining the optimal casting parameter, and performing the casting of the transformer winding, including:

[0041] Obtain a preset pouring adjustment step for adjusting the pouring parameters; calculate the bubble ratio of the first bubble parameter to the bubble quantity threshold value; multiply the mean value of the bubble ratio and the first distortion degree by the preset pouring adjustment step to obtain the pouring adjustment step; use the pouring adjustment step to adjust the first pouring parameter, continue to optimize the pouring parameters until convergence, output the pouring parameter with the maximum pouring fitness, obtain the optimal pouring parameter, and perform the pouring of the transformer winding.

[0042] Preferably, during the process of optimizing the pouring parameters, the system terminal first obtains a preset pouring adjustment step. This preset step is the default base value in the initial adjustment process and is determined based on historical experience and expert decision-making, and is used to control the amplitude of parameter update. Subsequently, calculate the bubble ratio according to the first bubble parameter and the set bubble quantity threshold value. This ratio reflects the deviation degree of bubbles in the current pouring process. To ensure that the adjustment step can dynamically adapt to the current pouring effect, the system terminal combines the mean value of the bubble ratio and the first distortion degree to correct the preset step. Specifically, when the bubble parameter and the distortion degree deviate greatly from the ideal range, the pouring effect is poor, and at this time the step will be enlarged to jump out of the current poor pouring parameter area and quickly approach the potential optimal pouring parameter. On the contrary, when the bubble parameter and the distortion degree gradually approach the ideal value, it indicates that the pouring effect is good, and at this time the step will be reduced to perform fine adjustment near the optimal parameter to find the possible better solution. The specific way to obtain the pouring adjustment step is to multiply the mean value of the bubble ratio and the distortion degree by the preset pouring adjustment step to obtain the pouring adjustment step required for the current iteration. Through this dynamic adjustment, the self-adaptive change of the step can be realized, which can not only quickly jump out of the poor solution area but also perform precise optimization near the optimal solution. Then, use the calculated pouring adjustment step to update the current pouring parameter. After the parameter is updated, evaluate the pouring effect again and repeat the above process. Through multiple rounds of iteration, continuously optimize the pouring parameters until the parameter adjustment converges, that is, the pouring fitness reaches the expected fitness. Finally, output the pouring parameter with the maximum pouring fitness as the optimal solution. The obtained optimal pouring parameter will be used in the actual pouring process of the transformer winding. This parameter can ensure the uniform thickness of the insulating layer while minimizing the generation of bubbles to achieve the best pouring effect and meet the requirements of high-quality production.

[0043] In summary, the embodiments of the present application have at least the following technical effects:

[0044] In the embodiment of the present application, by constructing a three-dimensional model of the winding and a pouring parameter space, combining the bubble prediction and the method for generating the insulation layer thickness distribution, the pouring quality is optimized. The pouring fitness is calculated through the distortion degree of the bubble parameters and the insulation layer thickness distribution, and the thickness distribution is predicted by using a generative adversarial network. Furthermore, the step size of the pouring parameters is dynamically adjusted, and the pouring process is continuously optimized. Finally, the optimal pouring parameters are obtained to ensure the uniformity of the insulation layer and the consistency of the quality, and improve the pouring efficiency and product performance. These technical effects jointly solve the technical problems of bubble generation and uneven insulation layer thickness during the pouring process of the transformer winding, and achieve the technical effect of avoiding the generation of bubbles and improving the pouring quality and performance of the transformer winding by dynamically adjusting the parameters of the pouring process.

[0045] Embodiment 2, based on the same inventive concept as the pouring optimization method for a transformer winding in the foregoing embodiment, as Figure 2 shown, the present application provides a pouring optimization system for a transformer winding. The system includes: a parameter space acquisition module 1: constructing a three-dimensional model of the winding of the transformer winding to be poured, and acquiring a pouring parameter space; a parameter distribution obtaining module 2: randomly generating a first pouring parameter within the pouring parameter space, combining the three-dimensional model of the winding, predicting the bubbles after pouring and generating the insulation layer thickness distribution, and obtaining a first bubble parameter and a first insulation layer thickness distribution; a parameter distribution analysis module 3: performing distortion recognition analysis on the first insulation layer thickness distribution to obtain a first distortion degree, and calculating and obtaining a first pouring fitness in combination with the first bubble parameter; a transformer winding pouring module 4: configuring a pouring adjustment step size according to the first bubble parameter and the first distortion degree, adjusting the first pouring parameter, continuously optimizing the pouring parameter, obtaining the optimal pouring parameter, and pouring the transformer winding.

[0046] Furthermore, the parameter space acquisition module 1 is further used to execute the following method:

[0047] Collect the surface topography of the transformer winding to be poured, and construct a three-dimensional model of the winding of the transformer winding; acquire a pouring speed adjustment space for pouring the transformer winding, and construct it as a pouring parameter space.

[0048] Furthermore, the parameter distribution obtaining module 2 is further used to execute the following method:

[0049] Randomly generate the first casting parameters within the casting parameter space; input the first casting parameters and the three-dimensional winding model into a pre-constructed bubble prediction path to predict and obtain the first bubble parameters, where the bubble prediction path is trained and constructed using a set of sample casting parameters, a set of sample three-dimensional winding models, and a set of sample bubble parameters, and each sample bubble parameter includes the number of sample bubbles; generate the insulation layer thickness distribution after casting according to the three-dimensional winding model and the first casting parameters to obtain the first insulation layer thickness distribution.

[0050] Furthermore, the parameter distribution obtaining module 2 is further configured to execute the following method:

[0051] Collect a set of sample casting parameters and a set of sample three-dimensional winding models according to the transformer winding casting detection data within the historical time, and collect the insulation layer thicknesses at multiple positions of the transformer winding after casting with different sample casting parameters and sample three-dimensional winding models to obtain a set of sample insulation layer thickness distributions; based on the generative adversarial network, construct an insulation layer thickness generation path, where the insulation layer thickness generation path includes a generator and a discriminator; use the set of sample casting parameters, the set of sample three-dimensional winding models, and the set of sample insulation layer thickness distributions as supervised training data to train the generator and the discriminator until convergence to obtain the insulation layer thickness generation path; input the three-dimensional winding model and the first casting parameters into the insulation layer thickness generation path to generate and obtain the first insulation layer thickness distribution.

[0052] Furthermore, the parameter distribution analysis module 3 is further configured to execute the following method:

[0053] Extract the maximum depth difference between the concave region and the convex region of the transformer winding according to the three-dimensional winding model; calculate the difference between the minimum insulation layer thickness and the maximum insulation layer thickness according to the first insulation layer thickness distribution to obtain the first insulation layer thickness difference; calculate the ratio of the first insulation layer thickness difference to the maximum depth difference as the first distortion degree; calculate the first casting fitness of the first casting parameters according to the first distortion degree and the first bubble parameters.

[0054] Furthermore, the parameter distribution analysis module 3 is further configured to execute the following method:

[0055] Calculate the first casting fitness of the first casting parameters according to the first distortion degree and the first bubble parameters as follows: ; where JZF is the casting fitness, is the bubble weight, is the distortion weight, The sum of and is 1, The threshold value of the number of air bubbles cast for the transformer winding, K is the thickness variance of the thickness distribution of the first insulating layer, and J is the distortion degree.

[0056] Further, the transformer winding casting module 4 is further configured to execute the following method:

[0057] Obtain a preset casting adjustment step for adjusting the casting parameters; calculate the air bubble ratio of the first air bubble parameter to the air bubble number threshold value; multiply the average value of the air bubble ratio and the first distortion degree by the preset casting adjustment step to obtain a casting adjustment step; use the casting adjustment step to adjust the first casting parameter, continue to optimize the casting parameters until convergence, output the casting parameter with the maximum casting fitness, obtain the optimal casting parameter, and perform the casting of the transformer winding.

[0058] It should be noted that the above sequence of embodiments of the present application is only for description and does not represent the superiority or inferiority of the embodiments. And the above describes a specific embodiment of this specification. The processes depicted in the drawings do not necessarily require the specific order and continuous order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0059] The above are only the preferred embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0060] This specification and the drawings are only exemplary illustrations of the present application and are considered to have covered any and all modifications, variations, combinations, or equivalents within the scope of the present application. Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalent technologies, the present application is intended to include these changes and modifications.

Claims

1. A pouring optimization method for a transformer winding, characterized in that, The method includes: Construct a three-dimensional model of the transformer winding to be cast and obtain the casting parameter space; Randomly generate a first casting parameter within the casting parameter space, and combine it with the three-dimensional model of the winding to perform bubble prediction and insulation layer thickness distribution generation after casting, obtaining a first bubble parameter and a first insulation layer thickness distribution, including: Randomly generate a first casting parameter within the casting parameter space; Input the first casting parameter and the three-dimensional model of the winding into a pre-constructed bubble prediction path to predict and obtain a first bubble parameter, where the bubble prediction path is constructed by training with a set of sample casting parameters, a set of sample three-dimensional models of the winding, and a set of sample bubble parameters, and each sample bubble parameter includes the number of sample bubbles; Generate the insulation layer thickness distribution after casting according to the three-dimensional model of the winding and the first casting parameter to obtain a first insulation layer thickness distribution, including: Collect a set of sample casting parameters and a set of sample three-dimensional models of the winding according to the transformer winding casting detection data within a historical time, and collect the insulation layer thicknesses at multiple positions of the transformer winding after casting with different sample casting parameters and sample three-dimensional models of the winding to obtain a set of sample insulation layer thickness distributions; Based on the generative adversarial network, construct an insulation layer thickness generation path, where the insulation layer thickness generation path includes a generator and a discriminator; Use the set of sample casting parameters, the set of sample three-dimensional models of the winding, and the set of sample insulation layer thickness distributions as supervised training data to train the generator and the discriminator until convergence to obtain the insulation layer thickness generation path; Input the three-dimensional model of the winding and the first casting parameter into the insulation layer thickness generation path to generate and obtain a first insulation layer thickness distribution; Perform distortion recognition analysis on the first insulation layer thickness distribution to obtain a first distortion degree, and calculate a first casting fitness in combination with the first bubble parameter, including: Extract the maximum depth difference between the concave region and the convex region of the transformer winding according to the three-dimensional model of the winding; Calculate the difference between the minimum insulation layer thickness and the maximum insulation layer thickness according to the first insulation layer thickness distribution to obtain a first insulation layer thickness difference; Calculate the ratio of the first insulation layer thickness difference to the maximum depth difference as the first distortion degree; Calculate the first casting fitness of the first casting parameter according to the first distortion degree and the first bubble parameter as follows: ； Among them, JZF is the pouring fitness, is the bubble weight, is the distortion weight, and The sum of is 1, is the bubble parameter, is the threshold value of the number of bubbles in the transformer winding pouring, K is the thickness variance of the first insulation layer thickness distribution, and J is the distortion degree; Configure a casting adjustment step size according to the first bubble parameter and the first distortion degree, adjust the first casting parameter, continue to optimize the casting parameter, obtain the optimal casting parameter, and perform the casting of the transformer winding.

2. The casting optimization method of the transformer winding according to claim 1, characterized in that, Construct a three-dimensional model of the transformer winding to be cast and obtain the casting parameter space, including: Collect the surface topography of the transformer winding to be cast and construct a three-dimensional model of the transformer winding; Obtain the casting speed adjustment space for casting the transformer winding and construct it as the casting parameter space.

3. The pouring optimization method of the transformer winding according to claim 1, wherein Configure the pouring adjustment step according to the first bubble parameter and the first distortion degree, adjust the first pouring parameter, continue to optimize the pouring parameter, obtain the optimal pouring parameter, and perform the pouring of the transformer winding, including: Obtain the preset pouring adjustment step for adjusting the pouring parameter; Calculate the bubble ratio of the first bubble parameter to the bubble quantity threshold; Multiply the mean value of the bubble ratio and the first distortion degree by the preset pouring adjustment step to obtain the pouring adjustment step; Use the pouring adjustment step to adjust the first pouring parameter, continue to optimize the pouring parameter until convergence, output the pouring parameter with the maximum pouring fitness, obtain the optimal pouring parameter, and perform the pouring of the transformer winding.

4. An optimized casting system for a transformer winding, characterized in that, Steps for implementing the pouring optimization method of a transformer winding according to any one of claims 1 to 3, including: Parameter space acquisition module: Construct a three-dimensional winding model of the transformer winding to be poured and obtain the pouring parameter space; Parameter distribution acquisition module: Randomly generate the first pouring parameter within the pouring parameter space, combine the three-dimensional winding model, perform bubble prediction and insulation layer thickness distribution generation after pouring, and obtain the first bubble parameter and the first insulation layer thickness distribution; Parameter distribution analysis module: Perform distortion recognition analysis on the first insulation layer thickness distribution to obtain the first distortion degree, and calculate the first pouring fitness in combination with the first bubble parameter; Transformer winding pouring module: Configure the pouring adjustment step according to the first bubble parameter and the first distortion degree, adjust the first pouring parameter, continue to optimize the pouring parameter, obtain the optimal pouring parameter, and perform the pouring of the transformer winding.

Images