Solid medium electrical tree development path prediction method

Abstract

The invention relates to a solid medium electrical tree development path prediction method, and belongs to the field of reliability analysis and life prediction of high-voltage insulating materials. The method comprises the following steps: carrying out numerical simulation on a solid medium electrical tree development path by adopting a phase field method under different load grades; constructing a data set based on the generated time sequence image, and carrying out grouping storage on the data set by taking a load grade as a label; processing the classified data set to obtain a training set and a data set; constructing and training a machine learning model used for predicting an electrical tree development path, and learning change characteristics of the electrical tree development path under different load grades; and obtaining an image of a solid medium in an actual scene and load information, inputting the image and the load information into the trained machine learning model, predicting an electrical tree development path of the solid medium under the load, and generating an electrical tree development image sequence. According to the method, the prediction precision of the electric tree extension path is improved, the applicability of the model is expanded, and the method can adapt to complex working conditions of different load grades.

Description

Technical Field

[0001] The present invention belongs to the field of reliability analysis and life prediction of high-voltage insulating materials, relates to the prediction of electrical tree extension in solid insulating materials, and in particular to a method for predicting the development path of electrical tree in solid dielectrics by integrating numerical simulation and machine learning. Background Art

[0002] Polymer insulation materials have been widely used in electrical equipment due to their excellent insulation properties, corrosion resistance and processing flexibility. In particular, polymer materials have become the first choice in key components such as high-voltage cables, transformers and insulators due to their high dielectric strength and good mechanical properties. However, during long-term operation, polymer insulation materials may still be affected by various environmental factors and electric fields, causing their insulation properties to gradually decrease, thereby affecting the stable operation of the equipment. Therefore, studying the aging and damage mechanism of polymer insulation materials, especially the development process of electrical dendrites inside polymers, is crucial to improving the reliability of electrical equipment.

[0003] Under the action of high electric fields and different types of mechanical loads, electrical dendrites may gradually form inside polymer insulating materials. Electrical dendrites are branch-like damage structures that usually show obvious nonlinear characteristics in their expansion process as the load accumulates and acts. Load factors such as temperature, electric field strength, and mechanical stress can significantly affect the development path and rate of electrical dendrites. If electrical dendrites are not effectively controlled, their expansion may eventually lead to local breakdown of the insulating material, and in severe cases, even cause failure of the entire electrical equipment. Therefore, the prediction and control of the expansion of electrical dendrites under load conditions is the key to improving the life of insulating materials and equipment safety.

[0004] In order to ensure the long-term safe operation of electrical equipment, it is particularly important to grasp the expansion status of electrical trees in real time. Traditional methods for predicting the development path of electrical trees are usually based on numerical simulation (such as the phase field method), which can visualize and simulate the expansion of electrical trees under laboratory conditions. However, such methods are limited by computing resources and are difficult to apply in real time to actual working conditions. With the development of data-driven technology, prediction methods that combine numerical simulation with machine learning can provide faster and more accurate predictions of the development path of electrical trees, and are expected to achieve real-time monitoring and prediction of the expansion process of electrical trees. Summary of the invention

[0005] In view of this, the purpose of the present invention is to provide a method for predicting the development path of electrical trees in solid dielectrics. By integrating numerical simulation and machine learning to predict the development path of electrical trees, the expansion path of electrical trees can be efficiently predicted under complex load conditions, thereby providing important technical support for the safe maintenance of electrical equipment.

[0006] In order to achieve the above object, the present invention provides the following technical solutions:

[0007] A method for predicting the development path of electrical tree in solid dielectrics, the method comprising:

[0008] S1. Under different load levels, the phase field method is used to numerically simulate the development path of electrical tree in solid medium, and a time series image containing the development path of electrical tree is generated;

[0009] S2, constructing a data set based on the generated image data, and grouping and storing the data set with load levels as labels;

[0010] S3, uniformly sample the frames of the classified data set, convert the grayscale image into a three-channel RGB image and normalize it, then group and sample the image sequence according to the preceding and following order to construct a training set and a data set;

[0011] S4, constructing a machine learning model for predicting the development path of electrical trees, and using the training set to train the model so that the model learns the changing characteristics of the development path of electrical trees under different load levels;

[0012] S5. Obtain images of solid media in actual scenes and load information, and input them into the trained machine learning model to predict the electrical tree development path of the solid media under the load and generate an electrical tree development image sequence after several time steps in the future.

[0013] Furthermore, in step S1, at different load levels, the phase field method is used to perform numerical simulation on the development path of electrical tree in solid dielectrics, including:

[0014] 1) Determine simulation parameters and phase field parameters of solid media; calculate field variables based on energy distribution and physical laws;

[0015] The simulation parameters include the relative dielectric constant of the air gap and the dielectric phase, the breakdown electric field strength and the electric tree development energy barrier, and the phase field parameters of the solid medium include the simulation step size and the mesh size set according to the target solid medium size;

[0016] 2) Apply the phase field dynamics equation to update the order parameters and obtain the development state of the electrical tree at each moment;

[0017] 3) Iterate the simulation until the electrical tree causes the breakdown of the solid medium, forming a complete time series image.

[0018] Furthermore, in step S1, the different load levels are different load levels formed by combined loads applied to the material under different loading voltages, applied mechanical stresses, and applied temperatures.

[0019] Furthermore, in step S4, the constructed machine learning model includes a spatial encoder, a temporal module and a spatial decoder. The spatial encoder is used to extract the spatial features of the input image sequence and encode them into a latent space representation, which is then input into the temporal module; the temporal module converts the encoded data into temporal information that can be used by the spatial decoder, captures the time-dependent characteristics of the electrical tree extension process, and enhances the perception of temporal information; the spatial decoder decodes the data output by the temporal module, integrates the feature information of each time step, and generates the ability to learn and predict the global features of electrical tree development.

[0020] The beneficial effect of the present invention is that the present invention not only significantly improves the prediction accuracy of the extension path of the electrical tree by combining numerical simulation and machine learning, but also expands the applicability of the model to adapt to complex working conditions of different load levels. In addition, the introduction of simulation data makes model training more effective, and the timing prediction capability of machine learning significantly enhances the ability to capture the nonlinear extension path of the electrical tree, providing a new technical means for the reliability research of high-voltage insulating materials.

[0021] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below in conjunction with the accompanying drawings, wherein:

[0023] Figure 1 A schematic diagram of a flow chart of a method for predicting the development path of electrical tree in solid dielectrics provided by an embodiment of the present invention;

[0024] Figure 2 Examples of time series images of electrical tree development paths generated for numerical simulations;

[0025] Figure 3 It is the preprocessing process of image data;

[0026] Figure 4 Example of predicting electrical tree propagation paths for a machine learning model. DETAILED DESCRIPTION

[0027] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0028] like Figure 1 As shown, a method for predicting the development path of electrical tree branches by integrating numerical simulation and machine learning is provided in one embodiment of the present invention. The method is specifically as follows:

[0029] 1. Under different load levels, the development path of electrical trees in solid media is numerically simulated based on the phase field method to generate time series images reflecting the development path of electrical trees.

[0030] Different load levels refer to the combined loads applied to the material under different loading voltages, applied mechanical stresses, and applied temperatures, thus forming different load levels.

[0031] Phase field simulation initializes the simulation by setting the material load and physical properties (such as relative dielectric constant, breakdown electric field strength, energy barrier, etc.). By calculating the system and local energy, the minimum energy path for the extension of electrical trees is obtained according to the fracture variational theory. Phase field simulation includes:

[0032] 1) Determine the simulation parameters and phase field parameters of the solid medium, including the relative dielectric constant of the air gap and the dielectric phase, the breakdown electric field strength, the energy barrier for the development of electrical trees, and also set the simulation step size and the mesh size according to the target dielectric size.

[0033] The relative dielectric constant is calculated by the following formula:

[0034] ε(S)=ε a ·s 3 (10-15s+6s 2 )+ε b ·(1-s 3 (10-15s+6s 2 ))

[0035] In the formula, S is the order parameter, ε(S) is the dielectric constant at the current position, and ε a is the dielectric constant of the material at complete breakdown, ε b is the dielectric constant of the initial material.

[0036] 2) Calculate field variables based on energy distribution and physical laws:

[0037] F=∫[f joule (s)+f mechanical (s)+f elec (s)+f gradient (s)+f sep (s)]dV

[0038] In the formula, f joule is the Joule heat energy density, f mechanical is the mechanical stress isoenergy density, f elec is the electrostatic energy density, f gradient is the gradient energy density, f sep is the phase separation energy density, and V is the unit volume of the solid medium.

[0039] 3) Apply the phase field dynamics equation to update the order parameters and obtain the state of electrical tree development at each moment; the phase field dynamics equation is expressed as:

[0040]

[0041] In the formula, f crictical is the breakdown threshold energy density, L0 is the updateable coefficient of interface mobility, H is the step function, and its independent variable is F.

[0042] The above-mentioned energy density and related material parameters can be characterized by phase field order parameters, which reflects the unity brought by the phase field method and can also be better adapted to machine learning.

[0043] 4) Iterate the simulation until the electrical tree causes the breakdown of the polymer solid medium, forming a complete time series image data set, such as Figure 2 shown.

[0044] 2. After generating the time series images of the electrical tree development paths under different load levels, a data set is constructed, and the data set is grouped and stored using the load level as a label. The data set constructed in this way can provide well-labeled training data for the machine learning model.

[0045] 3. Further process the grouped image datasets, such as Figure 3 As shown, it mainly includes:

[0046] (1) Uniformly sample frames from the time series image to ensure data consistency;

[0047] (2) Convert the grayscale image into a three-channel RGB image and perform normalization so that the image value range is normalized to between 0 and 1;

[0048] (3) The image sequences are grouped and sampled according to the pre-order and post-order order to construct the training set and validation set of the machine learning model.

[0049] 4. Based on the processed data set, a machine learning model with time series modeling capabilities is designed and trained to predict the extension of the electrical tree path and learn the changing characteristics of the electrical tree development path under different load levels, thereby improving the prediction accuracy.

[0050] In this embodiment, since the development path of the electrical tree has obvious time series characteristics, its path changes gradually accumulate with time and load conditions, presenting a nonlinear process. Therefore, the machine learning model selects a time series prediction model, which realizes the prediction output of the target time point through the mapping function shown in the following formula:

[0051] F Θ :X t,T →Y t+1,T′

[0052] That is, by inputting a sequence of initial images of several frames, the development path of electrical dendrites after the target time point is predicted.

[0053] like Figure 4 As shown, the time series prediction model includes a spatial encoder, a time series module and a spatial decoder. The spatial encoder is represented as:

[0054] z i =σ(Norm2d(Conv2d(z i -1 ))),1≤i≤N s

[0055] The spatial decoder is represented as:

[0056] z k =σ(Norm2d(unConv2d(z k-1 ))),N s +N t ≤k≤2N s

[0057] The timing module is a timing prediction module, such as an RNN network.

[0058] The training process of the time series prediction model includes:

[0059] 1) Input the frame-sampled image sequence into the spatial encoder, use operations such as convolution and normalization to extract spatial features, and encode the image information into a latent space representation to facilitate the subsequent learning of temporal features;

[0060] 2) The extracted features are input into the timing module to enhance the perception of timing information; at the same time, the timing module acts as a translator to convert the encoded data into timing information that can be used by the decoder, thereby capturing the time-dependent characteristics of the electrical tree extension process;

[0061] 3) The decoder decodes the data output by the timing module and integrates the feature information of each time step, thereby generating the ability to learn and predict the global characteristics of electrical tree development.

[0062] 5. After the model training is completed, the model parameters with the smallest loss function are saved to the local end. In practical applications, the input data containing load level information and initial crack images are passed into the trained model, and the model outputs a sequence of electrical tree development images after several time steps in the future to predict the development path of electrical trees under the load conditions, effectively assisting the state prediction and maintenance of electrical equipment.

[0063] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.

Claims

1. A method for predicting the development path of electrical tree in solid dielectrics, characterized in that: Under different load levels, the phase field method is used to numerically simulate the development path of electrical tree in solid medium, and a time series image containing the development path of electrical tree is generated. Build a data set based on the generated image data, and group and store the data set using load levels as labels; Perform uniform frame sampling on the classified data set, convert the grayscale image into a three-channel RGB image and normalize it, then group and sample the image sequence according to the preceding and following order to construct the training set and data set; Constructing a machine learning model for predicting the development path of electrical tree branches, and using the training set to train the model so that the model learns the changing characteristics of the development path of electrical tree branches under different load levels; The image and load information of the solid medium in the actual scene are obtained and input into the trained machine learning model to predict the electrical tree development path of the solid medium under the load and generate an image sequence of the electrical tree development after several time steps in the future.

2. The method according to claim 1, characterized in that The phase field method is used to perform numerical simulation on the development path of electrical tree in solid medium, including: determining simulation parameters and phase field parameters of solid medium; calculating field variables based on energy distribution and physical laws; applying phase field dynamic equations to update sequence parameters and obtain the development state of electrical tree at each moment; iterating simulation until electrical tree causes breakdown of solid medium, forming a complete time series image.

3. The method according to claim 2, characterized in that The simulation parameters include the relative dielectric constant of the air gap and the dielectric phase, the breakdown electric field strength and the electric tree development energy barrier, and the phase field parameters of the solid medium include the simulation step size and the subdivision grid size set according to the target solid medium size.

4. The method according to claim 1, characterized in that The different load levels are different load levels formed by combined loads applied to the material under different loading voltages, applied mechanical stresses and applied temperatures.

5. The method according to claim 1, characterized in that: The machine learning model includes a spatial encoder, a temporal module and a spatial decoder; the spatial encoder is used to extract spatial features of an input image sequence and encode them into a latent space representation, which is then input into the temporal module; The timing module converts the coded data into timing information used by the spatial decoder, captures the time-dependent characteristics of the electrical tree expansion process, and enhances the perception of the timing information; The spatial decoder decodes the data output by the timing module, integrates the characteristic information of each time step, and generates the learning and prediction capabilities of the global characteristics of electrical tree development.

Images