Dry-type reactor transient process prediction and evaluation method, system, equipment and medium
By establishing multi-granularity equivalent circuit models and AI prediction models, the problems of large computational load and long time consumption in traditional methods are solved, enabling rapid and accurate prediction of transient processes in high-voltage dry-type reactors, optimizing design schemes, and improving the reactor's shock resistance.
Patent Information
- Application Number
- CN202510926102.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-07
- Publication Date
- 2025-10-21
AI Technical Summary
Traditional methods for transient overvoltage analysis of high-voltage dry reactors involve large computational loads and long processing times, making it difficult to achieve online real-time evaluation and rapid response. Furthermore, they do not effectively combine artificial intelligence technology for rapid inference.
A multi-granularity equivalent circuit model of a dry reactor is established, and simulation is performed in conjunction with transient boundary conditions. A transient voltage and electric field AI prediction model is constructed, and a multi-layer feedforward neural network and a Transformer-CNN-GAN fusion model are used for training to achieve rapid prediction.
It significantly reduces computation time, improves the accuracy and speed of voltage and electric field distribution prediction, and can output the voltage and electric field change trends of the reactor transient process in real time, helping to optimize the design scheme and improve the reactor's shock resistance.
Smart Images
Figure CN120822369A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method, system, equipment and medium for predicting and evaluating the transient process of a dry-type reactor. Background Art
[0002] When a high-voltage dry-type reactor is subjected to transient overvoltages such as lightning strikes and grid switching, the calculation of the voltage distribution and electric field strength within the coil is crucial for insulation testing and operation monitoring of the reactor.
[0003] Traditional electromagnetic transient simulation methods (such as finite element electromagnetic field simulation, EMTP / PSCAD, etc.) are computationally intensive and require model refinement and extremely small step sizes to capture high-speed transient processes. As a result, a single simulation takes hours or even days, making it difficult to use for rapid response in online real-time evaluation or design iteration.
[0004] Currently, artificial intelligence (AI) technology has not been effectively integrated into reactor transient analysis. Traditional methods rely primarily on physical modeling and numerical solutions, requiring individual modeling and simulation for transient performance under different design parameters, lacking rapid inference capabilities. While AI applications in electrical engineering are rapidly developing, the prediction of high-voltage reactor transients remains a niche topic. Summary of the Invention
[0005] To solve the above problems, the first aspect of the present invention provides a method for predicting and evaluating a transient process of a dry-type reactor, comprising:
[0006] Establishing a multi-granularity equivalent circuit model of a dry-type reactor, applying transient boundary conditions to the multi-granularity equivalent circuit model for simulation, obtaining transient voltage distribution data of the multi-granularity equivalent circuit model, and generating a transient voltage distribution data set;
[0007] Construct a transient voltage AI prediction model. The structural parameters of the dry-type reactor coil, transient boundary conditions, and transient voltage distribution dataset are combined into training sample set 1. The transient voltage AI prediction model is trained to obtain the trained transient voltage AI prediction model and the transient voltage distribution prediction data output by the model.
[0008] Construct a transient electric field AI prediction model. Calculate the spatial electric field intensity distribution based on the transient voltage distribution prediction data output by the transient voltage AI prediction model. Combine the dry-type reactor coil's structural parameters, transient voltage distribution prediction data, and spatial electric field intensity distribution into a second training sample set. Train the transient electric field AI prediction model to obtain a trained transient electric field AI prediction model.
[0009] The trained transient voltage AI prediction model and transient electric field AI prediction model are deployed in the server system. The structural parameters and transient boundary conditions of the dry-type reactor coil are input to obtain the voltage distribution and electric field distribution prediction results of the dry-type reactor during the transient process.
[0010] The specific method of establishing the multi-granularity equivalent circuit model of the dry-type reactor is as follows:
[0011] The coil of the dry-type reactor is modeled in segments or turns, and the mutual inductance and coupling capacitance between each segment of the coil are modeled; each segment of the coil is a π-type circuit containing equivalent resistance, equivalent inductance, and equivalent capacitance;
[0012] According to the geometric dimensions of the dry-type reactor coil, the equivalent inductance and equivalent resistance are calculated using analytical formulas;
[0013] All π-type circuits are sequentially spliced together to form a series structure, and the inter-segment mutual inductance and coupling capacitance are embedded to obtain the multi-granularity equivalent circuit of the dry-type reactor.
[0014] In the server system, the input of the transient voltage AI prediction model is the structural parameters and transient boundary conditions of the dry-type reactor coil; the output is the transient voltage distribution prediction data of each section of the dry-type reactor coil.
[0015] The training of the transient voltage AI prediction model is specifically to obtain coarse-grained data and fine-grained data based on the first division of the training sample set, and to perform transfer learning training on the transient voltage AI prediction model from coarse-grained data to fine-grained data through a machine learning algorithm.
[0016] The specific process of the transfer learning training includes:
[0017] Using coarse-grained data to train a transient voltage AI prediction model, the transient voltage AI prediction model uses a multi-layer feedforward neural network, the multi-layer feedforward neural network including an input layer, a plurality of hidden layers, and an output layer; wherein the multi-layer feedforward neural network is trained using a preset first learning rate until convergence;
[0018] Fine-grained data is used to train the last two to three layers of the hidden layer close to the output layer of the multi-layer feedforward neural network, wherein the hidden layer of the multi-layer feedforward neural network is trained with a preset second learning rate until convergence.
[0019] The transient electric field AI prediction model adopts a Transformer-CNN-GAN fusion model. In the server system, the Transformer-CNN-GAN fusion model inputs the structural parameters of the dry-type reactor coil and the transient voltage distribution data; the output is the prediction result of the transient electric field intensity distribution of the dry-type reactor.
[0020] The prediction processing process of the Transformer-CNN-GAN fusion model includes extracting the sequence features of the transient voltage distribution along the coil axis through the Transformer module; extracting the electric field influencing factors in the local space through the CNN module; and fusing the extracted sequence features with the electric field influencing factors to generate a prediction result of the transient electric field intensity distribution.
[0021] A second aspect of the present invention further provides a dry-type reactor transient process prediction and evaluation system, which is used to implement the above-mentioned dry-type reactor transient process prediction and evaluation method, comprising:
[0022] An equivalent circuit module is used to establish a multi-granularity equivalent circuit model of a dry-type reactor, apply transient boundary conditions to the multi-granularity equivalent circuit model for simulation, obtain transient voltage distribution data of the multi-granularity equivalent circuit model, and generate a transient voltage distribution data set;
[0023] The model building module is used to build a transient voltage AI prediction model. The structural parameters of the dry-type reactor coil, transient boundary conditions, and transient voltage distribution data set are combined into a training sample set 1. The transient voltage AI prediction model is trained to obtain the trained transient voltage AI prediction model and the transient voltage distribution prediction data output by the model.
[0024] The model construction module is also used to construct a transient electric field AI prediction model, calculate the spatial electric field intensity distribution based on the transient voltage distribution prediction data output by the transient voltage AI prediction model, and form a training sample set 2 with the structural parameters of the dry-type reactor coil, the transient voltage distribution prediction data, and the spatial electric field intensity distribution. The transient electric field AI prediction model is trained to obtain a trained transient electric field AI prediction model;
[0025] The prediction module is used to deploy the trained transient voltage AI prediction model and transient electric field AI prediction model in the server system, input the structural parameters and transient boundary conditions of the dry-type reactor coil, and obtain the voltage distribution and electric field distribution prediction results of the dry-type reactor during the transient process.
[0026] The equivalent circuit module establishes a multi-granularity equivalent circuit model of the dry-type reactor, which is specifically:
[0027] The coil of the dry-type reactor is modeled in segments or turns, and the mutual inductance and coupling capacitance between each segment of the coil are modeled; each segment of the coil is a π-type circuit containing equivalent resistance, equivalent inductance, and equivalent capacitance;
[0028] According to the geometric dimensions of the dry-type reactor coil, the equivalent inductance and equivalent resistance are calculated using analytical formulas;
[0029] All π-type circuits are sequentially spliced together to form a series structure, and the inter-segment mutual inductance and coupling capacitance are embedded to obtain the multi-granularity equivalent circuit of the dry-type reactor.
[0030] The transient voltage AI prediction model constructed by the model construction module has as input the structural parameters and transient boundary conditions of the dry-type reactor coil; and outputs the transient voltage distribution prediction data of each section of the dry-type reactor coil.
[0031] The model construction module trains the transient voltage AI prediction model, specifically by dividing the training sample set into coarse-grained data and fine-grained data, and performing transfer learning training on the transient voltage AI prediction model from coarse-grained data to fine-grained data through a machine learning algorithm.
[0032] The specific process of the transfer learning training includes:
[0033] Using coarse-grained data to train a transient voltage AI prediction model, the transient voltage AI prediction model uses a multi-layer feedforward neural network, the multi-layer feedforward neural network including an input layer, a plurality of hidden layers, and an output layer; wherein the multi-layer feedforward neural network is trained using a preset first learning rate until convergence;
[0034] Fine-grained data is used to train the last two to three layers of the hidden layer close to the output layer of the multi-layer feedforward neural network, wherein the hidden layer of the multi-layer feedforward neural network is trained with a preset second learning rate until convergence.
[0035] The transient electric field AI prediction model constructed by the model construction module adopts a Transformer-CNN-GAN fusion model. In the server system, the Transformer-CNN-GAN fusion model inputs the structural parameters and transient voltage distribution data of the dry-type reactor coil; and outputs the prediction result of the transient electric field intensity distribution of the dry-type reactor.
[0036] The prediction processing process of the Transformer-CNN-GAN fusion model includes extracting the sequence features of the transient voltage distribution along the coil axis through the Transformer module; extracting the electric field influencing factors in the local space through the CNN module; and fusing the extracted sequence features with the electric field influencing factors to generate a prediction result of the transient electric field intensity distribution.
[0037] A third aspect provides a dry-type reactor transient process prediction and evaluation device, comprising a processor and a memory, wherein the processor implements the above-mentioned dry-type reactor transient process prediction and evaluation method when executing the computer program stored in the memory.
[0038] A fourth aspect provides a computer-readable storage medium for storing a computer program, wherein when the computer program is executed by a processor, the above-mentioned method for predicting and evaluating a transient process of a dry-type reactor is implemented.
[0039] Beneficial effects: The present invention provides a method, system, equipment and medium for predicting and evaluating the transient process of a dry-type reactor. The transient response of the dry-type reactor is obtained through multi-granularity equivalent circuit modeling, and the transient voltage distribution and transient electric field intensity distribution are quickly predicted using artificial intelligence model distribution, significantly reducing the calculation time. By integrating advanced deep learning algorithms, a model with high fidelity for electromagnetic response is trained, and measured data is incorporated through transfer learning to further improve the accuracy, which can reliably reproduce the changing trends of voltage and electric field in the transient process of the reactor.
[0040] By introducing an AI prediction model, tedious numerical iteration of differential equations or field finite element solutions are avoided for each analysis. After model training is complete, results can be output in real time for new reactor parameters or transient operating conditions.
[0041] This invention rapidly evaluates transient voltage distribution under different winding schemes or insulation structures, promptly identifying potential risks such as uneven overvoltage distribution and excessive local field strength, thereby helping engineers optimize designs and improve the reactor's impact resistance. Traditional methods for insulation coordination verification often require repeated model building and consume significant simulation time. However, this invention allows for rapid comparison of transient responses under various design parameter combinations, shortening the development cycle.
[0042] In summary, the technical solution of the present invention realizes the leap from "offline and time-consuming" to "online and fast" in transient process analysis of high-voltage dry-type reactors through computational acceleration and intelligent inference. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] By reading the detailed description of the preferred embodiment below, the scheme and advantages of the present invention will become clear to those skilled in the art. The accompanying drawings are only for illustrating the preferred embodiment and are not to be considered as limiting the present invention.
[0044] In the attached figure:
[0045] Figure 1 The flowchart of the transient process prediction and evaluation method for dry-type reactors is shown in Figure 2. DETAILED DESCRIPTION
[0046] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings.
[0047] Example
[0048] The present invention provides a method for predicting and evaluating a transient process of a dry-type reactor.
[0049] See also Figure 1 A method for predicting and evaluating the transient process of a dry-type reactor is provided. The method is specifically implemented in the following steps:
[0050] Establishing a multi-granularity equivalent circuit model of a dry-type reactor, applying transient boundary conditions to the multi-granularity equivalent circuit model for simulation, obtaining transient voltage distribution data of the multi-granularity equivalent circuit model, and generating a transient voltage distribution data set;
[0051] Construct a transient voltage AI prediction model. The structural parameters of the dry-type reactor coil, transient boundary conditions, and transient voltage distribution dataset are combined into training sample set 1. The transient voltage AI prediction model is trained to obtain the trained transient voltage AI prediction model and the transient voltage distribution prediction data output by the model.
[0052] Construct a transient electric field AI prediction model. Calculate the spatial electric field intensity distribution based on the transient voltage distribution prediction data output by the transient voltage AI prediction model. Combine the dry-type reactor coil's structural parameters, transient voltage distribution prediction data, and spatial electric field intensity distribution into a second training sample set. Train the transient electric field AI prediction model to obtain a trained transient electric field AI prediction model.
[0053] The trained transient voltage AI prediction model and transient electric field AI prediction model are deployed in the server system. The structural parameters and transient boundary conditions of the dry-type reactor coil are input to obtain the voltage distribution and electric field distribution prediction results of the dry-type reactor during the transient process.
[0054] The dry-type reactor transient process prediction and evaluation method provided by the embodiment of the present invention obtains the transient response of the dry-type reactor through multi-granularity equivalent circuit modeling, and uses artificial intelligence model distribution to quickly predict the transient voltage distribution and transient electric field strength distribution, significantly reducing the calculation time; by integrating advanced deep learning algorithms, a model with high fidelity for electromagnetic response is trained, and the measured data is incorporated through transfer learning to further improve the accuracy, which can reliably reproduce the changing trends of voltage and electric field in the transient process of the reactor.
[0055] By introducing an AI prediction model, tedious numerical iteration of differential equations or field finite element solutions are avoided for each analysis. After model training is complete, results can be output in real time for new reactor parameters or transient operating conditions.
[0056] By quickly evaluating the transient voltage distribution under different winding schemes or insulation structures, hidden dangers such as uneven overvoltage distribution and excessive local field strength can be discovered in a timely manner, thereby helping engineers optimize design solutions and improve the impact resistance of reactors. In insulation coordination verification, traditional methods often require repeated model building and consume a lot of simulation time. However, the use of this invention can compare transient responses under various design parameter combinations in a short period of time, shortening the R&D cycle. Through computational acceleration and intelligent inference, the technology of the present invention has achieved a leap from "offline, time-consuming" to "online, fast" in the transient process analysis of high-voltage dry-type reactors.
[0057] The following describes each of the above steps.
[0058] A multi-granularity equivalent circuit model of a dry-type reactor is established, and transient boundary conditions are applied to the multi-granularity equivalent circuit model for simulation to obtain transient voltage distribution data of the multi-granularity equivalent circuit model and generate a transient voltage distribution data set.
[0059] The specific method of establishing the multi-granularity equivalent circuit model of the dry-type reactor is as follows:
[0060] The coil of the dry-type reactor is modeled in segments or turns, and the mutual inductance and coupling capacitance between each segment of the coil are modeled; each segment of the coil is a π-type circuit containing equivalent resistance, equivalent inductance, and equivalent capacitance;
[0061] The coil of the dry-type reactor is divided into sections or turns according to the spatial distribution, and each section of the coil or each turn of the coil is regarded as an independent substructure;
[0062] A π-type circuit based on equivalent resistance, equivalent inductance and equivalent capacitance is established for each coil segment. The π-type circuit is composed of the ground capacitance C on both sides. g , the middle series resistor R and inductor L.
[0063] The mutual inductance model between segments is constructed using the mutual inductance matrix M ij Under high-frequency conditions, there will be electromagnetic induction coupling between the coil segments, that is, the current change in one coil segment will induce an electromotive force in other coil segments. This effect is represented by the mutual inductance matrix M. ij The mutual inductance matrix M ij It satisfies the symmetry and positivity conditions, and its construction specifically includes:
[0064] Between adjacent coil segments, the mutual inductance matrix is estimated based on the wire spacing and relative position, which can be estimated using the Neumann formula or the elliptic approximation method. Between distant coil segments, the magnetic flux is calculated based on the finite element method, or a mutual inductance attenuation function is introduced for modeling.
[0065] The coupling capacitance model is modeled using the coupling capacitance matrix C ijThere are parasitic capacitances between coil segments (especially in multi-layer winding structures), which mainly come from the boundary electric field between adjacent conductor surfaces and the polarization of the intermediate insulating material. These capacitances are represented as transverse coupling capacitances.
[0066] The coupling capacitance is obtained by numerical solution: the finite element method is used to perform electrostatic field simulation on the electric field distribution of the coil, the capacitance values between adjacent conductors and between conductors and ground are calculated, and the capacitance parameters required for the multi-granularity equivalent circuit model are accurately obtained.
[0067] According to the geometric dimensions of the dry-type reactor coil, the equivalent inductance and equivalent resistance are calculated using analytical formulas;
[0068] Based on the geometric dimensions of the coil, which include the average diameter, number of turns, and wire cross-section, the equivalent inductance and equivalent resistance of each coil segment are calculated using analytical formulas.
[0069] The analytical formula for the equivalent inductance is:
[0070] L=4πμ0R 3 n 2 (f(R,R,H)-f(R,R,0))
[0071] The formula for calculating coaxial mutual inductance is:
[0072]
[0073] In the above formula, R represents the radius of the coil; n represents the turn density of the self-inductance coil; H represents the height of the coil; S represents the axial distance between the centers of the two coils; μ0 represents the magnetic permeability; Z represents the axial distance between the two basic calculation units; θ represents the integration angle, which is 0 to 360° for circular coils; L is the equivalent inductance value; and M12 is the mutual inductance value between coil one and coil two.
[0074] All π-type circuits are sequentially spliced together to form a series structure, and the inter-segment mutual inductance and coupling capacitance are embedded to obtain the multi-granularity equivalent circuit of the dry-type reactor.
[0075] A series structure, formed by connecting several π-shaped circuits end to end, is used to simulate the transmission delay and distributed parameter effects of the entire coil during long-distance transmission and high-frequency transients. This structure is suitable for representing transient voltage waves propagating along the coil's axis and can capture the attenuation and reflection characteristics of the signal at different physical locations. Simultaneously, the mutual inductance matrix of the inter-segment mutual inductance and the capacitance matrix of the coupling capacitors are embedded to represent the near-far coupling of the magnetic field and the electrostatic coupling between the ground and the segments, respectively. This results in a multi-granular equivalent circuit for the dry-type reactor. This multi-granular equivalent circuit not only preserves the spatial non-uniformity within the coil but also reflects the complex electromagnetic wave propagation characteristics under high-frequency lightning strikes.
[0076] The multi-granularity equivalent circuit model is simulated by applying transient boundary conditions, wherein the transient boundary conditions include a lightning impulse waveform and a switching operation voltage step. By applying transient boundary conditions to the modeled multi-granularity equivalent circuit, the voltage and current responses of each coil segment of the dry-type reactor corresponding to the multi-granularity equivalent circuit over time are obtained based on the data solution.
[0077] The lightning impulse waveform has a 1.2 / 50μs characteristic waveform, defined as the leading edge time T1: the time it takes for the voltage to rise from 10% to 90% is approximately 1.2μs; the half-peak time T2: the time it takes for the voltage to fall from the peak to 50% is approximately 50μs, and the overshoot and oscillation at the peak value Um does not exceed 5%. The lightning impulse waveform can be represented by a double exponential function, the formula of which is:
[0078] U(t)=-U o (e -αt -e -βt )
[0079] Where U o represents voltage; α and β are waveform steepness and attenuation characteristic parameters respectively, β>α.
[0080] In order to make the double exponential function meet the standard waveform requirements, the fitting conditions are set as follows: U(0.1T1)=0.1Up; U(T1)=0.9U p ; U(T2)=0.5U p , numerical fitting determines α, β, U p Indicates peak voltage.
[0081] A transient voltage AI prediction model is constructed, and the structural parameters, transient boundary conditions and transient voltage distribution data sets of the dry-type reactor coil are combined into a training sample set 1. The transient voltage AI prediction model is trained to obtain the trained transient voltage AI prediction model and the transient voltage distribution prediction data output by the model.
[0082] In the server system, the input of the transient voltage AI prediction model is the structural parameters and transient boundary conditions of the dry-type reactor coil; the output is the transient voltage distribution prediction data of each section of the dry-type reactor coil.
[0083] The structural parameters include at least one of coil turns, coil size, number of coil layers and distance between turns; the transient boundary conditions are specifically transient overvoltage waveform parameters, including amplitude, rising edge steepness and duration.
[0084] The training of the transient voltage AI prediction model is specifically to obtain coarse-grained data and fine-grained data based on the first division of the training sample set, and to perform transfer learning training on the transient voltage AI prediction model from coarse-grained data to fine-grained data through a machine learning algorithm, so that the transient voltage AI prediction model can learn the nonlinear mapping relationship from structural parameters to transient response.
[0085] The coarse-grained data are simulation data of a layer multi-grained equivalent circuit model of a dry-type reactor; the fine-grained data are simulation data of a turn multi-grained equivalent circuit model of a dry-type reactor.
[0086] The specific process of the transfer learning training includes:
[0087] Coarse-grained data is used to train a transient voltage AI prediction model. The transient voltage AI prediction model uses a multi-layer feedforward neural network (MLP), which is composed of a deep belief network (DNN) and a deep neural network (DNN). The multi-layer feedforward neural network includes an input layer, several hidden layers, and an output layer. The multi-layer feedforward neural network is trained using a preset first learning rate until convergence.
[0088] Using fine-grained data to train the last two to three hidden layers of a multi-layer feedforward neural network, freezing the first n hidden layers of the multi-layer feedforward neural network, and only training the last two to three hidden layers close to the output layer, wherein the hidden layers of the multi-layer feedforward neural network are trained using a preset second learning rate until convergence;
[0089] During the transfer learning training process, the first learning rate is greater than the second learning rate, and the learning rate is reduced to prevent overfitting.
[0090] A transient electric field AI prediction model is constructed, and the spatial electric field intensity distribution is calculated based on the transient voltage distribution prediction data output by the transient voltage AI prediction model. The structural parameters of the dry-type reactor coil, the transient voltage distribution prediction data, and the spatial electric field intensity distribution are combined into a second training sample set. The transient electric field AI prediction model is trained to obtain a trained transient electric field AI prediction model.
[0091] The transient electric field AI prediction model adopts a Transformer-CNN-GAN fusion model. In the server system, the input of the Transformer-CNN-GAN fusion model is the structural parameters of the dry-type reactor coil and the transient voltage distribution data; the output is the prediction result of the transient electric field intensity distribution of the dry-type reactor.
[0092] The transient voltage distribution prediction data output by the transient voltage AI prediction model is used to calculate the spatial electric field intensity distribution through numerical simulation methods. The numerical simulation method is a commonly used calculation method for solving physical models, so it will not be described in detail.
[0093] The prediction processing of the Transformer-CNN-GAN fusion model includes extracting the sequence features of the voltage distribution along the coil axis through the Transformer module; extracting the electric field influencing factors in the local space through the CNN module; and fusing the extracted sequence features with the electric field influencing factors to generate a preliminary prediction result of the electric field distribution.
[0094] The generative adversarial network (GAN) generates preliminary predictions as generator outputs and constructs a discriminator to distinguish them from the "real" electric field. Adversarial training optimizes the parameters of the GAN. After training, the Transformer + CNN fusion model can be used independently for inference, outputting predictions of the transient electric field intensity distribution of dry-type reactors when given new inputs.
[0095] The Transformer module extracts the sequence features of the voltage distribution along the coil axis, and its processing operations are as follows:
[0096] The coil is divided into discrete points along its axis to obtain a number of sampling points. The preliminary features of each sampling point include the estimated voltage and electric field value.
[0097] Map the preliminary features of each sampling point to a high-dimensional vector; add position encoding information to each sampling point;
[0098] The mutual influence between the positions of each sampling point is calculated through the self-attention mechanism, and long-range dependencies are captured through multi-layer stacking;
[0099] The axial features are output to obtain the feature vector of each sampling point, which is the sequence feature of the voltage distribution along the axial direction of the coil.
[0100] The CNN module extracts the electric field influence factor in the local space by: encoding the coil cross-sectional geometry into an image / matrix; extracting local features through small-window convolution; extracting small-scale local features from the input image through multi-layer convolution operations, and integrating the small-scale local features to form large-scale features, such as the overall stacking effect of the coil; reducing the resolution through the pooling layer to retain important regional features; and outputting a local feature map. The electric field influence factor in the local space is the interaction between different features, that is, a vector is output for each spatial position, representing the local electric field characteristics of that position.
[0101] The trained transient voltage AI prediction model and transient electric field AI prediction model are deployed in the server system. The structural parameters and transient boundary conditions of the dry-type reactor coil are input to obtain the voltage distribution and electric field distribution prediction results of the dry-type reactor during the transient process.
[0102] Then, this embodiment also provides a dry-type reactor transient process prediction and evaluation system, including:
[0103] An equivalent circuit module is used to establish a multi-granularity equivalent circuit model of a dry-type reactor, apply transient boundary conditions to the multi-granularity equivalent circuit model for simulation, obtain transient voltage distribution data of the multi-granularity equivalent circuit model, and generate a transient voltage distribution data set;
[0104] The model building module is used to build a transient voltage AI prediction model. The structural parameters of the dry-type reactor coil, transient boundary conditions, and transient voltage distribution data set are combined into a training sample set 1. The transient voltage AI prediction model is trained to obtain the trained transient voltage AI prediction model and the transient voltage distribution prediction data output by the model.
[0105] The model construction module is also used to construct a transient electric field AI prediction model, calculate the spatial electric field intensity distribution based on the transient voltage distribution prediction data output by the transient voltage AI prediction model, and form a training sample set 2 with the structural parameters of the dry-type reactor coil, the transient voltage distribution prediction data, and the spatial electric field intensity distribution. The transient electric field AI prediction model is trained to obtain a trained transient electric field AI prediction model;
[0106] The prediction module is used to deploy the trained transient voltage AI prediction model and transient electric field AI prediction model in the server system, input the structural parameters and transient boundary conditions of the dry-type reactor coil, and obtain the voltage distribution and electric field distribution prediction results of the dry-type reactor during the transient process.
[0107] The equivalent circuit module establishes a multi-granularity equivalent circuit model of the dry-type reactor, which is specifically:
[0108] The coil of the dry-type reactor is modeled in segments or turns, and the mutual inductance and coupling capacitance between each segment of the coil are modeled; each segment of the coil is a π-type circuit containing equivalent resistance, equivalent inductance, and equivalent capacitance;
[0109] According to the geometric dimensions of the dry-type reactor coil, the equivalent inductance and equivalent resistance are calculated using analytical formulas;
[0110] All π-type circuits are sequentially spliced together to form a series structure, and the inter-segment mutual inductance and coupling capacitance are embedded to obtain the multi-granularity equivalent circuit of the dry-type reactor.
[0111] The transient voltage AI prediction model constructed by the model construction module has as input the structural parameters and transient boundary conditions of the dry-type reactor coil; and outputs the transient voltage distribution prediction data of each section of the dry-type reactor coil.
[0112] The model construction module trains the transient voltage AI prediction model, specifically by dividing the training sample set into coarse-grained data and fine-grained data, and performing transfer learning training on the transient voltage AI prediction model from coarse-grained data to fine-grained data through a machine learning algorithm.
[0113] The specific process of the transfer learning training includes:
[0114] Using coarse-grained data to train a transient voltage AI prediction model, the transient voltage AI prediction model uses a multi-layer feedforward neural network, the multi-layer feedforward neural network including an input layer, a plurality of hidden layers, and an output layer; wherein the multi-layer feedforward neural network is trained using a preset first learning rate until convergence;
[0115] Fine-grained data is used to train the last two to three layers of the hidden layer close to the output layer of the multi-layer feedforward neural network, wherein the hidden layer of the multi-layer feedforward neural network is trained with a preset second learning rate until convergence.
[0116] The transient electric field AI prediction model constructed by the model construction module adopts a Transformer-CNN-GAN fusion model. In the server system, the Transformer-CNN-GAN fusion model inputs the structural parameters and transient voltage distribution data of the dry-type reactor coil; and outputs the prediction result of the transient electric field intensity distribution of the dry-type reactor.
[0117] The prediction processing process of the Transformer-CNN-GAN fusion model includes extracting the sequence features of the transient voltage distribution along the coil axis through the Transformer module; extracting the electric field influencing factors in the local space through the CNN module; and fusing the extracted sequence features with the electric field influencing factors to generate a prediction result of the transient electric field intensity distribution.
[0118] In addition, a dry-type reactor transient process prediction and evaluation device is also provided, comprising a processor and a memory, wherein the processor implements the above-mentioned dry-type reactor transient process prediction and evaluation method when executing the computer program stored in the memory.
[0119] Finally, a computer-readable storage medium is also provided for storing a computer program, wherein when the computer program is executed by a processor, the above-mentioned method for predicting and evaluating a transient process of a dry-type reactor is implemented.
[0120] It will be understood by those skilled in the art that the embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present invention may be implemented in various computer languages, for example, the object-oriented programming language Java and the interpreted scripting language JavaScript.
[0121] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0122] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.
[0123] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.
[0124] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.
[0125] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.
Claims
1. A method for predicting and evaluating the transient process of a dry-type reactor, characterized in that: include: Establishing a multi-granularity equivalent circuit model of a dry-type reactor, applying transient boundary conditions to the multi-granularity equivalent circuit model for simulation, obtaining transient voltage distribution data of the multi-granularity equivalent circuit model, and generating a transient voltage distribution data set; Construct a transient voltage AI prediction model. The structural parameters of the dry-type reactor coil, transient boundary conditions, and transient voltage distribution dataset are combined into training sample set 1. The transient voltage AI prediction model is trained to obtain the trained transient voltage AI prediction model and the transient voltage distribution prediction data output by the model. Construct a transient electric field AI prediction model. Calculate the spatial electric field intensity distribution based on the transient voltage distribution prediction data output by the transient voltage AI prediction model. Combine the dry-type reactor coil's structural parameters, transient voltage distribution prediction data, and spatial electric field intensity distribution into a second training sample set. Train the transient electric field AI prediction model to obtain a trained transient electric field AI prediction model. The trained transient voltage AI prediction model and transient electric field AI prediction model are deployed in the server system. The structural parameters and transient boundary conditions of the dry-type reactor coil are input to obtain the voltage distribution and electric field distribution prediction results of the dry-type reactor during the transient process.
2. The method for predicting and evaluating the transient process of a dry-type reactor according to claim 1, characterized in that: The specific method of establishing the multi-granularity equivalent circuit model of the dry-type reactor is as follows: The coil of the dry-type reactor is modeled in segments or turns, and the mutual inductance and coupling capacitance between each segment of the coil are modeled; each segment of the coil is a π-type circuit containing equivalent resistance, equivalent inductance, and equivalent capacitance; According to the geometric dimensions of the dry-type reactor coil, the equivalent inductance and equivalent resistance are calculated using analytical formulas; All π-type circuits are sequentially spliced together to form a series structure, and the inter-segment mutual inductance and coupling capacitance are embedded to obtain the multi-granularity equivalent circuit of the dry-type reactor.
3. The method for predicting and evaluating the transient process of a dry-type reactor according to claim 1, characterized in that: In the server system, the input of the transient voltage AI prediction model is the structural parameters and transient boundary conditions of the dry-type reactor coil; the output is the transient voltage distribution prediction data of each section of the dry-type reactor coil.
4. The method for predicting and evaluating the transient process of a dry-type reactor according to claim 1, characterized in that: The training of the transient voltage AI prediction model is specifically to obtain coarse-grained data and fine-grained data based on the first division of the training sample set, and to perform transfer learning training on the transient voltage AI prediction model from coarse-grained data to fine-grained data through a machine learning algorithm.
5. The method for predicting and evaluating the transient process of a dry-type reactor according to claim 4, characterized in that: The specific process of the transfer learning training includes: Using coarse-grained data to train a transient voltage AI prediction model, the transient voltage AI prediction model uses a multi-layer feedforward neural network, the multi-layer feedforward neural network including an input layer, a plurality of hidden layers, and an output layer; wherein the multi-layer feedforward neural network is trained using a preset first learning rate until convergence; The fine-grained data is used to train the last two to three layers of the hidden layer close to the output layer of the multi-layer feedforward neural network trained with the coarse-grained data, wherein the hidden layer of the multi-layer feedforward neural network is trained with a preset second learning rate until convergence.
6. The method for predicting and evaluating transient processes of dry-type reactors according to claim 1, characterized in that: The transient electric field AI prediction model adopts a Transformer-CNN-GAN fusion model. In the server system, the Transformer-CNN-GAN fusion model inputs the structural parameters of the dry-type reactor coil and the transient voltage distribution data; the output is the prediction result of the transient electric field intensity distribution of the dry-type reactor.
7. The method for predicting and evaluating the transient process of a dry-type reactor according to claim 6, characterized in that: The prediction processing process of the Transformer-CNN-GAN fusion model includes extracting the sequence features of the transient voltage distribution along the coil axis through the Transformer module; extracting the electric field influencing factors in the local space through the CNN module; and fusing the extracted sequence features with the electric field influencing factors to generate a prediction result of the transient electric field intensity distribution.
8. A dry-type reactor transient process prediction and evaluation system, used to implement the dry-type reactor transient process prediction and evaluation method according to claim 1, characterized in that: include: An equivalent circuit module is used to establish a multi-granularity equivalent circuit model of a dry-type reactor, apply transient boundary conditions to the multi-granularity equivalent circuit model for simulation, obtain transient voltage distribution data of the multi-granularity equivalent circuit model, and generate a transient voltage distribution data set; The model building module is used to build a transient voltage AI prediction model. The structural parameters of the dry-type reactor coil, transient boundary conditions, and transient voltage distribution data set are combined into a training sample set 1. The transient voltage AI prediction model is trained to obtain the trained transient voltage AI prediction model and the transient voltage distribution prediction data output by the model. The model construction module is also used to construct a transient electric field AI prediction model, calculate the spatial electric field intensity distribution based on the transient voltage distribution prediction data output by the transient voltage AI prediction model, and form a training sample set 2 with the structural parameters of the dry-type reactor coil, the transient voltage distribution prediction data, and the spatial electric field intensity distribution. The transient electric field AI prediction model is trained to obtain a trained transient electric field AI prediction model; The prediction module is used to deploy the trained transient voltage AI prediction model and transient electric field AI prediction model in the server system, input the structural parameters and transient boundary conditions of the dry-type reactor coil, and obtain the voltage distribution and electric field distribution prediction results of the dry-type reactor during the transient process.
9. A dry-type reactor transient process prediction and evaluation device, characterized in that: The method comprises a processor and a memory, wherein when the processor executes the computer program stored in the memory, the method for predicting and evaluating the transient process of a dry-type reactor as claimed in any one of claims 1 to 7 is implemented.
10. A computer-readable storage medium, characterized in that Used to store a computer program, wherein when the computer program is executed by a processor, a method for predicting and evaluating a transient process of a dry-type reactor according to any one of claims 1 to 7 is implemented.