CN116451544B - Intelligent optimization design method for high-power high-frequency transformer - Google Patents
Intelligent optimization design method for high-power high-frequency transformer Download PDFInfo
- Publication number
- CN116451544B CN116451544B CN202310633234.9A CN202310633234A CN116451544B CN 116451544 B CN116451544 B CN 116451544B CN 202310633234 A CN202310633234 A CN 202310633234A CN 116451544 B CN116451544 B CN 116451544B
- Authority
- CN
- China
- Prior art keywords
- transformer
- parameter
- parameters
- short
- neural network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 89
- 238000013461 design Methods 0.000 title claims abstract description 46
- 238000005457 optimization Methods 0.000 title claims abstract description 30
- 238000013528 artificial neural network Methods 0.000 claims abstract description 76
- 238000012549 training Methods 0.000 claims abstract description 66
- 238000012360 testing method Methods 0.000 claims abstract description 45
- 238000004364 calculation method Methods 0.000 claims abstract description 28
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 22
- 238000004458 analytical method Methods 0.000 claims abstract description 8
- 230000006872 improvement Effects 0.000 claims abstract description 8
- 238000010206 sensitivity analysis Methods 0.000 claims abstract description 8
- 238000010606 normalization Methods 0.000 claims abstract description 7
- 238000012795 verification Methods 0.000 claims abstract description 7
- 238000010276 construction Methods 0.000 claims abstract description 5
- 230000009466 transformation Effects 0.000 claims abstract description 5
- 239000010410 layer Substances 0.000 claims description 45
- 238000004804 winding Methods 0.000 claims description 41
- 230000006870 function Effects 0.000 claims description 38
- 230000008569 process Effects 0.000 claims description 28
- 239000011159 matrix material Substances 0.000 claims description 26
- 239000003795 chemical substances by application Substances 0.000 claims description 20
- 238000012937 correction Methods 0.000 claims description 17
- 238000012545 processing Methods 0.000 claims description 14
- 230000004913 activation Effects 0.000 claims description 12
- 239000013598 vector Substances 0.000 claims description 12
- 210000002569 neuron Anatomy 0.000 claims description 10
- 238000013507 mapping Methods 0.000 claims description 9
- 238000003062 neural network model Methods 0.000 claims description 9
- 238000009413 insulation Methods 0.000 claims description 7
- 230000000694 effects Effects 0.000 claims description 5
- 238000011156 evaluation Methods 0.000 claims description 5
- 238000005516 engineering process Methods 0.000 claims description 4
- 230000035945 sensitivity Effects 0.000 claims description 4
- XEEYBQQBJWHFJM-UHFFFAOYSA-N Iron Chemical group [Fe] XEEYBQQBJWHFJM-UHFFFAOYSA-N 0.000 claims description 3
- 230000008859 change Effects 0.000 claims description 3
- 238000002790 cross-validation Methods 0.000 claims description 3
- 238000000354 decomposition reaction Methods 0.000 claims description 3
- 230000001419 dependent effect Effects 0.000 claims description 3
- 238000012938 design process Methods 0.000 claims description 3
- 238000002474 experimental method Methods 0.000 claims description 3
- 238000011423 initialization method Methods 0.000 claims description 3
- 239000011229 interlayer Substances 0.000 claims description 3
- 239000000463 material Substances 0.000 claims description 3
- 238000011160 research Methods 0.000 claims description 3
- 238000010200 validation analysis Methods 0.000 claims description 3
- 230000007547 defect Effects 0.000 description 2
- 230000005672 electromagnetic field Effects 0.000 description 2
- 230000002068 genetic effect Effects 0.000 description 2
- 150000001768 cations Chemical class 0.000 description 1
- 239000008358 core component Substances 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 230000002500 effect on skin Effects 0.000 description 1
- 101150050759 outI gene Proteins 0.000 description 1
- 230000000630 rising effect Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/23—Design optimisation, verification or simulation using finite element methods [FEM] or finite difference methods [FDM]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2111/00—Details relating to CAD techniques
- G06F2111/10—Numerical modelling
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Abstract
The invention belongs to the technical field of transformers, and particularly relates to an intelligent optimization design method for a high-power high-frequency transformer. And obtaining an original data set of the short-circuit parameters of the high-power high-frequency transformer by adopting a verified three-dimensional finite element method. And performing sensitivity analysis, dimensionless, normalization and scaling transformation on the short-circuit parameter original data set to form a transformer short-circuit parameter data set, and dividing the transformer short-circuit parameter data set into a training set and a testing set. And constructing an artificial neural network model, and training and optimizing the artificial neural network by using the transformer short-circuit parameter training set and an error back propagation algorithm. And (5) performing accuracy verification on the artificial neural network by using the transformer short-circuit parameter test set. And combining the network model with the short-circuit parameter analysis calculation model to form a high-power high-frequency transformer short-circuit parameter proxy model. According to the construction mode of the transformer short-circuit parameter proxy model, the transformer short-circuit parameter proxy model is embedded into a high-power high-frequency transformer optimal design algorithm, so that intelligent improvement of an optimal design method is realized.
Description
Technical Field
The invention belongs to the technical field of transformers, and particularly relates to an intelligent optimization design method for a high-power high-frequency transformer.
Background
The high-power high-frequency transformer is a core component of an electromagnetic coupling link of the power electronic transformer, and part of researchers are also called as a high-power intermediate-frequency transformer. When the high-power high-frequency transformer is optimally designed, the limitation brought by the front-stage and back-stage power electronic converters on the design of the transformer is required to be comprehensively planned, and the two are matched with each other: (1) The high-power high-frequency transformer has the characteristics of low voltage and high current due to the voltage withstand level of the power electronic converter, and the winding alternating current resistance value is obviously increased along with the rising of frequency, so that the winding loss of the high-power high-frequency transformer is far greater than the core loss; (2) As part of the power electronic converter circuit, the leakage inductance of the transformer needs to be matched with the converter whether phase-shifting control or resonance control is adopted. And the transformer short-circuit parameters are closely related to the winding loss and leakage inductance. As a new type of transformer, high power high frequency transformers have not formed a unified design standard. At present, the optimal design of the high-power high-frequency transformer is mainly carried out by searching an optimal solution in a plurality of groups of schemes. As an important electromagnetic parameter of a high-power high-frequency transformer, efficient and accurate calculation of a short-circuit parameter is a key basis for the optimal design work of the transformer.
The skin effect and the proximity effect brought by the high-frequency vortex enable the electromagnetic field distribution of the transformer winding area to show extremely strong nonlinear characteristics, and enable the calculation of the short-circuit parameters of the high-power high-frequency transformer to be more difficult compared with the power frequency. The methods currently mainly used can be divided into two categories: (1) an analytical calculation formula based on a boolean simplified model. When the high-power high-frequency transformer is optimally designed, the analytical formula is substituted into the transformer optimal design program, and the short-circuit parameter calculation is very rapid. However, the disadvantage is that the dongle makes a certain simplifying assumption in the process of establishing the high-power high-frequency transformer model, and the accuracy is not high. (2) According to the finite element numerical calculation method, a transformer model is built in electromagnetic finite element software, and split operation is carried out on the model to obtain electromagnetic field distribution in the whole model, so that key electromagnetic parameters of the transformer are obtained. The accuracy of the finite element numerical calculation method is higher, but the method has the defect that the method cannot be directly implanted into a transformer optimal design program, and aiming at a plurality of groups of design schemes, a transformer model is required to be built respectively, then the subdivision operation is carried out, and a large amount of time and calculation resources are required to be consumed.
The literature one (high voltage technology, 2017, volume 43, phase 1, pages 210 to 217) discloses a high-power high-frequency transformer applied to a 300kW power unit of an electric and electronic traction transformer, a transformer design flow is established through a free parameter scanning method, and the most effective scheme for considering transformer loss, leakage inductance and weight is selected according to comprehensive evaluation coefficients. The second literature (report of motor engineering in China, 2018, volume 38, 5, pages 1348 to 1355) is based on a constrained multi-objective genetic optimization algorithm, and provides a winding structure optimization design method which takes into consideration constraint conditions such as insulation, loss and leakage inductance of a transformer and optimization targets in the design of the high-frequency transformer, and the optimization design is carried out on a 3.52kW high-frequency transformer model machine through a non-dominant ordering genetic algorithm. The literature three (Design and Experimental Veri fi cation of Three-Phase Medium-Frequency Transformers for High-Power DC-DC Applications) (IEEJ Transactions on Electrical and Electronic Engineering, 16, 12, 1581 to 1593) adopts a free parameter scanning method to optimally design a 15kW high-frequency transformer in a three-Phase double-active bridge, and performs finite element simulation and experimental verification on leakage inductance, loss and temperature rise of a design prototype.
Because the finite element numerical value calculation method cannot be directly implanted into the transformer optimal design program, in the process of the high-power high-frequency transformer optimal design, analytical calculation formulas based on a Peer simplified model are adopted for the calculation of the transformer short-circuit parameters, and the accuracy of the transformer optimal design result is limited by errors of the formulas. Therefore, the accuracy, the high efficiency and the usability of the transformer short-circuit parameter calculation cannot be considered in the prior art, the technical popularization and the application of the high-power high-frequency transformer are affected to a certain extent, and the method has certain limitation.
Disclosure of Invention
Aiming at the defects existing in the prior art, the invention provides an intelligent optimization design method for a high-power high-frequency transformer.
In order to achieve the aim of the invention, the technical scheme adopted by the invention is as follows:
the intelligent optimization design method for the high-power high-frequency transformer is characterized by comprising the following steps of:
step 1, calculating short-circuit parameters of a high-power high-frequency transformer prototype based on a three-dimensional finite element method, and comparing and analyzing a finite element calculation result with a transformer prototype short-circuit parameter test result to verify the accuracy of the three-dimensional finite element method;
step 2, determining the value range of each parameter of the high-power high-frequency transformer, and further considering the number of samples required by training and testing of the artificial neural network to determine the value step length of each parameter of the transformer; constructing a numerical model of the high-power high-frequency transformer according to the value range and the value step length of each parameter, and acquiring an original data set of the short-circuit parameters of the high-power high-frequency transformer under the condition of each parameter through subdivision calculation;
step 3, performing sensitivity analysis, dimensionless treatment, normalization and scaling transformation on the original data set of the short-circuit parameters of the high-power high-frequency transformer to form a transformer short-circuit parameter data set, and dividing the transformer short-circuit parameter data set into a training set and a testing set;
step 4, determining the form, the number of hidden layers, the number of neurons of each layer and the connection relation among the neurons of the artificial neural network; the network parameter initialization comprises the steps of determining initial values of a learning rate, a regularization coefficient, a weight matrix and a bias matrix; the network function initialization comprises the steps of determining the forms of an activation function and a loss function;
step 5, training the artificial neural network parameters by utilizing sample data in a transformer short-circuit parameter training set and adopting an error back propagation algorithm, and continuously updating a weight matrix and a bias matrix in the network through a gradient descent process in the training process until the error rate of the network model on the transformer short-circuit parameter testing set is no longer reduced; because the initialization process of the artificial neural network has certain randomness, when the initial network training does not reach the ideal effect, namely the average error rate of all samples on the test setOptimizing the network; the network optimization method mainly comprises the following steps: training parameter adjustment and network parameter optimization;
step 6, combining the trained and tested artificial neural network and the transformer short-circuit parameter analysis model to form a high-power high-frequency transformer short-circuit parameter intelligent agent model; and designing a data interface of the short-circuit parameter intelligent agent model and the high-power high-frequency transformer optimal design algorithm, and embedding the short-circuit parameter intelligent agent model based on the artificial neural network into the existing transformer optimal design algorithm to realize improvement of the high-power high-frequency transformer optimal design algorithm.
Further as a preferable technical scheme of the invention, the specific steps of the step 1 are as follows:
step 1.1, according to prototype parameters of two high-power high-frequency transformers designed and manufactured in earlier stage research, constructing a prototype three-dimensional numerical model in electromagnetic finite element software, and carrying out subdivision solution on the prototype model to obtain a prototype short-circuit parameter finite element numerical solution;
step 1.2, carrying out short-circuit experiments on two high-power high-frequency transformer prototypes under different frequency operation conditions, short-circuiting the secondary side of the transformer prototypes, and measuring and collecting primary side voltage and current waveform data; carrying out Fourier technology decomposition treatment on primary side voltage and current waveform data to obtain short-circuit parameter experimental test values of a prototype under different frequency operation conditions;
and 1.3, comparing and analyzing the experimental test value with the three-dimensional finite element numerical value calculation value to ensure the correctness and reliability of the three-dimensional finite element numerical model construction and solving method.
Further as a preferable technical scheme of the invention, the specific steps of the step 2 are as follows:
step 2.1, consulting and summarizing the prototype parameters of the high-power high-frequency transformer shown in the prior literature, and determining the value range of each parameter of the high-power high-frequency transformer based on the prototype parameters, wherein the method specifically comprises two types of parameters of the transformer: geometric and electrical parameters; the geometric parameters include: the number of turns, the number of layers and related dimensional parameters of the transformer winding, the width and the height of an iron core window, and the inter-layer insulation and inter-turn insulation dimensions of each layer; the electrical parameters include: transformer current frequency and winding material conductivity;
step 2.2, considering the data sample size required by artificial neural network training and testing of the intelligent agent model of the transformer short-circuit parameters, and determining the value step length of various parameters of the transformer within the parameter range; respectively taking values of all parameters according to the value range and the step length of all parameters of the transformer;
and 2.3, establishing a three-dimensional numerical model of the high-power high-frequency transformer in electromagnetic finite element software according to the values of the parameters, and finally solving the three-dimensional numerical model of the transformer by adopting a three-dimensional finite element solving method verified in the previous step to obtain a transformer short-circuit parameter numerical solution under the conditions of the parameters. Finally, a high-power high-frequency transformer short-circuit parameter data set, namely a winding alternating-current resistance data set, is formedAnd transformer leakage inductance dataset +.>Wherein x is (n) For the n-th set of transformer parameter vectors, R F (n) Calculating the three-dimensional finite element value L of the alternating current resistance of the transformer in the nth group F (n) For the N-th set of transformer leakage inductance three-dimensional finite element calculated values, the data set contains N samples in total.
Further as a preferable technical scheme of the invention, the specific steps of the step 3 are as follows:
step 3.1, in the original data set, the single transformer parameter x i The changed data are classified by the transformer winding alternating current resistance R F Leakage inductance L of transformer F For single transformer parameter x i Deviation determination guide In the mode of (a), for R F 、L F And x i Sensitivity between the two is analyzed; eliminating transformer parameters with smaller partial derivative values in the data set, namely, eliminating the transformer parameters with the target parameters R F 、L F The transformer parameters with weak association degree are used for reducing the dimension of a transformer parameter vector x in the original data set;
step 3.2, reserving the target parameter R by sensitivity analysis F 、L F The transformer parameters with higher association degree are subjected to dimensionless treatment, so that the dimension of a transformer parameter vector x is further reduced, and the generalization capability of the intelligent agent model of the final transformer short-circuit parameters is improved;
step 3.3, further carrying out normalization processing on the transformer parameters subjected to dimensionless processing, and enabling the parameter values to be in a sensitive interval of an artificial neural network activation function derivative on the premise that the data characteristics of the parameters are converted into the same scale so as to improve the training efficiency of the follow-up neural network parameters; a standardized method is adopted, and each dimension parameter characteristic is adjusted to be 0 as a mean value and 1 as a variance; the number of samples in the data set is N, for each dimension of parameter X i Its mean value mu i Sum of variancesThe method comprises the following steps:
parameter X i (n) Subtracting the mean value and dividing the mean value by the variance to obtain a new normalized parameter
Step 3.4, output variable R in transformer short-circuit parameter original data set F 、L F Respectively transformer alternating currentThe three-dimensional finite element calculated value of the resistance and the leakage inductance is subjected to scaling change processing on the output variable in the original data set in a mode of comparing the three-dimensional finite element calculated value with the calculated value of the analytic formula, namely: v (v) R ＝R F /R A 、ν L ＝L F /L A Wherein R is A 、L A Analytical formulas of alternating current resistance and leakage inductance of transformer winding are calculated to obtain values v R 、ν L The correction coefficient between the three-dimensional finite element calculated value and the analytic formula calculated value is used for the alternating current resistance coefficient and leakage inductance of the transformer winding;
step 3.5, the transformer short-circuit parameter data set for training the artificial neural network parameters is formed by analyzing and processing the input variables and the output variables in the transformer short-circuit parameter original data setDividing the two data sets into training sets +.> And test set->The specific dividing mode is as follows: the original dataset of the dataset is divided into 5 groups of non-repeated subsets in average by adopting a cross-validation mode, 4 groups of subsets are selected each time as training sets, and the rest group of subsets are used as validation sets, namely N 1 =0.8n; and 5 times of artificial neural network training are carried out, 5 network models are obtained, and the average value of error rates of the 5 models on respective verification sets is used as the evaluation basis of classification.
Further as a preferable technical scheme of the invention, the specific steps of the step 4 are as follows:
step 4.1, realizing the complex of an input space and an output space without time sequence characteristics by the full-connection feedforward neural network through the multiple complex of simple nonlinear functionsHybrid mapping, network model characteristics and the core features of the mapping problem are consistent, so the full-connection feedback artificial neural network model is adopted to perform +.>Is constructed by an artificial neural network;
step 4.2, initializing a winding alternating current resistivity and transformer leakage inductance full-connection feedback artificial neural network model respectively; for calculating the correction coefficient v of the alternating current resistance of the winding R For example, describes a network initialization process for calculating v of a transformer leakage inductance correction coefficient L The artificial neural network initialization process is similar to the artificial neural network initialization process; proposed network hidden layer number m h Number of neurons in each layer n=2 l Respectively input layer n 1 =5, hidden layer first layer n 2 =5, hidden layer second layer n 3 =5, output layer n 4 =1, learning rate α=0.5, regularization coefficient λ=0.1;
step 4.3, initializing a network weight matrix W and a bias matrix b by adopting a parameter initialization method based on fixed variance; the activation function selects a Tanh function in the Sigmoid function, and the function expression is as follows:
the range of the Tanh function is (-1, 1), and the derivative sensitive interval is (-2, 2); the loss function is selected from square loss functions, and the function expression is as follows:
wherein,for the functional expression of the neural network model, the neural network model outputs +.>The parameter θ includes a neural network weight matrix w and a bias matrix b.
Further as a preferable technical scheme of the invention, the specific steps of the step 5 are as follows:
step 5.1, randomly selecting samples from the training setFeedforward calculates the net input Z for each layer of the neural network (l) And an activation value a (l) Until the last layer:
Z (l) ＝W (l) a (l-1) +b (l) (6)
a (l) ＝f(Z (l) ) (7)
step 5.2, calculating a total network error E:
step 5.3, counter-propagating to calculate the error delta of each layer (l) And updates the network parameter W (l) 、b (l) ：
W (l) ←W (l) -α(δ (l) (a (l-1) ) T +λW (l) ) (11)
b (l) ←b (l) -αδ (l) (12)
Step 5.4, substituting the sample independent variables in the test set into the artificial neural network model with updated parameters, calculating the error rate sigma between the output of the network model and the corresponding target value of the sample dependent variables in the test set, and circulating the steps 5.1-5.3 until the average error rate of all samples of the network model on the test setNo further descent;
step 5.5, because of certain randomness in the initialization process of the network, for example, the original network passes through parameter training, the average error rate of all samples on the test set is less than ideal, namelyAnd (3) optimizing the network by adopting training parameter adjustment, network parameter initialization and other super parameter optimization methods, and re-training the network parameters until the expected target is reached.
Further as a preferable technical scheme of the invention, the specific steps of the step 6 are as follows:
step 6.1, training and optimizing through an artificial neural network to respectively obtainThe neural network model can accurately realize the processed transformer parameter vector X and the winding alternating current resistance correction coefficient v R V of transformer leakage inductance correction coefficient L A nonlinear mapping relationship between the two; according to the processing procedure of the transformer parameters in the step 3, combining the transformer alternating current resistance and leakage inductance analysis calculation model to form a high-power high-frequency transformer short-circuit parameter intelligent agent model;
and 6.2, applying the intelligent transformer short-circuit parameter agent to winding loss and leakage inductance calculation in the high-power high-frequency transformer optimal design process, and realizing improvement of a high-power high-frequency transformer optimal design algorithm based on a free parameter scanning method.
Compared with the prior art, the intelligent optimization design method for the high-power high-frequency transformer has the following technical effects:
the invention provides an intelligent optimization design method of a high-power high-frequency transformer based on a short-circuit parameter proxy model, which has the characteristics of high accuracy and low consumption of computing resources.
Drawings
FIG. 1 is a flow chart of the method of the present invention;
fig. 2 is a flow chart of an improved algorithm of transformer optimization design based on a short-circuit parameter intelligent agent model.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings.
As shown in fig. 1, the invention provides an intelligent optimization design method for a high-power high-frequency transformer, which comprises the following steps:
step 1, calculating short-circuit parameters of a high-power high-frequency transformer prototype based on a three-dimensional finite element method, and comparing and analyzing a finite element calculation result with a transformer prototype short-circuit parameter test result to verify the accuracy of the three-dimensional finite element method;
and 2, determining the value range of each parameter of the high-power high-frequency transformer, and further considering the number of samples required by training and testing of the artificial neural network to determine the value step length of each parameter of the transformer. Constructing a numerical model of the high-power high-frequency transformer according to the value range and the value step length of each parameter, and acquiring an original data set of the short-circuit parameters of the high-power high-frequency transformer under the condition of each parameter through subdivision calculation;
step 3, performing sensitivity analysis, dimensionless treatment, normalization and scaling transformation on the original data set of the short-circuit parameters of the high-power high-frequency transformer to form a transformer short-circuit parameter data set, and dividing the transformer short-circuit parameter data set into a training set and a testing set;
step 4, determining the form, the number of hidden layers, the number of neurons of each layer and the connection relation among the neurons of the artificial neural network; the network parameter initialization comprises the steps of determining initial values of a learning rate, a regularization coefficient, a weight matrix and a bias matrix; the network function initialization comprises the steps of determining the forms of an activation function and a loss function;
and 5, training the artificial neural network parameters by using sample data in the transformer short-circuit parameter training set and adopting an error back propagation algorithm, and continuously updating a weight matrix and a bias matrix in the network through a gradient descent process in the training process until the error rate of the network model on the transformer short-circuit parameter testing set is not reduced. Because of certain randomness in the initialization process of the artificial neural network, for example, the training of the initial network cannot achieve the ideal effect (average error rate of all samples on the test set)) The network is optimized. The network optimization method mainly comprises the following steps: training parameter adjustment, network parameter optimization, and other super parameter optimization;
and 6, combining the trained and tested artificial neural network and the transformer short-circuit parameter analysis model to form the high-power high-frequency transformer short-circuit parameter intelligent agent model. And designing a data interface of the short-circuit parameter intelligent agent model and the high-power high-frequency transformer optimal design algorithm, and embedding the short-circuit parameter intelligent agent model based on the artificial neural network into the existing transformer optimal design algorithm to realize improvement of the high-power high-frequency transformer optimal design algorithm.
The specific steps of the step 1 are as follows:
step 1.1, according to prototype parameters of two high-power high-frequency transformers designed and manufactured in earlier stage research, constructing a prototype three-dimensional numerical model in electromagnetic finite element software, and carrying out subdivision solution on the prototype model to obtain a prototype short-circuit parameter finite element numerical solution;
and 1.2, carrying out short-circuit experiments on two high-power high-frequency transformer prototypes under different frequency operation conditions, shorting the secondary side of the transformer prototypes, and measuring and collecting primary side voltage and current waveform data. Carrying out Fourier technology decomposition treatment on primary side voltage and current waveform data to obtain short-circuit parameter experimental test values of a prototype under different frequency operation conditions;
and 1.3, comparing and analyzing the experimental test value with the three-dimensional finite element numerical value calculation value to ensure the correctness and reliability of the three-dimensional finite element numerical model construction and solving method.
The specific steps of the step 2 are as follows:
step 2.1, consulting and summarizing the prototype parameters of the high-power high-frequency transformer shown in the prior literature, and determining the value range of each parameter of the high-power high-frequency transformer based on the prototype parameters, wherein the method specifically comprises two types of parameters of the transformer: geometric parameters and electrical parameters. The geometric parameters include: the number of turns, the number of layers and related dimensional parameters of the transformer winding, the width and the height of the iron core window, and the inter-layer insulation and inter-turn insulation dimensions of each layer. The electrical parameters include: transformer current frequency and winding material conductivity;
and 2.2, considering the data sample size required by artificial neural network training and testing of the intelligent proxy model of the transformer short-circuit parameters, and determining the value step length of various parameters of the transformer within the parameter range. Respectively taking values of all parameters according to the value range and the step length of all parameters of the transformer;
and 2.3, establishing a three-dimensional numerical model of the high-power high-frequency transformer in electromagnetic finite element software according to the values of the parameters, and finally solving the three-dimensional numerical model of the transformer by adopting a three-dimensional finite element solving method verified in the previous step to obtain a transformer short-circuit parameter numerical solution under the conditions of the parameters. Finally, a high-power high-frequency transformer short-circuit parameter data set, namely a winding alternating-current resistance data set, is formedAnd transformer leakage inductance dataset +.>Wherein x is (n) For the n-th set of transformer parameter vectors, R F (n) Alternating current for the n-th group of transformersResistance three-dimensional finite element calculated value, L F (n) For the N-th set of transformer leakage inductance three-dimensional finite element calculated values, the data set contains N samples in total.
The specific steps of the step 3 are as follows:
step 3.1, in the original data set, the single transformer parameter x i Data changed (other parameters unchanged) are classified and passed through transformer winding ac resistor R F Leakage inductance L of transformer F For single transformer parameter x i Deviation determination guideIn the mode of (a), for R F 、L F And x i Sensitivity between the two is analyzed. Eliminating transformer parameters with smaller partial derivative values in the data set, namely, eliminating the transformer parameters with the target parameters R F 、L F The transformer parameters with weak association degree are used for reducing the dimension of a transformer parameter vector x in the original data set;
step 3.2, reserving the target parameter R by sensitivity analysis F 、L F And carrying out dimensionless processing on the transformer parameters with higher association degree so as to further reduce the dimension of a transformer parameter vector x and improve the generalization capability of the intelligent proxy model of the final transformer short-circuit parameters. Transformers contain both geometric and electrical parameters. For the same kind of geometric parameters, one of the geometric parameters is taken as a reference value, such as the thickness t of the transformer r And obtaining dimensionless parameters of the transformer geometric class by means of calculating the ratio of other geometric parameters to the reference value. For electrical parameters, the skin depth delta of the winding can be measured by the physical quantity (and the winding conductivity sigma w Winding current frequency f n Correlation), electrical parameter winding conductivity sigma w Winding current frequency f n Converting into geometric parameters, and performing dimensionless treatment in a mode of calculating a ratio with a geometric standard value;
step 3.3, further normalizing the dimensionless transformer parameters, and enabling the parameter values to be in artificial neural network activation on the premise that the data characteristics of the parameters are converted into the same scaleIn the sensitive interval of the function derivative, the training efficiency of the follow-up neural network parameters is improved. And (3) adjusting the characteristic of each dimension parameter to be 0 as a mean value and 1 as a variance by adopting a standardized method. The number of samples in the data set is N, for each dimension of parameter X i Its mean value mu i Sum of variancesThe method comprises the following steps:
parameters are setSubtracting the mean value and dividing by the variance to obtain a new normalized parameter +.>
Step 3.4, output variable R in transformer short-circuit parameter original data set F 、L F The three-dimensional finite element calculated values of the alternating current resistance and the leakage inductance of the transformer are respectively obtained, and scaling change processing is carried out on the output variables in the original data set in a mode of comparing the three-dimensional finite element calculated values with the calculated values of the analytic formula, namely: v (v) R ＝R F /R A 、ν L ＝L F /L A Wherein R is A 、L A Analytical formulas of alternating current resistance and leakage inductance of transformer winding are calculated to obtain values v R 、ν L The correction coefficient between the three-dimensional finite element calculated value and the analytic formula calculated value is used for the alternating current resistance coefficient and leakage inductance of the transformer winding;
step 3.5, the transformer short-circuit parameter data set for training the artificial neural network parameters is formed by analyzing and processing the input variables and the output variables in the transformer short-circuit parameter original data setDividing the two data sets into training sets +.> And test set->The specific dividing mode is as follows: the original dataset of the dataset is divided into 5 groups of non-repeated subsets in average by adopting a cross-validation mode, 4 groups of subsets are selected each time as training sets, and the rest group of subsets are used as validation sets, namely N 1 =0.8n. Therefore, the artificial neural network training can be carried out for 5 times, 5 network models are obtained, the average value of error rates of the 5 models on respective verification sets is used as the evaluation basis of classification, and the influence of randomness in the process of dividing the training set and the test set on the evaluation result can be effectively avoided.
The specific steps of the step 4 are as follows:
step 4.1, the fully connected feedforward neural network can well realize complex mapping of an input space and an output space without time sequence characteristics through multiple complex of simple nonlinear functions, and the network model characteristics of the fully connected feedforward neural networkThe core characteristics of the mapping problem are consistent, so that the full-connection feedback artificial neural network model is adopted to respectively carry outIs constructed by artificial neural network.
And 4.2, initializing a winding alternating-current resistivity and transformer leakage inductance full-connection feedback artificial neural network model respectively. For calculating the correction coefficient v of the alternating current resistance of the winding R For example, describes a network initialization process for calculating v of a transformer leakage inductance correction coefficient L The artificial neural network initialization process is similar thereto. Proposed network hidden layer number m h Number of neurons in each layer n=2 l Respectively input layer n 1 =5, hidden layer first layer n 2 =5, hidden layer second layer n 3 =5, output layer n 4 =1, learning rate α=0.5, regularization coefficient λ=0.1;
and 4.3, initializing a network weight matrix W and a bias matrix b by adopting a parameter initialization method based on fixed variance so as to ensure that the distinguishing property between different neurons is better. The activation function selects a Tanh function in the Sigmoid function, and the function expression is as follows:
the range of the Tanh function is (-1, 1), and the derivative sensitivity interval is (-2, 2). The loss function is selected from square loss functions, and the function expression is as follows:
wherein,for the functional expression of the neural network model, the neural network model outputs +.>The parameter θ includes a neural network weight matrix w and a bias matrix b.
Step 5, after the artificial neural network is built, training by parameters by using sample data in a training set and adopting an error back propagation algorithmAnd updating the weight matrix and the bias matrix in the network in the gradient descent process in the process until the error rate of the network model on the test set is no longer reduced. For calculating the correction coefficient v of the alternating current resistance of the winding R For example, describes a network parameter training process for calculating v of a transformer leakage inductance correction coefficient L The artificial neural network parameter training process is similar to the artificial neural network parameter training process, and comprises the following specific steps:
step 5.1, randomly selecting samples from the training setFeedforward calculates the net input Z for each layer of the neural network (l) And an activation value a (l) Until the last layer:
Z (l) ＝W (l) a (l-1) +b (l) (6)
a (l) ＝f(Z (l) ) (7)
step 5.2, calculating a total network error E:
step 5.3, counter-propagating to calculate the error delta of each layer (l) And updates the network parameter W (l) 、b (l) ：
W (l) ←W (l) -α(δ (l) (a (l-1) ) T +λW (l) ) (11)
b (l) ←b (l) -αδ (l) (12)
Step 5.4, substituting the sample independent variables in the test set into the artificial neural network model with updated parameters, calculating the error rate sigma between the output of the network model and the corresponding target value of the sample dependent variables in the test set, and circulating the steps 5.1-5.3 until the average error rate of all samples of the network model on the test setNo further descent;
step 5.5, because of certain randomness in the initialization process of the network, for example, the original network passes through parameter training, the average error rate of all samples on the test set is less than idealAnd (3) optimizing the network by adopting training parameter adjustment, network parameter initialization and other super parameter optimization methods, and re-training the network parameters until the expected target is reached.
The specific steps of the step 6 are as follows:
step 6.1, training and optimizing through an artificial neural network to respectively obtainThe neural network model can accurately realize the processed transformer parameter vector +.>Correction coefficient v of alternating current resistance of winding R V of transformer leakage inductance correction coefficient L Nonlinear mapping relation between the two. According to the processing procedure of the transformer parameters in the step 3, combining the transformer alternating current resistance and leakage inductance analysis calculation model to form a high-power high-frequency transformer short-circuit parameter intelligent agent model;
and 6.2, as shown in fig. 2, applying the intelligent transformer short-circuit parameter agent to winding loss and leakage inductance calculation in the high-power high-frequency transformer optimal design process, and realizing improvement of a high-power high-frequency transformer optimal design algorithm based on a free parameter scanning method.
The invention discloses an intelligent optimization design method for a high-power high-frequency transformer. Firstly, obtaining an original data set of the short-circuit parameters of a high-power high-frequency transformer by adopting a three-dimensional finite element method after verification. And combining the physical characteristics of the short-circuit parameters of the transformer with the characteristics of the artificial neural network method, performing sensitivity analysis, dimensionless, normalization and scaling transformation on the original data set of the short-circuit parameters to form a data set of the short-circuit parameters of the transformer, and dividing the data set of the short-circuit parameters into a training set and a testing set. Secondly, building a transformer short-circuit parameter artificial neural network model, specifically comprising network model form selection and network initialization, and training and optimizing the initialized artificial neural network by utilizing the obtained transformer short-circuit parameter training set and an error back propagation algorithm. And (3) performing accuracy verification on the trained and optimized artificial neural network by using the transformer short-circuit parameter test set. And combining the verified transformer short-circuit parameter artificial neural network model with the short-circuit parameter analysis calculation model to form a high-power high-frequency transformer short-circuit parameter proxy model. Finally, according to the construction mode of the transformer short-circuit parameter proxy model, embedding the transformer short-circuit parameter proxy model into a high-power high-frequency transformer optimal design algorithm to realize intelligent improvement of an optimal design method.
While the foregoing is directed to embodiments of the present invention, other and further details of the invention may be devised without departing from the basic scope thereof, and the scope thereof is determined by the claims that follow.
Claims (5)
1. The intelligent optimization design method for the high-power high-frequency transformer is characterized by comprising the following steps of:
step 1, calculating short-circuit parameters of a high-power high-frequency transformer prototype based on a three-dimensional finite element method, and comparing and analyzing a finite element calculation result with a transformer prototype short-circuit parameter test result to verify the accuracy of the three-dimensional finite element method;
step 2, determining the value range of each parameter of the high-power high-frequency transformer, and further considering the number of samples required by training and testing of the artificial neural network to determine the value step length of each parameter of the transformer; constructing a numerical model of the high-power high-frequency transformer according to the value range and the value step length of each parameter, and acquiring an original data set of the short-circuit parameters of the high-power high-frequency transformer under the condition of each parameter through subdivision calculation;
step 3, performing sensitivity analysis, dimensionless treatment, normalization and scaling transformation on the original data set of the short-circuit parameters of the high-power high-frequency transformer to form a transformer short-circuit parameter data set, and dividing the transformer short-circuit parameter data set into a training set and a testing set;
step 4, determining the form, the number of hidden layers, the number of neurons of each layer and the connection relation among the neurons of the artificial neural network; the network parameter initialization comprises the steps of determining initial values of a learning rate, a regularization coefficient, a weight matrix and a bias matrix; the network function initialization comprises the steps of determining the forms of an activation function and a loss function;
step 5, training the artificial neural network parameters by utilizing sample data in a transformer short-circuit parameter training set and adopting an error back propagation algorithm, and continuously updating a weight matrix and a bias matrix in the network through a gradient descent process in the training process until the error rate of the network model on the transformer short-circuit parameter testing set is no longer reduced; because the initialization process of the artificial neural network has certain randomness, when the initial network training does not reach the ideal effect, namely the average error rate of all samples on the test setOptimizing the network; the network optimization method mainly comprises the following steps: training parameter adjustment and network parameter optimization;
step 6, combining the trained and tested artificial neural network and the transformer short-circuit parameter analysis model to form a high-power high-frequency transformer short-circuit parameter intelligent agent model; designing a data interface of the short-circuit parameter intelligent agent model and a high-power high-frequency transformer optimal design algorithm, embedding the short-circuit parameter intelligent agent model based on the artificial neural network into the existing transformer optimal design algorithm, and realizing improvement of the high-power high-frequency transformer optimal design algorithm;
the specific steps of the step 3 are as follows:
step 3.1, in the original data set, the single transformer parameter x i The changed data are classified by the transformer winding alternating current resistance R F Leakage inductance L of transformer F For single transformer parameter x i Deviation determination guideIn the mode of (a), for R F 、L F And x i Sensitivity between the two is analyzed; eliminating transformer parameters with smaller partial derivative values in the data set, namely, eliminating the transformer parameters with the target parameters R F 、L F The transformer parameters with weak association degree are used for reducing the dimension of a transformer parameter vector x in the original data set;
step 3.2, reserving the target parameter R by sensitivity analysis F 、L F The transformer parameters with higher association degree are subjected to dimensionless treatment, so that the dimension of a transformer parameter vector x is further reduced, and the generalization capability of the intelligent agent model of the final transformer short-circuit parameters is improved;
step 3.3, further carrying out normalization processing on the transformer parameters subjected to dimensionless processing, and enabling the parameter values to be in a sensitive interval of an artificial neural network activation function derivative on the premise that the data characteristics of the parameters are converted into the same scale so as to improve the training efficiency of the follow-up neural network parameters; a standardized method is adopted, and each dimension parameter characteristic is adjusted to be 0 as a mean value and 1 as a variance; the number of samples in the data set is N, for each dimension of parameter X i Its mean value mu i Sum of variancesThe method comprises the following steps:
parameters are setSubtracting the mean value and dividing by the variance to obtain a new normalized parameter +.>
Step 3.4, output variable R in transformer short-circuit parameter original data set F 、L F The three-dimensional finite element calculated values of the alternating current resistance and the leakage inductance of the transformer are respectively obtained, and scaling change processing is carried out on the output variables in the original data set in a mode of comparing the three-dimensional finite element calculated values with the calculated values of the analytic formula, namely: v (v) R ＝R F /R A 、ν L ＝L F /L A Wherein R is A 、L A Analytical formulas of alternating current resistance and leakage inductance of transformer winding are calculated to obtain values v R 、ν L The correction coefficient between the three-dimensional finite element calculated value and the analytic formula calculated value is used for the alternating current resistance coefficient and leakage inductance of the transformer winding;
step 3.5, the transformer short-circuit parameter data set for training the artificial neural network parameters is formed by analyzing and processing the input variables and the output variables in the transformer short-circuit parameter original data set Dividing the two data sets into training sets +.>And test setThe specific dividing mode is as follows: the original dataset of the dataset is divided into 5 groups of non-repeated subsets in average by adopting a cross-validation mode, 4 groups of subsets are selected each time as training sets, and the rest group of subsets are used as validation sets, namely N 1 =0.8n; training an artificial neural network for 5 times, obtaining 5 network models, and taking the average value of error rates of the 5 models on respective verification sets as a classification evaluation basis;
the specific steps of the step 6 are as follows:
step 6.1, training and optimizing through an artificial neural network to respectively obtainThe neural network model can accurately realize the processed transformer parameter vector +.>Correction coefficient v of alternating current resistance of winding R V of transformer leakage inductance correction coefficient L A nonlinear mapping relationship between the two; according to the processing procedure of the transformer parameters in the step 3, combining the transformer alternating current resistance and leakage inductance analysis calculation model to form a high-power high-frequency transformer short-circuit parameter intelligent agent model;
and 6.2, applying the intelligent transformer short-circuit parameter agent to winding loss and leakage inductance calculation in the high-power high-frequency transformer optimal design process, and realizing improvement of a high-power high-frequency transformer optimal design algorithm based on a free parameter scanning method.
2. The intelligent optimization design method of the high-power high-frequency transformer according to claim 1, wherein the specific steps of the step 1 are as follows:
step 1.1, according to prototype parameters of two high-power high-frequency transformers designed and manufactured in earlier stage research, constructing a prototype three-dimensional numerical model in electromagnetic finite element software, and carrying out subdivision solution on the prototype model to obtain a prototype short-circuit parameter finite element numerical solution;
step 1.2, carrying out short-circuit experiments on two high-power high-frequency transformer prototypes under different frequency operation conditions, short-circuiting the secondary side of the transformer prototypes, and measuring and collecting primary side voltage and current waveform data; carrying out Fourier technology decomposition treatment on primary side voltage and current waveform data to obtain short-circuit parameter experimental test values of a prototype under different frequency operation conditions;
and 1.3, comparing and analyzing the experimental test value with the three-dimensional finite element numerical value calculation value to ensure the correctness and reliability of the three-dimensional finite element numerical model construction and solving method.
3. The intelligent optimization design method of the high-power high-frequency transformer according to claim 1, wherein the specific steps of the step 2 are as follows:
step 2.1, consulting and summarizing the prototype parameters of the high-power high-frequency transformer shown in the prior literature, and determining the value range of each parameter of the high-power high-frequency transformer based on the prototype parameters, wherein the method specifically comprises two types of parameters of the transformer: geometric and electrical parameters; the geometric parameters include: the number of turns, the number of layers and related dimensional parameters of the transformer winding, the width and the height of an iron core window, and the inter-layer insulation and inter-turn insulation dimensions of each layer; the electrical parameters include: transformer current frequency and winding material conductivity;
step 2.2, considering the data sample size required by artificial neural network training and testing of the intelligent agent model of the transformer short-circuit parameters, and determining the value step length of various parameters of the transformer within the parameter range; respectively taking values of all parameters according to the value range and the step length of all parameters of the transformer;
step 2.3, pressingAccording to the values of all parameters, a three-dimensional numerical model of the high-power high-frequency transformer is established in electromagnetic finite element software, and finally, the three-dimensional numerical model of the transformer is solved by adopting a three-dimensional finite element solving method verified in the previous step, so that a transformer short-circuit parameter numerical solution under all parameter conditions is obtained; finally, a high-power high-frequency transformer short-circuit parameter data set, namely a winding alternating-current resistance data set, is formedAnd transformer leakage inductance dataset +.>Wherein x is (n) For the n-th set of transformer parameter vectors, R F (n) Calculating the three-dimensional finite element value L of the alternating current resistance of the transformer in the nth group F (n) For the N-th set of transformer leakage inductance three-dimensional finite element calculated values, the data set contains N samples in total.
4. The intelligent optimization design method of the high-power high-frequency transformer according to claim 1, wherein the specific steps of the step 4 are as follows:
step 4.1, the fully connected feedforward neural network realizes complex mapping of input space and output space without time sequence characteristic through multiple complex of simple nonlinear function, and the network model characteristic and the output space are the same as those of the fully connected feedforward neural networkThe core features of the mapping problem are consistent, so the full-connection feedback artificial neural network model is adopted to perform +.>Is constructed by an artificial neural network;
step 4.2, initializing a winding alternating current resistivity and transformer leakage inductance full-connection feedback artificial neural network model respectively; for calculating the correction coefficient v of the alternating current resistance of the winding R Is the artificial neural network of (a)For example, a network initialization process is described for calculating v of a transformer leakage inductance correction coefficient L The artificial neural network initialization process is similar to the artificial neural network initialization process; proposed network hidden layer number m h Number of neurons in each layer n=2 l Respectively input layer n 1 =5, hidden layer first layer n 2 =5, hidden layer second layer n 3 =5, output layer n 4 =1, learning rate α=0.5, regularization coefficient λ=0.1;
step 4.3, initializing a network weight matrix W and a bias matrix b by adopting a parameter initialization method based on fixed variance; the activation function selects a Tanh function in the Sigmoid function, and the function expression is as follows:
the range of the Tanh function is (-1, 1), and the derivative sensitive interval is (-2, 2); the loss function is selected from square loss functions, and the function expression is as follows:
wherein,for the functional expression of the neural network model, the neural network model outputs +.>The parameter θ includes a neural network weight matrix w and a bias matrix b.
5. The intelligent optimization design method of the high-power high-frequency transformer according to claim 1, wherein the specific steps of the step 5 are as follows:
step 5.1, randomly selecting samples from the training setFeedforward calculates the net input Z for each layer of the neural network (l) And an activation value a (l) Until the last layer:
Z (l) ＝W (l) a (l-1) +b (l) (6)
a (l) ＝f(Z (l) ) (7)
step 5.2, calculating a total network error E:
step 5.3, counter-propagating to calculate the error delta of each layer (l) And updates the network parameter W (l) 、b (l) ：
W (l) ←W (l) -α(δ (l) (a (l-1) ) T +λW (l) ) (11)b (l) ←b (l) -αδ (l) (12)
Step 5.4, substituting the sample independent variables in the test set into the artificial neural network model with updated parameters, calculating the error rate sigma between the output of the network model and the corresponding target value of the sample dependent variables in the test set, and circulating the steps 5.1-5.3 until the average error rate of all samples of the network model on the test setNo further descent;
step 5.5, because of certain randomness in the initialization process of the network, when the original network passes through parameter training, all the network are on the test setThe average error rate of the samples is less than idealAnd (3) optimizing the network by adopting training parameter adjustment, network parameter initialization and other super parameter optimization methods, and re-training the network parameters until the expected target is reached.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310633234.9A CN116451544B (en) | 2023-05-31 | 2023-05-31 | Intelligent optimization design method for high-power high-frequency transformer |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310633234.9A CN116451544B (en) | 2023-05-31 | 2023-05-31 | Intelligent optimization design method for high-power high-frequency transformer |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116451544A CN116451544A (en) | 2023-07-18 |
| CN116451544B true CN116451544B (en) | 2023-11-07 |
Family
ID=87127618
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310633234.9A Active CN116451544B (en) | 2023-05-31 | 2023-05-31 | Intelligent optimization design method for high-power high-frequency transformer |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116451544B (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102570392A (en) * | 2012-01-17 | 2012-07-11 | 上海电力学院 | Method for identifying exciting inrush current of transformer based on improved probability neural network |
| CN108039271A (en) * | 2017-10-20 | 2018-05-15 | 北京首钢股份有限公司 | A kind of method of optimization EI transformer materials |
| CN111105332A (en) * | 2019-12-19 | 2020-05-05 | 河北工业大学 | Highway intelligent pre-maintenance method and system based on artificial neural network |
| CN111597749A (en) * | 2020-05-28 | 2020-08-28 | 南通大学 | Method for calculating alternating current resistance coefficient of hollow winding of high-frequency transformer |
| CN112182943A (en) * | 2020-08-27 | 2021-01-05 | 河海大学 | Parameter optimization method for improving short circuit between poles of DC/DC direct-current transformer |
| CN112287583A (en) * | 2020-10-29 | 2021-01-29 | 广东电网有限责任公司电力科学研究院 | Noise simulation prediction method for transformer |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11481664B2 (en) * | 2018-09-05 | 2022-10-25 | General Electric Company | Methods and systems for generating device-specific machine learning model |
| CA3126521A1 (en) * | 2021-03-31 | 2022-09-30 | Metlab Research Inc. | A scalable simulation platform for power transformers rating, loading policy, and thermal performances evaluation |
-
2023
- 2023-05-31 CN CN202310633234.9A patent/CN116451544B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102570392A (en) * | 2012-01-17 | 2012-07-11 | 上海电力学院 | Method for identifying exciting inrush current of transformer based on improved probability neural network |
| CN108039271A (en) * | 2017-10-20 | 2018-05-15 | 北京首钢股份有限公司 | A kind of method of optimization EI transformer materials |
| CN111105332A (en) * | 2019-12-19 | 2020-05-05 | 河北工业大学 | Highway intelligent pre-maintenance method and system based on artificial neural network |
| CN111597749A (en) * | 2020-05-28 | 2020-08-28 | 南通大学 | Method for calculating alternating current resistance coefficient of hollow winding of high-frequency transformer |
| CN112182943A (en) * | 2020-08-27 | 2021-01-05 | 河海大学 | Parameter optimization method for improving short circuit between poles of DC/DC direct-current transformer |
| CN112287583A (en) * | 2020-10-29 | 2021-01-29 | 广东电网有限责任公司电力科学研究院 | Noise simulation prediction method for transformer |
Non-Patent Citations (1)
| Title |
|---|
| 一种基于人工神经网络的反馈式神经元数优化方法;王宁 等;《电路与系统学报》;第18卷(第02期);第41-47页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116451544A (en) | 2023-07-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Guillod et al. | Artificial neural network (ANN) based fast and accurate inductor modeling and design | |
| CN106443285B (en) | Multiple-harmonic-source harmonic responsibility quantitative analysis method based on total least square method | |
| Zang et al. | Adaptive method for harmonic contribution assessment based on hierarchical K‐means clustering and Bayesian partial least squares regression | |
| Ma et al. | A hybrid dynamic equivalent using ANN-based boundary matching technique | |
| CN108019206B (en) | With boring electromagnetic wave resistivity instrument Range Extension method under a kind of high-k | |
| CN112417727B (en) | High-frequency transformer leakage inductance parameter calculation method considering end effect | |
| Chen et al. | Modeling and optimization of EMI filter by using artificial neural network | |
| CN109284541A (en) | A kind of more Method of Physical Modeling of neural network for microwave passive component | |
| CN112596016A (en) | Transformer fault diagnosis method based on integration of multiple one-dimensional convolutional neural networks | |
| CN111639471B (en) | Electromagnetic interference filter design method based on recurrent neural network | |
| Rojas-Dueñas et al. | Black-box modeling of DC–DC converters based on wavelet convolutional neural networks | |
| Rahimpour et al. | Applying artificial optimization methods for transformer model reduction of lumped parameter models | |
| Yang et al. | Multi-objective optimization of contactor’s characteristics based on RBF neural networks and hybrid method | |
| Wang et al. | Fault location of transmission line based on CNN-LSTM double-ended combined model | |
| CN116451544B (en) | Intelligent optimization design method for high-power high-frequency transformer | |
| CN111178528B (en) | Elite genetic algorithm improvement method applied to wireless charging system | |
| Zhu et al. | Modeling of line impedance stabilization network impedance characteristic based on genetic algorithm | |
| Yin et al. | Switching converter EMI conduction modelling and verification | |
| Kül et al. | Coreloss Estimation via Long Short-Term Memory Model (LSTM) of Dry-Type Transformer based on FEA | |
| CN116256402A (en) | Valve side sleeve defect detection method based on frequency domain dielectric spectrum data | |
| Wang et al. | A synthesis-analysis machine with self-inspection mechanism for automatic design of on-chip inductors based on artificial neural networks | |
| CN112307670A (en) | Design method of pressurized water reactor core parameter prediction model based on bagging integrated neural network | |
| Gu | Innovative Sensor Design Method of Industrial Ceramic Product Modeling Based on Immune Optimization Algorithm | |
| Ye et al. | Simulation Application of Interconnected Structure of Microsystems Based on Artificial Neural Network Technology | |
| Wang et al. | Surrogate-Assisted Multiobjective Optimization of Double-D Coil for Inductive Power Transfer System With LCC–LCC Compensation Network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |