CN114896852A - Converter transformer scaling model vibration parameter prediction method based on PSO-BP neural network - Google Patents

Converter transformer scaling model vibration parameter prediction method based on PSO-BP neural network Download PDF

Info

Publication number
CN114896852A
CN114896852A CN202210661243.4A CN202210661243A CN114896852A CN 114896852 A CN114896852 A CN 114896852A CN 202210661243 A CN202210661243 A CN 202210661243A CN 114896852 A CN114896852 A CN 114896852A
Authority
CN
China
Prior art keywords
neural network
converter transformer
training
iron core
data
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.)
Pending
Application number
CN202210661243.4A
Other languages
Chinese (zh)
Inventor
张黎
王栋
王昊
王东晖
王磊磊
张嵩阳
孙优良
邹亮
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shandong University
Electric Power Research Institute of State Grid Henan Electric Power Co Ltd
Original Assignee
Shandong University
Electric Power Research Institute of State Grid Henan Electric Power Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Shandong University, Electric Power Research Institute of State Grid Henan Electric Power Co Ltd filed Critical Shandong University
Priority to CN202210661243.4A priority Critical patent/CN114896852A/en
Publication of CN114896852A publication Critical patent/CN114896852A/en
Pending legal-status Critical Current

Links

Images

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
    • G06F30/25—Design optimisation, verification or simulation using particle-based methods
    • 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
    • G06N3/004—Artificial life, i.e. computing arrangements simulating life
    • G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
    • 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
    • G06N3/045—Combinations of networks
    • 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
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00—Details relating to CAD techniques
    • G06F2111/04—Constraint-based CAD
    • 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 provides a converter transformer scaling model vibration parameter prediction method based on a PSO-BP neural network, which comprises the following steps: constructing a converter transformer scaling model, and giving any input parameter based on the scaling model to obtain vibration output information corresponding to the converter transformer as training data; constructing a PSO-BP neural network suitable for training a converter transformer scaling model, taking the input of neural network data as the voltage, the size and the number of turns of the converter transformer, and outputting the input data as the acceleration of an iron core, the deformation displacement of the iron core and the stress of the iron core, and finishing the training of the neural network based on the training data; and inputting the size, voltage and turns of the converter transformer to be predicted by using the trained neural network, and predicting the acceleration, deformation and displacement of the iron core and stress of the iron core.

Description

Converter transformer scaling model vibration parameter prediction method based on PSO-BP neural network
Technical Field
The invention belongs to the technical field of converter transformer scale model vibration parameter prediction, and particularly relates to a converter transformer scale model vibration parameter prediction method based on a PSO-BP neural network.
Background
The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art.
Meanwhile, the problems of huge volume, strict requirement on ground insulation, high manufacturing cost and the like of the converter transformer make researchers difficult to conveniently and effectively research the converter transformer on some problems, and particularly relate to the non-linear problems of structural mechanics and the like of internal devices of the converter transformer. A reliable solution to the above problem could be found if it could be scaled to a laboratory-acceptable volume while still ensuring its operational characteristics. The similarity theory is a description of similarity rules among various objects and a theory for researching the application of the similarity rules among the objects, and provides an idea for solving the problem 5. However, the derivation of the scaling rule of the converter transformer must strictly adhere to the relevant empirical formulas of electromagnetism and structural mechanics, which may cause the derived scaling rule to be difficult to satisfy the practical operation and design requirements of the converter transformer, such as the number of winding turns and the frequency of the operating power supply. At present, a converter transformer scaling model which can be matched with an original machine in the aspects of electromagnetism and solid mechanics does not exist.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides the converter transformer scaling model vibration parameter prediction method based on the PSO-BP neural network, which can realize the prediction of the converter transformer vibration output information, thereby providing guidance and basis for the design of the converter transformer scaling model.
In order to achieve the above object, one or more embodiments of the present invention provide the following technical solutions:
in a first aspect, a converter transformer scaling model vibration parameter prediction method based on a PSO-BP neural network is disclosed, and comprises the following steps:
constructing a converter transformer scale model, and giving any input parameter based on the scale model to obtain vibration output information corresponding to the converter transformer as training data;
constructing a PSO-BP neural network suitable for training a converter transformer scaling model, taking the input of neural network data as the voltage, the size and the number of turns of the converter transformer, and outputting the input data as the acceleration of an iron core, the deformation displacement of the iron core and the stress of the iron core, and finishing the training of the neural network based on the training data;
and inputting the size, voltage and turns of the converter transformer to be predicted by using the trained neural network, and predicting the acceleration, deformation and displacement of the iron core and stress of the iron core.
As a further technical scheme, when a converter transformer scaling model is constructed, a plurality of converter transformers which are combined to operate are equivalent to a three-phase transformer, and converter transformer finite element simulation models in two different iron core forms are established in finite element simulation software.
As a further technical scheme, in the modeling process, the parts of the converter transformer are simplified: omitting the components for fixing and replacing by 'fixed constraint' boundary conditions; and material definition is carried out on each part of the model, the iron core is made of soft iron material, the winding is made of copper material, and other areas in the box body are made of transformer oil material.
As a further technical scheme, the input voltage, the geometric dimension and the winding turns of the converter transformer scaling model are used as three groups of random variables, and the stress amplitude values of the model iron cores and the windings and the displacement amplitude values of the iron cores and the windings under different combination conditions of the plurality of groups are recorded;
and carrying out data segmentation processing on the multiple groups of data groups, taking input voltage, geometric dimension and winding turns as input data, and taking the stress amplitude of the model iron core and the model winding corresponding to the input data and the displacement amplitude of the iron core and the model winding as output data to be used as neural network training data.
As a further technical scheme, the PSO-BP neural network model is established by firstly using a particle swarm algorithm to search an optimal solution globally, using the optimal solution as an initial parameter of the neural network, and then training the BP neural network: the optimal threshold and weight of the BP neural network are searched by using a particle swarm optimization algorithm, and then the optimal threshold and the optimal weight are used as initial parameters of the network to perform network training.
As a further technical scheme, the PSO-BP neural network is a particle swarm optimization BP neural network, and the detailed steps are established as follows:
(1) randomly generating an initial population of particles and initializing P i And P g ；
(2) Regarding each particle as a group of parameter values to form a parameter network;
(3) inputting training sample data in training data into a parameter network for training;
(4) p is performed by calculating a fitness value for each particle i And P g Determination of (1);
(5) if the termination condition of the program is not satisfied at this time, the operation goes to (6), and if the termination condition is satisfied, the operation goes to (7);
(6) updating each particle in the Weilian group according to the formula, and turning to (2);
(7) and taking the optimal particles as a group of optimal parameters, wherein the optimal particles are the optimal results, and ending the algorithm.
As a further technical scheme, the specific steps for constructing the PSO-BP neural network are as follows:
determining the number of network layers;
training input data are voltage, size and turns of the converter transformer, output data are deformation displacement of the iron core and the winding and stress of the iron core, the training input data are divided into training data and testing data, and the training data and the testing data are respectively subjected to normalization processing;
determining a number of hidden layer neurons;
by setting training parameters, the network performs training and learning to achieve the aim of setting error values
And (3) testing a sample for prediction: and predicting the size, voltage and turns of the converter transformer of the test sample, comparing the predicted size, voltage and turns with output characteristics, winding displacement deformation and stress in a simulation result, and checking the accuracy of the network.
In a second aspect, a converter transformer scaling model vibration parameter prediction system based on a PSO-BP neural network is disclosed, which includes:
a scaling model construction module configured to: constructing a converter transformer scaling model, and giving any input parameter based on the scaling model to obtain vibration output information corresponding to the converter transformer as training data;
a neural network construction module configured to: constructing a PSO-BP neural network suitable for training a converter transformer scaling model, taking the input of neural network data as the voltage, the size and the number of turns of the converter transformer, and outputting the input data as the acceleration of an iron core, the deformation displacement of the iron core and the stress of the iron core, and finishing the training of the neural network based on the training data;
a prediction module configured to: and inputting the size, voltage and turns of the converter transformer to be predicted by using the trained neural network, and predicting the acceleration, deformation and displacement of the iron core and stress of the iron core.
The above one or more technical solutions have the following beneficial effects:
the invention inputs the voltage, the size proportion and the number of turns of the winding of the converter transformer scaling model which is designed in advance to the input end of the network, and the vibration parameters of the scaling model which is prepared in the future can be accurately estimated. The particle swarm optimization-based BP neural network provides support for the reliability of the vibration scaling model of the converter transformer, and can avoid a large number of unnecessary trial and error.
The feasibility and the accuracy of the application of the particle swarm optimization-based BP neural network in the prediction of the vibration parameters of the scale model of the converter transformer are verified. The BP neural network trained by the optimization algorithm and a large amount of data can provide prediction for vibration parameters of a pre-prepared converter transformer scaling model.
Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, are included to provide a further understanding of the invention, and are incorporated in and constitute a part of this specification, illustrate exemplary embodiments of the invention and together with the description serve to explain the invention and not to limit the invention.
FIG. 1 is a schematic diagram of a converter transformer model according to an embodiment of the present invention;
FIG. 2 is a diagram of a BP neural network architecture;
fig. 3 is a schematic flow chart of a particle swarm optimization BP neural network algorithm.
Detailed Description
It is to be understood that the following detailed description is exemplary and is intended to provide further explanation of the invention as claimed. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of exemplary embodiments according to the invention.
The embodiments and features of the embodiments of the present invention may be combined with each other without conflict.
Example one
The embodiment discloses a converter transformer scaling model vibration parameter prediction method based on a PSO-BP neural network. And changing input according to the simulation model to obtain a large amount of training data, taking the input of neural network data as the voltage, the size and the number of turns of the converter transformer, and outputting the voltage, the size and the number of turns of the converter transformer as the acceleration, the deformation and the displacement and the stress of the iron core. After algorithm training, a BP network combination model is established to predict the vibration output information of the converter transformer, and verification is carried out through actually measured data of the converter transformer. And finally, the BP network is optimized by using a particle swarm optimization algorithm, so that the prediction error is reduced.
The artificial intelligence algorithm can obtain a network combination model through training data, and input is given to predict output. The particle swarm optimization back propagation network (PSO-BP) is applied to the research of the converter transformer scaling model, the three-dimensional multi-field coupling physical model of the converter transformer is changed and input to obtain a plurality of groups of training data, the input is the voltage, the size and the number of turns of the converter transformer, and the output is the deformation displacement of an iron core and a winding and the stress of the iron core.
The power supply voltage, the size and the number of winding turns in the model established through algorithm training are used as input, the vibration information of the converter transformer is used as output, and the model can be regarded as a vibration scaling model of the converter transformer established based on a particle swarm optimization BP network. Any input parameter is given based on the scaling model, so that the vibration output information corresponding to the converter transformer can be obtained, and the method can be used for the research related to the vibration of the converter transformer.
In a specific implementation example, regarding a three-dimensional physical model of a multi-field coupling converter transformer: according to drawing and data of a certain type of transformer, three converter transformers which are operated in a combined mode are equivalent to a three-phase transformer, and a converter transformer finite element simulation model is established in finite element simulation software. The cylindrical iron core is a common iron core structure form in transformer simulation modeling. The laminated core can remarkably reduce eddy current loss, a 20-laminated equivalent core converter transformer vibration model is established according to the existing laminated transformer structure diagram, the circular space in the coil is fully utilized, and the stepped rectangular sizes at all levels are reasonably selected, so that the sectional area of the core column is gradually reduced as much as possible, and the top end of the core column is close to the winding. As shown in fig. 1, the ideal model of the converter transformer has 4 core legs connected by a core yoke, and the two core legs in the middle are respectively sleeved with a winding-continuous-winding primary side winding and a spiral secondary side winding.
In the modeling process, considering that the convergence speed of model simulation calculation needs to be improved, parts of converter transformer components are simplified. The components such as iron core clamps, bolts and the like which play a role in fixing are omitted, and the boundary conditions of 'fixing constraint' are used instead. The material definition is carried out on each part of the model, the iron core is made of soft iron material, the winding is made of copper material, and other areas in the box body are made of transformer oil material. The part material property parameters are shown in table 1 below.
TABLE 1 Material Properties
Figure BDA0003690963770000061
In a reasonable numerical range, the input voltage, the geometric dimension and the number of turns of the winding of the model are used as three groups of random variables, and the stress amplitude of the iron core and the winding of the model and the displacement amplitude of the iron core and the winding of 100 groups of the model under different combination conditions are recorded. And (3) carrying out data segmentation on 100 groups of data groups, taking input voltage, geometric dimension and winding turns as input data, and taking the stress amplitude of the model iron core and the model winding corresponding to the input data and the displacement amplitude of the iron core and the model winding as output data to be used as neural network training data of the next step. Table 2 shows 20 of the 100 sets of data.
Table 220 set of training data
Figure BDA0003690963770000062
Figure BDA0003690963770000071
The PSO-BP neural network construction method is suitable for training of the converter scaling model:
the BP neural network adopts a back propagation algorithm to adjust the weight, can realize any nonlinear mapping, and the essence is to calculate the minimum value of an error function so as to adjust the weight of the multilayer feedforward neural network. The structure of the BP neural network comprises an input layer, a hidden layer and an output layer. The signal flow is transmitted from the input layer to the hidden layer by layer and then is finally output by the output layer; if the expected result is not achieved, the error is reversely propagated from the output layer to the front layer by layer through the middle layers, and the connection weight of the network is continuously corrected; finally, the error of the network is smaller and smaller, and the optimal output result is obtained. The network structure is shown in fig. 2.
BP learning algorithm is based on LMS algorithm, assuming p i Is the output of the sample, t i To desired output, y i For the network output of each sample, the network performance index mean square error is calculated to be minimum, namely:
minE(e T e)＝minE[(t-y) T (t-y)] (1)
using the same LMS algorithm
Figure BDA0003690963770000072
To approximate the mean square error:
Figure BDA0003690963770000073
the sensitivity is defined as:
Figure BDA0003690963770000081
in the formula, W is the network weight, and m is the number of network layers. Sensitivity was propagated backwards using the chain rule:
Figure BDA0003690963770000082
Figure BDA0003690963770000083
the final weight value adjustment formula is:
W m (k+1)＝W m (k)-αs m (y m-1 ) T (6)
b m (k+1)＝b m (k)-αs m (7)
in the formula, α is a learning rate.
The BP network has good mapping capacity on the nonlinear relation, but has some limitations which are difficult to overcome, which is shown in that the BP neural network adopts a gradient descent algorithm, the main defects of the BP algorithm are that the global search capacity is poor, the main advantage is that the local search capacity is strong, and the main advantage of the particle swarm algorithm is that the global search capacity is strong.
The embodiment combines the advantages of the two algorithms to make up the defects of the two algorithms, thereby achieving the purpose of complementing the advantages of the two algorithms. The establishment of the particle swarm optimization BP neural network model is that the particle swarm algorithm is used for global search of an optimal solution, the optimal solution is used as an initial parameter of the neural network, and then the BP neural network is trained. The optimal threshold and weight of the BP neural network are found by using a particle swarm optimization algorithm, then the optimal threshold and weight are used as initial parameters of the network, and then network training is carried out, so that the classification capability and the speed of the model are improved.
Set in the D-dimensional search space, a population X ═ X (X) composed of n particles 1 ,X 2 ,...,X n ) Wherein the ith particle is a vector X in D dimension i ＝(x i1 ,x i2 ,...,x iD ) T Representing a potential solution to the solution problem, each particle position X can be calculated from the objective function i A corresponding fitness value. The training error is used to calculate the fitness value f (x), which is given by:
Figure BDA0003690963770000084
wherein p is the sample size, y p Is the actual output value, t p Is the output in the sample.
The velocity of the ith particle is V i ＝(V i1 ,V i2 ,...,V iD ) T Extreme value of individual is P i ＝(P i1 ,P i2 ,...,P iD ) T Global extreme value of P g ＝(P g1 ,P g2 ,...,P gD ) T . The formula is updated every iteration as follows:
Figure BDA0003690963770000091
Figure BDA0003690963770000092
in the formula, omega is an inertia weight; d1, 2, ·, D; 1,2, n; k is the current iteration number; v id Is the velocity of the particle; c. C 1 And c 2 Is an acceleration factor, which is a non-negative constant; r is 1 And r 2 Are random numbers distributed between 0 and 1.
And optimizing the initial weight of the BP neural network by the particle swarm formula. The particles represent initial weights, and the initial weights of the network are dynamically updated through a training effect.
The detailed steps of establishing the particle swarm optimization BP neural network model are as follows:
(1) randomly generating an initial population of particles and initializing P i And P g Pi is an individual extremum, and Pg is a global extremum;
(2) regarding each particle as a group of parameter values to form a parameter network;
(3) inputting training sample data into a parameter network for training;
(4) by adaptation to each particleCalculation of the value of P i And P g Determination of (1);
(5) if the termination condition of the program is not satisfied at this time, the operation goes to (6), and if the termination condition is satisfied, the operation goes to (7);
(6) updating each particle in the Weilian group according to the formula, and turning to (2);
(7) and taking the optimal particles as a group of optimal parameters, wherein the optimal particles are the optimal results, and ending the algorithm.
Fig. 3 shows a flow schematic of the particle swarm optimization BP neural network algorithm:
obtaining a prediction value of a scaling parameter: training samples pass through a PSO-BP neural network according to a gradient descent method, and a gradient search technology is utilized so as to minimize the mean square error of the actual output value and the expected output value of the network.
Residual errors: and (4) the difference between the PSO-BP network scaling parameter prediction value and the output parameter in the test sample.
And (3) a prediction result of the scaling parameter: namely, the test sample is input into the corresponding scaled model output, including the stress and displacement of the iron core and the winding.
Parameters required for the PSO algorithm: group size, inertial weight, acceleration constant, maximum algebra, solution space. Determined by performing a grid search on the desired parameters.
The process is mainly divided into three parts: and (4) building, training and predicting the neural network. Firstly, modeling a system, determining the number of network layers and the number of nodes of each layer, and constructing a proper neural network; then initializing the weight and threshold of the network, setting network training parameters, selecting a proper training algorithm and a proper transfer function, carrying out the next step if the training is finished and the target value is reached, and returning to retrain if the training is not reached to the expected value; and finally, obtaining a final prediction result by using the trained network. The network construction mainly comprises the following steps:
(1) the number of network layers is determined. Because the data are all one-dimensional numerical values, a three-layer BP network is adopted, namely only one hidden layer is included, and the operation amount can be greatly reduced while the accuracy is not inaccurate.
(2) Sample data is prepared and normalized. The data are normalized mainly to cancel the difference of the magnitude of each variable, so that the subsequent data can be uniformly processed and the training efficiency of the neural network can be improved. The training input data prepared in the method is the voltage, the size and the number of turns of the converter transformer, and the output data is the deformation displacement of the iron core and the winding and the stress of the iron core, and the total number of the training input data is 100. The test data are 20 groups of vibration data of the converter transformer under the condition of changing the input parameters in the simulation model. And respectively carrying out normalization processing on the training data and the test data, and normalizing the data to be between 0 and 1. All input and output parameters are normalized as follows:
Figure BDA0003690963770000101
in the formula, X gi Represents the normalized parameter value, x, of the ith sample i Parameter value, x, representing the ith sample min Represents the minimum value of the parameter, x max Represents the maximum value of the parameter.
(3) The number of hidden layer neurons is determined, which is a key step in the process of building a neural network. Too many or too few neurons are set, which reduces the fault tolerance and generalization capability of the network, and leads to the reduction of the network prediction capability. At present, the number of the neurons in the hidden layer is determined without a unified standard, and a formula is summarized according to the experience of predecessors:
Figure BDA0003690963770000111
in the formula, l is the number of hidden layer nodes, m is the number of input layer nodes, n is the number of output layer nodes, and a is a constant between 1 and 10. According to the formula, the number of hidden layer nodes is determined by adopting a trial and error method. After many tests, this example was taken to be 6.
(4) By setting the training parameters, the network performs training and learning to achieve the goal of setting error values. In the embodiment, the training function adopts train lm, the learning rate is 0.01, the expected error is 0.02, and the transfer functions of the hidden layer and the output layer are logsig and purelin respectively.
(5) The test samples were used for prediction. And predicting the size, voltage and turns of the converter transformer of the test sample, comparing the predicted size, voltage and turns with output characteristics, winding displacement deformation and stress in a simulation result, and checking the accuracy of the network.
Reliability analysis on the prediction of the PSO-BP neural network:
when the particle swarm optimization is used for optimizing the structure of the BP neural network, the weight and the threshold of the BP neural network are mainly optimized, and the optimal initial weight and the threshold of the BP neural network are searched by utilizing the particle swarm optimization. Firstly, parameter setting is carried out on a particle swarm optimization algorithm: the number of particles is set to 10, the length of the particles is 4 as the number of input variables, and the maximum velocity v of the particles max 0.5, learning factor c 1 ＝c 2 The minimum error allowed is 0.001 ═ 2. The inertial weight ω ranges from (0.3,0.9) and the number of iterations is set to 200. The iteration number tends to be stable after 80 times, and is set to be 100 times in order to reduce the interference of random errors and improve the iteration speed.
Training and testing a particle swarm optimization neural network: there were 120 sets of input and output data as samples to be studied. The training set and the test set were 100 and 20 groups, respectively, at a ratio of 5: 1. The remaining 20 sets of input and output data are used for evaluating and testing the prediction performance of the particle swarm optimization neural network model. After data training, a model can be obtained, and the vibration parameters of the converter transformer can be obtained by inputting the voltage, the size and the number of winding turns of a group of converter transformers at will.
The accuracy of the scaling model can be obtained through the prediction and comparison of test data, and the model error analysis and evaluation are necessary, and two commonly used error indexes are as follows:
mean Absolute Error (Mean Absolute Error, MAE)
Figure BDA0003690963770000121
Root Mean Square Error (Root Mean Square Error, RMSE)
Figure BDA0003690963770000122
And predicting the size, voltage and impedance of the converter transformer according to the test data, comparing the predicted size, voltage and impedance with the output acceleration, displacement deformation and stress in the simulation result, and checking the accuracy of the network. Two error indexes are calculated by the formulas (13) and (14), and data obtained by comparing the sizes of the indexes before and after particle swarm optimization are shown in the following table:
TABLE 3 optimization of two error indicators
Figure BDA0003690963770000123
The particle swarm optimization algorithm has certain optimization effect on the prediction result of the BP neural network, so that the prediction error becomes smaller, the precision is higher, and the vibration information of the converter transformer under different input quantities can be obtained more accurately.
Example two
The present embodiment is directed to a computer device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method when executing the computer program.
EXAMPLE III
An object of the present embodiment is to provide a computer-readable storage medium.
A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the steps of the above-mentioned method.
Example four
The purpose of this embodiment is to provide a converter transformer scaling model vibration parameter prediction system based on a PSO-BP neural network, including:
a scaling model construction module configured to: constructing a converter transformer scaling model, and giving any input parameter based on the scaling model to obtain vibration output information corresponding to the converter transformer as training data;
a neural network construction module configured to: constructing a PSO-BP neural network suitable for training a converter transformer scaling model, taking the input of neural network data as the voltage, the size and the number of turns of the converter transformer, and outputting the input data as the acceleration of an iron core, the deformation displacement of the iron core and the stress of the iron core, and finishing the training of the neural network based on the training data;
a prediction module configured to: and inputting the size, voltage and turns of the converter transformer to be predicted by using the trained neural network, and predicting the acceleration, deformation and displacement of the iron core and stress of the iron core.
The steps involved in the apparatuses of the above second, third and fourth embodiments correspond to the first embodiment of the method, and the detailed description thereof can be found in the relevant description of the first embodiment. The term "computer-readable storage medium" should be taken to include a single medium or multiple media containing one or more sets of instructions; it should also be understood to include any medium that is capable of storing, encoding or carrying a set of instructions for execution by a processor and that cause the processor to perform any of the methods of the present invention.
Those skilled in the art will appreciate that the modules or steps of the present invention described above can be implemented using general purpose computer means, or alternatively, they can be implemented using program code that is executable by computing means, such that they are stored in memory means for execution by the computing means, or they are separately fabricated into individual integrated circuit modules, or multiple modules or steps of them are fabricated into a single integrated circuit module. The present invention is not limited to any specific combination of hardware and software.
Although the embodiments of the present invention have been described with reference to the accompanying drawings, it is not intended to limit the scope of the present invention, and it should be understood by those skilled in the art that various modifications and variations can be made without inventive efforts by those skilled in the art based on the technical solution of the present invention.

Claims (10)

1. The converter transformer scaling model vibration parameter prediction method based on the PSO-BP neural network is characterized by comprising the following steps:
constructing a converter transformer scaling model, and giving any input parameter based on the scaling model to obtain vibration output information corresponding to the converter transformer as training data;
constructing a PSO-BP neural network suitable for training a converter transformer scaling model, taking the input of neural network data as the voltage, the size and the number of turns of the converter transformer, and outputting the input data as the acceleration of an iron core, the deformation displacement of the iron core and the stress of the iron core, and finishing the training of the neural network based on the training data;
and inputting the size, voltage and turns of the converter transformer to be predicted by using the trained neural network, and predicting the acceleration, deformation and displacement of the iron core and stress of the iron core.
2. The PSO-BP neural network-based converter transformer scaling model vibration parameter prediction method as claimed in claim 1, wherein when the converter transformer scaling model is constructed, a plurality of converter transformers operated in combination are equivalent to a three-phase transformer, and two converter transformer finite element simulation models in different iron core forms are established in finite element simulation software.
3. The PSO-BP neural network-based converter transformer scaling model vibration parameter prediction method as claimed in claim 1, characterized in that in the modeling process, the converter transformer part components are simplified: omitting the components for fixing and replacing by 'fixed constraint' boundary conditions; and material definition is carried out on each part of the model, the iron core is made of soft iron material, the winding is made of copper material, and other areas in the box body are made of transformer oil material.
4. The PSO-BP neural network-based converter transformer scaling model vibration parameter prediction method as claimed in claim 1, characterized in that the input voltage, the geometric dimension and the winding turns of the converter transformer scaling model are taken as three groups of random variables, and the stress amplitude of a plurality of groups of model iron cores and windings and the displacement amplitude of the iron cores and the windings under different combination conditions are recorded;
and carrying out data segmentation processing on the multiple groups of data groups, taking the input voltage, the geometric dimension and the number of turns of the winding as input data, and taking the stress amplitude of the model iron core and the corresponding winding and the displacement amplitude of the iron core and the corresponding winding as output data to be used as neural network training data.
5. The PSO-BP neural network-based converter transformer scaling model vibration parameter prediction method as claimed in claim 1, wherein the PSO-BP neural network model is established by using a particle swarm algorithm to search an optimal solution globally, using the optimal solution as an initial parameter of the neural network, and then training the BP neural network: the optimal threshold and weight of the BP neural network are searched by using a particle swarm optimization algorithm, and then the optimal threshold and the optimal weight are used as initial parameters of the network to perform network training.
6. The PSO-BP neural network-based converter transformer scaling model vibration parameter prediction method as claimed in claim 1, wherein the PSO-BP neural network is a particle swarm optimization BP neural network, and the detailed steps are established as follows:
(1) randomly generating an initial population of particles and initializing P i And P g ；
(2) Regarding each particle as a group of parameter values to form a parameter network;
(3) inputting training sample data in training data into a parameter network for training;
(4) p is performed by calculating a fitness value for each particle i And P g Determination of (1);
(5) if the termination condition of the program is not satisfied at this time, the operation goes to (6), and if the termination condition is satisfied, the operation goes to (7);
(6) updating each particle in the Weilian group according to the formula, and turning to (2);
(7) and taking the optimal particles as a group of optimal parameters, wherein the optimal particles are the optimal results, and ending the algorithm.
7. The PSO-BP neural network-based converter transformer scaling model vibration parameter prediction method as claimed in claim 1, wherein the PSO-BP neural network is constructed by the specific steps of:
determining the number of network layers;
training input data are voltage, size and turns of the converter transformer, output data are deformation displacement of the iron core and the winding and stress of the iron core, the training input data are divided into training data and testing data, and the training data and the testing data are respectively subjected to normalization processing;
determining a number of hidden layer neurons;
by setting training parameters, the network performs training and learning to achieve the aim of setting error values
And (3) testing a sample for prediction: and predicting the size, voltage and turns of the converter transformer of the test sample, comparing the predicted size, voltage and turns with output characteristics, winding displacement deformation and stress in a simulation result, and checking the accuracy of the network.
8. The converter transformer scaling model vibration parameter prediction system based on the PSO-BP neural network is characterized by comprising the following steps:
a scaling model construction module configured to: constructing a converter transformer scaling model, and giving any input parameter based on the scaling model to obtain vibration output information corresponding to the converter transformer as training data;
a neural network construction module configured to: constructing a PSO-BP neural network suitable for training a converter transformer scaling model, taking the input of neural network data as the voltage, the size and the number of turns of the converter transformer, and outputting the input data as the acceleration of an iron core, the deformation displacement of the iron core and the stress of the iron core, and finishing the training of the neural network based on the training data;
a prediction module configured to: and inputting the size, voltage and turns of the converter transformer to be predicted by using the trained neural network, and predicting the acceleration, deformation and displacement of the iron core and stress of the iron core.
9. A computer arrangement comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the steps of the method as claimed in any one of claims 1 to 7 are performed by the processor when executing the program.
10. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, is adapted to carry out the steps of the method according to any one of the preceding claims 1 to 7.
CN202210661243.4A 2022-06-13 2022-06-13 Converter transformer scaling model vibration parameter prediction method based on PSO-BP neural network Pending CN114896852A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210661243.4A CN114896852A (en) 2022-06-13 2022-06-13 Converter transformer scaling model vibration parameter prediction method based on PSO-BP neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210661243.4A CN114896852A (en) 2022-06-13 2022-06-13 Converter transformer scaling model vibration parameter prediction method based on PSO-BP neural network

Publications (1)

Publication Number Publication Date
CN114896852A true CN114896852A (en) 2022-08-12

Family

ID=82727271

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210661243.4A Pending CN114896852A (en) 2022-06-13 2022-06-13 Converter transformer scaling model vibration parameter prediction method based on PSO-BP neural network

Country Status (1)

Country Link
CN (1) CN114896852A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115222157A (en) * 2022-09-06 2022-10-21 江阴市晶磁电子有限公司 Intelligent prediction platform for iron core data of current transformer

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115222157A (en) * 2022-09-06 2022-10-21 江阴市晶磁电子有限公司 Intelligent prediction platform for iron core data of current transformer
CN115222157B (en) * 2022-09-06 2023-11-07 江阴市晶磁电子有限公司 Intelligent prediction device for iron core data of current transformer

Similar Documents

Publication Publication Date Title
US20210383235A1 (en) Neural networks with subdomain training
Gu et al. Identification and U‐control of a state‐space system with time‐delay
Wang et al. Multi-objective optimization of turbomachinery using improved NSGA-II and approximation model
CN114503121A (en) Resource constrained neural network architecture search
CN108694467A (en) A kind of method and system that Line Loss of Distribution Network System rate is predicted
CN111784061B (en) Training method, device and equipment for power grid engineering cost prediction model
CN107085633B (en) Device and method for multi-point vibration response frequency domain prediction based on support vector machine
Vieira et al. Bio‐inspired algorithms for the optimization of offshore oil production systems
CN104678988A (en) Engine ECU (electronic control unit) circuit fault diagnosis method
CN109508498A (en) Rubber shock absorber formula designing system and method based on BP artificial neural network
CN114896852A (en) Converter transformer scaling model vibration parameter prediction method based on PSO-BP neural network
US20210295167A1 (en) Generative networks for physics based simulations
Abdullah An Improvement in LQR Controller Design based on Modified Chaotic Particle Swarm Optimization and Model Order Reduction.
Mao et al. Intelligent modeling and control for nonlinear systems with rate-dependent hysteresis
KR20220032861A (en) Neural architecture search method and attaratus considering performance in hardware
CN115878963B (en) Method, system, terminal and medium for predicting metering error of capacitive voltage transformer
CN116861256A (en) Furnace temperature prediction method, system, equipment and medium for solid waste incineration process
Fanni et al. A neural inverse problem approach for optimal design
Xu et al. Circuit tolerance design using belief rule base
CN117043778A (en) Generating a learned representation of a digital circuit design
Xu et al. Electromagnetic buffer optimization based on Nash game
Wang et al. Neural network and PSO-based structural approximation analysis for blade of wind turbine
Khumprom et al. A hybrid evolutionary CNN-LSTM model for prognostics of C-MAPSS aircraft dataset
Liu et al. KP model for hysteresis of piezoelectric ceramic actuators
Cao et al. Traffic flow prediction model using an integrated framework of improved intelligent optimization algorithms and deep learning models

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