CN115034111A - GIS electrostatic field calculation method based on U-net convolution neural network - Google Patents

GIS electrostatic field calculation method based on U-net convolution neural network Download PDF

Info

Publication number
CN115034111A
CN115034111A CN202210639149.9A CN202210639149A CN115034111A CN 115034111 A CN115034111 A CN 115034111A CN 202210639149 A CN202210639149 A CN 202210639149A CN 115034111 A CN115034111 A CN 115034111A
Authority
CN
China
Prior art keywords
gis
gis equipment
neural network
electric field
dimensional simulation
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
CN202210639149.9A
Other languages
Chinese (zh)
Inventor
张宇娇
徐斌
黄雄峰
陈志伟
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.)
Hefei University of Technology
Original Assignee
Hefei University of Technology
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 Hefei University of Technology filed Critical Hefei University of Technology
Priority to CN202210639149.9A priority Critical patent/CN115034111A/en
Publication of CN115034111A publication Critical patent/CN115034111A/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
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/29—Geographical information databases
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/10—Geometric CAD
    • 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
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
    • G06T17/20—Finite element generation, e.g. wire-frame surface description, tesselation
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Geometry (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Computer Hardware Design (AREA)
  • Software Systems (AREA)
  • Databases & Information Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Mathematical Physics (AREA)
  • General Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Computational Linguistics (AREA)
  • Computing Systems (AREA)
  • Biophysics (AREA)
  • Remote Sensing (AREA)
  • Computer Graphics (AREA)
  • Computational Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

A GIS electrostatic field calculation method based on a U-net convolutional neural network comprises the steps of firstly constructing a GIS equipment three-dimensional simulation model with partial discharge defects based on the actual running state of GIS equipment in a transformer substation, calculating the electric field intensity and electric potential distribution of the GIS equipment by adopting a finite element method in a calculation area of the whole GIS equipment three-dimensional simulation model, then taking the geometric parameters and boundary conditions of the constructed GIS equipment three-dimensional simulation model as input data, taking the electric field intensity and electric potential distribution obtained through calculation as output data, putting the output data into the U-net convolutional neural network for training to obtain a deep learning model with prediction accuracy meeting requirements, and then inputting the actually acquired geometric parameters and boundary conditions of the GIS equipment to be tested into the deep learning model to predict the electric field intensity and electric potential distribution of the GIS equipment to be tested. The method realizes the rapid and accurate calculation of the partial discharge electrostatic field of the GIS equipment.

Description

GIS electrostatic field calculation method based on U-net convolution neural network
Technical Field
The invention belongs to the field of physical field simulation calculation, and particularly relates to a GIS electrostatic field calculation method based on a U-net convolutional neural network.
Background
GIS equipment becomes the important component part of electric power distribution network because of its characteristics such as have operating stability height, occupation of land space are little, dismantle simple, the maintenance is examined in fortune, but GIS equipment very easily produces foreign matter such as metal particle, metal burr in production, transportation, equipment in-process very easily inside, leads to the partial discharge phenomenon to appear in the operation process to influence the safe operation of GIS equipment, harm fortune dimension personnel's personal safety and electric power system's steady operation.
At present, a finite element method is mostly adopted for electrostatic field analysis of GIS equipment, but because the finite element method is long in time consumption in the solving process, results cannot be generated in time, and troubleshooting and state monitoring of equipment faults can be influenced under certain conditions.
Disclosure of Invention
The invention aims to solve the problems in the prior art and provides a GIS electrostatic field calculation method based on a U-net convolution neural network, which can realize quick calculation.
In order to achieve the above purpose, the technical scheme of the invention is as follows:
a GIS electrostatic field calculation method based on a U-net convolution neural network sequentially comprises the following steps:
step A, constructing a GIS equipment three-dimensional simulation model with partial discharge defects based on the actual running state of GIS equipment in a transformer substation;
b, calculating the electric field intensity and the electric potential distribution of the GIS equipment by adopting a finite element method in the calculation region of the whole GIS equipment three-dimensional simulation model;
step C, taking the geometric parameters and boundary conditions of the GIS equipment three-dimensional simulation model constructed in the step A as input data, taking the electric field intensity and electric potential distribution obtained by calculation in the step B as output data, and putting the output data into a U-net convolution neural network for training to obtain a deep learning model with the prediction accuracy rate meeting the requirement;
and D, inputting the actually acquired geometric parameters and boundary conditions of the GIS equipment to be tested into a deep learning model with the prediction accuracy meeting the requirement, and predicting to obtain the electric field intensity and the electric potential distribution of the GIS equipment to be tested.
The step C comprises the following steps in sequence:
step C1, dividing an x-y-z coordinate system of the constructed three-dimensional simulation model of the GIS equipment into uniform grid points, generating grid point data according to the geometric parameters and boundary conditions of the three-dimensional simulation model of the GIS equipment, and corresponding electric field intensity and electric potential distribution data of the three-dimensional simulation model of the GIS equipment and exporting the grid point data into a matrix array;
c2, constructing a U-net convolutional neural network architecture based on 3D convolution operation, and dividing the matrix array into a training data set and a test data set;
step C3, taking the geometric parameters and boundary condition data in the training data set as the input of the U-net convolutional neural network, taking the electric field intensity and electric potential distribution data in the training data set as the output of the U-net convolutional neural network, training the U-net convolutional neural network by using an Adam optimization algorithm until the loss function value is reduced to the minimum, and obtaining a trained deep learning model at the moment;
step C4, inputting the test data set into the trained deep learning model for testing to obtain the predicted values of the electric field intensity and the electric potential distribution, and then calculating the prediction accuracy of the deep learning model:
and C5, judging whether the prediction accuracy of the deep learning model meets the requirement, if not, adjusting the network parameters and returning to the step C3 for training again until the prediction accuracy of the deep learning model meets the requirement.
In step C3, the training uses a learning rate decay method represented by the following formula:
Figure BDA0003681741630000021
in the above formula, /) i For the learning rate of the i-th iteration cycle, l max 、l min Maximum and minimum learning rates, N, respectively E The total number of iteration cycles.
In step C3, the loss function value is calculated by using the following formula:
Figure BDA0003681741630000022
in the above formula, L (x, y) is the loss function value corresponding to the predicted value and the real value of a single batch, N BS Is a batch number, x j 、y j Respectively obtaining predicted values and actual values of the electric field intensity and the electric potential of the jth sample in a single batch, wherein delta is a hyper-parameter;
in step C4, the prediction accuracy of the deep learning model is calculated by the following formula:
Figure BDA0003681741630000023
in the above formula, P Acc For prediction accuracy, M is the total number of grid points divided, x m 、y m The predicted value and the actual value of the electric field intensity and the electric potential at the mth grid point are respectively.
The method also comprises an accuracy evaluation step of the GIS equipment three-dimensional simulation model, which is positioned between the step B and the step C;
the accuracy evaluation of the GIS equipment three-dimensional simulation model specifically comprises the following steps: a GIS equipment partial discharge experiment platform is built to carry out a partial discharge experiment, and a critical loading voltage value U when partial discharge is generated in the experiment process is measured 0 And then applying U in the three-dimensional simulation model of the GIS equipment 0 And obtaining the maximum electric field intensity value E of the GIS equipment three-dimensional simulation model max And SF under pMPa pressure 6 Gas critical breakdown electric fieldIntensity value E 0 And C, comparing, if the absolute value of the deviation between the two values does not exceed the set threshold value, judging that the GIS equipment three-dimensional simulation model has high accuracy, and entering the step C.
P and E 0 Satisfies the relationship shown in the following formula:
Figure BDA0003681741630000031
in step C, D, the geometric parameters include the total length of the GIS tank, the radius of the central guide rod and the tank, and the thickness of the basin-type insulator, and the boundary conditions include the voltage value of the central guide rod when the GIS equipment is loaded and actually operated, and the integral grounding of the metal shell outside the GIS tank.
In the step B, the electric field intensity and the electric potential distribution of the GIS equipment are calculated by adopting the following formulas:
Figure BDA0003681741630000032
Figure BDA0003681741630000033
Figure BDA0003681741630000034
where v is the Hamiltonian, ε is the relative dielectric constant,
Figure BDA0003681741630000035
is the phasor form of the electric field strength + 2 In order to be the laplacian operator,
Figure BDA0003681741630000036
is an electrical potential.
Compared with the prior art, the invention has the beneficial effects that:
1. the GIS electrostatic field calculation method based on the U-net convolutional neural network firstly constructs a GIS equipment three-dimensional simulation model with partial discharge defects based on the actual running state of GIS equipment in a transformer substation, calculates the electric field intensity and potential distribution of the GIS equipment by adopting a finite element method in a calculation area of the whole GIS equipment three-dimensional simulation model, then takes the geometric parameters and boundary conditions of the constructed GIS equipment three-dimensional simulation model as input data, calculates the obtained electric field intensity and potential distribution as output data, puts the output data into the U-net convolutional neural network for training to obtain a deep learning model with the prediction accuracy meeting the requirement, then inputs the actually acquired geometric parameters and boundary conditions of the GIS equipment to be measured into the deep learning model with the prediction accuracy meeting the requirement, and can predict the electric field intensity and potential distribution of the GIS equipment to be measured, the method has the advantages of high calculation speed and high prediction precision. Therefore, the method and the device realize the rapid and accurate calculation of the partial discharge electrostatic field of the GIS equipment.
2. The GIS electrostatic field calculation method based on the U-net convolutional neural network further comprises an accuracy evaluation step of a GIS equipment three-dimensional simulation model, a GIS equipment partial discharge experiment platform is built to carry out a partial discharge experiment, and an experiment result is compared with a simulation calculation result to verify the accuracy of the GIS equipment three-dimensional simulation model. Therefore, the method and the device are beneficial to improving the prediction accuracy of the deep learning model obtained by later training.
Drawings
FIG. 1 is a flow chart of the present invention.
Fig. 2 is a schematic diagram of grid point division in embodiment 1.
Fig. 3 is a schematic structural diagram of a U-net convolutional neural network based on 3D convolution operation in embodiment 1.
Detailed Description
The present invention will be described in further detail with reference to the following detailed description and accompanying drawings.
Referring to fig. 1, the method for calculating the GIS electrostatic field based on the U-net convolutional neural network sequentially comprises the following steps:
a, constructing a GIS equipment three-dimensional simulation model with partial discharge defects based on the actual running state of GIS equipment in a transformer substation;
b, calculating the electric field intensity and the electric potential distribution of the GIS equipment by adopting a finite element method in the calculation region of the whole GIS equipment three-dimensional simulation model;
step C, taking the geometric parameters and boundary conditions of the GIS equipment three-dimensional simulation model constructed in the step A as input data, taking the electric field intensity and electric potential distribution obtained by calculation in the step B as output data, and putting the output data into a U-net convolution neural network for training to obtain a deep learning model with the prediction accuracy rate meeting the requirement;
and D, inputting the actually acquired geometric parameters and boundary conditions of the GIS equipment to be tested into a deep learning model with the prediction accuracy meeting the requirement, and predicting to obtain the electric field intensity and the electric potential distribution of the GIS equipment to be tested.
The step C comprises the following steps in sequence:
step C1, dividing an x-y-z coordinate system of the constructed three-dimensional simulation model of the GIS equipment into uniform grid points, generating grid point data according to the geometric parameters and boundary conditions of the three-dimensional simulation model of the GIS equipment, and corresponding electric field intensity and electric potential distribution data of the three-dimensional simulation model of the GIS equipment and exporting the grid point data into a matrix array;
c2, constructing a U-net convolutional neural network architecture based on 3D convolution operation, and dividing the matrix array into a training data set and a test data set;
step C3, taking the geometric parameters and boundary condition data in the training data set as the input of the U-net convolutional neural network, taking the electric field intensity and electric potential distribution data in the training data set as the output of the U-net convolutional neural network, training the U-net convolutional neural network by using an Adam optimization algorithm until the loss function value is reduced to the minimum, and obtaining a trained deep learning model at the moment;
step C4, inputting the test data set into the trained deep learning model for testing to obtain the predicted values of the electric field intensity and the electric potential distribution, and then calculating the prediction accuracy of the deep learning model:
and C5, judging whether the prediction accuracy of the deep learning model meets the requirement, if not, adjusting the network parameters and returning to the step C3 for training again until the prediction accuracy of the deep learning model meets the requirement.
In step C3, the training uses a learning rate decay method as shown in the following formula:
Figure BDA0003681741630000051
in the above formula, /) i For the learning rate of the i-th iteration cycle, l max 、l min Maximum and minimum learning rates, N, respectively E Is the total number of iteration cycles.
In step C3, the loss function value is calculated by using the following formula:
Figure BDA0003681741630000052
in the above formula, L (x, y) is the loss function value corresponding to the predicted value and the true value of a single batch, and N is BS Is the number of batches, x j 、y j Respectively obtaining predicted values and actual values of the electric field intensity and the electric potential of the jth sample in a single batch, wherein delta is a hyper-parameter;
in step C4, the prediction accuracy of the deep learning model is calculated by the following formula:
Figure BDA0003681741630000053
in the above formula, P Acc For prediction accuracy, M is the total number of grid points divided, x m 、y m The predicted value and the actual value of the electric field intensity and the electric potential at the mth grid point are respectively.
The method also comprises an accuracy evaluation step of the GIS equipment three-dimensional simulation model, which is positioned between the step B and the step C;
the accuracy of the three-dimensional simulation model of the GIS equipmentThe sexual evaluation specifically comprises the following steps: a GIS equipment partial discharge experiment platform is built to carry out a partial discharge experiment, and a critical loading voltage value U when partial discharge is generated in the experiment process is measured 0 And then applying U in the three-dimensional simulation model of the GIS equipment 0 Obtaining the maximum electric field intensity value E of the GIS equipment three-dimensional simulation model max And SF under pMPa pressure 6 Critical breakdown electric field strength of gas E 0 And C, comparing, if the absolute value of the deviation between the two values does not exceed the set threshold value, judging that the GIS equipment three-dimensional simulation model has high accuracy, and entering the step C.
P and E 0 Satisfies the relationship shown in the following formula:
Figure BDA0003681741630000061
in step C, D, the geometric parameters include the total length of the GIS tank, the radius of the central guide rod and the tank, and the thickness of the basin-type insulator, and the boundary conditions include the voltage value of the central guide rod when the GIS device is loaded and actually operated, and the integral grounding of the metal shell outside the GIS tank.
In the step B, the electric field intensity and the electric potential distribution of the GIS equipment are calculated by adopting the following formulas:
Figure BDA0003681741630000062
Figure BDA0003681741630000063
Figure BDA0003681741630000064
wherein ^ is Hamiltonian, epsilon is a relative dielectric constant,
Figure BDA0003681741630000065
is the phasor form of the electric field strength + 2 In order to be a laplacian of operator,
Figure BDA0003681741630000066
is an electrical potential.
The principle of the invention is illustrated as follows:
the invention provides a GIS electrostatic field calculation method based on a U-net convolutional neural network, which comprises the steps of firstly carrying out simulation calculation through a finite element method, obtaining the electric field intensity and the electric potential distribution of GIS equipment under different defect conditions only by modifying the parameters of a GIS equipment partial discharge defect model, generating a large amount of data for training, and further saving unnecessary cost generated for obtaining the data; then, a GIS partial discharge experiment platform is set up to perform a partial discharge experiment, the experiment result is compared with a finite element method simulation calculation result under the same condition, the accuracy of the simulation model is verified, and then the deep learning model obtained by performing network training by using the data acquired by the simulation model can be judged to accurately calculate the electrostatic field of the actual GIS equipment; finally, model parameters and boundary conditions which are actually acquired are input into the trained deep learning model, so that the electrostatic field distribution of the GIS equipment can be quickly obtained, and the quick calculation of the electrostatic field of the GIS is realized.
Example 1:
referring to fig. 1, the calculation method of the GIS electrostatic field based on the U-net convolutional neural network is sequentially performed according to the following steps:
1. constructing a GIS equipment three-dimensional simulation model with point discharge defects based on the actual running state of GIS equipment in a transformer substation, wherein the GIS equipment three-dimensional simulation model comprises a GIS tank body and internal SF 6 The model comprises gas, a central guide rod, a basin-type insulator and point discharge defects, wherein the geometric parameters of the model comprise 133cm of the total length of a GIS tank, 2.25cm of the radius of the central guide rod, 17.7cm of the radius of the tank and 4cm of the thickness of the basin-type insulator, and the boundary conditions of the model comprise the voltage value of the central guide rod when the GIS equipment is loaded to actually run and the integral grounding of a metal shell outside the GIS tank;
2. the central guide rod loads the voltage value of the GIS equipment during actual operation, the metal shell outside the GIS tank is integrally grounded, and the electric field intensity and the electric potential distribution of the GIS equipment are calculated in a calculation region of a three-dimensional simulation model of the whole GIS equipment by adopting a finite element method (because the GIS equipment is in a stable operation state, free charges do not exist in the whole model region, an electrostatic field control equation and a Laplace equation can be solved by adopting the finite element method):
Figure BDA0003681741630000071
Figure BDA0003681741630000072
Figure BDA0003681741630000073
where ^ is the Hamiltonian, i.e., the differential operator of the vector, ε is the relative permittivity,
Figure BDA0003681741630000074
is the phasor form of the electric field strength + 2 For the laplacian, i.e. the second order differential operator,
Figure BDA0003681741630000075
is an electric potential;
3. the method comprises the steps of building a GIS equipment partial discharge experiment platform to perform a point discharge experiment according to a simulation environment for building a GIS equipment three-dimensional simulation model with a point discharge defect, wherein the GIS equipment partial discharge experiment platform adopts West lake electronic XD5936 type GIS partial discharge experiment equipment and comprises a partial discharge experiment tank body (comprising a tank body main body, a point discharge model and a model control rod), a measuring instrument (comprising a pulse current partial discharge detector and a built-in ultrahigh frequency sensor), a power supply control box and an isolation filter, wherein SF6 gas with the constant pressure of pMPa is pre-filled into the GIS tank body during the point discharge experiment, the power supply control box provides stable and controllable loading voltage, the isolation filter avoids high-frequency interference generated during switching of a power supply, and the model control rod controls the point discharge to perform point dischargeGradually increasing the loading voltage value by the distance d between the electric model and the central guide rod, observing the generation of partial discharge by using a measuring instrument, and recording the critical loading voltage value U when the partial discharge is generated 0 ；
4. After the point discharge experiment is finished, the same loading voltage U is applied to the three-dimensional simulation model of the GIS equipment 0 Obtaining the maximum electric field intensity value E of the GIS equipment three-dimensional simulation model max And SF under pMPa pressure 6 Critical breakdown electric field strength of gas E 0 In comparison, the results are shown in table 1:
TABLE 1E max And E 0 Comparative results
Figure BDA0003681741630000081
As can be seen from Table 1, E max And E 0 The absolute value of the deviation is not more than 2%, so that the GIS equipment three-dimensional simulation model is judged to have high accuracy and meet the actual condition;
wherein, the p and E 0 Satisfies the relationship shown in the following formula:
Figure BDA0003681741630000082
5. firstly, grid points with the size of 32 multiplied by 32 are divided from an x-y-z coordinate system of a constructed GIS equipment three-dimensional simulation model, then 3920 groups of data consisting of geometric parameters and boundary conditions of the GIS equipment three-dimensional model, corresponding electric field intensity and electric potential distribution respectively generate corresponding grid point data according to the divided grid points and are led out to be matrix arrays with the size of 4 multiplied by 32, as shown in figure 2, different areas of the model are numbered, the grid point data value in the area 0 is 0, the grid point data value in the area 1 is 1, and the like;
6. referring to fig. 3, a U-net convolutional neural network architecture based on 3D convolutional operation is constructed, and a matrix array is divided into a training data set and a test data set, where the network includes a decoder-encoder structure and a similar residual error structure, the decoder is responsible for model feature extraction, and the encoder combines the similar residual error structure to fuse shallow features and deep features, so as to obtain more features while ensuring that the output shape of the neural network is unchanged;
7. taking geometric parameters and boundary condition data in the training data set as the input of the U-net convolutional neural network, taking electric field intensity and electric potential distribution data in the training data set as the output of the U-net convolutional neural network, training the U-net convolutional neural network by using an Adam optimization algorithm based on the following formula until a loss function value is reduced to the minimum, and obtaining a trained deep learning model at the moment:
Figure BDA0003681741630000091
Figure BDA0003681741630000092
in the above formula, /) i For the learning rate of the i-th iteration cycle, l max 、l min Maximum and minimum learning rates, N, respectively E For the total number of iteration cycles, L (x, y) is the loss function value corresponding to the predicted and actual values of a single batch, N BS Is the number of batches, x j 、y j Respectively obtaining predicted values and actual values of the electric field intensity and the electric potential of the jth sample in a single batch, wherein delta is a hyper-parameter and is used as a threshold value for determining the switching of a loss function between an average error loss function and a mean square error loss function;
8. inputting the test data set into a trained deep learning model for testing to obtain predicted values of electric field intensity and electric potential distribution, and then calculating the prediction accuracy of the deep learning model:
Figure BDA0003681741630000093
in the above formula, P Acc For prediction accuracy, M is the total number of grid points divided, x m 、y m Respectively obtaining predicted values and real values of the electric field intensity and the electric potential at the mth grid point;
9. judging whether the prediction accuracy of the deep learning model meets the requirement, if not, adjusting the learning rate, the batch size, the total iteration cycles and the over-parameter delta, and returning to the step 7 for training again until the prediction accuracy of the deep learning model meets the requirement;
10. and inputting the actually acquired geometric parameters and boundary conditions of the GIS equipment to be tested into the trained deep learning model, and predicting to obtain the electric field intensity and electric potential distribution of the GIS equipment to be tested.
The U-net convolutional neural network used in example 1 of the present invention is compared with models obtained by training using other machine learning methods for prediction accuracy and calculation speed, and the results are shown in table 2:
TABLE 2 comparison of prediction accuracy and computation speed of different machine learning methods
Figure BDA0003681741630000094
Figure BDA0003681741630000101
As can be seen from table 2, the U-net convolutional neural network adopted in the present invention is superior to FCN8s, FCN16s, and FCN32s in both prediction accuracy and calculation speed.

Claims (8)

1. The GIS electrostatic field calculation method based on the U-net convolution neural network is characterized in that:
the method sequentially comprises the following steps:
a, constructing a GIS equipment three-dimensional simulation model with partial discharge defects based on the actual running state of GIS equipment in a transformer substation;
b, calculating the electric field intensity and the electric potential distribution of the GIS equipment by adopting a finite element method for the calculation region of the whole GIS equipment three-dimensional simulation model;
step C, taking the geometric parameters and boundary conditions of the GIS equipment three-dimensional simulation model constructed in the step A as input data, taking the electric field intensity and electric potential distribution obtained by calculation in the step B as output data, and putting the output data into a U-net convolution neural network for training to obtain a deep learning model with the prediction accuracy rate meeting the requirement;
and D, inputting the actually acquired geometric parameters and boundary conditions of the GIS equipment to be tested into a deep learning model with the prediction accuracy meeting the requirement, and predicting to obtain the electric field intensity and the electric potential distribution of the GIS equipment to be tested.
2. The GIS electrostatic field calculation method based on U-net convolutional neural network of claim 1, characterized in that:
the step C comprises the following steps in sequence:
step C1, dividing an x-y-z coordinate system of the constructed three-dimensional simulation model of the GIS equipment into uniform grid points, generating grid point data according to the geometric parameters and boundary conditions of the three-dimensional simulation model of the GIS equipment, and corresponding electric field intensity and electric potential distribution data of the three-dimensional simulation model of the GIS equipment and exporting the grid point data into a matrix array;
c2, constructing a U-net convolutional neural network architecture based on 3D convolution operation, and dividing the matrix array into a training data set and a test data set;
step C3, taking the geometric parameters and boundary condition data in the training data set as the input of the U-net convolutional neural network, taking the electric field intensity and electric potential distribution data in the training data set as the output of the U-net convolutional neural network, training the U-net convolutional neural network by using an Adam optimization algorithm until the loss function value is reduced to the minimum, and obtaining a trained deep learning model at the moment;
step C4, inputting the test data set into the trained deep learning model for testing to obtain the predicted values of the electric field intensity and the electric potential distribution, and then calculating the prediction accuracy of the deep learning model:
and C5, judging whether the prediction accuracy of the deep learning model meets the requirement, if not, adjusting the network parameters and returning to the step C3 for training again until the prediction accuracy of the deep learning model meets the requirement.
3. The GIS electrostatic field calculation method based on U-net convolutional neural network of claim 2, characterized in that:
in step C3, the training uses a learning rate decay method as shown in the following formula:
Figure FDA0003681741620000021
in the above formula, /) i For the learning rate of the i-th iteration cycle, l max 、l min Maximum and minimum learning rates, N, respectively E Is the total number of iteration cycles.
4. The GIS electrostatic field calculation method based on U-net convolutional neural network of claim 2, characterized in that:
in step C3, the loss function value is calculated by using the following formula:
Figure FDA0003681741620000022
in the above formula, L (x, y) is the loss function value corresponding to the predicted value and the true value of a single batch, and N is BS Is the number of batches, x j 、y j Respectively obtaining predicted values and actual values of the electric field intensity and the electric potential of the jth sample in a single batch, wherein delta is a hyper-parameter;
in step C4, the prediction accuracy of the deep learning model is calculated by the following formula:
Figure FDA0003681741620000023
in the above formula, P Acc For prediction accuracy, M is the total number of grid points divided, x m 、y m Electric field strength and potential at m-th grid pointPredicted values and true values.
5. The GIS electrostatic field calculation method based on U-net convolutional neural network of any one of claims 1-4, characterized in that:
the method also comprises an accuracy evaluation step of the GIS equipment three-dimensional simulation model, which is positioned between the step B and the step C;
the accuracy evaluation of the GIS equipment three-dimensional simulation model specifically comprises the following steps: a GIS equipment partial discharge experiment platform is built to carry out a partial discharge experiment, and a critical loading voltage value U when partial discharge is generated in the experiment process is measured 0 And then applying U in the three-dimensional simulation model of the GIS equipment 0 And obtaining the maximum electric field intensity value E of the GIS equipment three-dimensional simulation model max And SF under pMPa pressure 6 Critical breakdown electric field strength of gas E 0 And C, comparing, if the absolute value of the deviation between the two values does not exceed the set threshold value, judging that the GIS equipment three-dimensional simulation model has high accuracy, and entering the step C.
6. The GIS electrostatic field calculation method based on U-net convolutional neural network of claim 5, characterized in that:
p and E 0 Satisfies the relationship shown in the following formula:
Figure FDA0003681741620000031
7. the GIS electrostatic field calculation method based on the U-net convolutional neural network is characterized in that according to any one of claims 1-4: in step C, D, the geometric parameters include the total length of the GIS tank, the radius of the central guide rod and the tank, and the thickness of the basin-type insulator, and the boundary conditions include the voltage value of the central guide rod when the GIS device is loaded and actually operated, and the integral grounding of the metal shell outside the GIS tank.
8. The GIS electrostatic field calculation method based on U-net convolutional neural network of any one of claims 1-4, characterized in that:
in the step B, the electric field intensity and the electric potential distribution of the GIS equipment are calculated by adopting the following formulas:
Figure FDA0003681741620000032
Figure FDA0003681741620000033
Figure FDA0003681741620000034
in the above formula, the first and second carbon atoms are,
Figure FDA0003681741620000035
is a Hamiltonian, epsilon is a relative dielectric constant,
Figure FDA0003681741620000036
in the form of a phasor of the electric field strength,
Figure FDA0003681741620000037
in order to be the laplacian operator,
Figure FDA0003681741620000038
is an electrical potential.
CN202210639149.9A 2022-06-07 2022-06-07 GIS electrostatic field calculation method based on U-net convolution neural network Pending CN115034111A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210639149.9A CN115034111A (en) 2022-06-07 2022-06-07 GIS electrostatic field calculation method based on U-net convolution neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210639149.9A CN115034111A (en) 2022-06-07 2022-06-07 GIS electrostatic field calculation method based on U-net convolution neural network

Publications (1)

Publication Number Publication Date
CN115034111A true CN115034111A (en) 2022-09-09

Family

ID=83122993

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210639149.9A Pending CN115034111A (en) 2022-06-07 2022-06-07 GIS electrostatic field calculation method based on U-net convolution neural network

Country Status (1)

Country Link
CN (1) CN115034111A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117709170A (en) * 2024-02-05 2024-03-15 合肥工业大学 Magnetic field rapid calculation method based on improved depth operator network

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117709170A (en) * 2024-02-05 2024-03-15 合肥工业大学 Magnetic field rapid calculation method based on improved depth operator network
CN117709170B (en) * 2024-02-05 2024-04-19 合肥工业大学 Magnetic field rapid calculation method based on improved depth operator network

Similar Documents

Publication Publication Date Title
CN105699843A (en) Electric cable running state diagnosis method and system
CN112200694B (en) Dominant instability mode identification model construction and application method based on graph neural network
CN115034111A (en) GIS electrostatic field calculation method based on U-net convolution neural network
CN112540316B (en) Complicated battery impedance spectrum analysis method
CN105975713A (en) Method for predicting electromagnetic radiation of electronic circuit through source reconstruction
CN103440497A (en) GIS insulation defect partial discharge atlas pattern recognition method
CN112307677A (en) Power grid oscillation mode evaluation and safety active early warning method based on deep learning
Zhou et al. Identification of XLPE cable insulation defects based on deep learning
CN111693825B (en) Method and system for determining rain intensity of corona cage rain device under breeze condition
CN110738009B (en) Setting method of analog charges in conducting wire in electric field calculation of power transmission line
Wang et al. Transmission line fault diagnosis based on wavelet packet analysis and convolutional neural network
Akram et al. Prediction of partial discharge inception voltage for electric vehicle motor insulation using deep learning
CN114254481B (en) Bubble defect analysis method and equipment for transformer bushing
CN105912771A (en) Ansys-based analysis method for optimization of influence of insulator detecting and cleaning robot on insulator string distribution voltage
CN110737998A (en) equalizing ring optimization design method based on finite element and depth belief network
CN112147465B (en) GIS partial discharge optical diagnosis method based on multi-fractal and extreme learning machine
CN114744631A (en) Data driving voltage estimation method based on non-PMU power distribution network
Sun et al. A Novel Fault Diagnosis of GIS Partial Discharge Based on Improved Whale Optimization Algorithm
CN114355125A (en) Transmission cable defect judgment device and method based on electric field analysis
CN114167133A (en) Harmonic voltage evaluation method and device for newly-built station in power transmission network
CN108090311A (en) Air puncture voltage Forecasting Methodology based on route characteristic collection and support vector regression
Zhang et al. Three-dimensional reconstruction of the electric tree in oil-impregnated pressboard based on partial discharge location
Kim et al. Partial discharge pattern recognition using fuzzy-neural networks (FNNs) algorithm
Kang et al. Numerical Simulation of Vq Characteristics of Impulsive Corona for Thin Conductor in Coaxial Corona Cage
CN116125123A (en) Non-invasive voltage measurement method based on near-end integral inversion

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