CN114065594B - Single-post insulator electrical performance optimization method for GIS based on neural network model - Google Patents
Single-post insulator electrical performance optimization method for GIS based on neural network model Download PDFInfo
- Publication number
- CN114065594B CN114065594B CN202111442270.4A CN202111442270A CN114065594B CN 114065594 B CN114065594 B CN 114065594B CN 202111442270 A CN202111442270 A CN 202111442270A CN 114065594 B CN114065594 B CN 114065594B
- Authority
- CN
- China
- Prior art keywords
- neural network
- insulator
- post insulator
- post
- layer
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 239000012212 insulator Substances 0.000 title claims abstract description 161
- 238000000034 method Methods 0.000 title claims abstract description 40
- 238000003062 neural network model Methods 0.000 title claims abstract description 39
- 238000005457 optimization Methods 0.000 title claims abstract description 34
- 239000002245 particle Substances 0.000 claims abstract description 67
- 230000005684 electric field Effects 0.000 claims abstract description 34
- 238000013528 artificial neural network Methods 0.000 claims abstract description 26
- 238000011156 evaluation Methods 0.000 claims abstract description 26
- 238000012549 training Methods 0.000 claims abstract description 26
- 238000004458 analytical method Methods 0.000 claims abstract description 22
- 230000002068 genetic effect Effects 0.000 claims abstract description 12
- 230000006870 function Effects 0.000 claims description 44
- 239000002184 metal Substances 0.000 claims description 17
- 238000009396 hybridization Methods 0.000 claims description 15
- 230000035772 mutation Effects 0.000 claims description 9
- 230000008859 change Effects 0.000 claims description 7
- 238000004364 calculation method Methods 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 6
- 230000009191 jumping Effects 0.000 claims description 5
- 230000008569 process Effects 0.000 claims description 5
- 238000012545 processing Methods 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 230000015572 biosynthetic process Effects 0.000 claims description 3
- 238000003786 synthesis reaction Methods 0.000 claims description 3
- 239000002131 composite material Substances 0.000 claims description 2
- 230000003247 decreasing effect Effects 0.000 claims description 2
- 239000000758 substrate Substances 0.000 claims 1
- 238000013461 design Methods 0.000 abstract description 8
- 230000008901 benefit Effects 0.000 description 4
- 238000009826 distribution Methods 0.000 description 4
- 230000005540 biological transmission Effects 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000009413 insulation Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000007789 sealing Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/23—Design optimisation, verification or simulation using finite element methods [FEM] or finite difference methods [FDM]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/30—Circuit design
- G06F30/39—Circuit design at the physical level
- G06F30/398—Design verification or optimisation, e.g. using design rule check [DRC], layout versus schematics [LVS] or finite element methods [FEM]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2111/00—Details relating to CAD techniques
- G06F2111/04—Constraint-based CAD
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2111/00—Details relating to CAD techniques
- G06F2111/06—Multi-objective optimisation, e.g. Pareto optimisation using simulated annealing [SA], ant colony algorithms or genetic algorithms [GA]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2113/00—Details relating to the application field
- G06F2113/04—Power grid distribution networks
Abstract
A method for optimizing electrical performance of a single-post insulator for GIS based on a neural network model comprises the following steps: establishing an electric field finite element analysis model of the single-pillar insulator in three-dimensional CAD software; determining structural parameters to be optimized of an electric field finite element analysis model of the single-pillar insulator; selecting an electrical performance evaluation index of the single-pillar insulator; obtaining the influence rule of structural parameters to be optimized of the single-post insulator on the electrical performance evaluation index in a parameterized scanning mode, and obtaining a training set of the deep neural network; establishing a four-layer deep neural network, and training the model by using a training set to obtain an equivalent model of the three-dimensional finite element model; setting an objective function value function; performing global optimization on the equivalent model by using a genetic particle swarm algorithm; the invention greatly reduces the time required for calculating the electrical performance index of the single-post insulator at a time, obviously reduces the design period and the cost of the single-post insulator, and provides a new thought for optimizing the structure of the ultra-high voltage GIS single-post insulator.
Description
Technical Field
The invention belongs to the technical field of insulating structures of power equipment, and particularly relates to a method for optimizing electrical performance of a single-post insulator for GIS based on a neural network model.
Background
The single-post insulator is an important component in GIS equipment, and the structural rationality and electric field distribution uniformity of the single-post insulator play an important role in the insulating performance of the single-post insulator. In actual production, severe operating conditions cause insulation failure of the GIS equipment. When the structure design of the single-post insulator is carried out, if only a certain area with high field intensity is optimized, the field intensity of other areas is easily distorted.
The safe and reliable operation of GIS equipment is an important factor for ensuring stable power transmission. The GIS equipment has the advantages of complete sealing of elements, no interference from external environment, high operation reliability, strong arc extinguishing capability, long maintenance period, high operation reliability and the like. With the continuous development of the power system in China, the transmission capacity is continuously increased, and the more severe operation conditions cause insulation faults of GIS equipment, so that the traditional local optimization method has great limitation.
The structural optimization design of the single-post insulator relates to a plurality of variables such as the shape of an umbrella skirt of the insulator, the size of the insulator, the surface shape of an insert and the like, and the performance requirements comprise the surface field intensity of the insert, the surface synthesized field intensity of the insulator and the tangential field intensity. This strong coupling system of multiple inputs and multiple outputs belongs to the problem of multi-variable multiple constraint optimal values. The single-post insulator has a complex structure, the field intensity influence of different positions on key positions is different, the traditional optimization method has the defects of long single operation time and difficulty in obtaining a global optimal solution, and the parameter design of the traditional method is difficult to meet the global optimal requirement.
Disclosure of Invention
In order to overcome the defects in the prior art, the invention aims to provide the single-post insulator electrical performance optimization method for the GIS based on the neural network model, which can meet the requirements of the whole electrical performance and the local electrical performance of the single-post insulator for the GIS through the structural optimization of the single-post insulator and optimize and improve the electrical performance of the single-post insulator for the GIS.
In order to achieve the above purpose, the invention is realized by adopting the following technical scheme:
the method for optimizing the electrical performance of the single-post insulator for the GIS based on the neural network model comprises the following steps:
step 1, establishing an electric field finite element analysis model of a single-post insulator by using three-dimensional CAD software, determining structural parameters to be optimized of the electric field finite element analysis model of the single-post insulator according to the characteristics of the single-post insulator, and selecting the electric field intensity of a region with the electric field intensity exceeding a maximum allowable value as an electric performance evaluation index of the single-post insulator;
and 2, in the electric field finite element analysis model of the single-post insulator established in the step 1, a control variable method is used, so that structural parameters to be optimized are changed within a certain range, a change result of an electrical performance index when the parameters are changed is calculated, a deep neural network training set is obtained, input variables of the training set are parameters to be optimized of the single-post insulator, and output variables are electrical performance evaluation indexes of the single-post insulator.
Step 3, establishing a four-layer deep BP neural network model comprising an input layer, two hidden layers and an output layer, carrying out model training on the model by the deep neural network training set obtained in the step 2 to obtain a deep neural network model for evaluating the electric performance of the single-post insulator, taking the model as an equivalent model of a single-post insulator electric field finite element analysis model, solving a weight of the deep BP neural network model, and taking the weight as a characteristic parameter of the deep BP neural network model;
step 4, carrying out normalization processing on the electric performance evaluation index of the single-post insulator, calculating a weighted average value of the electric performance evaluation index of the single-post insulator, and optimizing the weighted average value as an objective function value;
and 5, taking an equivalent model of the electric field finite element analysis model of the single-post insulator in the step 3 as an optimization object, performing global optimization on the structural parameters to be optimized of the electric field finite element analysis model of the single-post insulator determined in the step 1 by using a genetic particle swarm algorithm, and obtaining the optimal structural parameters after iterative optimization.
The structural parameters to be optimized of the electric field finite element analysis model of the single-post insulator comprise: radius of metal insert R 1 Metal insert fillet radius R 2 Radius of insulator R 3 Metal insert depth L 1 Depth L of insulator umbrella skirt 2 Outer radius r of insulator umbrella skirt 1 Insulator umbrella skirt chamfer radius r 2 Radius r of bottom fillet of umbrella skirt of insulator 3 Distance H between insulator umbrella skirt and top end 1 Tangential distance H of insulator umbrella skirt 2 。
The variation range of the optimized structural parameters in the step 2 is as follows:
wherein R is 1 ，R 2 ，R 3 ，L 2 ，r 1 ，r 2 ，r 3 ，H 1 ，H 2 To be optimized for the parameters, R 1min ，R 1max ，R 2min ，R 2max ，R 3min ，R 3max ，r 1min ，r 1max ，r 2min ，r 2max ，r 3min ，r 3max ，H 1min ，H 1max ，H 2min ，H 2max ，L 1min ，L 1max ，L 2min ，L 2max Is the variation range of the parameters to be optimized. In the dynamic change of the above variables, there is still a need to satisfy the dynamic adjustment constraint of the following formula (2), where L 0 Is the total height of the single-post insulator;
the electrical performance evaluation indexes of the single-post insulators in the step 1 and the step 2 are selected by taking the value range of the structural parameters to be optimized as constraint conditions, wherein the constraint conditions comprise the maximum value E 'of the surface synthesized field intensity of the metal insert' 1 Surface synthesized field intensity maximum E 'of single-pillar insulator' 2 Maximum value E 'of tangential field intensity of single-pillar insulator' 3 The method comprises the steps of carrying out a first treatment on the surface of the The electrical performance evaluation of the unipost insulator without optimization is defined as the composite field intensity E of the surface of the metal insert 1 Maximum value E of surface synthesis field intensity 2 And maximum value of tangential field strength E 3 。
The specific method of the step 3 is as follows:
step 3.1, determining the number of nodes of an input layer, the number of nodes of an output layer and the number of nodes of an hidden layer, wherein the number of nodes of the input layer is a single-post insulatorThe number of the structural parameters to be optimized, the number of the nodes of the output layer is the number of the electrical performance evaluation indexes of the single-post insulator, the number of the nodes of the hidden layer is selected according to experience, the number of the nodes of the first layer is between the number of the input layer and the number of the output layer, and the number of the nodes of the second layer is equal to the sum of the numbers of the nodes of the input layer and the output layer; initializing an initial weight omega, a learning coefficient eta, a momentum coefficient alpha, a target error epsilon and a learning frequency epoch max Wherein the weight ω includes the weight ω from the input layer to the first hidden layer 0i Weight ω of first hidden layer to second hidden layer ij Second hidden layer to output layer weight omega jk ；
Step 3.2, the output variable of the input layer is the input variable of the deep BP neural network model, and the activation function of the BP neural network model is set as formula (3), wherein e is a natural constant
The hidden layer and the output layer are calculated using a BP neural network algorithm, in the following equation (4),for the nth iteration, the first hidden layer output variable, the second hidden layer output variable, and the output layer output variable,
calculating error of output variable of BP neural network model to obtain training error, wherein y is output variable of neural network training set in nth iteration
Step 3.3, adjusting the weight, in the nth iteration,for inputting the weight adjustment value of the layer to the first hidden layer,/>Adjusting the value for the weights of the first hidden layer to the second hidden layer,/for the first hidden layer>The value is adjusted for the second hidden layer to output layer weights. Finding the error E when the number of iterations is n (n) For->Taking the partial derivative of (2) as a new weight adjustment value, and calculating the weight in n+1 iterations, wherein the weight is represented by a formula (6)
Recording error E (n) For a pair ofAs the weight adjustment value of n+1 iterations, the partial derivative of (2) is represented by the formula (7),
step 3.4, if the current training error value is smaller than the target error, finishing learning, continuing to carry out step 3.5, otherwise, jumping to step 3.2, and continuing to iterate;
and 3.5, saving the learned weight as a characteristic parameter of the deep BP neural network model for the subsequent optimization process.
The normalization processing of the electrical performance index in the step 4 is determined by a formula (8),
wherein the coefficient alpha 1 、α 2 And alpha 3 Weights corresponding to the maximum field strengths of the three positions respectively, f s Is an objective function value function under the current structural parameters.
The specific method of the step 5 is as follows:
step 5.1, according to the principle of the genetic particle swarm algorithm, assigning initial values to the following parameters: population number n and spatial dimension n dim Size limit X of each dimension parameter limit Speed limit V for each dimension parameter limit Inertial weight c 1 Self-learning factor c 2 Group learning factor c 3 Probability of hybridization eta 1 Probability of variation eta 2 Maximum number of iterations n ger And an objective function value tolerance epsilon;
step 5.2, randomly generating particles with different parameters, calculating an objective function value corresponding to each particle by using a trained depth BP neural network model, and recording a population history optimal position and a population history optimal fitness for an initial population;
step 5.3, calculating inertial weight according to the maximum iteration times and the current iteration times in step 5.1, wherein the calculating method is a weight linear decreasing strategy, and the formula (9) is shown
Updating the speeds and positions of all particles in the particle swarm, wherein the iterative formula of the speeds and positions of the particle swarm algorithm is as follows:
rand 1 and rand 2 Is in the value range of [0,1 ]]Random number of (1), pbest i As the position of the individual optimal value of each particle in the current iteration number, gbest i The position of the global optimal value of each particle history is located; benefit (benefit)Updating the position of the particles by using a speed updating formula;
step 5.4, carrying out hybridization and mutation operation on the particles subjected to speed and position updating in step 5.3, wherein hybridization refers to randomly exchanging parameters of the same kind in any two particles, and mutation refers to randomly reassigning a certain parameter in a certain particle;
step 5.5, aiming at the particle swarm subjected to hybridization and mutation in step 5.4, calculating an objective function value corresponding to each particle by using a trained depth BP neural network model, solving the objective function value of each particle, and recording the optimal position of the population history and the optimal fitness of the population history;
and 5.6, if the optimal fitness of the population history is smaller than the tolerance of the objective function value or the current iteration number is larger than or equal to the maximum iteration number, stopping calculation, and if the optimal fitness is not reached, jumping to the step 5.3, and adding one to the iteration number.
The characteristics of the single-post insulator in the step 1 comprise the supporting structural characteristics of the single-post insulator and the shape structural characteristics of the metal insert of the single-post insulator.
The hybridization in the step 5.4 is to randomly exchange the same kind of parameters in any two particles, and the mutation is to randomly reassign a certain parameter in a certain particle.
The optimal position of the population history in the step 5.2 and the step 5.5 represents the structural parameter of the single-post insulator, and the optimal fitness of the population history represents the objective function value function of the single-post insulator.
Compared with the prior art, the invention has the following beneficial effects:
according to the method for optimizing the electrical performance of the single-post insulator for the GIS based on the neural network model, firstly, an electric field finite element analysis model of the single-post insulator is established, structural parameters to be optimized of the electric field finite element analysis model of the single-post insulator are determined through analyzing electric field distribution of an initial model, and meanwhile, evaluation indexes of the electrical performance of the single-post insulator are selected. And carrying out parameterized scanning on the selected structural parameters, taking the original structure as a reference, carrying out parameterized scanning in a certain range taking the original model parameters as the reference on the premise of conforming to the structural rationality, taking the structural parameters in the parameterized scanning process as input variables, taking the electrical performance evaluation index as output variables to form a deep neural network training set, and determining the variation range and constraint conditions of the parameters. And (3) establishing a four-layer deep neural network, training the model by using the training set to obtain a deep neural network model for evaluating the electrical performance of the single-post insulator, and taking the model as an equivalent model of a single-post insulator electric field finite element analysis model. Defining the electrical performance evaluation index of the single-post insulator as an objective function value function, carrying out normalization processing on parameters, setting an optimization priority function, and configuring the weight of the electrical performance evaluation index of the single-post insulator to obtain the objective function value function capable of describing the electrical performance optimization degree. And finally, using an equivalent model of the single-post insulator as an optimization object, optimizing an electric field finite element equivalent model of the single-post insulator by using a genetic particle swarm algorithm, and solving a structure optimization problem with a structure parameter as an input variable and an objective function value function as an fitness function.
In summary, the invention starts from the whole structure of the single-post insulator, selects the structural parameters of the insulator, considers the electrical performance indexes of key parts of multiple parts of the single-post insulator, obtains a deep neural network training set by using a parameterized scanning mode, establishes a four-layer neural network to fit a complex single-post insulator electric field model, obtains an equivalent model of the single-post insulator electric field finite element analysis model, optimizes the parameters by using a genetic particle swarm algorithm for the model, and greatly reduces the design period of the single-post insulator with complex structure.
Further, when the structural parameters of the insulator are selected, the shapes of all areas of the single-pillar insulator are parameterized according to the structural characteristics of the single-pillar insulator, and the influence degree of the structural parameters on the field intensity of the key areas is evaluated on the premise that the structural rationality condition is met, so that the change range of the structural parameters and the constraint conditions are determined.
Furthermore, the four-layer deep neural network is selected as an equivalent model building method, and compared with the conventional three-layer neural network, the four-layer deep neural network has stronger generalization capability and better fitting capability for special and harder learning samples. The method has the advantages that a limited number of evaluation indexes capable of describing the electrical performance of the single-post insulator are selected, the indexes of a plurality of areas are weighted and normalized, the optimization priority sequence of each area can be considered in a balanced mode, the number of structural parameters to be optimized can be reduced as much as possible on the premise that the optimization effect is met, and the overall electrical performance of the single-post insulator is comprehensively improved.
Furthermore, the invention comprehensively considers the electrical performance evaluation indexes at different positions, and reasonably distributes the weight of the electrical performance indexes, so that the final optimization result can integrally meet the electrical performance requirements in engineering, and the design period of designers in the design of the single-post insulator is saved.
Furthermore, the invention combines the genetic algorithm and the particle swarm algorithm, improves the optimizing capability of the particle swarm algorithm by using the idea of the genetic algorithm, and the calculation times of the genetic particle swarm algorithm are larger than those of the conventional particle swarm algorithm from the operation point of view, but because the optimization object of the invention is an equivalent model of a deep neural network, the single calculation time is far shorter than the one-time electric field finite element calculation time, so the whole optimization time is greatly shortened.
Drawings
FIG. 1 is a schematic flow chart of the present invention.
FIG. 2 is a schematic diagram of a training process of the deep neural network model according to the present invention.
FIG. 3 is a schematic flow chart of the genetic particle swarm algorithm of the present invention.
Fig. 4 is a schematic diagram of the method of the present invention for selecting optimization parameters for a single post insulator.
Detailed Description
The invention is described in further detail below with reference to the drawings and examples.
Examples
As shown in fig. 1 to 4, the method for optimizing the electrical performance of the single-post insulator for the GIS based on the neural network model comprises the following steps:
step 1, using three-dimensional CAD software to establish a geometric model of a single-post insulator for GIS, using the geometric model of the single-post insulator as an electric field finite element analysis model, determining key structural parameters of the single-post insulator finite element analysis model affecting the electric field of the single-post insulator, wherein the structural parameters comprise structural parameters describing the shape of an insert of the single-post insulator and structural parameters describing the shape of the post of the single-post insulator, wherein the shape parameters describing the insert of the single-post insulator are the embedding depth, diameter and surface fillet radius of a metal insert, the structural parameters describing the shape of the post of the single-post insulator are the radius of the insulator, the depth of an insulator umbrella skirt, the outer radius of the insulator umbrella skirt, the chamfering radius of the insulator umbrella skirt, the bottom fillet radius of the insulator umbrella skirt, the distance between the insulator umbrella skirt and the top end, and the tangential distance of the insulator umbrella skirt, and taking all the structural parameters as structural parameters to be optimized;
selecting the maximum value E 'of the synthesized field intensity of the surface of the metal insert' 1 Surface synthesized field intensity maximum E 'of single-pillar insulator' 2 Maximum value E 'of tangential field intensity of single-pillar insulator' 3 As electrical performance evaluation fingers; r is R 1 Radius of metal insert, R 2 R 2 R is the radius of a metal insert fillet 3 Is the radius of the insulator, L 1 For metal insert depth, L 2 Is the depth of the umbrella skirt of the insulator, r 1 Is the outer radius of the umbrella skirt of the insulator, r 2 The radius r of the chamfer of the umbrella skirt of the insulator 3 Is the radius of the rounded corner at the bottom of the umbrella skirt of the insulator, H 1 H is the distance between the umbrella skirt and the top end of the insulator 2 Is the tangential distance of the umbrella skirt of the insulator;
step 2, determining the variation range of parameters to be optimized, and defining constraint conditions of dynamic variation of the parameters, wherein the variation range of the single-post insulator is as follows:
wherein R is 1 ，R 2 ，R 3 ，L 2 ，r 1 ，r 2 ，r 3 ，H 1 ，H 2 To be optimized for the parameters, R 1min ，R 1max ，R 2min ，R 2max ，R 3min ，R 3max ，r 1min ，r 1max ，r 2min ，r 2max ，r 3min ，r 3max ，H 1min ，H 1max ，H 2min ，H 2max ，L 1min ，L 1max ，L 2min ，L 2max The variation range of the parameters to be optimized; in the dynamic variation of the variables described above, there is still a need to satisfy the dynamic adjustment constraint, where L 0 Is the total height of the single post insulator:
and performing parameterization scanning on the parameters by using a control variable method to obtain a deep neural network training set consisting of the single-post insulator structural parameters and the electrical performance evaluation indexes.
Step 3, building a four-layer deep neural network to train the equivalent model in step 2, wherein the steps are as follows:
and 3.1, determining the number of nodes of an input layer, the number of nodes of an output layer and the number of nodes of an hidden layer, wherein the number of nodes of the input layer is the number of structural parameters to be optimized of the single-post insulator, the number of nodes of the output layer is the number of electrical performance evaluation indexes of the single-post insulator, the number of nodes of the hidden layer is selected empirically, the number of nodes of the first layer is 80% of the number of nodes of the input layer, and the number of nodes of the second layer is equal to the sum of the number of nodes of the input layer and the number of nodes of the output layer. Initializing an initial weight omega, a learning coefficient eta, a momentum coefficient alpha, a target error epsilon and a learning frequency epoch max Wherein the weight ω includes the weight ω from the input layer to the first hidden layer 0i Weight ω of first hidden layer to second hidden layer ij Second hidden layer to output layer weight omega jk (these parameters are all parameters of the nature of the neural network);
step 3.2, the output variable of the input layer is the input variable of the deep BP neural network, and the activation function of the BP neural network is set as shown in a formula (3);
the hidden layer and the output layer are calculated using a BP neural network algorithm, in the following equation (4),for the nth iteration, the first hidden layer output variable, the second hidden layer output variable, and the output layer output variable,
calculating an error of an output variable of the BP neural network model to obtain a training error, wherein y is an output variable of a neural network training set in the nth iteration, as shown in the following formula (5);
step 3.3, adjusting the weight, in the nth iteration,for inputting the weight adjustment value of the layer to the first hidden layer,/>Adjusting the value for the weights of the first hidden layer to the second hidden layer,/for the first hidden layer>The value is adjusted for the second hidden layer to output layer weights. Finding the error E when the number of iterations is n (n) For->Taking the partial derivative of (2) as a new weight adjustment value, and calculating the weight in n+1 iterations, wherein the formula (6) is shown;
recording error E (n) For a pair ofTaking the partial derivative of (2) as a weight adjustment value of n+1 iterations, wherein the weight adjustment value is shown in a formula (7);
step 3.4, if the current training error value is smaller than the target error, finishing learning, continuing to carry out step 3.5, otherwise, jumping to step 3.2, and continuing to iterate;
step 3.5, saving the learned weight as a characteristic parameter of the deep BP neural network for the subsequent optimization process;
step 4: construction of optimized objective function value function
Taking the parameter to be optimized of the single-post insulator as an independent variable, and writing the parameter into a vector format:
X＝[R 1 ,R 2 ,R 3 ,r 1 ,r 2 ,H 1 ,H 2 ,L 1 ,L 2 ] T
the electrical evaluation index of the single post insulator is used as a dependent variable and written as a vector format:
E＝[E’ 1 ,E’ 2 ,E’ 3 ] T
taking a finite element electric field solving model as a mapping relation, and constructing the following functional relation:
E＝f p (X)
normalizing the electrical performance index, wherein,coefficient alpha 1 、α 2 And alpha 3 The weights corresponding to the maximum field intensity of the three positions are respectively analyzed according to the importance degrees of different parts of the single-post insulator, the optimized weight distribution of the three parts can be changed by changing the size relation of the coefficients, and an objective function value function f capable of measuring the performance index of the whole electric field by the structural parameters can be established through normalization and weight distribution s ；
Step 5, aiming at the single-post insulator electric field finite element equivalent model which is trained and fitted in step 3, the genetic particle swarm optimization is used for optimizing the single-post insulator, and the specific method for solving the structural optimization problem with the structural parameters as input variables and the objective function value function as fitness function comprises the following steps:
step 5.1, initializing population number n and space dimension n dim Size limit X of each dimension parameter limit Speed limit V for each dimension parameter limit Inertial weight c 1 Self-learning factor c 2 Group learning factor c 3 Probability of hybridization eta 1 Probability of variation eta 2 Maximum number of iterations n ger And an objective function value tolerance epsilon;
step 5.2, initializing a particle swarm, randomly generating particles with different positions, calculating a local optimal value corresponding to each particle, and recording a population history optimal position and a population history optimal fitness; the position of the particle represents the structural parameter of the single-post insulator, and the fitness index of the population represents the objective function value function of the single-post insulator;
step 5.3, calculating inertia weight according to the maximum iteration times and the current iteration times, wherein the calculating method is a weight linear decrementing strategy,
the velocity and position of all particles in the population of particles are updated. The iterative formula of the speed and position of the particle swarm algorithm is as follows:
rand 1 and rand 2 Is in the value range of [0,1 ]]Random number of (1), pbest i As the position of the individual optimal value of each particle in the current iteration number, gbest i The global optimum is located for each particle history. Updating the position of the particles by using a speed updating formula;
step 5.4, hybridizing and mutating the particles, replacing parameters of the offspring and parent particles,
the parent particles are ordered according to the fitness, the hybridization particle number is determined according to the hybridization probability, each parent particle to be hybridized is subjected to the following operation,refers to randomly taking out the r-th dimension parameter of any particle in the hybridization pool, giving the parameter to the child particle to finish hybridization,
performing mutation, randomly mutating a certain proportion of particles in the sub-population, and randomly resetting parameters of random dimension of the mutated particles, wherein the specific mode is as followsAs the r-th dimension parameter of the variant particle, x rdmax And x rdmin Is the parameter range of the r-th dimension parameter, and rand is the random reset probability;
step 5.5, calculating the current particle swarm, solving the fitness of each particle, and recording the optimal position of the population history and the optimal fitness of the population history;
and 5.6, judging whether the tolerance condition of the objective function value is reached or the maximum iteration number is reached, stopping calculation if the tolerance condition of the objective function value is reached, and jumping to the step 5.3 if the tolerance condition of the objective function value is not reached.
The invention comprehensively considers the electrical performance indexes of a plurality of key positions of the single-post insulator, can provide a method for optimizing the electrical performance structure of the single-post insulator with a complex structure, effectively shortens the design period of the single-post insulator, and has good practicability and economy.
Claims (8)
1. The method for optimizing the electrical performance of the single-post insulator for the GIS based on the neural network model is characterized by comprising the following steps of:
step 1, establishing an electric field finite element analysis model of a single-post insulator by using three-dimensional CAD software, determining structural parameters to be optimized of the electric field finite element analysis model of the single-post insulator according to the characteristics of the single-post insulator, and selecting the electric field intensity of a region with the electric field intensity exceeding a maximum allowable value as an electric performance evaluation index of the single-post insulator;
step 2, in the electric field finite element analysis model of the single-post insulator established in the step 1, a control variable method is used to enable structural parameters to be optimized to change within a certain range, a change result of an electric performance index when the parameters change is calculated, a deep neural network training set is obtained, input variables of the training set are parameters to be optimized of the single-post insulator, and output variables are electric performance evaluation indexes of the single-post insulator;
the variation range of the optimized structural parameters is as follows:
wherein R is 1 ，R 2 ，R 3 ，L 2 ，r 1 ，r 2 ，r 3 ，H 1 ，H 2 To be optimized for the parameters, R 1min ，R 1max ，R 2min ，R 2max ，R 3min ，R 3max ，r 1min ，r 1max ，r 2min ，r 2max ，r 3min ，r 3max ，H 1min ，H 1max ，H 2min ，H 2max ，L 1min ，L 1max ，L 2min ，L 2max The variation range of the parameters to be optimized; in the dynamic change of the above variables, there is still a need to satisfy the dynamic adjustment constraint of the following formula (2), where L 0 Is the total height of the single-post insulator;
step 3, building a four-layer deep BP neural network model comprising an input layer, two hidden layers and an output layer, carrying out model training on the model by the deep neural network training set obtained in the step 2 to obtain a deep neural network model for evaluating the electric performance of the single-post insulator, taking the model as an equivalent model of a single-post insulator electric field finite element analysis model, solving a weight of the deep BP neural network model, and taking the weight as characteristic parameters of the deep BP neural network model:
3.1 determining the number of nodes of an input layer, the number of nodes of an output layer and the number of nodes of an hidden layer, wherein the number of nodes of the input layer is the number of structural parameters to be optimized of the single-post insulator, the number of nodes of the output layer is the number of electrical performance evaluation indexes of the single-post insulator, the number of nodes of the hidden layer is selected according to experience, the number of nodes of the first layer is between the number of the input layer and the number of the output layer, and the number of nodes of the second layer is equal to the sum of the number of nodes of the input layer and the number of the output layer; initializing an initial weight omega, a learning coefficient eta, a momentum coefficient alpha, a target error epsilon and a learning frequency epoch max Wherein the weight ω includes the weight ω from the input layer to the first hidden layer 0i Weight ω of first hidden layer to second hidden layer ij Second hiddenLayer-to-output layer weight ω jk ；
3.2 the output variable of the input layer is the input variable of the deep BP neural network model, the activation function of the BP neural network model is set as formula (3), wherein e is a natural constant
The hidden layer and the output layer are calculated using a BP neural network algorithm, in the following equation (4),for the nth iteration, the first hidden layer output variable, the second hidden layer output variable, and the output layer output variable,
calculating error of output variable of BP neural network model to obtain training error, wherein y is output variable of neural network training set in nth iteration
3.3 adjusting the weights such that, in the nth iteration,for inputting the weight adjustment value of the layer to the first hidden layer,/>Adjusting the value for the weights of the first hidden layer to the second hidden layer,/for the first hidden layer>Is the second hiddenLayer-containing to output layer weight adjustment values; finding the error E when the number of iterations is n (n) For->Taking the partial derivative of (2) as a new weight adjustment value, and calculating the weight in n+1 iterations, wherein the weight is represented by a formula (6)
Recording error E (n) For a pair ofAs the weight adjustment value of n+1 iterations, the partial derivative of (2) is represented by the formula (7),
3.4, finishing learning if the current training error value is smaller than the target error, continuing to carry out the step 3.5, otherwise, continuing to iterate the step 3.2;
3.5, saving the learned weight as a characteristic parameter of the deep BP neural network model for the subsequent optimization process;
step 4, carrying out normalization processing on the electric performance evaluation index of the single-post insulator, calculating a weighted average value of the electric performance evaluation index of the single-post insulator, and optimizing the weighted average value as an objective function value;
and 5, taking an equivalent model of the electric field finite element analysis model of the single-post insulator in the step 3 as an optimization object, performing global optimization on the structural parameters to be optimized of the electric field finite element analysis model of the single-post insulator determined in the step 1 by using a genetic particle swarm algorithm, and obtaining the optimal structural parameters after iterative optimization.
2. The substrate according to claim 1The method for optimizing the electrical performance of the single-post insulator for the GIS of the neural network model is characterized in that the structural parameters to be optimized of the electric field finite element analysis model of the single-post insulator comprise: radius of metal insert R 1 Metal insert fillet radius R 2 Radius of insulator R 3 Metal insert depth L 1 Depth L of insulator umbrella skirt 2 Outer radius r of insulator umbrella skirt 1 Insulator umbrella skirt chamfer radius r 2 Radius r of bottom fillet of umbrella skirt of insulator 3 Distance H between insulator umbrella skirt and top end 1 Tangential distance H of insulator umbrella skirt 2 。
3. The optimization method of electrical performance of single-post insulators for GIS based on neural network model as claimed in claim 1, wherein the electrical performance evaluation index of the single-post insulators in step 1 and step 2 is selected by taking the range of values of structural parameters to be optimized as constraint conditions, including maximum value E 'of surface synthesis field intensity of metal inserts' 1 Surface synthesized field intensity maximum E 'of single-pillar insulator' 2 Maximum value E 'of tangential field intensity of single-pillar insulator' 3 The method comprises the steps of carrying out a first treatment on the surface of the The electrical performance evaluation of the unipost insulator without optimization is defined as the composite field intensity E of the surface of the metal insert 1 Maximum value E of surface synthesis field intensity 2 And maximum value of tangential field strength E 3 。
4. The optimization method of electrical performance of single post insulators for GIS based on neural network model as set forth in claim 1, wherein the normalization of electrical performance index in step 4 is determined by equation (8),
wherein the coefficient alpha 1 、α 2 And alpha 3 Weights corresponding to the maximum field strengths of the three positions respectively, f s Is an objective function value function under the current structural parameters.
5. The method for optimizing the electrical performance of the single-post insulator for the GIS based on the neural network model according to claim 1, wherein the specific method of the step 5 is as follows:
step 5.1, according to the principle of the genetic particle swarm algorithm, assigning initial values to the following parameters: population number n and spatial dimension n dim Size limit X of each dimension parameter limit Speed limit V for each dimension parameter limit Inertial weight c 1 Self-learning factor c 2 Group learning factor c 3 Probability of hybridization eta 1 Probability of variation eta 2 Maximum number of iterations n ger And an objective function value tolerance epsilon;
step 5.2, randomly generating particles with different parameters, calculating an objective function value corresponding to each particle by using a trained depth BP neural network model, and recording a population history optimal position and a population history optimal fitness for an initial population;
step 5.3, calculating inertial weight according to the maximum iteration times and the current iteration times in step 5.1, wherein the calculating method is a weight linear decreasing strategy, and the formula (9) is shown
Updating the speeds and positions of all particles in the particle swarm, wherein the iterative formula of the speeds and positions of the particle swarm algorithm is as follows:
rand 1 and rand 2 Is in the value range of [0,1 ]]Random number of (1), pbest i As the position of the individual optimal value of each particle in the current iteration number, gbest i The position of the global optimal value of each particle history is located; updating the position of the particles by using a speed updating formula;
step 5.4, carrying out hybridization and mutation operation on the particles subjected to speed and position updating in step 5.3, wherein hybridization refers to randomly exchanging parameters of the same kind in any two particles, and mutation refers to randomly reassigning a certain parameter in a certain particle;
step 5.5, aiming at the particle swarm subjected to hybridization and mutation in step 5.4, calculating an objective function value corresponding to each particle by using a trained depth BP neural network model, solving the objective function value of each particle, and recording the optimal position of the population history and the optimal fitness of the population history;
and 5.6, if the optimal fitness of the population history is smaller than the tolerance of the objective function value or the current iteration number is larger than or equal to the maximum iteration number, stopping calculation, and if the optimal fitness is not reached, jumping to the step 5.3, and adding one to the iteration number.
6. The method for optimizing electrical performance of single-post insulators for GIS based on neural network model according to claim 1, wherein the characteristics of the single-post insulators in step 1 include support structure characteristics of the single-post insulators and shape structure characteristics of metal inserts of the single-post insulators.
7. The optimization method of electrical performance of a single post insulator for GIS based on a neural network model according to claim 5, wherein the hybridization in step 5.4 is to randomly exchange the same kind of parameters in any two particles, and the mutation is to randomly reassign a certain parameter in a certain particle.
8. The method for optimizing electrical performance of single-post insulators for GIS based on neural network model according to claim 5, wherein the optimal position of population history in step 5.2 and step 5.5 represents structural parameters of the single-post insulators, and the optimal fitness of population history represents objective function value functions of the single-post insulators.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111442270.4A CN114065594B (en) | 2021-11-30 | 2021-11-30 | Single-post insulator electrical performance optimization method for GIS based on neural network model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111442270.4A CN114065594B (en) | 2021-11-30 | 2021-11-30 | Single-post insulator electrical performance optimization method for GIS based on neural network model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114065594A CN114065594A (en) | 2022-02-18 |
| CN114065594B true CN114065594B (en) | 2024-04-02 |
Family
ID=80277458
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111442270.4A Active CN114065594B (en) | 2021-11-30 | 2021-11-30 | Single-post insulator electrical performance optimization method for GIS based on neural network model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114065594B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018072351A1 (en) * | 2016-10-20 | 2018-04-26 | 北京工业大学 | Method for optimizing support vector machine on basis of particle swarm optimization algorithm |
| CN108984978A (en) * | 2018-08-29 | 2018-12-11 | 西安交通大学 | Tri- support insulator electric property optimum design method of GIL based on particle swarm algorithm |
| CN110189872A (en) * | 2019-05-13 | 2019-08-30 | 平高集团有限公司 | A kind of fixed one-mast support insulator, pipeline bus and its bus unit |
-
2021
- 2021-11-30 CN CN202111442270.4A patent/CN114065594B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018072351A1 (en) * | 2016-10-20 | 2018-04-26 | 北京工业大学 | Method for optimizing support vector machine on basis of particle swarm optimization algorithm |
| CN108984978A (en) * | 2018-08-29 | 2018-12-11 | 西安交通大学 | Tri- support insulator electric property optimum design method of GIL based on particle swarm algorithm |
| CN110189872A (en) * | 2019-05-13 | 2019-08-30 | 平高集团有限公司 | A kind of fixed one-mast support insulator, pipeline bus and its bus unit |
Non-Patent Citations (1)
| Title |
|---|
| 房柯夫 ; 吴光亚 ; 刘志强 ; .1000kV交流变电站支柱和空心绝缘子污秽外绝缘设计的探讨.电瓷避雷器.2007,(01),全文. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114065594A (en) | 2022-02-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110110930B (en) | Recurrent neural network short-term power load prediction method for improving whale algorithm | |
| CN110046408B (en) | Interval uncertainty optimization method based on BP neural network differentiation and interval analysis | |
| CN105608295B (en) | The multi-objective genetic algorithm of coking furnace pressure and RBF neural Optimization Modeling method | |
| CN105427241B (en) | Distortion correction method for large-view-field display equipment | |
| CN108984978B (en) | Particle swarm algorithm-based GIL three-post insulator electrical performance optimization method | |
| CN108167802B (en) | Multi-model intelligent optimizing and predicting control method for boiler load under low load | |
| CN111355633A (en) | Mobile phone internet traffic prediction method in competition venue based on PSO-DELM algorithm | |
| CN108764449A (en) | A method of improving PSO Algorithm white body assemble welding line balance problem | |
| CN113138555A (en) | GRNN electric spindle thermal error modeling method based on genetic algorithm optimization | |
| CN112497216A (en) | Industrial robot pose precision compensation method based on deep learning | |
| CN114065594B (en) | Single-post insulator electrical performance optimization method for GIS based on neural network model | |
| CN105976029B (en) | A kind of overhead crane neural network modeling approach of cuckoo behavior RNA-GA | |
| Wu et al. | Many-objective brain storm optimization algorithm | |
| CN113762602A (en) | Short-term wind speed prediction method for wind power plant | |
| CN105930574A (en) | Genetic and neural network algorithm based design method for air intake port model of internal combustion engine | |
| CN113095477A (en) | Wind power prediction method based on DE-BP neural network | |
| CN117056763A (en) | Community discovery method based on variogram embedding | |
| CN107871034A (en) | Tolerance assignment multi-objective optimization design of power method based on mutative scale learning aid algorithm | |
| CN111222799A (en) | Assembly sequence planning method based on improved particle swarm optimization | |
| CN112528556B (en) | Micro-electro-mechanical system design optimization method based on integrated model assisted social learning particle swarm algorithm | |
| Yin et al. | Improved hybrid fireworks algorithm-based parameter optimization in high-order sliding mode control of hypersonic vehicles | |
| CN113219840B (en) | Self-adaptive sliding mode cross-coupling contour control method for three-axis motion platform | |
| CN112966423B (en) | Construction method of IPSO-IAGA-BPNN algorithm-based space primary mirror structure response prediction model | |
| CN112183843B (en) | Load optimization distribution method for thermal power plant based on hybrid intelligent algorithm | |
| CN113807005A (en) | Bearing residual life prediction method based on improved FPA-DBN |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |