CN117272756A - Design method of equalizing ring for GIL combined type ultrahigh-voltage casing pipe - Google Patents

Abstract

A design method of a grading ring for a GIL combined type ultra-high voltage sleeve. Relates to a grading ring of a power transmission and transformation circuit. Comprising the following steps: step 100, determining the original ring diameter R, the pipe diameter R and the lifting height H structural parameters of the equalizing ring through finite element model simulation calculation; taking the maximum bearing field intensity E2 which does not form corona of the equalizing ring as a basis and the maximum field intensity E1 which is borne by the surface of the insulator as a basis, wherein the maximum bearing field intensity E2 which does not form corona of the equalizing ring is less than or equal to 2.2KV/mm, the ring diameter R, the pipe diameter R and the lifting height H of the equalizing ring are used as variables, and the constraint condition of a finite element simulation model is set to E1max=minf (R, R and H) which is less than or equal to 0.60KV/mm, and E2=g (R, R and H) which is less than or equal to 2.2KV/mm; calculating and determining the corresponding ring diameter R and the original basic structural parameters of the pipe diameter R of the equalizing ring by taking the lifting distance H as a reference through a finite element simulation model; step 200, determining main parameters affecting the performance of the equalizing ring through an orthogonal test, and optimizing the main parameters with larger variation; the invention not only can obtain more optimized parameters, but also can greatly shorten the optimization time.

Description

Design method of equalizing ring for GIL combined type ultrahigh-voltage casing pipe

Technical Field

The invention relates to the design of an equalizing ring on a power transmission and transformation circuit, in particular to a method for designing an equalizing ring for a GIL combined ultra-high voltage sleeve.

Background

The grading ring is used as one of key components of the power transmission and transformation circuit, is widely applied to the ultra-high voltage power transmission circuit, and mainly plays a role in uniform electric field. The advantages and disadvantages of the equalizing ring performance directly affect the operation safety of the whole circuit. In use, in order to reduce the electric field intensity and uniformly distribute the electric field intensity of the insulator at the tail end of the high-voltage sleeve, the grading ring with reasonable structural parameters is particularly important.

The design of the equalizing ring is usually carried out by combining self experience with finite element simulation software or adopting a genetic algorithm and a neural network algorithm singly, and mainly has the following problems:

1. the equalizing rings of different models have different nonlinear degrees, different design experience and higher requirements on designers;

2. because the equalizing ring simulation is continuously modified and requires repeated labor, the comparison takes time for the designer.

3. The equalizing ring is designed by simply utilizing a neural network algorithm and a genetic algorithm and combining finite element secondary development and the like, and certain limitations exist in the optimization process.

Therefore, how to improve the design accuracy of equalizing rings with different models and specifications is a technical problem to be solved in the field.

Disclosure of Invention

Aiming at the problems, the invention provides a design method of the equalizing ring for the GIL combined type ultra-high voltage sleeve, which improves the design efficiency and quality of equalizing rings of different types.

The technical scheme of the invention is as follows:

a design method of a grading ring for a GIL combined type ultra-high voltage sleeve comprises the following steps:

step 100, determining the original ring diameter R, the pipe diameter R and the lifting height H structural parameters of the equalizing ring through finite element model simulation calculation;

step 200, determining main parameters affecting the performance of the equalizing ring through an orthogonal test, and optimizing the main parameters with larger variation;

and 300, continuing to optimize by using an algorithm combining the neural network and the genetics, and determining reasonable structural parameters of the grading ring.

Specifically, in the step 100, the constraint condition of a finite element simulation model is set to E1max=minf (R, R and H) which is less than or equal to 0.60KV/mm, and E2=g (R, R and H) which is less than or equal to 2.2KV/mm based on that the maximum bearing field intensity E2 which does not cause corona of the equalizing ring is less than or equal to 2.2KV/mm and the maximum field intensity E1 which is borne by the surface of the insulator is less than or equal to 0.60 KV/mm; and calculating and determining the corresponding ring diameter R and the original basic structural parameters of the pipe diameter R of the equalizing ring by taking the lifting distance H as a reference through a finite element simulation model.

Specifically, step 200 includes:

step 210, expanding the conventional specification of the corresponding ring diameter R and the pipe diameter R by taking the lifting distance H as a reference to form an orthogonal test table;

step 220, inputting parameters of each group of corresponding ring diameter R, lifting distance H and pipe diameter R into a finite element simulation model in sequence according to the orthogonal test table established in the step 210, simulating field intensity E2 generated on the surface of a corresponding equalizing ring, and recording;

step 230, calculating the field intensity E generated on the surface of the optimization target equalizing ring by taking the lifting distance H of at least two specifications as a reference _{Lifting device} Average value of (2);

calculating the field intensity E generated on the surface of the equalizing ring with corresponding specification by taking the ring diameter R with at least two specifications as a reference _Ring(s) Average value of (2);

calculating the field intensity E generated on the surface of the optimization target equalizing ring by taking the pipe diameters H of at least two specifications as the reference _Pipe Average value of (2);

step 240, respectively calculating the extreme values of the ring diameter, the lifting distance and the pipe diameter influence factors;

and 250, determining main parameters affecting the performance of the grading ring according to the magnitude of the corresponding extreme value, and recommending the orthogonal test optimization target arrangement.

Specifically, step 300 includes the steps of:

step 310, a neural network model is established, an input vector is a grading ring structural parameter, an output vector is bearing field intensity E1 of an insulator and surface field intensity E2 of the grading ring, and an activation function of an implicit layer and an output layer is a double tangent function;

step 320, data normalization preprocessing

Normalizing the structural parameters and optimization indexes of the equalizing ring to the interval (1, 1), and carrying out the following steps:

fe (x) =2f (x) -fmax-fmin/fmax-fmin (formula 1)

fe (x) is normalized data;

fmax and fmin are the maximum value and the minimum value of the original data before normalization respectively;

f (x) is original data before normalization;

the structural parameters and the output field intensity of the equalizing ring are normalized to be within the interval of [ -1,1], and the information of the neurons in the neural network is summarized through the double tangent activation function of the hidden layer and the output layer, wherein the double tangent activation function is expressed as formula 2:

f (y) =1-e-2 y/1+e-2y (formula 2)

f (y) is the output value of each neuron, y is the input value of each neuron, and e is a natural constant;

step 330, BP neural network training

Step 331, setting training parameters

The number of the neurons of the input layer is 3, and the neurons are respectively a circular diameter R, a pipe diameter R and a lifting height H; the number of hidden layer neurons is selected according to the rule of N, N is the number of nodes of an input layer, the number of neurons of an output layer is 2, and the number of neurons of the output layer is the bearing field intensity E1 and the surface field intensity E2 of the equalizing ring respectively; the training algorithm of the network is selected as a Levenberg-Marquardt learning algorithm, the learning rate is 0.01, the training times are 50 times, the error square sum index is 0.001, and the training times between two updating displays are 1;

step 332, training network

Step 3321, inputting the set training parameter data into a BP neural network training system for learning and training, fitting the BP neural network, and predicting system output by using the trained BP network according to system input;

step 3322, calculating the output of each unit of the output layer by transfer function, the process being calculated according to function 3;

(3)

Step 3323, checking whether the output of training reaches a preset target, if the output of training cannot reach the preset target, repeating the steps 3321-3323 until the network global error function E is smaller than a preset value or the training times reach a preset value, and ending the whole training process;

step 333, BP neural network predicted data

Determining data meeting preset optimization targets, which are learned and trained through a BP neural network algorithm, and finishing the data as an initial population of a genetic algorithm to wait for inputting the genetic algorithm;

and 333, sorting the predicted data of the BP neural network meeting the preset optimization target to serve as an initial population of the genetic algorithm.

Specifically, after step 333 is completed, the method further includes:

step 340, taking the predicted data obtained by the BP neural network learning and training as an initial population of a genetic algorithm, and inputting the initial population into the genetic algorithm;

step 341, genetic algorithm model parameter setting

Setting i as the number of equalizing rings in the mathematical model of equalizing ring optimization, wherein i=1, 2 and 3 are each equalizing ring in sequence according to the number, each equalizing ring corresponds to 3 optimization variables, and 50 initial populations are randomly generated; the random pairing is crossed to generate a new population, the new population and the old population are combined, and then individuals with high fitness are selected to enter the next generation according to the fitness; setting the individual with the largest fitness not to participate in mutation, and carrying out mutation on other individuals with the mutation probability of 0.01 to finish primary evolution from parent to offspring; after N generations of evolution, selecting an optimal solution from the finally obtained population;

step 3411, establishment of an initial population of genetic algorithm

Combining the moderate data values predicted by the neural network and meeting the optimization target to form an initial population farm=zeros (50, 3) of the genetic algorithm;

step 3412, select

Randomly selecting two individuals P1 and P2, then carrying out genetic variation on the P1 and P2 to obtain new individuals Pc1 and Pc2, repeating the steps until N individuals are obtained, and establishing a new population newfarm=zeros (50, 3);

step 3413, crossing

Crossing the parent population and the newly established population to ensure random pairing and generate a new population FARM= [ norm ] newnorm;

step 3414, replication

Selecting a mode of pairwise random pairing competition for replication, and reserving an individual with the optimal performance; according to convergence curve 1: LCI (counter+1) =maxfitness and convergence curve 2: LC2 (counter+1) =meanfitness, calculating optimal individual fitness values for each generation and group average fitness values for each generation;

step 3415, variation

The method comprises the steps of performing mutation on a population generated by crossover and replication, adding the population with the highest adaptability after one mutation, performing mutation with a mutation probability of 0.01, repeating the steps until N mutated individuals are obtained, and combining the N mutated individuals to obtain a new population farm (i.p1) =rand (UB-LB) +LB;

in step 3416, it is checked whether the data satisfies the condition, and if the data fails to satisfy the preset optimization condition, steps 3411 to 3415 should be repeated.

Specifically, step 100 includes:

taking the maximum bearing field intensity E2 of the equalizing ring, which does not generate corona, and the maximum field intensity E1 borne by the surface of the insulator as the basis, and taking the structural parameters of the equalizing ring, namely the ring diameter R, the pipe diameter R and the lifting height H, as variables;

the optimization model of the insulator and the grading ring containing constraint conditions is as follows: e1max=minf (R, R, H) E2=g (R, R, H) is less than or equal to 2.2KV/mm, and the original basic structural parameters of the equalizing ring are determined through finite element simulation calculation.

Specifically, the grading ring does not corona and has the maximum bearing field intensity E2 which is less than or equal to 2.2KV/mm.

Specifically, in step 230:

E _ring(s) Comprises an average value E1 with the diameter of 450mm as a reference, an average value E2 with the diameter of 500mm as a reference and an average value E3 with the diameter of 550mm as a reference;

E _{lifting device} Comprising an average value E4 based on a lifting distance of 220mm, an average value E5 based on a lifting distance of 300mm and an average value E6 based on a lifting distance of 350 mm;

E _pipe Comprises an average value E7 taking the pipe diameter as a reference, an average value E8 taking the pipe diameter as a reference and 55mm and an average value E9 taking the pipe diameter as a reference and 60mm.

The invention has the beneficial effects that:

1. when the structural parameters of the equalizing ring are optimized through orthogonal test design, the analysis of the results shows that the field intensity born by the surfaces of the insulator and the equalizing ring is related to the high-pitch (H), the ring diameter (R) and the pipe diameter (R) of the equalizing ring. The optimization objective is to require a set of H, R, r values so that the field strength experienced by the grading ring and insulator surfaces is minimized. However, the field strength and H, R, r experienced by the grading ring and insulator surface are non-linear, which is difficult or impossible to describe in terms of a explicit function. It is common practice to simply let H, R, r change in the respective definition fields to obtain a set of structural parameters of the grading ring, then solve for index values under these parameters, and finally compare to find the best combination of parameters that meets the design conditions. This exhaustive method is simple in concept, but is quite cumbersome to calculate, and is almost impossible to apply when the engineering of the calculation is complex or the level of variation of the variables is high. The nonlinear mapping relation between the optimization targets E and H, R, r is established by utilizing an algorithm combining an orthogonal test, a neural network and inheritance, a plurality of groups of data calculated by utilizing a finite element method are used for learning and checking by the neural network, the optimal structural parameters of the equalizing ring are calculated by the established network, and the calculation time of the network and inheritance can be basically ignored, so that more optimal parameters can be obtained, and the optimization time can be greatly shortened. The simulation technology adopted is used for carrying out orthogonal experiments, and the neural network genetic algorithm is reused to optimize the design method, so that the principle is simple, the method is effective, and the problem of complicated calculation process can be solved.

Drawings

Figure 1 is a schematic structural view of a grading ring,

figure 2 is a schematic diagram of a neural network model,

fig. 3 is a flow chart of a neural network model.

Detailed Description

Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to like or similar elements or elements having like or similar functions throughout. The embodiments described below by referring to the drawings are illustrative only and are not to be construed as limiting the invention.

A design method of a grading ring for a GIL combined type ultra-high voltage sleeve comprises the following steps:

step 100, determining the original ring diameter R, the pipe diameter R and the lifting height H structural parameters of the equalizing ring through finite element model simulation calculation;

taking the maximum bearing field intensity E2 which does not form corona of the equalizing ring as a basis and the maximum field intensity E1 which is borne by the surface of the insulator as a basis, wherein the maximum bearing field intensity E2 which does not form corona of the equalizing ring is less than or equal to 2.2KV/mm, the ring diameter R, the pipe diameter R and the lifting height H of the equalizing ring are used as variables, and the constraint condition of a finite element simulation model (a simulation model conventional in the field) is set as E1max=minf (R, R and H) which is less than or equal to 0.60KV/mm, and E2=g (R, R and H) which is less than or equal to 2.2KV/mm; calculating and determining the corresponding ring diameter R and the original basic structural parameters of the pipe diameter R of the equalizing ring by taking the lifting distance H as a reference through a finite element simulation model; namely, the basic values of the corresponding ring diameter R and the pipe diameter R are determined by taking the lifting distance H as the reference when the lifting distance H is 220mm, 300mm and 350mm respectively, and the obtained preliminary data are specifically shown in a table 1;

table 1:

parameters (parameters)	Elevation distance H (mm)	Ring diameter R (mm)	Pipe diameter r (mm)
				Level of variation	220、300、350	450、500、550	50、55、60

Step 200, determining main parameters affecting the performance of the equalizing ring through an orthogonal test, and optimizing the main parameters with larger variation;

step 210, expanding the conventional specification of the corresponding ring diameter R and the pipe diameter R by taking the lifting distance H (mm) as a reference to form an orthogonal test table, as shown in table 2;

TABLE 2

	Elevation distance H (mm)	Ring diameter R (mm)	Pipe diameter r (mm)
				Test 1	220	450	50
Test 2	220	500	60
				Test 3	220	550	55
Test 4	300	450	55
				Test 5	300	500	50
Test 6	300	550	60
				Test 7	350	450	60
Experiment 8	350	500	55
				Test 9	350	550	50

Step 220, inputting parameters of each group of corresponding ring diameter R, lifting distance H and pipe diameter R into a finite element simulation model in sequence according to the orthogonal test table established in step 210, simulating field intensity E2 (set as an optimization target) generated on the surface of a corresponding equalizing ring, and recording to obtain a table 3;

TABLE 3 Table 3

Step 230, calculating the field intensity E generated on the surface of the optimization target equalizing ring by taking the lifting distance H of at least two specifications as a reference _{Lifting device} Average value of (2);

calculating the field intensity E generated on the surface of the equalizing ring with corresponding specification by taking the ring diameter R with at least two specifications as a reference _Ring(s) Average value of (2);

calculating the field intensity E generated on the surface of the optimization target equalizing ring by taking the pipe diameters H of at least two specifications as the reference _Pipe Average value of (2);

further defined, E _Ring(s) Comprises an average value Ea1 based on the diameter of 450mm, an average value Ea2 based on the diameter of 500mm and an average value Ea3 based on the diameter of 550 mm;

E _{lifting device} Comprises an average value Ea4 based on a lifting distance of 220mm, an average value Ea5 based on a lifting distance of 300mm and an average value Ea6 based on a lifting distance of 350 mm;

E _pipe Comprises an average value Ea7 taking the pipe diameter as a standard, an average value Ea8 taking the pipe diameter as a standard and a average value Ea9 taking the pipe diameter as a standard, wherein the pipe diameter is 50 mm;

take table 2 as an example:

1) Calculating the average value of an optimization target Ea1 by taking the diameter of the ring as a reference, wherein Ea 1= (1.65403+1.56585+1.54280)/(3= 1.58756);

2) Calculating the average value of an optimization target Ea2 with the diameter of the ring being 500mm as a reference, wherein Ea 2= (1.59501+1.55041.53812)/(3= 1.56118);

3) Based on the loop diameter of 550mm, the average value of the optimization target Ea3 is calculated, ea 3= (1.57010+1.52335+1.51305)/(3= 1.53550).

4) Calculating the average value of an optimization target Ea4 with the elevation distance of 220mm as a reference, wherein Ea 4= (1.65403+1.59501+1.57010)/(3= 1.60638);

5) Calculating the average value of an optimization target Ea5 based on the elevation distance of 300mm, wherein Ea 5= (1.56585+1.55041.52335)/(3= 1.54654);

6) On the basis of the elevation distance of 350mm, the average value of the optimization target Ea6 is calculated, ea 6= (1.54280+1.53812+1.51305)/(3= 1.53132).

7) Calculating the average value of an optimization target Ea7 based on the pipe diameter of 50mm, wherein Ea 7= (1.65403+1.55041.51305)/(3= 1.57250);

8) Calculating the average value of an optimization target Ea8 based on the pipe diameter of 55mm, wherein Ea 8= (1.56585+1.53812+1.57010)/(3= 1.55802);

9) Based on the pipe diameter of 60mm, the average value of the optimization target Ea9 is calculated, ea 9= (1.54280+1.59501+1.52335)/(3= 1.55372).

Step 240, calculating the range values of the ring diameter, the lifting distance and the pipe diameter influence factors respectively, and utilizing the range=emax-Emin of the formula E;

further defined, the range of the ring diameter influencing factors = Ea1-ea3 = 1.58756-1.53550 = 0.05206; the range of elevation influencing factors = Ea4-ea6 = 1.60638-1.53132 = 0.07056; the range of pipe diameter influencing factors = Ea7-ea9 = 1.57250-1.55372 = 0.01878; the relevant data were statistically shown in table 4 below:

TABLE 4 Table 4

	Based on the ring diameter R	Based on the elevation distance H	Based on pipe diameter r
				Mean 1	1.58756	1.60638	1.57250
Mean value 2	1.56118	1.54654	1.55802
				Mean 3	1.53550	1.53132	1.55372
Difference value of extremely	0.05206	0.07056	0.01878

Step 250, determining main parameters affecting the performance of the equalizing ring according to the magnitude of the corresponding extreme value and recommending orthogonal test optimization target arrangement;

from Table 3, it can be seen that the data set 350, 550, 50 in the orthogonal test result test 9 is the best combination data;

it can be seen from table 4 that the maximum influence of the elevation distance H on the optimization target E is 0.07056, the diameter R of the ring is minimum, and the structural parameters of the equalizing ring are optimally combined by referring to table 3 and table 4, so that the optimal combination of the elevation distance H, the diameter R of the ring and the diameter R of the ring is determined; the optimal combination is that the height distance H=350 mm, the ring diameter R=550 mm and the pipe diameter r=60 mm

By carrying out simulation experiments on the optimal combination determined according to the optimization targets, the simulation results are greatly improved when the optimal combination is not optimized according to the optimization targets, the relation is difficult or impossible to describe in the form of a display function, and the non-uniformity coefficient is low. It is recommended that the arrangement is optimized, namely, the lifting distance H is 350mm, the ring diameter R is 550mm, and the pipe diameter R is 60mm.

Although the orthogonal test design method can well optimize the structural parameters of the equalizing ring, certain defects still exist; the method is characterized in that the influence of each factor cannot be comprehensively reflected on test arrangement, and in addition, the influence of the prediction effect of the orthogonal test design method on the level change of the design factors is large. Therefore, the scheme further utilizes the algorithm combining the neural network and the genetics to establish the nonlinear mapping relation between the optimization targets E and H, R, r, utilizes 50 groups of data calculated by the finite element method to learn and check the neural network, and then uses the established genetic model to calculate the optimal structural parameters of the equalizing ring, and the calculation time of the network and the genetics can be basically ignored, so that more optimized parameters can be obtained, and the optimization time can be greatly shortened.

Step 300, neural network genetic algorithm

Step 310, using a neural network model (conventional model in the art), as shown in fig. 1, respectively introducing the elevation distance H, the ring diameter R and the pipe diameter R of the optimal combination structural parameters of the equalizing ring into input vectors of the neural network model, and obtaining output vectors from an output layer of the neural network model, wherein the output vectors are respectively the bearing field intensity E1 of the insulator and the surface field intensity E2 of the equalizing ring;

step 320, data normalization preprocessing

In order to improve the optimization effect of the neural network, the input and output vectors should be normalized to dimensionless vectors respectively; meanwhile, in order to fully utilize the characteristic of an activation function, structural parameters (lifting distance H, ring diameter R and pipe diameter R) of the equalizing ring and optimization indexes (bearing field intensity E1 of an insulator and surface field intensity E2 of the equalizing ring) are normalized to the section of-1, 1 according to the formula (1), so that later optimization is facilitated;

fe (x) =2f (x) -fmax-fmin/fmax-fmin (formula 1)

fe (x) is normalized data;

fmax and fmin are the maximum value and the minimum value of the original data before normalization respectively;

f (x) is the original data before normalization

For example:

normalization of ring diameter 450mm data is to substituting the original data 450mm of the ring diameter, the maximum value 550mm of the ring diameter and the minimum value 450mm of the ring diameter into formula 1 respectively for normalization calculation to obtain normalized fe (x) data, namely fe (x) =2×450-550-450/550-450= -100/100= -1.

Normalization of ring diameter 500mm data is to respectively substituting the original ring diameter data 500mm, the maximum ring diameter data 550mm and the minimum ring diameter data 450mm into formula 1 to perform normalization calculation, and obtain normalized fe (x) data, namely fe (x) =2×500-550-450/550-450=0/100=0.

Normalization of ring diameter 550mm data is to respectively substituting the original data of 500mm of the ring diameter, the maximum value of 550mm of the ring diameter and the minimum value of 450mm into formula 1 to perform normalization calculation, and obtain normalized fe (x) data, namely fe (x) =2×550-550-450/550-450=100/100=1.

The structural parameters and the output field intensity of the equalizing ring are normalized to be within the range of [ -1,1], so that the activation functions of the hidden layer and the output layer are double tangent activation functions, and the double tangent activation functions are used for summarizing information of neurons in a neural network and converting the information into new output end signals to be transmitted to the next neurons, so that the optimization speed is higher. The specific bi-tangent activation function is equation 2:

f (y) =1-e-2 y/1+e-2y (formula 2)

f (y) is the output value of each neuron, y is the input value of each neuron, and e is a natural constant.

Step 330, BP neural network training

Step 331, setting training parameters

Leading parameters (R, R, h) required to be optimized of the equalizing ring into an input layer; setting the number of hidden layer neurons as N (N is the node number of an input layer), and referring to a table 5, the number of hidden layer neurons is 10; the number of the neurons of the output layer is 2, and the number of the neurons of the output layer is respectively the bearing field intensity E1 of the insulator and the surface field intensity E2 of the equalizing ring; the network training algorithm is selected as a Levenberg-Marquardt learning algorithm, the learning rate is 0.01, the training times are 50 times, the error square sum index is 0.001, and the training times between two updating displays are 1.

Step 332, training network

Step 3321, inputting the equalizing ring parameters of the set training parameter table 5 into a BP neural network training system for learning, training and fitting, and predicting the optimization target output by the system by using the trained BP network according to the table 5 data input by the system.

Table 5: finite element simulation data for neural network training and verification

Step 3322, calculating the bearing field intensity of each unit of the output layer by using the transfer function in the BP neural network training system, wherein the bearing field intensity is E1 and E2 respectively, and the transfer function is formula 3.

(3)

Xi is the input to the ith neuron;

wi is the current weight of the ith neuron;

θ is the threshold of the neuron.

Step 3323, checking whether the bearing field intensity outputted from the training in step 3322 reaches a preset target, if the bearing field intensity cannot reach the preset target, repeating the steps 3321-3323 until the network global error function f (Ei) is smaller than a preset value (network error) or the training times reach a preset value, and ending the whole training process.

And 333, sorting the predicted data of the BP neural network meeting the preset optimization target to serve as an initial population of the genetic algorithm.

In the traditional BP network learning, the network approximation is very slow, and when the approximation error does not reach the theoretical requirement value, the approximation error is easy to be trapped into local minimum, so that the network cannot learn. In fact, BP networks based on gradient methods tend to affect solution quality due to too slow convergence. In view of the above, the invention refers to a genetic algorithm, and the data obtained by the BP neural network learning training is input into the genetic algorithm, so that the equalizing ring parameter optimization target is more accurate and quicker.

Step 340, taking the predicted data obtained by the BP neural network learning and training as an initial population of a genetic algorithm, and inputting the initial population into the genetic algorithm;

step 341, genetic algorithm model parameter setting

Setting i in the genetic algorithm model as the number of equalizing rings, wherein i=1, 2 and 3 are each equalizing ring in sequence according to the number, each equalizing ring corresponds to 3 optimized variables (R, H, r), and 50 initial populations are randomly generated; the random pairing is crossed to generate a new population, the new population and the old population are combined, and then individuals with high fitness are selected to enter the next generation according to the fitness; the individuals with the greatest fitness are not involved in mutation, and other individuals are mutated with the mutation probability of 0.01, so that primary evolution from the parent to the offspring is completed; after N generations of evolution, selecting an optimal solution from the finally obtained population;

step 3411, establishment of an initial population of genetic algorithm

Moderate data values predicted by the neural network to meet the optimization objective are combined to form an initial population of genetic algorithm farm=zeros (50, 3).

Step 3412, selecting;

two individuals P1 and P2 were randomly selected, then P1 and P2 were subjected to genetic variation to obtain new individuals Pc1 and Pc2, and this step was repeated until N individuals were obtained, creating a new population newfarm=zeros (50, 3).

Step 3413, crossing;

the parent population and the newly established population are crossed to ensure random pairing, and a new population FARM= [ norm; newnorm ] is generated.

Step 3414, replication

The selective replication adopts a pairwise random pairing competition mode, and the individuals with the optimal performance are reserved. According to convergence curve 1: LCI (counter+1) =maxfitness and convergence curve 2: LC2 (counter+1) =meanfitness, and each generation of optimal individual fitness values and each generation of population average fitness values are calculated.

Step 3415, mutating;

and (3) mutating the population generated by crossover and replication, adding the population with the highest fitness after one mutation, mutating with a mutation probability of 0.01, repeating the steps until N mutated individuals are obtained, and combining the N mutated individuals to obtain a new population farm (i.p1) =rand (UB-LB) +LB.

In step 3416, it is checked whether the data satisfies the condition, and if the data fails to satisfy the preset optimization condition, steps 3411 to 3415 should be repeated.

Step 400, implementation of neural network genetic algorithm

1. And writing a neural network genetic algorithm program by using MATLAB language.

2. And optimizing the structural parameters of the equalizing ring by using a written neural network genetic program.

1) Data normalization preprocessing

XX=[220:50:450;220:450:55;220:450:60;220:500:55;220:550:55;220:550:60;220:50:550;];

YY=[1.39693，0.574;1.36308,0.528;1.36948,0.516;1.34792,0.501;1.34427,0.493;1.36951,0.481;1.33603,0.474]';

XX=premnmx(XX);

YY=premnmx(YY);

YY

2) Creating a network

Net=newff(minmax(XX),[1.33603,0.574],{'tansig','tansig','purelin'},'trainlm');

3) Setting training parameters

net.trainParam.show=1;

net.trainParam.lr=0.01;

net.trainParam.epochs=50;

net.trainParam.goal=0.001;

4) Training network

net=train（net,XX,YY）

a=sim(net，XX）

minYY=[1.33603,0.474]

maxYY=[1.39693,0.574]

y=postmnmx(a,minYY,maxYY)

5) Preservation network save BPnet net function

[Yp,Xp,LC1,LC2]=MYGA(bpnet,M,N,Pm,LB,UB,XX,YY)

First step, variable initialization

LC1 = zeros (1, 50); convergence curve 1

LC2 = LC1; convergence curve 2

Second step, randomly generating initial population

Farm=zeros(50,3);

fori=1:50

forj=1:3

farm(ij)=rand*[UB(j)-LB(j))]+LB(j);

end

end

Counter=0, set iteration Counter

while counter<50

Third step, crossing

Newfarm=zeros (50, 3), new population

Ser=randperm (50); guaranteed random pairing

Fori=1:49

Aa=farm (Ser (i)): two parent individuals to be crossed

BB=farm(Ser(i+1),:);

Pl=rand

p2=rand;

pos=unidrnd(4);

A=[p1*AA(1:pos)+(1-p1)*BB(1:pos),p2*AA((pos+1):3)+(1-p2)*BB((pos+1):3)];

B=[p2*AA(1:pos)+(1-p2)*BB(1:pos),p1*AA((pos+1):3)+(1-p2)*BB((pos+1):3)];

newfarm(i,:)=A;

newfarm(i+1,:)=B;

end

New and old population combination

FARM=[farm;newfarm];

Fourth step, select copy

FITNESS=zeros(1,50);

Fitness=zeros(1,50);

Fori=1:(50)

X=FARM(i,:);

FITNESS(i)=1/NETSIM(bpnet,X',XX,YY);

end

Ser=randperm (50), and the selective replication adopts a mode of pairwise random pairing competition, and has the capacity of retaining the optimal body.

fori=1:50

f1=FITNESS(Ser(2*i-1)）;

f2=FITNESS(Ser(2*i));

if f1>=f2

farm(i,：)=FARM(Ser(2*i-1),:);

fitness(i)=FITNESS(Ser(2*i-1));

else

farm(i,:)=FARM(Ser(2*i);：);

fitness(i)=FITNESS(Ser(2*i));

end

end

Recording optimal individuals and Convergence curves

Maxfitness=max(fitness);

Meanfitness=mean(fitness);

LCI (counter+1) =maxfitness; convergence Curve 1, record of optimal individual fitness values for each generation

LC2 (counter+1) =meanfitness; convergence Curve 2, record of average fitness of populations of each generation

Pos=find(fitness==maxfitness);

Xp=farm(pos(1),:);

Yp=maxfitness;

Fifth step, mutation

for i=1:50

if 0.01> rand &i & -pos (1), the variation probability is 0.01

X=farm(i,:);

p1=unidrnd(3);

farm(i,:)=X;

farm(i.p1)=rand(UB-LB)+LB

end

end

counter=counter+1;

disp(counter);

end。

Step 500, neural network genetic algorithm optimizing result

The neural network inheritance is used for solving and optimizing, so that when the equalizing ring lifting distance H= 345.98mm, the ring diameter R=515.12 mm and the pipe diameter r=56.76 mm, the minimum value of the equalizing ring surface field intensity optimization target is E2= 1.27522KV/mm, and the voltage distribution of the insulator is uniform.

Step 600, simulation test verification of neural network genetic optimization result

Simulation tests are carried out by adopting the parameters of the equalizing ring, namely the lifting distance H=350 mm, the ring diameter R=500 mm and the pipe diameter r=60 mm, the field intensity distribution on the insulator calculated under the three-dimensional model after the equalizing ring is installed is obviously improved, and meanwhile, the cost of raw materials is also saved. The maximum electric field intensity of the surface of the grading ring with the configuration is 1.30888kV/mm less than or equal to 2.2kV/mm and is within an acceptable range, the maximum field intensity of the insulator is 0.47kV/mm less than or equal to 0.60kV/mm, the genetic optimization result of the neural network is compared with the experimental verification result, the relative error E% is E% = 1.30888 (experimental value) -1.27522/1.30888 =2.57%, the relative error is smaller, and the obtained result is more satisfactory.

Step 100 comprises:

taking the maximum bearing field intensity E2 of the equalizing ring, which does not generate corona, and the maximum field intensity E1 borne by the surface of the insulator as the basis, and taking the structural parameters of the equalizing ring, namely the ring diameter R, the pipe diameter R and the lifting height H, as variables;

the optimization model of the insulator and the grading ring containing constraint conditions is as follows: e1max=minf (R, R, H) E2=g (R, R, H). Ltoreq.2.2 KV/mm, wherein a one-to-one mapping relation exists between the two, and the original basic structural parameters (the original basic parameters R, R and H) of the equalizing ring are determined through finite element simulation calculation.

The maximum bearing field intensity E2 of the equalizing ring is less than or equal to 2.2KV/mm.

For the purposes of this disclosure, the following points are also described:

(1) The drawings of the embodiments disclosed in the present application relate only to the structures related to the embodiments disclosed in the present application, and other structures can refer to common designs;

(2) The embodiments disclosed herein and features of the embodiments may be combined with each other to arrive at new embodiments without conflict;

the above is only a specific embodiment disclosed in the present application, but the protection scope of the present disclosure is not limited thereto, and the protection scope of the present application shall be subject to the protection scope of the claims.

Claims (8)

1. The design method of the equalizing ring for the GIL combined type ultrahigh-voltage sleeve is characterized by comprising the following steps of:

step 100, determining the original ring diameter R, the pipe diameter R and the lifting height H structural parameters of the equalizing ring through finite element model simulation calculation;

step 200, determining main parameters affecting the performance of the equalizing ring through an orthogonal test, and optimizing the main parameters with larger variation;

and 300, continuing to optimize by using an algorithm combining the neural network and the genetics, and determining reasonable structural parameters of the grading ring.

2. The design method of the equalizing ring for the GIL combined type ultra-high voltage bushing, which is disclosed in claim 1, is characterized in that in the step 100, the constraint condition of a finite element simulation model is set to E1max=minf (R, R and H) which is less than or equal to 0.60KV/mm, and E2=g (R, R and H) which is less than or equal to 2.2KV/mm based on the maximum bearing field intensity E2 which is less than or equal to 2.2KV/mm of the equalizing ring and the maximum field intensity E1 which is less than or equal to 0.60KV/mm of the surface of the insulator; and calculating and determining the corresponding ring diameter R and the original basic structural parameters of the pipe diameter R of the equalizing ring by taking the lifting distance H as a reference through a finite element simulation model.

3. The method of designing a grading ring for GIL composite ultra-high voltage casing according to claim 1, wherein step 200 comprises:

step 210, expanding the conventional specification of the corresponding ring diameter R and the pipe diameter R by taking the lifting distance H as a reference to form an orthogonal test table;

step 220, inputting parameters of each group of corresponding ring diameter R, lifting distance H and pipe diameter R into a finite element simulation model in sequence according to the orthogonal test table established in the step 210, simulating field intensity E2 generated on the surface of a corresponding equalizing ring, and recording;

step 230, calculating the field intensity E generated on the surface of the optimization target equalizing ring by taking the lifting distance H of at least two specifications as a reference _{Lifting device} Average value of (2);

calculating the field intensity E generated on the surface of the equalizing ring with corresponding specification by taking the ring diameter R with at least two specifications as a reference _Ring(s) Average value of (2);

calculating the field intensity E generated on the surface of the optimization target equalizing ring by taking the pipe diameters H of at least two specifications as the reference _Pipe Average value of (2);

step 240, respectively calculating the extreme values of the ring diameter, the lifting distance and the pipe diameter influence factors;

and 250, determining main parameters affecting the performance of the grading ring according to the magnitude of the corresponding extreme value, and recommending the orthogonal test optimization target arrangement.

4. The method of designing a grading ring for GIL composite ultra-high voltage bushing of claim 1, wherein step 300 comprises the steps of:

step 310, a neural network model is established, an input vector is a grading ring structural parameter, an output vector is bearing field intensity E1 of an insulator and surface field intensity E2 of the grading ring, and an activation function of an implicit layer and an output layer is a double tangent function;

step 320, data normalization preprocessing

Normalizing the structural parameters and optimization indexes of the equalizing ring to the interval (1, 1), and carrying out the following steps:

fe (x) =2f (x) -fmax-fmin/fmax-fmin (formula 1)

fe (x) is normalized data;

fmax and fmin are the maximum value and the minimum value of the original data before normalization respectively;

f (x) is original data before normalization;

the structural parameters and the output field intensity of the equalizing ring are normalized to be within the interval of [ -1,1], and the information of the neurons in the neural network is summarized through the double tangent activation function of the hidden layer and the output layer, wherein the double tangent activation function is expressed as formula 2:

f (y) =1-e-2 y/1+e-2y (formula 2)

f (y) is the output value of each neuron, y is the input value of each neuron, and e is a natural constant;

step 330, BP neural network training

Step 331, setting training parameters

The number of the neurons of the input layer is 3, and the neurons are respectively a circular diameter R, a pipe diameter R and a lifting height H; the number of hidden layer neurons is selected according to the rule of N, N is the number of nodes of an input layer, the number of neurons of an output layer is 2, and the number of neurons of the output layer is the bearing field intensity E1 and the surface field intensity E2 of the equalizing ring respectively; the training algorithm of the network is selected as a Levenberg-Marquardt learning algorithm, the learning rate is 0.01, the training times are 50 times, the error square sum index is 0.001, and the training times between two updating displays are 1;

step 332, training network

Step 3321, inputting the set training parameter data into a BP neural network training system for learning and training, fitting the BP neural network, and predicting system output by using the trained BP network according to system input;

step 3322, calculating the output of each unit of the output layer by transfer function, the process being calculated according to function 3;

(3)

Step 3323, checking whether the output of training reaches a preset target, if the output of training cannot reach the preset target, repeating the steps 3321-3323 until the network global error function E is smaller than a preset value or the training times reach a preset value, and ending the whole training process;

step 333, BP neural network predicted data

Determining data meeting preset optimization targets, which are learned and trained through a BP neural network algorithm, and finishing the data as an initial population of a genetic algorithm to wait for inputting the genetic algorithm;

and 333, sorting the predicted data of the BP neural network meeting the preset optimization target to serve as an initial population of the genetic algorithm.

5. The method of designing a grading ring for GIL composite ultra-high voltage bushing of claim 4, further comprising, after step 333:

step 340, taking the predicted data obtained by the BP neural network learning and training as an initial population of a genetic algorithm, and inputting the initial population into the genetic algorithm;

step 341, genetic algorithm model parameter setting

Setting i as the number of equalizing rings in the mathematical model of equalizing ring optimization, wherein i=1, 2 and 3 are each equalizing ring in sequence according to the number, each equalizing ring corresponds to 3 optimization variables, and 50 initial populations are randomly generated; the random pairing is crossed to generate a new population, the new population and the old population are combined, and then individuals with high fitness are selected to enter the next generation according to the fitness; setting the individual with the largest fitness not to participate in mutation, and carrying out mutation on other individuals with the mutation probability of 0.01 to finish primary evolution from parent to offspring; after N generations of evolution, selecting an optimal solution from the finally obtained population;

step 3411, establishment of an initial population of genetic algorithm

Combining the moderate data values predicted by the neural network and meeting the optimization target to form an initial population farm=zeros (50, 3) of the genetic algorithm;

step 3412, select

Randomly selecting two individuals P1 and P2, then carrying out genetic variation on the P1 and P2 to obtain new individuals Pc1 and Pc2, repeating the steps until N individuals are obtained, and establishing a new population newfarm=zeros (50, 3);

step 3413, crossing

Crossing the parent population and the newly established population to ensure random pairing and generate a new population FARM= [ norm ] newnorm;

step 3414, replication

Selecting a mode of pairwise random pairing competition for replication, and reserving an individual with the optimal performance; according to convergence curve 1: LCI (counter+1) =maxfitness and convergence curve 2: LC2 (counter+1) =meanfitness, calculating optimal individual fitness values for each generation and group average fitness values for each generation;

step 3415, variation

The method comprises the steps of performing mutation on a population generated by crossover and replication, adding the population with the highest adaptability after one mutation, performing mutation with a mutation probability of 0.01, repeating the steps until N mutated individuals are obtained, and combining the N mutated individuals to obtain a new population farm (i.p1) =rand (UB-LB) +LB;

in step 3416, it is checked whether the data satisfies the condition, and if the data fails to satisfy the preset optimization condition, steps 3411 to 3415 should be repeated.

6. The method of designing a grading ring for GIL composite ultra-high voltage bushing of claim 1, wherein step 100 comprises:

taking the maximum bearing field intensity E2 of the equalizing ring, which does not generate corona, and the maximum field intensity E1 borne by the surface of the insulator as the basis, and taking the structural parameters of the equalizing ring, namely the ring diameter R, the pipe diameter R and the lifting height H, as variables;

the optimization model of the insulator and the grading ring containing constraint conditions is as follows: e1max=minf (R, R, H) E2=g (R, R, H) is less than or equal to 2.2KV/mm, and the original basic structural parameters of the equalizing ring are determined through finite element simulation calculation.

7. The design method of the equalizing ring for the GIL combined type ultra-high voltage bushing, which is disclosed in claim 5, is characterized in that the maximum bearing field intensity E2 of the equalizing ring is less than or equal to 2.2KV/mm.

8. The method of designing a grading ring for GIL composite ultra-high voltage bushing according to claim 1, wherein in step 230:

E _ring(s) Comprises an average value E1 with the diameter of 450mm as a reference, an average value E2 with the diameter of 500mm as a reference and an average value E3 with the diameter of 550mm as a reference;

E _{lifting device} Comprising an average value E4 based on a lifting distance of 220mm, an average value E5 based on a lifting distance of 300mm and an average value E6 based on a lifting distance of 350 mm;

E _pipe Comprises an average value E7 taking the pipe diameter as a reference, an average value E8 taking the pipe diameter as a reference and 55mm and an average value E9 taking the pipe diameter as a reference and 60mm.