Composite post insulator umbrella skirt parameter optimization method based on rain flashover characteristic and neural network

Abstract

The invention provides a composite post insulator umbrella skirt parameter optimization method based on a rain flashover characteristic and a neural network, and the method comprises the steps: carrying out the rain flashover test of a composite post insulator under different umbrella skirt structures, and obtaining the rain flashover voltage test data samples of different umbrella skirt structure parameters; umbrella skirt structure parameters in the test data samples serve as decision variables, the rain lightning voltage serves as an optimization target, and a BP neural network algorithm is adopted to establish a grey incidence relation between the decision variables and the optimization target; redefining a value range of a decision variable to be optimized, and inputting the value range into the trained BP neural network to obtain a rain lightning voltage objective function value under different umbrella skirt structure parameter combination conditions; and extracting an umbrella skirt structure parameter corresponding to the maximum value of the target function as an optimal parameter. According to the method, the optimal value range of the umbrella skirt parameters of the large-diameter composite post insulator is explored through optimization design geometric analysis and neural network analysis of the umbrella skirt parameters.

Description

Technical Field

[0001] The invention relates to the technical field of shed parameter optimization, and in particular to a shed parameter optimization method for a composite post insulator based on rain-flash characteristics and a neural network. Background Art

[0002] As the main insulating equipment of the converter station of the high-voltage direct current project, the post insulator plays the role of mechanical connection and electrical insulation. Composite post insulators have been widely used in my country's ultra-high voltage direct current transmission projects in recent years due to their excellent hydrophobicity and pollution resistance. Operation experience shows that post insulator flashover is the main problem of insulation equipment outside the converter station. Dirt, fog, rain, snowfall, and ice are all external factors that may cause surface discharge or even flashover of the insulator. With the commissioning of many ultra-high voltage direct current projects at home and abroad, it is not uncommon for the post insulators of the converter station to discharge along the surface or even flashover on rainy days. With the increase of voltage levels, the diameter of the post insulators of the converter station is getting larger and larger. The flashover problem of large-diameter composite post insulators caused by heavy rainfall has become an important technical problem that seriously threatens the operation of high-voltage direct current transmission projects.

[0003] At present, there are few studies on the flashover characteristics of large-diameter composite post insulators under rain conditions at home and abroad. The understanding of the raindrop deformation law of the insulator canopy and its influence on the discharge mechanism, as well as the influence of various factors on the flashover voltage, is still incomplete. There is still a lot of room for optimization of the canopy structural parameters of large-diameter composite post insulators in heavy rainfall areas.

[0004] The selection and design of composite post insulators in converter stations mainly considers their pollution flashover characteristics, but does not fully consider the impact of rain conditions, especially heavy rainfall, on their external insulation performance. The shed provides the main creepage distance for composite post insulators. If the shed structural parameters are not designed reasonably, rainwater or even arc bridging between the sheds is likely to occur, which will greatly reduce the overall flashover voltage of the composite post insulator. Therefore, optimizing the shed parameters of composite post insulators is an important way to improve their rain flashover performance, which can greatly increase their rain flashover voltage, and can reduce the amount of insulation materials used to a certain extent, saving resources and reducing investment. Limited by the test conditions, the rain flashover characteristics of large-size post insulators used in ultra-high voltage / ultra-high voltage projects are still less studied. Summary of the invention

[0005] In order to solve the technical problems existing in the above background technology, the present invention provides a method for optimizing the shed parameters of a composite post insulator based on rain-flash characteristics and a neural network, comprising: Conduct rain flashover tests on composite post insulators with different shed structures, and obtain rain flashover voltage test data samples with different shed structure parameters; Taking the shed structure parameters in the test data sample as decision variables and the rain-flashover voltage as the optimization target, a BP neural network algorithm is used to establish a grey correlation relationship between the decision variables and the optimization target; Redefine the value range of the decision variables to be optimized and input them into the trained BP neural network to obtain the rain-flashover voltage objective function value under different shed structure parameter combinations. The shed structural parameters corresponding to the maximum value of the objective function are extracted as the optimal parameters.

[0006] As a further explanation of the present invention, before performing the rain flashover test on the composite post insulator, a geometric analysis of the optimization of the shed parameters is required so that the shed structural parameters of the rain flashover test satisfy the following relationship:

[0007] Among them, λ is the extension difference between large and small umbrellas, P1 is the extension of large umbrella, P2 is the extension of small umbrella, S2 is the distance between large umbrella and large umbrella, and β is the inclination angle of the upper surface of the umbrella skirt.

[0008] As a further explanation of the present invention, the composite post insulator rain flashover test includes two conditions: one large and one small umbrella and one large and two small umbrellas, wherein: The shed structural parameters of the composite post insulator with one large and one small shed include four decision variables: average shed extension, difference between large and small shed extension, shed spacing and pole diameter. The shed structural parameters of the composite post insulator with one large shed and two small sheds include three decision variables: average shed extension, difference between large and small shed extensions, and shed spacing.

[0009] As a further explanation of the present invention, the optimization objectives include: the rain flash voltage U 50% , Rain flashover voltage E per unit insulation height h And unit creepage distance rain flash voltage E L .

[0010] As a further explanation of the present invention, the rain-flash voltage objective function value specifically includes a single objective function U 50% and U 50% 、E h and E L The weighted objective function Y of the three mul .

[0011] As a further illustration of the present invention, Y mul =0.4U 50% +0.3E h +0.3E L .

[0012] As a further explanation of the present invention, the gray correlation relationship between the decision variables and the optimization target is established by using the BP neural network algorithm, specifically including: The rain-flashover voltage test data samples under different shed structural parameters are divided into a training set, a validation set and a test set, and a BP neural network is established; Using the training set to train the BP neural network, using the trained BP neural network to perform simulation prediction on the test set, and using the verification set to perform verification; The error norm between the actual value in the validation set and the corresponding predicted value is used as a measure of the generalization ability of the network, so that the relative error between the actual value and the predicted value is less than the preset error, thereby meeting the network learning requirements.

[0013] As a further illustration of the present invention, the training is divided into three steps: forward propagation, calculation of loss and gradient, and back propagation to update model parameters; first, a BP neural network algorithm is used to establish a grey correlation relationship between the decision variables and the optimization target; then, the value range of the decision variables to be optimized is redefined and input into the BP neural network to obtain the rain-flashover voltage objective function value under different shed structural parameter combinations; finally, the structural parameters corresponding to the maximum value of the objective function are extracted as the optimal parameters.

[0014] As a further illustration of the present invention, the number of neurons in the input layer of the BP neural network model is 4, the number of neurons in the hidden layer is 6, and the number of neurons in the output layer is 5.

[0015] As a further illustration of the present invention, the hidden layer neurons use the S-type tangent function tansig() as the transfer function, the output layer neurons use the S-type logarithmic function logsig() as the transfer function, and the normalized data is between 0 and 1.

[0016] As a further illustration of the present invention, during the training process, the weights and thresholds are reversely learned and continuously iteratively corrected using the gradient descent method.

[0017] Compared with the prior art, the present invention has the following beneficial technical effects: The present invention takes the rain-flashover voltage and its decreasing gradient as the main evaluation indexes, and explores the optimal value range of the shed parameters of large-diameter composite post insulators through geometric analysis and neural network analysis of the optimized design of shed parameters.

[0018] Other features and advantages of the technical solution will be described in the subsequent description, and partly become apparent from the description, or understood by implementing the technical solution. The purpose and other advantages of the technical solution can be achieved and obtained through the structures specifically pointed out in the written description and the drawings.

[0019] The technical solution of the present technical solution is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The accompanying drawings are used to provide a further understanding of the technical solution and constitute a part of the specification. Together with the embodiments of the technical solution, they are used to explain the technical solution and do not constitute a limitation of the technical solution. In the accompanying drawings: Figure 1 A geometrical representation method for the design of the shed structure parameters in the present invention; Figure 2 The rain conditions under the conditions of different shed parameters in the present invention; Figure 3 The invention discloses a composite post insulator shed parameter optimization process based on rain-flash characteristics and neural network.

[0021] Figure 4 The large and small umbrella-shaped composite post insulator U of the present invention 50% 、E h and E L Comparison of predicted and actual values.

[0022] Figure 5 The large and small umbrella-shaped composite post insulator U of the present invention 50% and the predicted value under the weighted objective function.

[0023] Figure 6 The composite support insulator with one large and two small umbrellas in the present invention U 50% , E h and E L Comparison of predicted and actual values.

[0024] Figure 7 The composite support insulator with one large and two small umbrellas in the present invention U 50% and the predicted value under the weighted objective function. DETAILED DESCRIPTION

[0025] The preferred embodiments of the present technical solution are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present technical solution, and are not used to limit the present technical solution.

[0026] like Figure 3 As shown, the present invention provides a method for optimizing the parameters of composite post insulator sheds based on rain-flash characteristics and neural networks, comprising the following steps: S1. Carry out rain flashover tests on composite post insulators with different shed structures to obtain rain flashover voltage test data samples with different shed structure parameters.

[0027] In a feasible manner, before conducting the rain flashover test of the composite post insulator, a geometric analysis of the shed parameter optimization is required to make the shed structural parameters of the rain flashover test satisfy the following relationship:

[0028] Among them, λ is the extension difference between large and small umbrellas, P1 is the extension of large umbrella, P2 is the extension of small umbrella, S2 is the distance between large umbrella and large umbrella, and β is the inclination angle of the upper surface of the umbrella skirt.

[0029] Specifically, the geometric analysis process includes: Taking the "one large and one small" umbrella structure as an example, under certain conditions of shower angle and umbrella skirt inclination, the geometric relationship between the umbrella skirt structural parameters (including umbrella extension, umbrella spacing, umbrella inclination) and the shower direction can be expressed as Figure 1 Be described.

[0030] like Figure 1 As shown, let OE and KN be large umbrellas and FJ be small umbrellas. β is the inclination angle of the upper surface of the shed. That is: (1) After rearranging the above formula, we get: (2) The present invention believes that the umbrella skirt structure that satisfies or is close to the umbrella extension and umbrella spacing matching relationship described by formula (2) is more reasonable. In addition, Figure 1 The middle MN section is where rain can directly wet the umbrella, while KM is the dry area. η =MN / KN is used as a measure of the degree to which the umbrella is wetted by rain. From the figure, it can be deduced that: (3) η The value expresses the degree of wetting of the shed by rain. η The larger the value is, the larger the area of ​​the shed wetted by rainwater under certain rain and shed parameter conditions.

[0031] Increasing the distance between umbrellas makes it less likely that the air gap between umbrellas will be directly bridged by water droplets or water columns from the umbrella eaves, which can help improve the utilization rate of the insulation height and thus increase the rain flashover voltage.

[0032] In addition, the rain flashover characteristics of large-diameter post insulators are closely related to the wetting state of their surface when they encounter rain. When the rain rate is constant, if the insulating surface is more susceptible to the direct effect of rain, it will help the rain to gather on the surface of the shed, thereby increasing the flow and velocity of rain on the upper surface of the shed, and it is easier to form a bridged water curtain or water column between the sheds. For composite post insulators, when the insulator encounters rainfall, the upper shed has the function of shielding the shed below it, especially for insulators with large and small sheds, the small shed is shielded by the large shed above it and is not easily wetted by rain. If the extension and spacing of the large and small sheds can be reasonably designed, the surface area directly wetted by rain can be reduced when the large-diameter post insulator encounters rain, thereby improving the rain flashover characteristics. Figure 2 Three typical situations of umbrella extension and umbrella spacing are given, as well as the situation of the insulation surface being directly wetted by rainwater under different matching conditions.

[0033] If only the effect of umbrella extension is considered, that is, when the distance between the large umbrella and the large umbrella remains unchanged, the extension length of the small umbrella is increased, then the situation of rainwater wetting the insulator surface is changed to Figure 2 (a) Figure 2 (b) Gradually Figure 2 (c) Change.

[0034] In order to obtain the shed parameters corresponding to the maximum flashover voltage, this paper uses the BP neural network method to carry out the optimization analysis of shed parameters. The test results of the flashover characteristics of large-diameter composite post insulators with different shed structure parameters of one large and one small shed and one large and two small sheds are summarized, and the results are shown in Tables 1 and 2.

[0035] Table 1 Rain flash test results of one large and one small umbrella-shaped composite post insulator

[0036] Table 2 Rain flash test results of one large and two small umbrella-shaped composite post insulator

[0037] S2. Taking the shed structure parameters in the test data sample as decision variables and the rain-flashover voltage as the optimization target, the BP neural network algorithm is used to establish the grey correlation relationship between the decision variables and the optimization target.

[0038] As can be seen from Table 1, the optimization of the shed structural parameters of the composite post insulator with one large and one small umbrella includes four decision variables: average shed extension, difference between large and small shed extension, shed spacing, and pole diameter. As can be seen from Table 2, the optimization of the shed structural parameters of the composite post insulator with one large and two small umbrellas includes three decision variables: average shed extension, difference between large and small shed extension, and shed spacing. In terms of optimization objectives, if only the rain-flashover voltage is considered U50% , which is a single-objective optimization problem; if we want to consider the rain-flash voltage at the same time U 50% , Rain-flashover voltage per unit insulation height E h And unit creepage distance rain flash voltage E L , then it is a multi-objective optimization problem. By assigning different weights to different objective functions, the multi-objective optimization problem is converted into a single-objective optimization problem. U 50% , E h and E L The weight coefficients are 0.4, 0.3 and 0.3, that is, the optimization goal is Y mul =0.4 U 50% +0.3 E h +0.3 E L .

[0039] Since there is a nonlinear relationship between multiple decision variables and optimization targets and it cannot be described by analytical expressions, the present invention adopts BP neural network algorithm to perform parameter optimization.

[0040] The above-mentioned BP neural network algorithm is used to establish the grey correlation relationship between the decision variables and the optimization target, which specifically includes: The rain-flashover voltage test data samples under different shed structure parameters are divided into training set, validation set and test set, and a BP neural network is established. Use the training set to train the BP neural network, use the trained BP neural network to simulate and predict the test set, and use the verification set to verify; The error norm between the actual value in the validation set and the corresponding predicted value is used as a measure of the generalization ability of the network, so that the relative error between the actual value and the predicted value is less than the preset error, thereby meeting the network learning requirements.

[0041] Specifically, the training is divided into three steps: forward propagation, calculation of loss and gradient, and back propagation to update model parameters; first, the BP neural network algorithm is used to establish the grey correlation relationship between the decision variables and the optimization objectives; then, the value range of the decision variables to be optimized is redefined and input into the BP neural network to obtain the rain-flashover voltage objective function value under different shed structural parameter combinations; finally, the structural parameters corresponding to the maximum value of the objective function are extracted as the optimal parameters.

[0042] Specifically, the number of neurons in the input layer of the BP neural network model is 4, the number of neurons in the hidden layer is 6, and the number of neurons in the output layer is 5.

[0043] The error norm between the actual value in the validation set and the corresponding predicted value is used as a measure of the generalization ability of the network, and the fitness of the individual is calculated by the error norm. The hidden layer neurons use the S-type tangent function tansig() as the transfer function, and the output layer neurons use the S-type logarithmic function logsig() as the transfer function. The normalized data is between 0 and 1. During the training process, the weights and thresholds are reversely learned and continuously iterated and corrected using the gradient descent method.

[0044] S3. Redefine the value range of the decision variable to be optimized, input it into the trained BP neural network, and obtain the rain-flashover voltage objective function value under different shed structure parameter combinations.

[0045] S4. Extract the shed structural parameters corresponding to the maximum value of the objective function as the optimal parameters.

[0046] Optimization results and analysis of the above optimization methods (1) One large and one small umbrella-shaped composite post insulator Composite post insulators with one large and one small shed under different shed structure parameters obtained based on BP neural network U 50% , E h and E L The predicted and actual values ​​of Figure 4 As shown in the figure, the relative error between the network predicted value and the actual value is less than 0.1%, which meets the network learning requirements.

[0047] Redefine the value range and step size of the decision variables, and use the trained network model to set a single objective function for different structural parameter combinations U 50% And weighted objective function Y for prediction, the result is as follows Figure 5 As shown in the figure, we can get the single objective function U 50% The maximum value is 188.25 kV, and the corresponding average shed extension, difference between large and small shed extension, shed spacing and pole diameter are 92.27 mm, 26.77 mm, 90.12 mm and 559.45 mm respectively. U 50% , E h and E L Weighted objective function Y mul The maximum value is 124.48. U50% , E h and E L They are 180.05 kV, 144.04 kV / m and 30.83 kV / m respectively, and the corresponding average shed extension, difference in extension between large and small sheds, shed spacing and pole diameter are 90.26 mm, 29.28 mm, 88.56 mm and 549.1 mm respectively.

[0048] (2) One large and two small umbrella-shaped composite post insulator One large and two small umbrella-shaped composite post insulator under different umbrella skirt structural parameters obtained based on BP neural network U 50% , E h and E L The predicted and actual values ​​of Figure 6 As shown in the figure, the relative error between the network predicted value and the actual value is less than 0.1%, which meets the network learning requirements.

[0049] Redefine the value range and step size of the decision variables, and use the trained network model to set a single objective function for different structural parameter combinations U 50% And the weighted objective function Y mul Make predictions, and the results are as follows Figure 7 As shown in the figure, we can get the single objective function U 50% The maximum value is 239.36 kV, and the corresponding average shed extension, the difference between large and small shed extension and the shed spacing are 92.5 mm, 55 mm and 129 mm respectively; the weighted objective function Y mul The maximum value of is 164.339. U 50% , E h and E L They are 239.73 kV, 191.58 kV / m and 36.42 kV / m respectively, and the corresponding average shed extension, the extension difference between large and small sheds and the shed spacing are 92.7 mm, 46 mm and 111.2 mm respectively.

[0050] Combined with the rain flashover test results of large-diameter composite post insulators, the present invention uses the rain flashover voltage and its gradient as the main evaluation indicators, and recommends the range of values ​​of the shed parameters of large-diameter composite post insulators through traditional shed parameter optimization design geometric analysis and neural network analysis: 1) For large-diameter composite post insulators with a large-and-small umbrella structure, the recommended large umbrella spacing range is 90-92mm, and the average umbrella extension and umbrella extension difference ranges are 90-92mm and 29-31mm respectively.

[0051] 2) For large-diameter composite post insulators with a large-two-small umbrella structure, the recommended large umbrella spacing range is 108-110mm, and the average umbrella extension and umbrella extension difference ranges are 93-95mm and 45-47mm respectively.

[0052] Obviously, those skilled in the art can make various changes and modifications to the technical solution without departing from the spirit and scope of the technical solution. Thus, if these modifications and variations of the technical solution fall within the scope of the claims of the technical solution and their equivalents, the technical solution is also intended to include these modifications and variations.

Claims

1. A composite post insulator shed parameter optimization method based on rain-flash characteristics and neural network, characterized in that: include: Conduct rain flashover tests on composite post insulators with different shed structures, and obtain rain flashover voltage test data samples with different shed structure parameters; Taking the shed structural parameters in the test data sample as decision variables and the rain-flashover voltage as the optimization target, a BP neural network algorithm is used to establish a grey correlation relationship between the decision variables and the optimization target; Redefine the value range of the decision variables to be optimized and input them into the trained BP neural network to obtain the rain-flashover voltage objective function value under different shed structure parameter combinations. The shed structural parameters corresponding to the maximum value of the objective function are extracted as the optimal parameters.

2. The method for optimizing the parameters of composite post insulator sheds based on rain-flash characteristics and neural network according to claim 1, characterized in that: Before conducting the rain flashover test on composite post insulators, it is also necessary to conduct a geometric analysis of the shed parameters optimization so that the shed structural parameters of the rain flashover test meet the following relationship: ； Among them, λ is the extension difference between large and small umbrellas, P1 is the extension of large umbrella, P2 is the extension of small umbrella, S2 is the distance between large umbrella and large umbrella, and β is the inclination angle of the upper surface of the umbrella skirt.

3. The method for optimizing the parameters of composite post insulator sheds based on rain-flash characteristics and neural network according to claim 1, characterized in that: The composite post insulator rain flash test includes two cases: one large and one small umbrella and one large and two small umbrellas. The shed structural parameters of the composite post insulator with one large and one small shed include four decision variables: average shed extension, difference between large and small shed extension, shed spacing and pole diameter. The shed structural parameters of the composite post insulator with one large shed and two small sheds include three decision variables: average shed extension, difference between large and small shed extensions, and shed spacing.

4. The method for optimizing the parameters of composite post insulator sheds based on rain-flash characteristics and neural network according to claim 1, characterized in that: The optimization objectives include: rain flashover voltage U 50% , Rain flashover voltage E per unit insulation height h And unit creepage distance rain flash voltage E L .

5. The method for optimizing the parameters of composite post insulator sheds based on rain-flash characteristics and neural network according to claim 4, characterized in that: The rain-flash voltage objective function value specifically includes a single objective function U 50% and U 50% 、E h and E L The weighted objective function Y of the three mul .

6. The composite post insulator shed parameter optimization method based on rain-flash characteristics and neural network according to claim 5, characterized in that: Y mul =0.4U 50% +0.3E h +0.3E L 。 7. The composite post insulator shed parameter optimization method based on rain-flash characteristics and neural network according to claim 1, characterized in that: The use of the BP neural network algorithm to establish the grey association relationship between the decision variables and the optimization target specifically includes: The rain-flashover voltage test data samples under different shed structural parameters are divided into a training set, a validation set and a test set, and a BP neural network is established; Using the training set to train the BP neural network, using the trained BP neural network to perform simulation prediction on the test set, and using the verification set to perform verification; The error norm between the actual value in the validation set and the corresponding predicted value is used as a measure of the generalization ability of the network, so that the relative error between the actual value and the predicted value is less than the preset error, thereby meeting the network learning requirements.

8. The method for optimizing the parameters of composite post insulator sheds based on rain-flash characteristics and neural network according to claim 7, characterized in that: The training is divided into three steps: forward propagation, calculation of loss and gradient, and back propagation to update model parameters. First, the BP neural network algorithm is used to establish the gray correlation relationship between the decision variables and the optimization target. Then, the value range of the decision variables to be optimized is redefined and input into the BP neural network to obtain the rain-flashover voltage objective function value under different shed structural parameter combinations; finally, the structural parameters corresponding to the maximum value of the objective function are extracted as the optimal parameters.

9. The method for optimizing the parameters of composite post insulator sheds based on rain-flash characteristics and neural network according to claim 7, characterized in that: The BP neural network model has 4 neurons in the input layer, 6 neurons in the hidden layer, and 5 neurons in the output layer. The hidden layer neurons use the S-type tangent function tansig() as the transfer function, and the output layer neurons use the S-type logarithmic function logsig() as the transfer function. The normalized data is between 0 and 1.

10. The method for optimizing the parameters of composite post insulator sheds based on rain-flash characteristics and neural network according to claim 7, characterized in that: During the training process, the weights and thresholds are learned in reverse and are continuously iterated and corrected using the gradient descent method.

Images