Method and system for predicting the height of a jump during the de-icing of a power transmission line, and storage medium

Abstract

The application discloses a kind of transmission line ice shedding process jump height prediction method, system and storage medium, belong to power grid disaster prevention and mitigation field, method includes: with the jump height prediction network of well-trained jump height prediction;The training process includes: S1, obtains transmission line structure parameter and environmental load parameter;S2, based on the structure parameter and environmental load parameter, establishes the finite element model of transmission line, and the jump height in the ice shedding process of transmission line is calculated under single parameter change by control variable method;S3, it is judged whether the jump height exceeds the jump height threshold value set, if exceeds, the parameter corresponding to the jump height is handled into data set, input to jump height prediction network, and the loss training of jump height prediction network is carried out.The application can accurately predict the jump height of transmission line in the ice shedding process, improve the poor practicality, poor generalization ability and the shortcomings of large prediction error of empirical formula.

Description

Technical Field

[0001] This invention belongs to the field of power grid disaster prevention and mitigation, and more specifically, relates to a method, system and storage medium for predicting the jump height during the de-icing process of transmission lines. Background Technology

[0002] With the increasing density of power grid construction, many transmission lines inevitably pass through heavily icing areas with severe weather conditions such as high altitude, low temperature, and strong winds, leading to ice accumulation on the transmission lines. Under specific weather conditions or with manual de-icing, ice falling off the conductors can cause conductor jumps. If the distance between conductors is less than the minimum insulation gap, it can lead to conductor flashover, entanglement, or collision, causing tripping and potentially resulting in line breaks or collapses. The disasters caused by ice falling off and jumping on transmission lines pose a significant challenge to power grid construction.

[0003] Existing technologies primarily rely on statistical analysis of the jump height after de-icing of overhead transmission lines. A simplified calculation formula for the jump height, convenient for engineering applications, is derived and fitted using stress-sag variation relationships or energy conservation relationships, and then corrected using empirical coefficients. This formula only considers the influence of the sag difference before and after de-icing and the span distance on the jump height. This method considers only a single factor affecting the jump height, and because different transmission line systems have varying structures, the accuracy, practicality, and generalization performance of the calculated jump height are poor. More importantly, the sag after de-icing can only be obtained after the jump is completed, making it impossible for existing methods to predict the jump height during de-icing. Summary of the Invention

[0004] To address the shortcomings and improvement needs of existing technologies, this invention provides a method, system, and storage medium for predicting jump height during the de-icing process of transmission lines, with the aim of improving the accuracy and generalization ability of jump height prediction during the de-icing process of transmission lines.

[0005] To achieve the above objectives, according to one aspect of the present invention, a method for predicting the jump height during the de-icing process of a transmission line is provided, comprising:

[0006] The trained jump height prediction network is used to predict the jump height during the actual de-icing process of transmission lines; wherein, the jump height prediction network is trained through the following process:

[0007] S1. Obtain the structural parameters and environmental load parameters of the transmission line;

[0008] S2. Based on the structural parameters and environmental load parameters, a finite element model of the transmission line is established, and the jump height of the transmission line during the de-icing process is calculated by the control variable method under the change of a single parameter; wherein, the single parameter is any one of the structural parameters and environmental load parameters;

[0009] S3. Determine whether the jump height exceeds the set jump height threshold. If it does, process the parameters corresponding to the jump height into a dataset, input it into the jump height prediction network, and perform loss training on the jump height prediction network until the loss converges or the set number of training rounds is reached to obtain a trained jump height prediction network. The jump height prediction network is used to predict the jump height during the de-icing process of the transmission line.

[0010] Furthermore, in S3, the process of training the jump height prediction network using loss includes:

[0011] Divide the training set k in the dataset into equal parts;

[0012] Each training set is used as a validation set in turn, and the remaining k-1 sets are used as cross-validation training sets for cross-validation.

[0013] The average mean squared error between the jump height prediction network output and the validation set in each cross-validation is used as the loss function. The hyperparameters of the jump height prediction network are adjusted in reverse with the goal of minimizing the loss function until the loss converges or the set number of training rounds is reached, so as to obtain the optimal hyperparameters of the jump height prediction network and thus obtain the trained jump height prediction network.

[0014] Furthermore, the process of obtaining the optimal hyperparameters of the jump height prediction network includes:

[0015] S331. Determine the hyperparameter X that needs to be adjusted for the jump height prediction network;

[0016] S332. Randomly initialize multiple sets of the hyperparameters X, and train the jump height prediction network to obtain the value of the loss function f(X) in the corresponding training rounds;

[0017] S333. Fit the multiple sets of data combination points (X, f(X)) obtained in S332 to obtain a surrogate function that reflects the relationship between the loss function f(X) and the hyperparameter X.

[0018] S334. Calculate the minimum value of the surrogate function. Collect new data combination points in the vicinity of the minimum value and in the region that was not sampled during random initialization in S332. Refit the new data combination points with the multiple sets of data combination points (X, f(X)) obtained in S332 to obtain a new surrogate function.

[0019] S335. Using the new surrogate function as the surrogate function for the current iteration number, repeat S334 until the set iteration number is reached, and obtain the optimal hyperparameters of the jump height prediction network.

[0020] Furthermore, in S3, before processing the parameters corresponding to the jump height into a dataset, the following steps are also included:

[0021] The parameters corresponding to the jump height are evaluated for feature importance to obtain feature importance scores; parameters with feature importance scores below a set threshold are deleted to obtain parameters with a high degree of influence on the jump height.

[0022] Or / and also includes: calculating the correlation coefficient between any two parameters that have a high degree of influence on jump height; for two parameters whose correlation coefficient exceeds a set threshold, deleting one of the parameters.

[0023] Furthermore, the structural parameters of the transmission line include: conductor type, initial stress, suspension point height, number of conductor splits, span, and parameters of the insulator string; wherein, the parameters of the insulator string include the type, length, and included angle of the insulator string;

[0024] The environmental load parameters include ice thickness, de-icing rate, and pulsating wind load.

[0025] Furthermore, in S2, based on the aforementioned structural parameters and environmental load parameters, a finite element model of the transmission line is established, including:

[0026] S21. Based on the structural parameters, beam elements are used to simulate the conductor. The shape of the conductor under its own weight and the ice thickness is determined by iteratively applying a vertically downward gravitational acceleration, thereby achieving conductor shape finding.

[0027] S22. Generate a time series of fluctuating wind and convert the fluctuating wind speed into fluctuating wind pressure, which is then applied to the traverse after shape finding.

[0028] S23. Perform de-icing simulation of the transmission line according to the stated de-icing rate.

[0029] Furthermore, it also includes: a performance evaluation step for the jump height prediction network.

[0030] Calculate the difference between the predicted value output by the jump height prediction network and the jump height calculated in S2;

[0031] Based on the difference, the prediction performance of the jump height prediction network is evaluated using one or more evaluation indicators among the coefficient of variation, mean absolute error, mean square error, and coefficient of determination. Among them, the smaller the values ​​of mean absolute error and mean square error, the better the prediction performance; the closer the values ​​of coefficient of variation and coefficient of determination are to 1, the better the prediction performance.

[0032] Furthermore, after using the trained jump height prediction network to predict the jump height during the actual transmission line de-icing process, an early warning step is also included:

[0033] Determine the minimum safe distance between different phases of a transmission line;

[0034] If the predicted jump height is greater than or equal to the minimum safe distance, an early warning message will be issued.

[0035] According to another aspect of the present invention, a jump height prediction system for transmission line de-icing is provided, for performing the jump height prediction method for transmission line de-icing as described in any of the first aspects, comprising:

[0036] A jump height prediction module uses a trained jump height prediction network to predict the jump height during the actual de-icing process of transmission lines; wherein, the jump height prediction network is trained through the following modules:

[0037] The parameter acquisition module is used to acquire the structural parameters and environmental load parameters of the transmission line;

[0038] The calculation module is used to establish a finite element model of the transmission line based on the structural parameters and environmental load parameters, and to calculate the jump height of the transmission line during the de-icing process under the change of a single parameter using the controlled variable method; wherein, the single parameter is any one of the structural parameters and environmental load parameters;

[0039] The jump height prediction network training module is used to determine whether the jump height exceeds a set jump height threshold. If it does, the parameters corresponding to the jump height are processed into a dataset, input into the jump height prediction network, and the jump height prediction network is trained with loss until the loss converges or the set number of training rounds is reached to obtain a trained jump height prediction network. The jump height prediction network is used to predict the jump height during the de-icing process of transmission lines.

[0040] According to another aspect of the invention, a computer-readable storage medium is provided, including a stored computer program; when the computer program is executed by a processor, it controls the device in which the computer-readable storage medium is located to perform the method described in any of the first aspects.

[0041] In summary, the above-described technical solutions conceived in this invention can achieve the following beneficial effects:

[0042] (1) The method of the present invention determines the factors that affect the jump height during the de-icing process of transmission lines by establishing a finite element model, uses the corresponding influencing factors as the input dataset of the jump height prediction network, optimizes and trains the jump height prediction network, and uses the trained jump height prediction network to predict the jump height during the actual de-icing process of transmission lines. The present invention combines the theoretical calculation of the finite element model with the actual prediction, and the accuracy and generalization ability of the jump height prediction value obtained are high. At the same time, the transmission line structural parameters and environmental load parameters used in the present invention are all parameters before or during the de-icing process, which can realize the prediction of the jump height during the de-icing process of transmission lines.

[0043] (2) The loss function designed in this invention uses the average MSE of cross-validation of all folds as the loss function during training, which can make full use of the training set and avoid the problems of large training error and insufficient training caused by small samples. At the same time, using the average MSE of cross-validation of all folds as the loss function takes into account the MSE value obtained from training each fold, which can reflect the training effect of the model on the entire dataset.

[0044] (3) In the process of optimizing the hyperparameters of the jump height prediction network, this invention determines a surrogate function that reflects the relationship between the loss function f(X) and the hyperparameter X, uses the posterior distribution as the prior distribution for the next iteration, and continuously iterates to optimize the surrogate function so that the surrogate function continuously approximates the true functional relationship between the loss function f(X) and the hyperparameter X. Based on the determined loss function, the optimal hyperparameters are obtained. Compared with grid search, which requires exhaustively listing all hyperparameter combinations, this invention improves training efficiency. Compared with random search, which uses random sampling to select the optimal hyperparameter combination for all hyperparameter combinations, resulting in large differences in the optimization results each time, the model obtained by this invention has higher accuracy.

[0045] (4) As a preferred approach, by evaluating the feature importance of the parameters corresponding to the jump height, selecting the parameters that have a high degree of influence on the jump height, or / and performing feature correlation analysis on the parameters with a high degree of influence, and deleting redundant features before using them as the dataset, the prediction accuracy of the jump height prediction network can be further improved, and the computational efficiency of the method can be improved.

[0046] (5) When predicting the jump height, this invention considers the influence of various factors such as conductor type, initial stress, hanging point height, conductor split number, span, insulator string parameters, ice thickness, de-icing rate and pulsating wind load on the jump height, which can improve the accuracy of the prediction method.

[0047] (6) In the process of establishing the finite element model of the transmission line, considering that the conductor has certain bending and torsional resistance when the jumping motion occurs during the de-icing process, compared with the commonly used cable element to simulate the conductor, the present invention uses beam element to simulate the conductor, which has higher simulation accuracy.

[0048] (7) Preferably, the present invention can also evaluate the predictive performance of the jump height prediction network by performing performance evaluation on the jump height prediction network. Experiments have shown that the jump height prediction network of the present invention has a good predictive performance.

[0049] (8) Preferably, based on the method of the present invention, an early warning can also be issued. If the predicted jump height of the power transmission line during the de-icing process is greater than or equal to the minimum safe distance under specific wind speed, ice thickness and de-icing rate, an early warning message is issued.

[0050] In summary, this invention can quickly and accurately predict the jump height of transmission lines during the ice shedding process, overcoming the shortcomings of empirical formulas such as poor practicality, poor generalization ability, and large prediction errors. Attached Figure Description

[0051] Figure 1 This is a schematic diagram of the jump height prediction method for power transmission lines during de-icing provided by the present invention.

[0052] Figure 2 The flowchart of the method for predicting jump height during the de-icing process of transmission lines provided by the present invention is shown.

[0053] Figure 3 This is a schematic diagram illustrating the training process of the jump height prediction network of the present invention.

[0054] Figure 4 This is a schematic diagram of the cross-validation process of the present invention.

[0055] Figure 5 The figure shows the simulation results of an embodiment of the present invention.

[0056] Figure 6 This is a flowchart illustrating the practical application of the method of the present invention. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0058] like Figures 1-3As shown, the method for predicting jump height during the de-icing process of transmission lines according to the present invention mainly includes the following steps:

[0059] The trained jump height prediction network is used to predict the jump height during the actual de-icing process of transmission lines. Specifically, the actual transmission line structural parameters and environmental load parameters are input into the trained jump height prediction network to obtain the predicted jump height during the de-icing process.

[0060] The jump height prediction network is trained through the following steps:

[0061] S1. Obtain the structural parameters and environmental load parameters of the transmission line; In this embodiment of the invention, the structural parameters of the transmission line include: conductor type, initial stress, suspension point height, number of conductor splits, span, and parameters of the insulator string, wherein the parameters of the insulator string include the type, length, and included angle of the insulator string. The environmental load parameters include: ice load parameters and pulsating wind load, wherein the ice load includes ice thickness and de-icing rate.

[0062] S2. Based on the parameters obtained in S1 (transmission line structural parameters and environmental load parameters), a finite element model of the transmission line is established to simulate the dynamic response of the transmission line to de-icing under the influence of each parameter. The jump height during the de-icing process of the transmission line under the change of a single parameter is obtained by using the controlled variable method; where the single parameter is any one of the structural parameters and the environmental load parameters.

[0063] S3. Determine the degree of influence of each parameter on the jump height during the de-icing process of the transmission line. If the jump height during the de-icing process of the transmission line exceeds the set threshold under the change of a single parameter obtained in S2, it is considered that the parameter affects the jump height during the de-icing process of the transmission line; otherwise, it is considered that the parameter does not affect the jump height during the de-icing process of the transmission line.

[0064] After processing the parameters affecting the jump height during the de-icing process of transmission lines, the jump height prediction network is trained using the dataset with loss. The hyperparameters of the jump height prediction network are then adjusted in reverse until the loss converges or the set number of training rounds is reached, resulting in a trained jump height prediction network. The jump height prediction network is a deep neural network used to predict the jump height during the de-icing process of transmission lines.

[0065] Specifically, in S2, a finite element model of the transmission line is established based on the parameters obtained in S1, including:

[0066] S21. Based on the structural parameters of the transmission line, beam elements are used to simulate the conductor. By iteratively applying a vertically downward gravitational acceleration, the conductor's shape under its own weight and after icing is determined, and the maximum sag f of the conductor under its own weight and after icing is obtained. mThis achieves shape finding; among them, since the conductor has certain bending and torsional resistance when it jumps during the ice removal process, compared with the commonly used cable element to simulate the conductor, the embodiment of the present invention uses beam element to simulate the conductor, which has higher simulation accuracy.

[0067] S22. Generating a pulsating wind time series: In this embodiment of the invention, a series of trigonometric function sequences with random frequencies are used to simulate and generate a pulsating wind time series based on the harmonic synthesis method. The transmission line is divided into beam elements of equal length, and the pulsating wind speed is converted into pulsating wind pressure and applied to the transmission line after shape finding.

[0068] S23. Perform de-icing simulation of the transmission line according to the set de-icing rate.

[0069] Preferably, in step S3, before using the parameters affecting the jump height during the de-icing process of the transmission line as the dataset, the method further includes: evaluating the feature importance of the parameters affecting the jump height during the de-icing process of the transmission line to obtain the parameters with a high degree of influence on the jump height; and / or performing feature correlation analysis on the parameters with a high degree of influence obtained after the feature importance evaluation, deleting redundant features, and then using them as the dataset.

[0070] Specifically, the steps for evaluating the characteristic importance of parameters affecting the jump height during the de-icing process of transmission lines include:

[0071] S311. Using the Gain method in the XGBoost algorithm to determine the optimal splitting node, the feature importance of parameters affecting the jump height during the de-icing process of transmission lines is evaluated, and the feature importance score is obtained; where, the larger the Gain, the greater the splitting benefit, and the more important the feature corresponding to the parameter.

[0072] S312. Delete the parameters whose feature importance scores are lower than the set threshold to obtain the parameters that have a high degree of influence on jump height.

[0073] Specifically, for the generated pulsating wind time series, statistical features are first extracted from the pulsating wind time series, including: maximum wind speed, minimum wind speed, average wind speed, mean wind speed, variance, and average wind speed within a set time window; then, the importance of each statistical feature is evaluated. In this embodiment of the invention, the statistical features that have a high degree of influence on jump height include: the average wind speed in the pulsating wind time series and the average wind speed in the first 4.39 seconds.

[0074] In this embodiment of the invention, the threshold is set to 50. If the gain is greater than 50, the parameter is considered to have a high degree of influence on the jump height during the ice removal process. Furthermore, deleting parameters with scores less than 50 can further improve the accuracy of the model's predictions. In other embodiments, different thresholds can be selected based on the prediction accuracy. Additionally, in other embodiments, the weights from the XGBoost algorithm can also be used for feature importance evaluation.

[0075] The steps for performing feature correlation analysis on the parameters after feature importance evaluation include:

[0076] S321. Calculate the correlation coefficient between any two parameters that have a high degree of influence.

[0077] S322. For two parameters whose correlation coefficients exceed a set threshold, delete one of them to avoid redundant features and improve the computational efficiency of the method. Simultaneously, inputting highly correlated features amplifies the impact of noise, thus further improving prediction accuracy.

[0078] Specifically, in this embodiment of the invention, the correlation test threshold is set to 0.9, and the correlation coefficient between any two parameters with high influence is calculated using the Pearson correlation coefficient. In other embodiments, an appropriate test threshold is selected based on the prediction accuracy.

[0079] After the above-mentioned feature correlation analysis, the parameters obtained in this embodiment of the invention include conductor type, initial stress, hanging point height, number of conductor splits, span, parameters of insulator string, ice thickness, de-icing rate, average wind speed and average wind speed in the first 4.39 seconds.

[0080] The numerical features in the above parameters are normalized, and the symbolic features are encoded and then input into the jump height prediction network as a dataset. In this embodiment of the invention, the symbolic features include: the type of conductor and the type of insulator string. The symbolic features are converted into matrix-type numerical features using a one-hot encoding method.

[0081] The features of numerical classes are normalized using the following formula:

[0082]

[0083] Where, x i Let x represent the i-th sample value of feature x, min(x) represent the minimum value of the feature, and max(x) represent the maximum value of the feature.

[0084] Specifically, in S3, the dataset is divided into a supervised learning training set and a test set. During the loss training of the jump height prediction network, the training set is divided k equally, and each portion is used as a validation set in turn, while the remaining k-1 portions are used as cross-validation training sets. The average mean squared error between the predicted value and the validation set obtained in each cross-validation is used as the training optimization objective. Through repeated iterations, the minimum mean squared error is used as the objective function to determine the optimal hyperparameters, such as... Figure 4 As shown.

[0085] The loss function f(X) constructed in this invention is:

[0086]

[0087] Where k represents dividing the training set into k equal parts, MSE cv MSE represents the average mean squared error between the predicted value and the validation set obtained in each cross-validation step. i This represents the mean squared error between the predicted value and the validation set obtained in each cross-validation.

[0088] The optimization objective is to minimize the loss function.

[0089] As a preferred option, the jump height prediction network adopts the XGBoost neural network structure.

[0090] In this embodiment of the invention, taking the XGBoost neural network structure as an example, based on the loss function constructed above, the optimal hyperparameters of the jump height prediction network are determined using the following method:

[0091] S331. Determine the hyperparameters that need to be adjusted from all the hyperparameters of the XGBoost neural network, including:

[0092] For any given hyperparameter α, a range for the hyperparameter is randomly set, and different values ​​are taken within this range. The XGBoost neural network is then trained to obtain the learning curve (α, MSE) of the hyperparameter. i The learning curve has the following values: the horizontal axis represents the range of the hyperparameter, and the vertical axis represents the mean square error between the predicted value and the validation set in each cross-validation.

[0093] Calculate the magnitude of change of the learning curve corresponding to each hyperparameter within a set threshold. The hyperparameters whose magnitude of change exceeds the set threshold are the determined hyperparameters that need to be adjusted.

[0094] Compared to selecting hyperparameters to be adjusted based on existing experience, the hyperparameters determined by the method of this invention are more accurate. In this embodiment of the invention, through this step, six hyperparameters need to be adjusted, namely P1∈{1,2,3,…,500}, P2∈{1,2,3,…,30}, P3∈{1,2,3,…,20}, P4∈{0.01,0.02,0.03,…,0.3}, P5∈{0,0.1,0.2,…5+,P6∈{0.0.1,0.2,…,1}, and the optimization range of each parameter is initially determined, as shown in Table 1 below:

[0095] Table 1: Hyperparameters determined in the embodiments of the present invention and their initial value ranges

[0096]

[0097] In other embodiments, the corresponding hyperparameters that need to be optimized are determined based on the specific deep neural network selected by the jump height prediction network.

[0098] S332. Randomly initialize multiple sets of hyperparameters X = {P1, P2, P3, P4, P5, P6} to be adjusted, and train the jump height prediction network accordingly to obtain the value of the loss function f(X) for the corresponding training rounds.

[0099] S333. Fit the multiple sets of data combination points (X, f(X)) obtained in step S332 to obtain a surrogate function that reflects the relationship between the loss function f(X) and the hyperparameter X.

[0100] S334. Calculate the minimum value of the surrogate function. Collect multiple new data combination points in the vicinity of the minimum value and in the region that was not sampled during random initialization in step S332. Refit the new data combination points with the multiple sets of data combination points (X, f(X)) obtained in step S332 to obtain a new surrogate function.

[0101] S335. Using the new surrogate function as the surrogate function for the current iteration number, repeat step S334 until the set iteration number is reached to obtain the optimized hyperparameters.

[0102] In this embodiment of the invention, the number of iterations was set to 120, and the optimized values ​​of the six hyperparameters are shown in Table 2 below:

[0103] Table 2: Hyperparameter values ​​before and after optimization

[0104]

[0105] Using grid search and random search as a comparison, the experimental comparison data obtained under the same experimental conditions are shown in Table 3 below:

[0106] Table 3. MSE values ​​and time consumption of hyperparameters obtained by different optimization methods.

[0107]

[0108] In Table 3, the symbol " / " indicates a time consumption greater than 24 hours. It can be seen that when the number of hyperparameters to be optimized is small, the time and MSE values ​​of the three optimization methods are not significantly different. However, as the number of hyperparameters increases, grid search requires exhaustively listing all hyperparameter combinations, greatly reducing efficiency. Random search uses a random sampling method for all hyperparameter combinations to select the optimal combination, which is more efficient than grid search. However, due to the randomness of sampling, the results of each optimization vary considerably. The method of this invention is superior to the other two methods in both efficiency and performance. Furthermore, the method of this invention also includes a jump height prediction network performance evaluation step:

[0109] Calculate the difference between the predicted value output by the jump height prediction network and the jump height calculated by the finite element model in S2;

[0110] Based on this difference, the coefficients of variation (EVS), mean absolute error (MAE), mean squared error (MSE), and coefficient of determination (R²) are used to analyze the results. 2 One or more of the following are evaluation criteria for the jump height prediction network: MAE and MSE scores are lower, indicating better model prediction performance, and EVS and R... 2 The closer the value of is to 1, the better the performance of the prediction model.

[0111] The four evaluation indicators can be defined as follows:

[0112]

[0113] Where: y and These represent the actual value and the predicted value, respectively. The operator Var represents the variance.

[0114]

[0115]

[0116]

[0117] Where, N samples y represents the total number of samples in the training set. i and These represent the actual value (the jump height calculated using the finite element model in S2) and the predicted value for each sample, respectively. This represents the mean of the total sample.

[0118] The method of this invention can predict applicable structural parameters and environmental load parameters for different transmission lines, including different conductor types, initial stress, span, suspension point height; 4-split, 6-split, and 8-split conductors; "I" type insulator strings, "II" type insulator strings, and "V" type insulator strings; and the included angle, ice thickness, de-icing rate, and wind speed of different "V" type insulator strings.

[0119] The trained jump height prediction network is used to predict the jump height during actual power transmission line de-icing. In this embodiment of the invention, predictions are performed on 396 samples, and the results are as follows: Figure 5 As shown, the predicted value and the actual value are sufficiently close, and the evaluation index R... 2 The values ​​for EVS and MAE were 0.991 and 0.991, respectively, and the values ​​for MAE and MSE were 0.381 and 0.664, respectively.

[0120] Furthermore, the method of the present invention, after predicting the jump height during the actual de-icing process of transmission lines using a trained jump height prediction network, also includes an early warning step:

[0121] Determine the minimum safe distance between different phases of a transmission line based on different transmission line systems;

[0122] If the predicted jump height of the transmission line during de-icing is greater than or equal to the minimum safe distance under specific wind speeds, ice thicknesses, and de-icing rates, an early warning will be issued. Simultaneously, power grid maintenance personnel can also obtain the predicted jump height and early warning information by manually inputting the transmission line structural parameters and environmental load parameters into the prediction system. Figure 6 As shown.

[0123] According to another aspect of the present invention, a jump height prediction system for power transmission line de-icing process is provided, mainly comprising:

[0124] The parameter acquisition module is used to acquire the structural parameters of the transmission line and the environmental load parameters.

[0125] The calculation module is used to establish a finite element model of the transmission line based on structural parameters and environmental load parameters, and to calculate the jump height of the transmission line during the de-icing process under the change of a single parameter using the controlled variable method; where the single parameter is any one of the structural parameters and environmental load parameters.

[0126] The jump height prediction network training module is used to determine whether the jump height exceeds the set jump height threshold. If it does, the parameters corresponding to the jump height are processed into a dataset and input into the jump height prediction network. The jump height prediction network is then trained with loss. The hyperparameters of the jump height prediction network are adjusted in reverse until the loss converges or the set number of training rounds is reached, and a trained jump height prediction network is obtained. The jump height prediction network is a deep neural network used to predict the jump height during the de-icing process of transmission lines.

[0127] The jump height prediction module uses a trained jump height prediction network to predict the jump height during the actual de-icing process of transmission lines.

[0128] Each of the above modules is used to perform the corresponding steps in the jump height prediction method during the de-icing process of transmission lines in the above embodiments.

[0129] According to another aspect of the present invention, a computer-readable storage medium is provided, including a stored computer program; when the computer program is executed by a processor, it controls the device where the computer-readable storage medium is located to perform each step of the jump height prediction method during the de-icing process of the transmission line in the above embodiments.

[0130] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting jump height during the de-icing process of transmission lines, characterized in that, include: The trained jump height prediction network is used to predict the jump height during the actual de-icing process of transmission lines; wherein, the jump height prediction network is trained through the following process: S1. Obtain the structural parameters and environmental load parameters of the transmission line; S2. Based on the structural parameters and environmental load parameters, a finite element model of the transmission line is established, and the jump height of the transmission line during the de-icing process is calculated by the control variable method under the change of a single parameter; wherein, the single parameter is any one of the structural parameters and environmental load parameters; S3. Determine whether the jump height exceeds the set jump height threshold. If it does, process the parameters corresponding to the jump height into a dataset, input it into the jump height prediction network, and perform loss training on the jump height prediction network until the loss converges or the set number of training rounds is reached to obtain the trained jump height prediction network. The jump height prediction network is used to predict the jump height during the de-icing process of the transmission line. The structural parameters of the transmission line include: conductor type, initial stress, suspension point height, number of conductor splits, span, and parameters of the insulator string; wherein, the parameters of the insulator string include the type, length, and included angle of the insulator string; The environmental load parameters include ice thickness, de-icing rate, and pulsating wind load. In S2, based on the structural parameters and environmental load parameters, a finite element model of the transmission line is established, including: S21. Based on the structural parameters, beam elements are used to simulate the conductor. The shape of the conductor under its own weight and the ice thickness is determined by continuously applying a vertically downward gravitational acceleration, thereby achieving conductor shape finding. S22. Generate a time series of fluctuating wind and convert the fluctuating wind speed into fluctuating wind pressure, which is then applied to the traverse after shape finding. S23. Perform de-icing simulation of the transmission line according to the stated de-icing rate; In S3, the process of training the jump height prediction network using loss includes: Divide the training set k in the dataset into equal parts; Each training set is used as a validation set in turn, and the remaining k-1 sets are used as cross-validation training sets for cross-validation. The average mean square error between the jump height prediction network output by each cross-validation and the validation set is used as the loss function. The hyperparameters of the jump height prediction network are adjusted in reverse with the goal of minimizing the loss function until the loss converges or the set number of training rounds is reached, thus obtaining the optimal hyperparameters of the jump height prediction network. The process of obtaining the optimal hyperparameters of the jump height prediction network includes: S331. Determine the hyperparameter X that needs to be adjusted for the jump height prediction network; S332. Randomly initialize multiple sets of the hyperparameters X, and train the jump height prediction network to obtain the loss function for the corresponding training rounds. The value; S333, Combining multiple sets of data points obtained in S332 ( , By fitting the data, a loss function can be obtained. With hyperparameters A proxy function for the relationship between them; S334. Calculate the minimum value of the surrogate function. Collect new data combination points in the vicinity of the minimum value and in the region that was not sampled during random initialization in S332. Use the new data combination points and the multiple sets of data combination points obtained in S332 ( , The fitting process is repeated to obtain a new surrogate function; S335. Using the new surrogate function as the surrogate function for the current iteration number, repeat S334 until the set iteration number is reached, and obtain the optimal hyperparameters of the jump height prediction network.

2. The method according to claim 1, characterized in that, In S3, before processing the parameters corresponding to the jump height into a dataset, the following steps are also included: The parameters corresponding to the jump height are evaluated for feature importance to obtain feature importance scores; parameters with feature importance scores below a set threshold are deleted to obtain parameters with a high degree of influence on the jump height. Or / and also includes: calculating the correlation coefficient between any two parameters that have a high degree of influence on jump height; for two parameters whose correlation coefficient exceeds a set threshold, deleting one of the parameters.

3. The method according to claim 1, characterized in that, Also includes: The performance evaluation steps for the jump height prediction network are as follows: Calculate the difference between the predicted value output by the jump height prediction network and the jump height calculated in S2; Based on the difference, the prediction performance of the jump height prediction network is evaluated using one or more evaluation indicators among the coefficient of variation, mean absolute error, mean square error, and coefficient of determination. Among them, the smaller the values ​​of mean absolute error and mean square error, the better the prediction performance; the closer the values ​​of coefficient of variation and coefficient of determination are to 1, the better the prediction performance.

4. The method according to claim 1 or 3, characterized in that, After using the trained jump height prediction network to predict the jump height during the actual transmission line de-icing process, an early warning step is also included: Determine the minimum safe distance between different phases of a transmission line; If the predicted jump height is greater than or equal to the minimum safe distance, an early warning message will be issued.

5. A jump height prediction system for power transmission lines during de-icing, characterized in that, A method for predicting jump height during de-icing of transmission lines as described in any one of claims 1-4, comprising: A jump height prediction module uses a trained jump height prediction network to predict the jump height during the actual de-icing process of transmission lines; wherein, the jump height prediction network is trained through the following modules: The parameter acquisition module is used to acquire the structural parameters and environmental load parameters of the transmission line; The calculation module is used to establish a finite element model of the transmission line based on the structural parameters and environmental load parameters, and to calculate the jump height of the transmission line during the de-icing process under the change of a single parameter using the controlled variable method; wherein, the single parameter is any one of the structural parameters and environmental load parameters; The jump height prediction network training module is used to determine whether the jump height exceeds a set jump height threshold. If it does, the parameters corresponding to the jump height are processed into a dataset, input into the jump height prediction network, and the jump height prediction network is trained with loss until the loss converges or the set number of training rounds is reached to obtain a trained jump height prediction network. The jump height prediction network is used to predict the jump height during the de-icing process of transmission lines.

6. A computer-readable storage medium, characterized in that, Includes a stored computer program; when executed by a processor, the computer program controls the device containing the computer-readable storage medium to perform the method according to any one of claims 1-4.

Images