Method, system and equipment for predicting deicing jump response characteristic parameters of isolated power transmission line and medium

Abstract

The invention relates to the technical field of power transmission line deicing jump prediction, and discloses a prediction method, system, equipment and medium for isolated-span power transmission line deicing jump response characteristic parameters, and the method comprises the steps: constructing a finite element simulation model considering various characteristics, carrying out deicing jump simulation under various typical working conditions, and obtaining sample data; according to the method, input and output parameters are processed and subjected to feature extraction, a two-stage stacked integrated learning model is constructed and trained and optimized, and the problems that in the prior art, a power transmission line deicing jump prediction method is low in accuracy, large in uncertainty due to the influence of various factors, large in difference with an actual result, complex in calculation and the like are solved.

Description

Technical Field

[0001] This invention relates to the field of transmission line de-icing jump prediction technology, and in particular to a method, system, device and medium for predicting the de-icing jump response characteristic parameters of isolated span transmission lines. Background Technology

[0002] With my country's rapid economic development, residents' electricity demand and consumption have increased rapidly. To ensure residents' electricity needs are met, the country has vigorously developed ultra-high voltage (UHV) technology. In particular, the electricity needs of residents living in areas with complex terrain require UHV technology. However, areas with complex terrain generally have high atmospheric humidity and cold winters, creating an environment conducive to icing on transmission lines. This makes lines in areas with complex terrain prone to icing disasters.

[0003] When ice detaches from transmission lines, the lines may jump or even gallop due to stress changes, reducing the phase spacing. When this spacing is less than the minimum insulation gap length, it can easily lead to accidents such as insulator flashover, transmission line entanglement and collision, and power outages. In severe cases, it can even cause large-scale power disasters such as line breaks, tower collapses, aging and damage to insulators and hardware, and large-scale power outages. Therefore, predicting transmission line ice-shedding jumps is of great significance for preventing major ice disasters.

[0004] Currently, mainstream methods for predicting de-icing jumps in transmission lines include dynamic model research, growth model research, theoretical algorithm research, and dynamic response analysis. The dynamic model method adds concentrated mass to unit nodes to simulate icing, then uses finite element analysis and initial strain to find the initial configuration, resulting in high accuracy and detailed analysis of the de-icing response. The growth model method, based on an icing growth model with multiple parameters, is affected by climate and icing morphology, leading to higher uncertainty. The theoretical algorithm calculates the maximum de-icing jump height using a simplified model based on energy conservation and force analysis, but this method differs significantly from actual results and is computationally complex. Finally, the dynamic response analysis method uses finite element analysis to build the transmission line model, ultimately performing finite element analysis based on implicit dynamics. Summary of the Invention

[0005] In view of the aforementioned existing problems, the present invention is proposed.

[0006] Therefore, this invention provides a method, system, device, and medium for predicting the de-icing jump response characteristic parameters of isolated span transmission lines, which can solve the problems of low accuracy, large uncertainty due to various factors, large difference from actual results, and complex calculation in the existing methods for predicting de-icing jumps of transmission lines.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for predicting the de-icing jump response characteristic parameters of isolated span transmission lines, comprising: A finite element simulation model of an isolated span transmission line was constructed, taking into account the mechanical properties of the conductor-insulator string connection node, the non-uniform distribution of icing load, and the plastic deformation of the conductor. Based on the finite element simulation model, de-icing jump simulations were performed under various typical working conditions to obtain the corresponding response characteristic parameters as sample data. A global sensitivity analysis of the input parameters is performed to identify key input parameters, and multi-location, multi-dimensional feature extraction is performed on the output response characteristic parameters to form a multi-dimensional output index that includes dynamic unbalanced tension, static unbalanced tension, and maximum jumping height at multiple specified locations. Based on the key input parameters and multidimensional output indicators, a training dataset covering the actual parameter space of the project is generated, and the training dataset is preprocessed by outlier removal, feature expansion, stratified sampling and hybrid standardization. A two-level stacked ensemble learning model is constructed, and the model is trained using a preprocessed training dataset to obtain a predictive model for predicting the de-icing jump response characteristics of isolated span transmission lines.

[0008] As a preferred embodiment of the prediction method for the de-icing jump response characteristic parameters of isolated span transmission lines according to the present invention, the multidimensional output indicators include: The dynamic and static unbalanced tensions formed by the tension difference of the conductor before and after ice removal, and the maximum ice jumping heights corresponding to the quarter, half, three-quarters and suspension point positions, respectively; The multi-dimensional feature extraction includes smoothing the time history curves of tension and jumping height, and sampling to obtain key features in the dynamic response process.

[0009] As a preferred embodiment of the prediction method for the de-icing jump response characteristic parameters of isolated span transmission lines according to the present invention, the key input parameters include: Ice thickness, de-icing rate, span, elevation difference, conductor cross-sectional area, number of split conductors, and de-icing method; The feature expansion includes constructing interaction features and higher-order features between the key input parameters, expanding the original input feature space into a higher-dimensional feature set. The hybrid standardization preprocessing includes Z-score standardization for continuous physical parameters and one-hot encoding for classification parameters such as de-icing methods.

[0010] As a preferred embodiment of the prediction method for the de-icing jump response characteristic parameters of isolated span transmission lines according to the present invention, the two-level stacked ensemble learning model includes: Basic model layer and meta-model layer; The base model layer includes a random forest model, a gradient boosting tree model, and an extreme gradient boosting model, which output preliminary prediction results for the multidimensional output index in parallel. The meta-model layer uses an elastic network regression model to weight and fuse the prediction results output by the basic model layer to generate the final prediction value.

[0011] This preferred solution employs a two-level stacked ensemble learning model that fully leverages the advantages of different base models.

[0012] As a preferred embodiment of the prediction method for the de-icing jump response characteristic parameters of isolated span transmission lines according to the present invention, wherein: the training process of the two-level stacked ensemble learning model adopts a nested cross-validation strategy and combines an early stopping mechanism to control the number of training rounds; Training is automatically terminated when the error metric on the validation set does not show significant improvement within a set number of consecutive rounds to prevent overfitting.

[0013] As a preferred embodiment of the prediction method for the de-icing jump response characteristic parameters of isolated span transmission lines described in this invention, the method employs a Bayesian optimization algorithm to adaptively optimize the hyperparameters of the two-level stacked ensemble learning model, with the optimization objective being to maximize the average determination coefficient of the multidimensional output index on the validation set. The model performance is quantitatively evaluated using a comprehensive evaluation system that includes three categories of indicators: goodness of fit, error, and robustness.

[0014] As a preferred embodiment of the prediction method for the de-icing jump response characteristic parameters of isolated span transmission lines described in this invention, the various typical operating conditions include standard operating conditions, elevation difference operating conditions, and complex de-icing operating conditions.

[0015] Secondly, the present invention provides a prediction system for the de-icing jump response characteristic parameters of isolated span transmission lines, comprising: The simulation model building module is used to build a finite element simulation model of an isolated span transmission line that considers the mechanical properties of the conductor-insulator string connection node, the non-uniform distribution of ice load, and the plastic deformation of the conductor. The sample data acquisition module is used to perform ice-breaking jump simulations under various typical working conditions based on the finite element simulation model, and obtain the corresponding response characteristic parameters as sample data. The indicator acquisition module is used to perform global sensitivity analysis on input parameters, determine key input parameters, and extract multi-location and multi-dimensional features from the output response characteristic parameters to form a multi-dimensional output indicator that includes dynamic unbalance tension, static unbalance tension, and the maximum jumping height at multiple specified locations. The data processing module is used to generate a training dataset covering the actual parameter space of the project based on the key input parameters and multidimensional output indicators, and to perform outlier removal, feature expansion, stratified sampling and hybrid standardization preprocessing on the training dataset. The prediction model building module is used to build a two-level stacked ensemble learning model. The model is trained using a preprocessed training dataset to obtain a prediction model for predicting the de-icing jump response characteristic parameters of isolated span transmission lines.

[0016] Thirdly, the present invention provides an electronic device including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method described above.

[0017] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described above.

[0018] Compared with existing technologies, the beneficial effects of this invention are that it proposes a method for predicting the de-icing jump response characteristic parameters of isolated transmission lines. By constructing a finite element simulation model that considers multiple characteristics, sample data is obtained through de-icing jump simulations under various typical operating conditions. Input and output parameters are processed and features are extracted. A two-level stacked ensemble learning model is constructed, trained, and optimized. This overcomes the problems of low accuracy, large uncertainty due to multiple factors, large differences from actual results, and computational complexity in existing transmission line de-icing jump prediction methods. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 The flowchart illustrates a method for predicting the de-icing jump response characteristic parameters of an isolated span transmission line, as provided in an embodiment of the present invention.

[0021] Figure 2This diagram illustrates the key output parameter extraction for line de-icing based on feature extraction, which is part of a method for predicting the de-icing jump response characteristic parameters of an isolated span transmission line according to an embodiment of the present invention.

[0022] Figure 3 This is a two-level stacked structure model framework diagram of a method for predicting the de-icing jump response characteristic parameters of an isolated span transmission line according to an embodiment of the present invention.

[0023] Figure 4 This is an internal structural diagram of an electronic device for predicting the de-icing jump response characteristic parameters of an isolated span transmission line, as provided in an embodiment of the present invention. Detailed Implementation

[0024] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0025] Example 1, referring to Figure 1 This is the first embodiment of the present invention, which provides a method for predicting the de-icing jump response characteristic parameters of isolated span transmission lines, including: This invention provides a method that can effectively solve the problems mentioned above. The following will describe in detail how to predict the de-icing jump response characteristic parameters of the isolated span transmission line in conjunction with several embodiments. Figure 1 A flowchart of a method for predicting the de-icing jump response characteristic parameters of an isolated span transmission line is shown, including: S101, Construct a finite element simulation model of an isolated span transmission line that considers the mechanical properties of the conductor-insulator string connection node, the non-uniform distribution of icing load, and the plastic deformation of the conductor. In this embodiment of the invention, the finite element simulation model is constructed using COMSOL software, and a correction factor of 1.05–1.12 for non-uniform ice distribution and a correction term of 0.98 for conductor plastic deformation are introduced. The span is divided into 20 calculation units, and the segmented equivalent concentrated force method is used to simulate the ice load.

[0026] In some embodiments, when constructing a finite element simulation model of an isolated span transmission line, it is necessary to accurately reproduce the mechanical behavior of the connection node between the conductor and the insulator string. This node is subjected to complex tension, compression, bending and shear coupling during the ice-free jumping process. For example, under the standard working condition of a span of 475 meters and a height difference of 0 meters, the conductor is connected to the insulator string through a suspension clamp. There is a local stress concentration phenomenon at the connection. The model uses nonlinear spring elements to simulate the constraint stiffness of the clamp on the conductor and introduces rotational degrees of freedom to reflect the flexibility characteristics of the actual connection.

[0027] In some embodiments, a refined model is used to address the non-uniform distribution characteristics of icing loads. Icing does not cover the span uniformly, but rather thickens or thins locally due to wind direction, sunlight, and terrain. For example, in complex de-icing conditions, on a line with a span of 500 meters and a height difference of 30 meters, the ice thickness on the windward conductor can reach 25 mm, while it is only 18 mm on the leeward side. The model introduces a correction coefficient of 1.05 to 1.12 for non-uniform icing distribution, divides the entire span into 20 calculation units, assigns ice thickness to each unit independently based on azimuth and height, and applies the load using the segmented equivalent concentrated force method to ensure that the spatial distribution of the load is consistent with the actual meteorological conditions.

[0028] In some embodiments, the plastic deformation effect of the conductor under long-term icing and de-icing impact is considered. After the conductor material exceeds the yield strength, it will generate irreversible residual strain, which will affect the subsequent jump response. For example, in the case of elevation difference (span 300 meters, elevation difference 50 meters, icing 15 mm), the conductor tension drops sharply at the moment of de-icing, causing violent vibration. The stress in some sections briefly exceeds the yield limit of the aluminum stranded wire. The model introduces a conductor plastic deformation correction term of 0.98, and combines it with the improved iterative catenary equation for initial shape finding, so that the initial state of the simulation is closer to the actual geometric configuration after stress.

[0029] The conductor plastic deformation correction term refers to the proportional adjustment of the conductor's elastic modulus or initial sag during the form-finding stage, in order to equivalently characterize the influence of historical plastic strain on the current configuration.

[0030] It should be noted that step S101 constructs a high-fidelity finite element simulation model, which can accurately reflect the physical process of real ice-breaking jump, laying the foundation for the subsequent generation of reliable sample data.

[0031] S102, based on the finite element simulation model, performs ice-breaking jump simulations under various typical working conditions to obtain the corresponding response characteristic parameters as sample data; In the embodiments of the present invention, various typical working conditions include standard working conditions, elevation difference working conditions, and complex de-icing working conditions.

[0032] In some embodiments, based on the established finite element simulation model, the system conducts de-icing jump simulations under three typical operating conditions to comprehensively cover possible operating scenarios in actual engineering. For example, in the standard operating condition, the span is set to 475 meters, the elevation difference to 0 meters, the ice thickness to 20 millimeters, and the de-icing rate to 100%, simulating the symmetrical jump response caused by synchronous de-icing of the entire span; in the elevation difference operating condition, the span is set to 300 meters, the elevation difference to 50 meters, the ice thickness to 15 millimeters, and the de-icing rate to 70%, simulating the asymmetric tension release and skew jump behavior caused by the terrain elevation difference on mountain lines; in the complex de-icing operating condition, the span is set to 500 meters, the elevation difference to 30 meters, and the ice thickness to 25 millimeters, and a chain de-icing method is adopted, that is, de-icing occurs segment by segment from one end to the other, simulating the continuous de-icing process caused by local temperature changes or wind disturbances under actual meteorological conditions.

[0033] In some embodiments, the simulation model is driven by the above-mentioned working conditions to record the dynamic response data of key positions of the conductor in real time, including the time history curves of conductor tension before and after de-icing and the time history of vertical displacement at each monitoring point. This allows for the extraction of six response characteristic parameters: dynamic unbalanced tension, static unbalanced tension, and the maximum ice-jumping height at the quarter span, half span, three-quarter span, and suspension point. For example, in the simulation of complex de-icing conditions, the maximum ice-jumping height at the conductor midpoint (L / 2) reaches 2.83 meters, while the peak value of the dynamic unbalanced tension is 18.7 kN. These values ​​are all fully saved as sample labels for subsequent machine learning models.

[0034] Among them, the response characteristic parameters refer to key physical quantities that can quantitatively characterize the severity of the ice-breaking jump and the stress state of the structure, and directly determine whether the line will experience safety accidents such as phase-to-phase flashover or hardware damage.

[0035] It should be noted that step S102 involves simulating and acquiring sample data under various typical working conditions to ensure that the data covers the actual engineering scenario and provides a comprehensive and representative training basis for subsequent modeling.

[0036] S103 performs a global sensitivity analysis on the input parameters to determine the key input parameters, and extracts multi-location and multi-dimensional features from the output response characteristic parameters to form a multi-dimensional output index that includes dynamic unbalanced tension, static unbalanced tension, and the maximum jumping height at multiple specified locations. In this embodiment of the invention, the global sensitivity analysis employs the Sobol method combined with 10,000 Monte Carlo simulations.

[0037] In this embodiment of the invention, key input parameters include: Ice thickness, de-icing rate, span, elevation difference, conductor cross-sectional area, number of split conductors, and de-icing method; Feature expansion includes constructing interaction features and higher-order features between key input parameters, expanding the original input feature space into a higher-dimensional feature set; Hybrid normalization preprocessing includes Z-score normalization for continuous physical parameters and one-heat encoding for classification parameters such as de-icing methods.

[0038] In embodiments of the present invention, the multidimensional output indicators include: The dynamic and static unbalanced tensions formed by the tension difference of the conductor before and after ice removal, and the maximum ice jumping heights corresponding to the quarter, half, three-quarters and suspension point positions, respectively; Multidimensional feature extraction includes smoothing the time history curves of tension and jumping height, and sampling to obtain key features in the dynamic response process.

[0039] Specifically, in some embodiments, after obtaining a large number of ice-breaking jump simulation samples, a global sensitivity analysis based on the Sobol index is first performed on the input parameters to quantify the independent influence and interactive contribution of each input variable on the output response characteristics. For example, 1200 simulations were run within a parameter combination space of 300 to 500 meters span, 10 to 25 millimeters ice thickness, and 30% to 100% ice-breaking rate. The first-order sensitivity index of ice thickness to the maximum ice-breaking height was calculated to be 0.42, while its second-order interactive sensitivity index with the ice-breaking rate reached 0.18, indicating a significant coupling effect between the two. The total effect index of elevation difference on dynamic unbalanced tension was the highest, reaching 0.61, indicating that terrain factors are the core variable controlling tension abrupt changes.

[0040] In some embodiments, key input parameters are determined accordingly, including ice thickness, de-icing rate, span, elevation difference, conductor cross-sectional area, number of split conductors, and de-icing method. These parameters collectively govern the dynamic behavior of de-icing jump.

[0041] In some embodiments, multi-location and multi-dimensional feature extraction is further performed on the output response characteristic parameters. This not only records the static and dynamic unbalanced tension at the suspension point, but also simultaneously extracts the maximum vertical ice-jumping height at three typical positions: one-quarter (L / 4), the midpoint (L / 2), and three-quarters (3L / 4), forming a six-dimensional output index vector. For example, in a simulation with a height difference of 50 meters, 20 millimeters of ice cover, and an 80% ice removal rate, the output vector is [16.3kN, 9.7kN, 1.92m, 2.45m, 2.18m], corresponding to the dynamic unbalanced tension, the static unbalanced tension, and the ice-jumping heights at L / 4, L / 2, and 3L / 4, respectively.

[0042] In some embodiments, feature extensions are carried out on this basis to construct interactive features and higher-order features between key input parameters, expanding the original seven-dimensional input space into a higher-dimensional feature set. For example, 21 new features are generated, such as the product of ice thickness and de-icing rate, the square of span, and the ratio of elevation difference to the number of split conductors, to enhance the model's ability to capture nonlinear relationships.

[0043] In some embodiments, the expanded feature set is then subjected to hybrid normalization preprocessing. Continuous physical parameters such as ice thickness, span, height difference, and conductor cross-sectional area are normalized using Z-score to make their mean zero and standard deviation one. The classification parameter of de-icing method is encoded using one-hot encoding, which is converted into multiple binary indicator variables, such as "synchronous de-icing" encoded as [1, 0, 0], "chain de-icing" encoded as [0, 1, 0], and "random local de-icing" encoded as [0, 0, 1].

[0044] The hybrid normalization preprocessing here refers to adopting normalization strategies that are adapted to the data properties of different types of input variables, so as to ensure that the subsequent machine learning model can effectively learn physical laws and pattern associations on a unified scale.

[0045] It should be noted that step S103 uses global sensitivity analysis to screen key input parameters and extract multi-location, multi-dimensional output indicators, which effectively reduces input redundancy, clarifies prediction targets, and improves the learning efficiency and physical interpretability of subsequent models.

[0046] S104 generates a training dataset covering the actual parameter space of engineering based on key input parameters and multidimensional output indicators, and performs outlier removal, feature expansion, stratified sampling and hybrid standardization preprocessing on the training dataset. In some embodiments, based on the determined key input parameters and multidimensional output indicators, a comprehensive training dataset is generated in the actual engineering parameter space using the Latin hypercube sampling method. This parameter space is strictly limited to the range allowed by the current transmission line design specifications. For example, the ice thickness is 10 to 30 mm, the de-icing rate is 30% to 100%, the span is 200 to 600 meters, the elevation difference is 0 to 80 meters, the conductor cross-sectional area is 300 to 720 square millimeters, the number of split conductors is 2 to 4, and the de-icing method covers three types: synchronous de-icing, chain de-icing, and random local de-icing. A total of 2,000 sets of input combinations are generated, and the aforementioned finite element simulation model is called to calculate the corresponding six-dimensional output indicators one by one.

[0047] In some embodiments, outlier removal is first performed on the obtained training dataset. Unreasonable samples are identified and removed through a dual check of box plot criteria and physical rationality. For example, a sample shows a span of 300 meters and an ice thickness of only 12 millimeters, but produces an ice jumping height of 3.5 meters, which is more than twice the standard deviation of the average of similar working conditions and violates the theoretical upper limit under the conductor stiffness constraint. It is determined to be an outlier caused by numerical divergence and is removed. Another example is that the dynamic unbalanced tension has a negative value, which obviously violates the mechanical logic and is also excluded.

[0048] In some embodiments, after removing outliers, the remaining valid samples are feature-expanded to construct second-order interaction terms and square terms between key input parameters. For example, 18 higher-order features such as "ice thickness × de-icing rate", "span²", and "elevation difference / number of split conductors" are generated, expanding the input dimension from the original 7 dimensions to 25 dimensions to explicitly express the nonlinear coupling effect.

[0049] In some embodiments, stratified sampling is then performed on the expanded dataset, dividing the samples into three subsets based on the classification variable of the de-icing method. In each subset, training, validation and test sets are extracted proportionally to ensure that the distribution of various de-icing modes is balanced in each data subset, and to avoid model bias due to insufficient samples of a certain type of de-icing method.

[0050] In some embodiments, the data after stratified sampling is preprocessed using hybrid normalization. The ice thickness, span, height difference, conductor cross-sectional area and their derived continuous high-order features are uniformly normalized using Z-score to make them conform to a standard normal distribution. The de-icing method and its three binary variables after unique thermal encoding are kept in their original 0-1 form and are not numerically scaled.

[0051] In this context, stratified sampling refers to dividing the model into layers based on the de-icing method, and then randomly dividing each layer independently. This ensures that the model can fully learn the jump response patterns under different de-icing mechanisms during training, thereby improving generalization ability and engineering applicability.

[0052] It should be noted that step S104 involves outlier removal, feature expansion, stratified sampling, and mixed standardization of the dataset to improve data quality and distribution balance, making subsequent model training more stable and generalization more effective.

[0053] S105. Construct a two-level stacked ensemble learning model, train the model using the preprocessed training dataset, and obtain a predictive model for predicting the de-icing jump response characteristic parameters of isolated span transmission lines.

[0054] In an embodiment of the present invention, the two-level stacked ensemble learning model includes: Basic model layer and meta-model layer; The base model layer includes a random forest model, a gradient boosting tree model, and an extreme gradient boosting model, which output preliminary prediction results for multidimensional output indicators in parallel. The meta-model layer uses an elastic network regression model to weight and fuse the prediction results output by the basic model layer to generate the final prediction value.

[0055] In some embodiments, the constructed two-level stacked ensemble learning model consists of a base model layer and a meta-model layer, used for high-precision prediction of ice-breaking jump response characteristics. For example, a random forest model, a gradient boosting tree model, and an extreme gradient boosting model are deployed in parallel in the base model layer. All three use 25-dimensional input features after hybrid standardization and feature expansion as input, and independently output preliminary prediction results for six-dimensional multi-dimensional output indicators (including dynamic imbalance tension, static imbalance tension, and maximum ice-breaking height at L / 4, L / 2, and 3L / 4). Among them, the random forest model effectively suppresses overfitting and captures the nonlinear interaction between parameters by constructing 100 decision trees and introducing a feature subset random selection mechanism. The gradient boosting tree model uses a forward step-by-step addition strategy to optimize the residuals round by round, and has a strong fitting ability for jump mutations caused by the coupling of height difference and ice thickness. The extreme gradient boosting model uses second-order Taylor expansion and regularization term control to improve the prediction stability for small sample areas (such as high ice-breaking rate and low-gap combination) while ensuring training speed.

[0056] In some embodiments, the prediction results of the three basic models on the validation set are concatenated into a new 18-dimensional intermediate feature (i.e., each output index corresponds to the prediction values ​​of the three models), which is used as the input of the meta-model layer. For example, for a certain working condition sample, the random forest predicts the ice jumping height at L / 2 as 2.31 meters, the gradient boosting tree predicts it as 2.45 meters, and the extreme gradient boosting predicts it as 2.38 meters. Then the intermediate feature corresponding to this position is [2.31, 2.45, 2.38], and the six output indices form an 18-dimensional vector.

[0057] In some embodiments, the meta-model layer uses an elastic network regression model to weight and fuse these 18 intermediate features. This model combines L1 and L2 regularization terms to automatically select reliable base models while suppressing redundant weights. For example, when the gradient boosting tree performs significantly better than other models under elevation difference conditions, the elastic network will give it higher weights. Under standard symmetric conditions, if the performance of the three models is similar, balanced weights will be assigned, and finally, a fused prediction value of the six-dimensional output index will be generated.

[0058] The meta-model layer here refers to the intelligent integration of multiple model outputs by learning the complementarity and bias patterns between the prediction results of the base models, thereby significantly improving prediction accuracy and robustness while maintaining generalization ability.

[0059] In this embodiment of the invention, a nested cross-validation strategy is adopted during the training process of the two-level stacked ensemble learning model, and an early stopping mechanism is combined to control the number of training rounds. Training is automatically terminated when the error metric on the validation set does not show significant improvement within a set number of consecutive rounds to prevent overfitting.

[0060] In some embodiments, a nested cross-validation strategy is employed when training a two-level stacked ensemble learning model, combined with an early stopping mechanism for fine-grained control of the training process. For example, in the outer cross-validation, the entire training dataset is divided into 5 mutually exclusive folds, and 4 folds are selected as the outer training set each time, with the remaining 1 fold serving as the outer test set. In the inner cross-validation, the outer training set is further divided into 5 sub-folds for hyperparameter tuning and basic model training, ensuring that the model evaluation results are not affected by data partitioning bias and effectively simulating the model's generalization performance under unknown conditions.

[0061] In some embodiments, during each inner layer training process, the mean squared error of the three basic models—random forest, gradient boosting tree, and extreme gradient boosting—on the inner layer validation subset is monitored, and an early stopping mechanism is set: when the decrease in validation error is less than 0.001 kN (for tension indicators) or 0.005 m (for ice jumping height indicators) in 15 consecutive iterations, it is determined that there is no significant improvement, and the training round of the model is automatically terminated. For example, when training the extreme gradient boosting model, if the validation error of the dynamic imbalance tension remains between 1.23 and 1.24 kN between rounds 82 and 97 and does not exceed the preset improvement threshold, training is stopped after round 97 to avoid overfitting to training noise due to continued iteration.

[0062] In some embodiments, the elastic network regression of the meta-model layer is also incorporated with early stopping logic. The average absolute error of the model on the inner layer validation set is used as the monitoring index. Training is terminated if there is no improvement after 10 consecutive rounds. At the same time, the confidence interval of the final model performance is calculated using the multi-fold validation results obtained by nested cross-validation, providing a reliable basis for engineering applications.

[0063] Nested cross-validation refers to separating the hyperparameter selection and model performance evaluation processes through inner and outer cross-validation structures to prevent information leakage. The early stopping mechanism dynamically truncates training based on the convergence behavior of validation errors. The two work together to ensure that the model has both high accuracy and strong generalization ability under complex nonlinear mapping relationships.

[0064] In this embodiment of the invention, a Bayesian optimization algorithm is used to adaptively optimize the hyperparameters of the two-level stacked ensemble learning model, with the optimization objective being to maximize the average decision coefficient of the multidimensional output index on the validation set. The model performance is quantitatively evaluated using a comprehensive evaluation system that includes three categories of indicators: goodness of fit, error, and robustness.

[0065] In some embodiments, Bayesian optimization algorithms are used to adaptively optimize the hyperparameters of the two-level stacked ensemble learning model to improve the overall fitting ability of the model under multi-dimensional output tasks. For example, in the base model layer, the key hyperparameters of random forest include the number of decision trees, the maximum number of features, and the minimum number of leaf node samples; the key hyperparameters of gradient boosting tree include the learning rate, the maximum depth, and the subsampling ratio; and the extreme gradient boosting model requires optimization of the regularization coefficient, column sampling rate, and tree structure growth strategy. In the meta-model layer, the hyperparameters of resilient network regression mainly include the ratio of L1 to L2 regularization weights (i.e., the mixing ratio α) and the regularization strength λ. Bayesian optimization constructs a Gaussian process surrogate model, predicts the average coefficient of determination (R²) of the six-dimensional output index on the validation set under different hyperparameter combinations based on historical evaluation results, and iteratively selects the next most promising hyperparameter configuration for evaluation based on the expected improvement criterion. For example, in the 12th iteration, the algorithm recommended a combination with a learning rate of 0.08, a maximum depth of 7, and an elastic network α of 0.65. After inner-layer cross-validation, the average R² was 0.963, which was better than the previous best value of 0.958. This result was included in the historical observation set to update the surrogate model.

[0066] In some embodiments, the optimization objective is explicitly set as maximizing the average determination coefficient of the multidimensional output index on the validation set, that is, taking into account the proportion of the explained variance of dynamic unbalance tension, static unbalance tension and the jumping height at the three positions, so as to avoid prediction bias caused by a single index.

[0067] In some embodiments, after hyperparameter optimization, a comprehensive evaluation system is further constructed, including three categories of indicators: goodness of fit, error, and robustness, to quantitatively evaluate the performance of the final model. Goodness of fit indicators include the R² of each output dimension and the overall average R². Error indicators cover the root mean square error (RMSE), mean absolute error (MAE), and their proportions relative to engineering allowable limits. Robustness indicators are measured by the performance standard deviation of the outer test set on different subsets of ice removal methods, the upper limit of error under worst-case conditions, and the 95% confidence interval width of the R² obtained from nested cross-validation. For example, the final model achieved an average R² of 0.961 on 200 sets of outer test samples, an RMSE of 0.082 meters for the ice jump height at L / 2, a dynamic unbalanced tension MAE of 1.03 kN, and an R² standard deviation of no more than 0.015 for the three subsets of ice removal methods, indicating that the model not only has high accuracy but also performs stably under various working conditions.

[0068] The comprehensive evaluation system here refers to the collaborative characterization of the model's predictive ability, error control level, and generalization stability from multiple dimensions, ensuring that the constructed intelligent prediction model meets the engineering reliability requirements for the safety assessment of transmission line de-icing and jumping.

[0069] It should be noted that step S105 constructs and trains a two-level stacked ensemble learning model, which combines the advantages of multiple strong learners, significantly improving the prediction accuracy and robustness for complex nonlinear responses, and ultimately achieving highly reliable prediction of de-icing jump characteristic parameters.

[0070] Example 2: Based on the above examples, a specific implementation method for predicting the de-icing jump response characteristic parameters of an isolated span transmission line can be designed as follows: A high-precision prediction system is constructed through finite element simulation modeling, feature engineering optimization, multi-model fusion training, and adaptive hyperparameter tuning, including the following steps: A fully parametric simulation model was established using the finite element software COMSOL, focusing on reproducing the mechanical properties of the conductor-insulator string connection node, considering conductor elastic deformation and correction factors for non-uniform distribution of icing load (correction factors range from 1.05 to 1.12). The ice load simulation employed the piecewise equivalent concentrated force method, dividing the span into 20 calculation units, with each unit dynamically adjusting its load value according to the ice removal method. The initial form finding used an improved iterative catenary equation, introducing a correction term for conductor plastic deformation (correction factor 0.98). The model was validated using three typical scenarios: ① Standard working condition (475m span, 0m elevation difference, 20mm icing, 100% de-icing rate); ② Elevation difference working condition (300m span, 50m elevation difference, 15mm icing, 70% de-icing rate); ③ Complex de-icing condition (500m span, 30m elevation difference, 25mm icing, chain de-icing). Comparison with literature data showed: the relative error of vertical ice jump height ≤ 5.8%, and the tension time history curve agreement ≥ 94.3%, validating the model's stability in multiple scenarios. Furthermore, based on the Sobol global sensitivity analysis method, the sensitivity of seven candidate input parameters (ice thickness b, de-icing rate Q, span length L, elevation difference H, conductor cross-sectional area S, number of splits n, and de-icing method A) was ranked through 10,000 Monte Carlo simulations. The first-order sensitivity index and the total sensitivity index of each parameter were calculated, and the overall importance index of the parameters was determined as follows: de-icing method (0.23) > ice thickness (0.19) > de-icing rate (0.17) > span length (0.15) > conductor cross-sectional area (0.11) > elevation difference (0.09) > number of splits (0.06). The results show that the total sensitivity index of all seven parameters is >0.05, indicating that they are all key input variables and need to be included in the prediction model. Output parameter extraction employs a time-history curve multi-feature fusion method (e.g.) Figure 2As shown): ① Dynamic unbalanced tension (f1=f0-fmin), where f0 is the steady-state tension of the conductor before ice removal, and fmin is the minimum tension after ice removal; ② Static unbalanced tension (f2=f0-fs), where fs is the stable tension after ice removal; ③ Maximum ice jumping height at ML / 4 (h1); ④ Maximum ice jumping height at ML / 2 (h2); ⑤ Maximum ice jumping height at 3ML / 4 (h3); ⑥ Maximum ice jumping height at the suspension point (h4). The time history curves of tension versus ice jumping height were smoothed using cubic spline interpolation, with a sampling frequency of 100Hz to eliminate high-frequency noise interference. Simultaneously, the rate characteristics (v=dh / dt) at the first 5% of the time before the peak point were extracted as an auxiliary verification index to evaluate the accuracy of the prediction model in the dynamic response stage. Furthermore, based on the design requirements for heavy icing lines in actual engineering projects, and combined with the actual operating data of 110kV to 500kV lines in southern China, the value ranges of each input parameter are determined as follows: ice thickness [5-35mm] (covering light icing to extremely heavy icing areas), de-icing rate [20%-100%] (values ​​taken at 10% intervals), span [200-600m] (values ​​taken at 50m intervals), elevation difference [0-80m] (values ​​taken at 10m intervals), cross-sectional area [240-900mm²] (values ​​corresponding to typical conductor models), number of splits [1-6] (corresponding to different voltage levels), de-icing method [1-12] (1 for whole span de-icing, 2-8 for uneven de-icing, and 9-12 for chain de-icing).

[0071] Subsequently, 1800 sets of input data were generated using the Latin hypercube sampling method to ensure uniform coverage of the parameter space. Simulation calculations were performed using a validated finite element model to obtain the six output parameters corresponding to each set of inputs, thus constructing the basic dataset. The preprocessing process consists of four steps: ① Outlier detection: The IQR method (interquartile range ± 1.5 times) is used to screen the output parameters and remove abnormal samples, mainly abnormal ice jumping height data under extreme de-icing rates; ② Feature expansion: 14 derived features are generated, including interactive features (such as b×Q, L / H, S×n) and higher-order features (such as b², L³, Q²), expanding the feature space from 7 dimensions to 21 dimensions; ③ Stratified sampling: Stratified sampling is performed according to the de-icing method label (1-12) to ensure that the sample ratio of each layer is consistent, dividing the data into training set (70%, 1244 groups), validation set (15%, 267 groups), and test set (15%, 266 groups); ④ Hybrid standardization: Z-score standardization (x*=(x-μ) / σ) is used for physical quantity parameters (b, Q, L, etc.), and one-hot encoding is used for classification parameters (de-icing method A) to convert them into 12-bit binary feature vectors. Furthermore, the model architecture adopts a two-level stacked structure (such as...). Figure 3 As shown), prediction accuracy is improved through heterogeneous model fusion: The base model layer contains three heterogeneous learners with complementary performance: ① Optimized Random Forest (RF): An ensemble of 250 CART regression trees with a maximum depth of 25. It uses the Gini index as the splitting criterion, with a minimum number of samples per leaf node of 4. Each tree randomly selects samples and features for training to reduce model variance; ② Gradient Boosting Tree (GBDT): It uses an additive model for iterative training with a learning rate of 0.01, a tree depth of 8, 500 iterations, and a subsampling rate of 0.8 (randomly selecting 80% of the samples in each iteration). The loss function is mean squared error, and gradient descent is used to optimize model bias; ③ Extreme Gradient Boosting (XGBoost): Based on a regularized boosting framework, it sets gamma=0.1 (minimum loss reduction for node splitting), subsample=0.85 (sample sampling rate), colsample_bytree=0.9 (feature sampling rate), a learning rate of 0.02, and a tree depth of 7. Overfitting is controlled through L1 / L2 regularization. The meta-model layer uses ElasticNet regression, combining the advantages of L1 regularization (Lasso) and L2 regularization (Ridge). The regularization parameter α=0.005 and the L1 ratio 0.5 (to balance the strength of L1 and L2 regularization) are used to weight and fuse the three outputs of the basic model layer (a total of 3×6=18 dimensions of features) to output the final predicted value. The training process employs a nested 5-fold cross-validation strategy: the outer 5-fold validation layer assesses the overall generalization ability of the model, while the inner 5-fold validation layer optimizes the parameters for each base model. An early stopping mechanism (patience=50) is introduced to prevent overfitting; training automatically terminates when the validation set MSE shows no improvement (change < 1e-6) for 50 consecutive iterations, saving computational resources. The training process uses mini-batch gradient descent with a batch size of 32 and a maximum total number of iterations of 1000. Furthermore, a Bayesian optimization algorithm is employed to intelligently optimize the model's hyperparameters. A mapping relationship between hyperparameters and model performance is established based on a Gaussian process, and the next sampling point is selected using the Expected Improvement function. The search space includes: ①RF: number of trees [100-500], maximum depth [10-30], minimum number of samples per leaf node [2-10]; ②GBDT: learning rate [0.001-0.1], tree depth [5-15], number of iterations [200-1000]; ③XGBoost: gamma [0-0.5], subsample [0.7-0.95], colsample_bytree [0.7-0.95]; ④ meta-model: regularization parameter α [0.001-0.01], L1 ratio [0.1-0.9]. The optimization objective is to maximize the average R² of the validation set, with 100 iterations to ultimately obtain the globally optimal hyperparameter combination. The evaluation index system includes three categories and a total of seven indicators: ① Goodness of fit: R² (coefficient of determination), adjusted R² (corrected for the influence of degrees of freedom); ② Error indicators: MAE (mean absolute error), MSE (mean square error), RMSE (root mean square error); ③ Robustness indicators: MaxAE (maximum absolute error), SDE (standard deviation of error). Evaluation results on the test set: The R² values ​​for the six output parameters are 0.902 (f1), 0.908 (f2), 0.925 (h1), 0.937 (h2), 0.851 (h3), and 0.864 (h4), respectively; the MAE values ​​are 0.23 kN (f1), 0.12 kN (f2), 0.082 m (h1), 0.032 m (h2), 71.56 N (h3), and 47.38 N (h4), respectively; and the MSE values ​​are 0.0004, 0.00018, 0.00007, 0.000039, 35.26, and 22.47, respectively. Compared with the single model, the stacked model improves the average R² by 8.3% and reduces the MAE by 12.7%, with all indicators outperforming the single learner, verifying the effectiveness of the ensemble strategy.

[0072] It should be noted that a refined finite element simulation modeling and multi-scenario verification method for isolated span transmission lines is proposed. By restoring the mechanical properties of the conductor-insulator string connection node, introducing a non-uniform icing correction coefficient (1.05-1.12) and a conductor plastic deformation correction term (0.98), and combining verification under three typical working conditions (relative error of vertical ice jump height ≤ 5.8%, and consistency of tension time history curve ≥ 94.3%), the reliability of the model under multiple scenarios is ensured. Based on Sobol global sensitivity analysis, seven key input parameters (ice thickness, de-icing rate, etc.) are determined. Six core output parameters (dynamic / static unbalanced tension, maximum ice jump height at four locations) are extracted through the time history curve multi-feature fusion method, and interactive and high-order derived features are generated to expand to a 21-dimensional feature space to construct accurate parameter mapping relationships. A two-level stacked integrated prediction model architecture is constructed. The basic model layer uses three heterogeneous learners—optimized random forest (RF), gradient boosting tree (GBDT), and extreme gradient boosting (XGBoost)—for parallel training. The meta-model layer fuses 18-dimensional features through elastic network regression to improve prediction accuracy. A nested 5-fold cross-validation strategy and an early stopping mechanism (patience=50) are used to control the model training process. The hyperparameters are intelligently optimized by combining the Bayesian optimization algorithm (100 iterations). An evaluation system with 7 indicators in 3 categories, including goodness of fit, error, and robustness, is established to ensure the model's generalization ability and prediction stability.

[0073] Example 3, referring to Figure 4 This embodiment also provides a prediction system for the de-icing jump response characteristic parameters of isolated span transmission lines, including: The simulation model building module is used to build a finite element simulation model of an isolated span transmission line that considers the mechanical properties of the conductor-insulator string connection node, the non-uniform distribution of ice load, and the plastic deformation of the conductor. The sample data acquisition module is used to perform ice-breaking jump simulations under various typical working conditions based on the finite element simulation model, and to obtain the corresponding response characteristic parameters as sample data. The indicator acquisition module is used to perform global sensitivity analysis on input parameters, determine key input parameters, and extract multi-location and multi-dimensional features from the output response characteristic parameters to form a multi-dimensional output indicator that includes dynamic unbalance tension, static unbalance tension, and the maximum jumping height at multiple specified locations. The data processing module is used to generate a training dataset that covers the actual parameter space of the project based on key input parameters and multidimensional output indicators, and to perform outlier removal, feature expansion, stratified sampling and hybrid standardization preprocessing on the training dataset. The prediction model building module is used to build a two-level stacked ensemble learning model. The model is trained using a preprocessed training dataset to obtain a prediction model for predicting the de-icing jump response characteristic parameters of isolated span transmission lines.

[0074] The above-mentioned unit modules can be embedded in the processor of the electronic device in hardware form or independent of it, or they can be stored in the memory of the electronic device in software form, so that the processor can call and execute the corresponding operations of the above modules.

[0075] This embodiment also provides an electronic device, which can be a terminal, and its internal structure diagram can be as follows. Figure 4 As shown, the electronic device includes a processor, memory, communication interface, display screen, and input device connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. When executed by the processor, the computer program implements a method for predicting the de-icing jump response characteristic parameters of isolated transmission lines. The display screen can be an LCD screen or an e-ink screen. The input device can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the device's casing, or an external keyboard, touchpad, or mouse.

[0076] This embodiment also provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, it performs the following steps: A finite element simulation model of an isolated span transmission line was constructed, taking into account the mechanical properties of the conductor-insulator string connection node, the non-uniform distribution of ice load, and the plastic deformation of the conductor. Based on the finite element simulation model, de-icing jump simulations were carried out under various typical working conditions to obtain the corresponding response characteristic parameters as sample data. A global sensitivity analysis of the input parameters is performed to identify key input parameters. Multi-location and multi-dimensional feature extraction is performed on the output response characteristic parameters to form a multi-dimensional output index that includes dynamic unbalance tension, static unbalance tension, and maximum jumping height at multiple specified locations. Based on key input parameters and multidimensional output indicators, a training dataset covering the actual parameter space of engineering is generated, and the training dataset is preprocessed by outlier removal, feature expansion, stratified sampling and hybrid standardization. A two-level stacked ensemble learning model was constructed, and the model was trained using a preprocessed training dataset to obtain a predictive model for predicting the de-icing jump response characteristics of isolated span transmission lines.

[0077] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

[0078] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the invention.

[0079] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for predicting the de-icing jump response characteristic parameters of an isolated span transmission line, characterized in that, include: A finite element simulation model of an isolated span transmission line was constructed, taking into account the mechanical properties of the conductor-insulator string connection node, the non-uniform distribution of ice load, and the plastic deformation of the conductor. Based on the finite element simulation model, de-icing jump simulations were performed under various typical working conditions to obtain the corresponding response characteristic parameters as sample data. A global sensitivity analysis of the input parameters is performed to identify key input parameters. Multi-location and multi-dimensional feature extraction is performed on the output response characteristic parameters to form a multi-dimensional output index that includes dynamic unbalance tension, static unbalance tension, and maximum jumping height at multiple specified locations. Based on the key input parameters and multidimensional output indicators, a training dataset covering the actual parameter space of the project is generated, and the training dataset is preprocessed by outlier removal, feature expansion, stratified sampling and hybrid standardization. A two-level stacked ensemble learning model is constructed, and the model is trained using a preprocessed training dataset to obtain a predictive model for predicting the de-icing jump response characteristic parameters of isolated span transmission lines.

2. The method for predicting the de-icing jump response characteristic parameters of an isolated span transmission line as described in claim 1, characterized in that, The multidimensional output metrics include: The dynamic and static unbalanced tensions formed by the tension difference of the conductor before and after ice removal, and the maximum ice jumping heights corresponding to the quarter, half, three-quarters and suspension point positions, respectively; The multi-dimensional feature extraction includes smoothing the time history curves of tension and jumping height, and sampling to obtain key features in the dynamic response process.

3. The method for predicting the de-icing jump response characteristic parameters of an isolated span transmission line as described in claim 2, characterized in that, The key input parameters include: Ice thickness, de-icing rate, span, elevation difference, conductor cross-sectional area, number of split conductors, and de-icing method; The feature expansion includes constructing interaction features and higher-order features between the key input parameters, expanding the original input feature space into a higher-dimensional feature set. The hybrid standardization preprocessing includes Z-score standardization for continuous physical parameters and one-hot encoding for classification parameters such as de-icing methods.

4. The method for predicting the de-icing jump response characteristic parameters of an isolated span transmission line as described in claim 3, characterized in that, The two-level stacked ensemble learning model includes: Basic model layer and meta-model layer; The base model layer includes a random forest model, a gradient boosting tree model, and an extreme gradient boosting model, which output preliminary prediction results for the multidimensional output index in parallel. The meta-model layer uses an elastic network regression model to weight and fuse the prediction results output by the basic model layer to generate the final prediction value.

5. The method for predicting the de-icing jump response characteristic parameters of an isolated span transmission line as described in claim 4, characterized in that, The training process of the two-level stacked ensemble learning model adopts a nested cross-validation strategy and combines an early stopping mechanism to control the number of training rounds. Training is automatically terminated when the error metric on the validation set does not show significant improvement within a set number of consecutive rounds to prevent overfitting.

6. The method for predicting the de-icing jump response characteristic parameters of an isolated span transmission line as described in claim 5, characterized in that, The hyperparameters of the two-level stacked ensemble learning model are adaptively optimized using the Bayesian optimization algorithm, with the optimization objective being to maximize the average decision coefficient of the multidimensional output index on the validation set. The model performance is quantitatively evaluated using a comprehensive evaluation system that includes three categories of indicators: goodness of fit, error, and robustness.

7. The method for predicting the de-icing jump response characteristic parameters of an isolated span transmission line as described in claim 6, characterized in that, The various typical operating conditions include standard operating conditions, elevation difference operating conditions, and complex de-icing operating conditions.

8. A prediction system for the de-icing jump response characteristic parameters of an isolated span transmission line, using the method described in any one of claims 1 to 7, characterized in that, include: The simulation model building module is used to build a finite element simulation model of an isolated span transmission line that considers the mechanical properties of the conductor-insulator string connection node, the non-uniform distribution of ice load, and the plastic deformation of the conductor. The sample data acquisition module is used to perform ice-breaking jump simulations under various typical working conditions based on the finite element simulation model, and obtain the corresponding response characteristic parameters as sample data. The indicator acquisition module is used to perform global sensitivity analysis on input parameters, determine key input parameters, and extract multi-location and multi-dimensional features from the output response characteristic parameters to form a multi-dimensional output indicator that includes dynamic unbalance tension, static unbalance tension, and the maximum jumping height at multiple specified locations. The data processing module is used to generate a training dataset covering the actual parameter space of the project based on the key input parameters and multidimensional output indicators, and to perform outlier removal, feature expansion, stratified sampling and hybrid standardization preprocessing on the training dataset. The prediction model building module is used to build a two-level stacked ensemble learning model. The model is trained using a preprocessed training dataset to obtain a prediction model for predicting the de-icing jump response characteristic parameters of isolated span transmission lines.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method for predicting the de-icing jump response characteristic parameters of an isolated span transmission line according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for predicting the de-icing jump response characteristic parameters of an isolated span transmission line according to any one of claims 1 to 7.

Images