Method and system for analyzing the durability and vulnerability of power transmission and distribution facilities

Abstract

The application discloses a power transmission and transformation facility foundation durability and vulnerability analysis method and system, relates to the technical field of power transmission and transformation facility structure safety evaluation and disaster prevention and mitigation, and comprises the following steps: acquiring basic environment data and power transmission tower structure data, and constructing a basic database; screening parameters from the database, and performing spatial encryption on parameters with discontinuous spatial distribution to obtain a continuous basic evaluation parameter set; performing oversampling processing on a few class samples in the basic evaluation parameter set to generate an enhanced sample set; training a neural network by using the enhanced sample set, and globally optimizing initial weights and threshold values of the neural network to obtain a prediction model; establishing a flood reliability equation based on the prediction model, calculating reliability, evaluating failure risks in combination with vulnerability analysis, identifying weak links, and outputting optimization suggestions; and the application solves the problems of being unable to comprehensively simulate the influence of flood disasters, data imbalance and inaccurate foundation durability evaluation in traditional analysis.

Description

Technical Field

[0001] This invention relates to the field of structural safety assessment and disaster prevention and mitigation technology for power transmission and transformation facilities. More specifically, this invention relates to methods and systems for analyzing the durability and vulnerability of power transmission and transformation facility foundations. Background Technology

[0002] As a key component of the power system, power transmission and transformation facilities primarily undertake the functions of transmitting and distributing electrical energy. Their stability and reliability are crucial to the safe operation of the entire power system. Transmission towers and their foundations are important parts of these facilities, constantly exposed to complex natural environments, especially to hydrological and meteorological disasters (such as floods). These can lead to structural damage such as foundation settlement, cracks, and fractures, which in severe cases may cause equipment collapse or failure, resulting in widespread power outages or catastrophic accidents. Therefore, conducting durability and vulnerability analysis of power transmission and transformation facility foundations under flood conditions, identifying potential risks in advance, and providing risk warnings are of significant practical importance.

[0003] Currently, the durability analysis methods for power transmission and transformation facility foundations under flood disasters mainly rely on traditional mechanical analysis and empirical assessments. Commonly used techniques include structural stability analysis of the foundation, stress-strain calculations under load, and single assessments of environmental factors (such as groundwater level changes and soil erosion). These traditional methods are mostly limited to static assessments of the local state of the foundation and often fail to comprehensively reflect the long-term behavior and vulnerability of the foundation under dynamic disaster conditions.

[0004] Traditional analytical methods are mostly based on local or point-based basic environmental data for assessment, failing to fully consider variations under different geographical locations and environmental conditions. For example, the spatial distribution of factors such as soil type, groundwater level, and precipitation often neglects the impact on foundation durability. Furthermore, existing technologies typically cannot handle multidimensional spatial data, making it difficult to capture the influencing factors of spatially discontinuous distributions. Under the influence of floods, the load, displacement, and settlement experienced by foundations exhibit high uncertainty and dynamism. Existing methods fail to effectively simulate the long-term stress conditions and deterioration process of foundation structures under flood conditions. The limitations of the analysis result in a lack of sufficient disaster-related references in risk assessment results, leading to a lack of accuracy and reliability in the assessment of foundation durability and vulnerability. In disaster analysis, the frequency of certain foundation failures or major disaster events is low, resulting in imbalanced disaster sample data. Existing assessment methods struggle to accurately identify minority samples (such as foundation failure under extreme flooding). Imbalanced datasets often lead to poor identification of minority samples by predictive models, thus affecting overall prediction performance and disaster early warning capabilities. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a method and system for analyzing the durability and vulnerability of power transmission and transformation facility foundations, which solves the problems of traditional analysis, such as the inability to fully simulate the impact of flood disasters, data imbalance, and inaccurate foundation durability assessment.

[0006] To achieve the above objectives, the present invention provides the following technical solution: Firstly, this application provides a method for analyzing the durability and vulnerability of power transmission and transformation facility foundations. This method includes: acquiring foundation environmental data and transmission tower structural data to construct a foundation database; filtering parameters from the database and spatially encrypting parameters with discontinuous spatial distribution to obtain a continuous set of foundation evaluation parameters; oversampling the minority class samples in the foundation evaluation parameter set to generate an enhanced sample set; wherein the oversampling process includes: clustering the minority class samples, weighting them within each cluster based on kernel density probability, and sampling between the centroid and vertices of a triangle formed by three sample points to generate new samples; training a neural network with the enhanced sample set and globally optimizing its initial weights and thresholds to obtain a prediction model; establishing a flood reliability equation based on the prediction model, calculating reliability, assessing failure risk in conjunction with vulnerability analysis, identifying weak points, and outputting optimization suggestions.

[0007] In one embodiment, constructing a basic database includes: building a preliminary spatial database based on the locations of transmission tower foundations; establishing vertical correlation relationships for each tower foundation; establishing proximity influence correlation rules based on the spatial proximity relationships between towers; classifying the basic environmental data and establishing environmental gradient correlation rules with the structural data; integrating the vertical correlation, proximity influence correlation, and environmental gradient correlation into a multi-layer spatial correlation model, and generating a multi-dimensional spatial index containing vertical, proximity, and environmental gradient data for each basic node to obtain a spatial basic database.

[0008] In one embodiment, parameter selection from the database includes: establishing a mapping relationship between output indicators and input parameters in a spatial base database; generating a high-dimensional input parameter sample set using Sobol sequences to cover the input parameter space; performing basic stability analysis on the sample set to obtain simulated data; calculating the first-order sensitivity index and total sensitivity index of each input parameter using the Sobol sensitivity analysis method; sorting the parameters according to the first-order sensitivity index and total sensitivity index, and selecting parameters higher than a preset threshold as key evaluation parameters.

[0009] In one embodiment, spatial densification is performed on parameters with discontinuous spatial distribution to obtain a continuous set of basic evaluation parameters. This includes: identifying regions with discontinuous spatial distribution or sparse data for key evaluation parameters based on the density of neighboring points and the spatial change gradient; performing trend surface fitting on the entire spatial region to obtain trend surface values; calculating the residuals of key evaluation parameters at the original sampling points; performing local interpolation in the regions with discontinuous spatial distribution or sparse data based on the trend surface values ​​and residuals using an inverse distance weighted interpolation algorithm; and superimposing the local interpolation results with the trend surface values ​​to generate a continuous set of basic evaluation parameters.

[0010] In one embodiment, oversampling is performed on minority class samples in the basic evaluation parameter set to generate an enhanced sample set, including: constructing an original sample dataset based on the basic evaluation parameter set and historical damage records; identifying minority class samples and performing clustering to obtain the sample distribution within each cluster; generating new samples in each cluster using a probability density constrained sampling method, and adding the new samples to the original minority class samples to obtain the enhanced evaluation sample data.

[0011] In one embodiment, new samples are generated in each cluster using a probability density-constrained sampling method, including: constructing a joint probability density function of samples within each cluster using kernel density estimation; randomly selecting three sample points in each cluster to form a triangle and calculating the centroid of the triangle; and performing weighted random sampling according to the joint probability density function within the parameter space interval between the centroid and any vertex to generate new minority class samples, such that the sampling probability of high-density areas is higher than that of low-density areas.

[0012] In one embodiment, training a neural network with an augmented sample set includes: using the augmented sample set as training data, setting the number of neurons in the input layer and the number of neurons in the output layer; determining the number of neurons in the hidden layer based on the number of neurons in the input layer and the number of neurons in the output layer; selecting the activation function with the smallest mean squared error as the optimal function by comparing the mean squared errors of each activation function; determining the learning rate through trial and error to achieve training convergence; and training the network using the augmented sample set to obtain a BP neural network model.

[0013] In one embodiment, the initial weights and thresholds are globally optimized to obtain a prediction model, including: initializing the weights and thresholds of the BP neural network; setting a fitness function with mean squared error as the optimization objective; performing a global search by simulating sparrow foraging behavior, which includes a random leap strategy and a local leap strategy, used for global search and local refinement, respectively; sorting and updating individual positions according to fitness values; terminating optimization when the number of iterations reaches a preset value or the mean squared error drops to the target accuracy, thus obtaining optimized weights and thresholds; and training the BP neural network with the optimized weights and thresholds as initial values ​​to obtain an optimized basic durability and vulnerability prediction model.

[0014] In one embodiment, a flood reliability equation is established based on a prediction model, reliability is calculated, failure risk is assessed by combining vulnerability analysis, weak points are identified, and optimization suggestions are output. This includes: acquiring flood disaster-related data and foundation design parameters; constructing a foundation reliability state equation by combining an optimized foundation durability and vulnerability prediction model; calculating the foundation reliability index under different flood disaster conditions; performing vulnerability analysis based on the reliability index and the foundation's response characteristics under flood disasters, calculating failure risk indicators, and outputting a risk analysis report; identifying weak points in the foundation structure based on the risk analysis report, and outputting optimization suggestions for the weak points, including foundation design improvements, material optimization, or flood-resistant reinforcement measures.

[0015] Secondly, this application provides a system for analyzing the durability and vulnerability of power transmission and transformation facility foundations. This system includes: a database construction module for acquiring foundation environmental data and transmission tower structure data to construct a foundation database; a data processing module for filtering parameters from the database and spatially encrypting parameters with discontinuous spatial distribution to obtain a continuous set of foundation evaluation parameters; a sample optimization module for oversampling minority class samples in the foundation evaluation parameter set to generate an enhanced sample set; a model construction module for training a neural network with the enhanced sample set and globally optimizing its initial weights and thresholds to obtain a prediction model; and an analysis module for establishing a flood reliability equation based on the prediction model, calculating reliability, assessing failure risk in conjunction with vulnerability analysis, identifying weak points, and outputting optimization suggestions.

[0016] As can be seen from the above technical solutions, the embodiments of this application have the following advantages: By integrating multiple steps such as spatial database construction, key parameter screening, spatial encryption, probability density constrained oversampling, and intelligent predictive analysis, this approach can simultaneously address the problems of discontinuous spatial distribution of data, insufficient minority class samples, and weak generalization ability of traditional models in the evaluation of power transmission and transformation facility foundations. Specifically, spatial encryption of discontinuous parameters improves the continuity and accuracy of the foundation evaluation parameter set. An improved oversampling method based on kernel density probability weighting makes the generated new samples more consistent with the true spatial distribution of minority class samples, effectively avoiding the problem of marginalized samples generated by the traditional SMOTE algorithm, thereby improving the quality of the enhanced sample set. Furthermore, combining neural networks and global optimization methods enhances the convergence speed, prediction accuracy, and stability of model training. Finally, by combining flood reliability equations with vulnerability analysis, it achieves accurate identification of failure risks and location of weak points in power transmission and transformation facility foundations, providing a reliable basis for facility durability assessment, risk warning, and optimized maintenance decisions. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic flowchart of the method for analyzing the durability and vulnerability of power transmission and transformation facility foundations provided in the embodiments of this application.

[0019] Figure 2 This is a schematic diagram of the structure of the power transmission and transformation facility foundation durability and vulnerability analysis system provided in the embodiments of this application.

[0020] Figure 3 A schematic diagram illustrating the generation of new samples using the conventional SMOTE algorithm provided in this application embodiment.

[0021] Figure 4 A schematic diagram illustrating the generation of new samples using the improved SMOTE algorithm provided in this application embodiment. Detailed Implementation

[0022] To enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0023] Reference Figure 1 As shown in the schematic diagram, the method for analyzing the durability and vulnerability of power transmission and transformation facility foundations provided by this invention includes the following steps: S1: Obtain the basic environmental data and transmission tower structure data required for the durability and vulnerability analysis of power transmission and transformation facility foundations, and construct a spatial foundation database.

[0024] Based on the spatial distribution of power transmission and transformation facilities, basic environmental data and transmission tower structural data related to the service environment and structural characteristics of the power transmission and transformation facility foundations are acquired. The basic environmental data includes natural environmental data and engineering geological environmental data. The natural environmental data characterizes the external service conditions of the power transmission and transformation facility foundations, including: annual average temperature, extreme maximum and minimum temperatures, annual precipitation, rainfall intensity, freeze-thaw cycles, wind speed and direction, relative humidity, and corrosive atmospheric environmental parameters. Each natural environmental data point is expressed in spatial raster or vector form and corresponds one-to-one with the geographical location of the power transmission and transformation facility foundation. The engineering geological environmental data includes: stratigraphic structure type, physical and mechanical parameters of soil or rock mass, groundwater depth and chemical properties, seismic intensity zoning information, and information on the distribution of adverse geological processes. This engineering geological environmental data characterizes the foundation stability and deterioration risk of the power transmission and transformation facility foundations during long-term service.

[0025] The structural data of the transmission towers includes: tower type, tower height, foundation type, foundation depth, foundation geometry, foundation material type and strength grade, reinforcement configuration parameters, and design service life. This structural data characterizes the load-bearing capacity and degradation sensitivity of the transmission and transformation facility foundation under external environmental influences. The structural data also includes operation and maintenance information, such as: commissioning time, service life, maintenance records, types of defects and their occurrence times. This operation and maintenance information assists in assessing the actual durability of the transmission and transformation facility foundation.

[0026] In this embodiment, the construction of a spatial basic database includes: The basic environmental data and transmission tower structure data are standardized, including data alignment and data cleaning, and a preliminary spatial database is constructed based on the location of the transmission tower foundations. Based on the preliminary spatial database, a longitudinal correlation is established for each tower foundation, mapping the underground structure of the foundation to the surface environment data to describe the longitudinal impact of environmental factors on the foundation. Based on the spatial proximity relationship between towers, a proximity influence association rule is established to cross-correlate the environmental and structural data of upstream and downstream towers and adjacent tower foundations, in order to characterize the propagation effect of environmental changes or diseases in local areas. Basic environmental data are classified according to topographic slope, soil type and hydrological gradient, and environmental gradient association rules are established with the corresponding basic structural data to reflect the potential impact of environmental condition changes on foundation durability and vulnerability. By integrating vertical correlation, proximity influence correlation, and environmental gradient correlation into a unified multi-layer spatial correlation model, data overlay and interactive storage of different spatial scales and different influence types can be achieved. Based on the aforementioned multi-layered spatial association model, a multi-dimensional spatial index is generated for each basic node, including three dimensions: vertical, proximity, and environmental gradient. This index is used for quick querying and calculation of durability and vulnerability analysis indicators, resulting in a spatial basic database.

[0027] It should be noted that the advantage of constructing a spatial foundation database is that it can standardize, spatialize, and multi-layer correlate the environmental and structural data of transmission tower foundations to form a unified and highly structured database. This database can not only clearly describe the stress and deterioration relationships of the foundation in multiple dimensions such as longitudinal, proximity, and environmental gradient, but also enable the overlay and interactive analysis of data of different scales and types. This provides accurate and efficient data support for the durability assessment, vulnerability analysis, risk prediction, and optimized maintenance of power transmission and transformation facility foundations, thereby improving the scientificity and reliability of the analysis.

[0028] S2. Based on the aforementioned spatial database, key evaluation parameters that significantly affect the stability of power transmission and transformation facilities are selected, and key evaluation parameters with discontinuous spatial distribution are identified. Spatial encryption processing is performed using an inverse distance weighted interpolation algorithm to obtain a continuous and high-precision set of basic evaluation parameters.

[0029] In this embodiment, based on the aforementioned spatial database, key evaluation parameters that significantly impact the stability of power transmission and transformation facility foundations are selected, including: Obtain the output indicators of the basic stability analysis, including basic displacement, settlement, bearing capacity or vulnerability index, and establish a mapping relationship between the output indicators and the input parameters in the spatial foundation database, wherein the input parameters refer to the standardized basic environmental data and transmission tower structure data in the spatial foundation database; Based on the input parameters and output indicators, the value range and boundary conditions of each parameter are defined, and a high-dimensional input parameter sample set is generated using Sobol sequences to ensure uniform coverage of the input parameter space. At the same time, it quantifies the contribution rate of each input parameter and its interaction to the output indicators for subsequent sensitivity analysis. It should be noted that the Sobol sequence is a low-discrepancy sequence used to generate uniformly distributed sample points in a high-dimensional space, which can more evenly and systematically cover the entire parameter space compared to random sampling. In this method, the high-dimensional input parameter sample set generated by the Sobol sequence can fully explore the variation of each basic environmental parameter and structural parameter within its value range, thereby ensuring that sensitivity analysis not only quantifies the contribution of individual parameters to the output index, but also accurately reflects the impact of the interaction between parameters on the stability index of power transmission and transformation facilities.

[0030] A basic stability analysis calculation is performed on the high-dimensional input parameter sample set. The output index results corresponding to each set of input parameters are recorded as simulated data. The first-order sensitivity index and total sensitivity index of each basic environmental assessment parameter and structural assessment parameter are calculated by the Sobol sensitivity analysis method. Among them, the first-order sensitivity index represents the contribution rate of a single parameter to the variance of the output index, and the total sensitivity index represents the total variance contribution rate of the parameter and its interaction.

[0031] The specific formula for calculating the first-order sensitivity index is as follows:

[0032] The specific formula for calculating the overall sensitivity index is as follows:

[0033] In the formula, It is a first-order sensitivity index. Y is the overall sensitivity index, and Y is the output index. For input parameters, In order to indicate that, except All other input parameter sets besides To output Y pairs while keeping other parameters constant Conditional expectation, To be in a fixed Under the given conditions, the conditional expectation of the output Y with respect to other parameters. To output the total variance of index Y, For the output index Y, it consists of a single parameter The variance caused by the change, To indicate in a fixed Under the given conditions, the variance of the output index Y is the variance caused by changes in other parameters.

[0034] Based on the first-order sensitivity index and the total sensitivity index, all basic environmental evaluation parameters and structural evaluation parameters are sorted according to their sensitivity. A predetermined threshold is set, and parameters with a first-order sensitivity index or total sensitivity index higher than the threshold are selected as key evaluation parameters, thereby obtaining the key evaluation parameters that have a significant impact on the stability of power transmission and transformation facilities.

[0035] The predetermined threshold can be determined by sorting the sensitivity indices from largest to smallest, and taking the smallest sensitivity index corresponding to the top 20% of parameters as the threshold.

[0036] It should be noted that the advantage of using the Sobol sensitivity analysis method to screen key evaluation parameters for power transmission and transformation facility foundations is that it can systematically and quantitatively identify the single parameter that has the greatest impact on foundation stability and the interaction between parameters, thereby accurately screening the most critical environmental and structural evaluation parameters, providing reliable data support for high-precision stability analysis, vulnerability assessment and optimized maintenance.

[0037] Furthermore, addressing the issue of spatial discontinuous distribution of parameters such as soil and groundwater levels during floods, a specialized encryption step is introduced to generate a continuous parameter set, overcoming the shortcomings of traditional methods in effectively handling spatial heterogeneity. The identification of key evaluation parameters with spatial discontinuities employs an inverse distance weighted interpolation algorithm for spatial encryption processing, resulting in a continuous and high-precision set of basic evaluation parameters, including: Spatial distribution analysis of the key evaluation parameters is performed. First, the parameter values ​​are mapped in GIS or rasterized grid according to their spatial coordinates to obtain the density of neighboring points and spatial change gradient of each spatial unit or sampling point. By analyzing the areas where the density of neighboring points is lower than a preset threshold, the spatial change gradient is higher than a preset change threshold, or there are missing points, the spatial distribution of parameters is discontinuous or the data is sparse, which provides the target location and range basis for subsequent local spatial densification processing. Based on the spatial distribution of the key evaluation parameters, the variable trend kriging method is used to fit the trend surface of the entire spatial area. The trend term includes topographic slope, soil type and hydrological gradient, and the trend surface value is obtained to reflect the spatial variation law of the parameters with environmental factors. Specifically, the variable trend kriging method involves constructing a trend function using topographic slope, soil type, and hydrological gradient as covariates. This function typically employs a first- or second-order polynomial form; for example, the trend term can be represented as a linear combination of the covariates (e.g.,...). ), where the coefficient The estimation is performed using generalized least squares or regression analysis based on observational data. The remaining residuals are modeled using a common kriging model, with the semivariogram function selected from one of the spherical, exponential, or Gaussian models. The parameters of the variogram function (null value, sill value, range) are fitted using the least squares method. During fitting, the empirical semivariogram values ​​at different distance intervals are calculated first, and then the optimal parameters of the theoretical semivariogram model are obtained using the least squares or weighted least squares method. The trend term is then superimposed with the kriging interpolation results to obtain the final trend surface value.

[0038] The residuals of key evaluation parameters at the original sampling points are calculated, which are the differences between the actual observed values ​​and the trend surface values, representing local spatial variability information. The specific formula for calculating the residuals is as follows:

[0039] In the formula, The residuals of the key evaluation parameters at the original sampling points, These are actual observed values. This represents the trend value.

[0040] In areas with discontinuous spatial distribution or sparse data, local interpolation is performed using an inverse distance weighted interpolation algorithm based on trend surface values ​​and residuals. The specific calculation formula for this local interpolation is as follows:

[0041] In the formula, Point to be inserted Local interpolation of key evaluation parameters, Point to be inserted The trend value, For the i-th observation point, the interpolation point is... The weights, where, , The distance between the insertion point and the observation point. For power-order parameters, Let n be the residual of the i-th observation point, and n be the total number of observation points.

[0042] By overlaying local interpolation with trend surface values, a continuous, high-precision set of basic evaluation parameters, including both basic environmental evaluation parameter datasets and structural evaluation parameter datasets, is generated.

[0043] It should be noted that by first identifying key evaluation parameters with discontinuous spatial distribution or sparse data, and then combining trend-changing kriging trend surface fitting and local inverse distance weighted interpolation, the global variation trend of parameters with topography, soil type and hydrological gradient is preserved, while fully reflecting local non-smooth characteristics. This generates a continuous and high-precision dataset of basic environmental and structural evaluation parameters, providing scientific, reliable and directly applicable spatial data support for the durability analysis, vulnerability assessment and optimized maintenance of power transmission and transformation facility foundations.

[0044] S3. Based on the basic evaluation parameter set, construct a basic durability and fragility evaluation sample dataset, and use the improved SMOTE algorithm to oversample the minority class samples to generate an enhanced sample set, including: Based on a continuous and high-precision set of basic evaluation parameters, combined with the historical status and damage records of power transmission and transformation facility foundations, environmental, structural and operational parameters related to foundation durability and vulnerability are extracted, and the samples are labeled according to the foundation durability level to obtain the original sample dataset for foundation durability and vulnerability evaluation. The original sample dataset is subjected to category statistics and distribution analysis to identify minority class and majority class samples, and the number of samples in each class and their distribution characteristics in the parameter space are recorded to determine the target category and spatial distribution structure for subsequent data augmentation. For the identified minority class samples, the K-means clustering algorithm is used to cluster them to obtain the distribution of samples within each cluster. The samples within each cluster have high similarity to ensure that the newly generated samples are reasonably distributed in the local parameter space and avoid cross-cluster marginalization. Specifically, the clustering process involves: for the identified minority class samples, firstly, using their feature vectors (such as environmental parameters, structural parameters, etc.) in the basic evaluation parameter space as clustering input, and determining the optimal number of clusters using the elbow method or silhouette coefficient method, i.e., calculating the sum of squared errors within clusters (SSE) or the average silhouette coefficient under different optimal clustering values, and selecting the optimal clustering value corresponding to the inflection point or maximum value; then, randomly selecting the optimal number of clusters from the minority class samples as initial cluster centers, calculating the Euclidean distance from each sample to each cluster center, dividing the samples into the nearest clusters, and updating the cluster centers according to the mean of all samples in each cluster, repeating the iteration until the maximum number of iterations (e.g., 300 times), and finally outputting the set of samples in each cluster and their distribution characteristics.

[0045] Within each cluster, new samples are generated by random sampling using a probability density-constrained sampling method. The new samples are added to the original minority class samples to obtain an enhanced basic durability and fragility evaluation sample dataset, in which the number of minority class samples is effectively increased, the class distribution is more balanced, and the sample space coverage is higher.

[0046] Furthermore, within each cluster, new samples are generated through random sampling using a probability density-constrained sampling method, including: Within each cluster, based on the multidimensional parameter values ​​of the samples in the cluster, the kernel density estimation method is used to construct the joint probability density function of the samples in the cluster, in order to characterize the distribution law of the samples in the parameter space. The joint probability density function is calculated using the following formula:

[0047] In the formula, To generate the probability density value of a sample point in the parameter space, which reflects the likelihood that the location belongs to the sample distribution within the current cluster; the larger the probability density value, the closer the location is to the concentrated region of samples within the cluster. For samples within a cluster, Where N is the feature dimension and N is the number of samples within the cluster. Let h be the Gaussian kernel function, and let h be the bandwidth. The kernel function uses a Gaussian kernel, and the bandwidth is determined using the Silverman method or cross-validation. To generate sample parameter vectors.

[0048] In each cluster, three sample points are randomly selected to form a triangle, and the centroid of the triangle is calculated as the reference point for generating new samples, ensuring that the generation of new samples follows the local structure within the cluster. Within the parameter space interval between the centroid of the triangle and any vertex, random sampling is performed according to the constructed joint probability density function to generate new minority class samples. This ensures that the distribution of the generated samples follows the density distribution law within the cluster, avoiding deviation from the main cluster region or the generation of marginalized samples.

[0049] Within the parameter space interval between the centroid of the triangle and any vertex, a sampling interval is first defined along the vector direction from the centroid to the vertex. Then, this sampling interval is mapped to the range of the intra-cluster probability density function. Specifically, the sampling interval is discretized to obtain multiple discrete location points. Each discrete point is substituted into the intra-cluster joint probability density function to calculate the corresponding probability density value, thus establishing the correspondence between each location point and the probability density value within the sampling interval, achieving the mapping from the sampling interval to the intra-cluster probability density distribution. Next, the probability density values ​​of all discrete points are normalized to obtain the sampling probability corresponding to each location point. A cumulative probability distribution is then constructed, and random numbers are generated. The corresponding discrete points are selected as new sample generation locations using a cumulative probability matching method. This means that the probability of generating new samples is higher in high-probability-density regions and lower in low-probability-density regions, ensuring that the generated new minority class samples are both close to the cluster center and conform to the actual distribution law of the intra-cluster parameter space. This effectively avoids samples deviating from the cluster core region or falling into the edge region, achieving an enhanced effect that balances the directionality and spatial rationality of sample generation.

[0050] It should be noted that, Figure 3 The traditional SMOTE algorithm suffers from marginalization, resulting in newly generated samples that are not helpful for machine learning training. Therefore, this project uses an improved SMOTE algorithm for secondary data processing. The improvements primarily involve: determining the distribution of minority class samples based on K-means clustering; ensuring high similarity among samples in each cluster; and guaranteeing that each newly generated sample will only be within that cluster. This directly replaces the K-nearest neighbor generation step in the traditional SMOTE algorithm. Figure 4 As shown, the regions within the two elliptical dashed lines are two clusters; three points are randomly selected in each cluster to generate a triangle, and the sample data of the centroid position of triangle BCD is calculated and marked as black points. New samples are randomly generated between its centroid and vertex B. This step can effectively avoid marginalization and make the generation of new samples directional, moving closer to the centroid position.

[0051] S4. Using the enhanced sample set as training data, a BP neural network model is constructed with basic environmental evaluation parameters and structural evaluation parameters as inputs and the evaluation results of the durability or vulnerability of the power transmission and transformation facility foundation as outputs. An intelligent optimization algorithm is then used to globally optimize the initial weights and thresholds of the BP neural network model to obtain an optimized foundation durability and vulnerability prediction model.

[0052] In this embodiment, using the enhanced sample set as training data, a BP neural network model is constructed, with basic environmental evaluation parameters and structural evaluation parameters as inputs and the evaluation results of the durability or vulnerability of the power transmission and transformation facility foundation as outputs, including: Based on the enhanced sample set, key input features affecting the stability index of transmission towers are determined. These key input features include elevation, slope, distance from the fault zone, cohesion, friction angle, normalized vegetation index, rainfall, tower foundation load, and transmission tower type. The number of neurons in the input layer is set to 9 to correspond to each key input feature. The number of neurons in the output layer is determined and set to 1, representing the prediction result of the stability of the transmission line foundation, which is a numerical assessment of the foundation's durability or vulnerability. The number of hidden layer neurons affects the data processing performance of the BP neural network model. At the same time, under the premise of achieving better network performance and generalization ability, the number of hidden layer neurons should be reduced as much as possible to reduce training time and quickly obtain prediction results. Based on the number of input layer neurons and the number of output layer neurons, the number of hidden layer neurons can be initially determined using a preset empirical formula. The empirical formula, specifically the calculation formula, is as follows:

[0053] In the formula, denoted as the optimal number of hidden layer neurons, b as the number of input layer neurons, v as the number of output layer neurons, and a as any integer from 0 to 10, representing the number of hidden layers.

[0054] An activation function maps a model's input to its output. Nonlinear activation functions increase the model's nonlinearity, thereby enhancing its expressive power and enabling it to learn and process more complex data and information. The three most widely used activation functions are Log-Sigmoid, Tan-Sigmoid, and Purelin. Both the hidden and output layers can use one of these three activation functions. By experimenting with each activation function in the hidden and output layers, calculating the mean squared error (MSE) of each function, and selecting the activation function with the smallest MSE as the optimal function, the optimal activation function can be determined. The learning rate refers to the amount of weight updates during model training, ranging from 0 to 1. For example, a learning rate of 0.2 means that each time the weights are updated, 20% of the estimated error is updated. However, as training progresses, an excessively large learning rate can lead to significant fluctuations, even causing the loss function value to oscillate around its minimum, making it difficult to reach the optimal value. The learning rate interacts with many other aspects of the optimization process, and these interactions are non-linear. Contrary to expectations, generally, a lower learning rate should require more training time. On the other hand, a higher learning rate should require less training time. Furthermore, considering noisy error gradient inference, a smaller sample size is more suitable for a lower learning rate. After determining the activation function, the training results are tested with different learning rates to select an appropriate learning rate value. Through repeated trial and error, a learning rate suitable for the current data and model structure is determined to ensure the convergence and stability of the training process. The learning rate should be adjusted according to the noise characteristics of the training data to avoid overfitting or underfitting. Based on a defined learning rate, the number of neurons in the input layer, the number of neurons in the output layer, the number of neurons in the hidden layer, and the activation function, the BP neural network is trained using an augmented sample set. During the training process, the network weights and thresholds are continuously adjusted according to the loss function (e.g., mean squared error) until the network converges and reaches the best prediction accuracy, thus obtaining the BP neural network model.

[0055] Furthermore, BP neural networks possess strong learning capabilities, nonlinear mapping capabilities, and generalization tolerance, leading to their widespread application in practice. However, BP neural network models are sensitive to initial weights and hidden layer values, and since weights and thresholds are randomly assigned and lack logical consistency, the model's prediction results may be biased. Additionally, BP neural network models can exhibit local minima. This is because their core algorithm adjusts weights based on error gradient descent; when multiple local minima appear, the model's computation becomes trapped and unable to escape. The network also has strict requirements on the dimensionality of input samples, as the addition of new samples can affect the previously trained model. Therefore, it is necessary to optimize BP neural networks to overcome these limitations and enable nonlinear prediction of transmission line foundation stability.

[0056] In this embodiment, an intelligent optimization algorithm is used to globally optimize the initial weights and thresholds of the BP neural network model to obtain an optimized basic durability and vulnerability prediction model, including: The sparrow search algorithm is used to initialize the weights and thresholds of the BP neural network. Based on the foraging behavior of sparrows in nature, the sparrow search algorithm simulates their global search capability to determine the weights and thresholds between layers in the neural network, avoiding training errors caused by random assignment. Individual initialization in the sparrow search algorithm uses preset boundary conditions to ensure that each individual does not exceed a pre-defined reasonable range during the search process.

[0057] The fitness function of the sparrow search algorithm is configured to evaluate the prediction error of the current weight and threshold combination. Specifically, the fitness function uses the mean squared error (MSE) as the optimization objective; the smaller the MSE value, the higher the fitness, indicating that the weight and threshold combination performs better in prediction. A sparrow population is formed by randomly generating a preset number of sparrow individuals. Each sparrow individual contains a set of weights and thresholds, which are used as the initial solution input into the BP neural network model to train the network and calculate the prediction error (i.e., the error between the output result and the actual result). The optimal solution is found by simulating the foraging behavior of sparrows to perform a global search. In each iteration, the fitness value of each sparrow is calculated, and the individual with the highest fitness value is selected as the leader individual. Other individuals perform local searches based on the position of the leader individual and use update rules to jump within the search space to avoid getting trapped in local minima. The update rules of the sparrow search algorithm include two strategies: a random jump strategy and a local jump strategy. The random jump strategy is used for global search, and the local jump strategy is used to refine the search results. Sparrows are ranked according to their fitness values, and their positions (i.e., weights and thresholds) are updated by introducing two strategies from sparrow foraging behavior (i.e., random leaps and local leaps). After each round of updates, the fitness value of each individual is recalculated, and the optimal weights and thresholds are determined by comparing the fitness values. Set a stopping criterion for the optimization process. When the number of iterations of the sparrow search algorithm reaches a preset value, or the fitness value (i.e., mean squared error) drops to the target accuracy, the optimization process is terminated, and the final optimized BP neural network weights and thresholds are obtained. Using the optimized weights and thresholds as initial values, the BP neural network model is retrained to obtain the optimized basic durability and vulnerability prediction model.

[0058] The training process employs the backpropagation algorithm to continuously adjust the weights and thresholds between neurons until the network converges and reaches the preset prediction accuracy. The optimization effect of the sparrow search algorithm is verified by comparing the prediction results before and after optimization.

[0059] It should be noted that the sparrow search algorithm simulates sparrow foraging behavior and globally optimizes the weights and thresholds of the BP neural network during training, thereby improving the model's prediction accuracy and generalization ability. This method avoids the local minima problem of traditional BP neural network models, enabling more accurate assessment of the stability of transmission line foundations and providing more effective technical support for power engineering.

[0060] S5. Based on the prediction model, establish the reliability state equation of the power transmission and transformation facility foundation under flood disaster conditions, calculate the corresponding foundation reliability index, and quantitatively assess the durability and failure risk of the power transmission and transformation facility foundation under different flood disaster conditions based on the reliability index and vulnerability analysis results. Identify the weak links in the foundation structure and output corresponding optimization suggestions, including: Obtain flood disaster-related data, including historical rainfall, river flow, soil permeability, groundwater level, and other environmental parameters that affect the stability of power transmission and transformation facility foundations; Based on the flood disaster-related data and combined with the basic design parameters of power transmission and transformation facilities (such as foundation type, geological conditions, and tower load), the key factors affecting foundation reliability are identified and determined. By utilizing key factors and combining the prediction results of the optimized foundation durability and vulnerability prediction model, a foundation reliability state equation is constructed. This equation is used to describe the relationship between various variables affecting foundation reliability under different flood disaster conditions, and the foundation reliability is calculated based on the changes of these variables. Based on the aforementioned reliability state equation and combined with specific flood disaster conditions, the corresponding foundation reliability index is calculated. Specifically, the Monte Carlo simulation analysis method is used to evaluate the reliability of the foundation under the action of flood disaster.

[0061] The specific calculation formula for the aforementioned basic reliability index is as follows:

[0062] In the formula, It is a basic reliability index, that is, the probability of stability of a foundation under specific flood disaster conditions. The total number of simulated samples, The base stability value for the i-th simulation is obtained by predicting it using the optimized base durability and fragility prediction model. For indicator functions, if but Otherwise, it is 0. The threshold for basic failure.

[0063] Based on the aforementioned basic reliability indices, and considering the structural characteristics and load requirements of the power transmission and transformation facilities, a vulnerability analysis of the foundation is conducted. The vulnerability analysis should consider the foundation's response characteristics under different flood disaster conditions, such as settlement, tilting, and fracture. Structural characteristics include the foundation's dimensions, materials, shape, and soil bearing capacity, while load requirements include the weight of the equipment, external environmental loads (such as wind and snow loads), and additional loads caused by flood disasters (such as water pressure and changes in soil moisture). Then, using probabilistic analysis methods, such as Monte Carlo simulation and Latin hypercube sampling, a large amount of sample data is generated by randomly sampling multiple input variables (such as precipitation, river flow, and soil moisture) to evaluate the foundation's response characteristics under the influence of these random variables. Through simulation and analysis, the failure probability of the foundation under specific flood disaster conditions can be calculated, i.e., the probability of the foundation experiencing failure modes such as settlement, tilting, or fracture.

[0064] Based on the vulnerability analysis results and reliability indicators, calculate the failure risk index of the foundation under different flood disaster conditions, and output a foundation failure risk analysis report. The report should include the failure risk assessment results under different flood disaster conditions, and provide the potential impact range, failure mode and probability distribution of foundation failure. When outputting the foundation failure risk analysis report, it is first necessary to summarize the failure risk indicators calculated during the vulnerability analysis process, and combine them with the reliability assessment results of the foundation under different flood disaster conditions. The report should include the failure modes of the foundation (such as settlement, tilting, and fracture), the probability of occurrence of each failure mode under different disaster conditions, and its potential impact range. For example, if the foundation settlement caused by flooding exceeds a predetermined threshold, the report needs to specify the probability of this phenomenon and further assess its impact on the normal operation of power transmission and transformation facilities, such as equipment damage and service interruption. The report should also list in detail the main factors that may cause failure, such as soil moisture, groundwater level changes, and rainfall, and provide the contribution of these factors to failure based on the probability analysis results. At the same time, the report should include failure risk curves under different flood disaster conditions, showing the risk level of the foundation under different environmental conditions, to help decision-makers assess the threat of various disasters to the stability of the foundation. Finally, the report should provide corresponding risk control and optimization recommendations, such as reinforcing high-risk areas, adjusting foundation design, or implementing disaster prevention measures to reduce failure risk and improve the long-term stability of the foundation.

[0065] The specific calculation formula for the failure risk index is as follows:

[0066] In the formula, This is a foundation failure risk indicator, which measures the probability of foundation failure under specific flood disaster conditions. ,like but Otherwise, it is 0.

[0067] Based on the foundation failure risk analysis report, and combined with the foundation reliability and vulnerability analysis results, weak links in the foundation structure were identified. First, it's crucial to comprehensively analyze the various failure risk indicators and vulnerability analysis data provided in the report, especially those risk points exhibiting high failure probabilities under different flood disaster conditions. For example, if the analysis shows that certain foundation locations or design features (such as changes in soil moisture or rising groundwater levels) lead to a higher risk of settlement or tilting under specific flood conditions, these areas may be weak points. Furthermore, if the report indicates a high probability of failure modes (such as excessive settlement or structural cracks), it's necessary to further identify problems in the foundation design or construction process, such as inappropriate material selection, construction quality issues, or insufficient soil bearing capacity. By combining reliability data with failure modes, it's possible to pinpoint which design parameters, structural components, or operating conditions are most prone to failure, effectively identifying the weak points most susceptible to failure under disaster conditions. Ultimately, these weak points can help engineers prioritize and implement targeted reinforcement measures, such as optimizing design, improving construction quality, or adjusting material selection, to enhance the overall safety and disaster resistance of the foundation structure.

[0068] For the identified weaknesses, corresponding optimization measures will be developed based on reliability analysis and engineering experience. These measures may include improving foundation design, optimizing material selection, strengthening construction quality control, and enhancing flood resistance. Targeted optimization suggestions will be proposed for different foundation types and weaknesses.

[0069] It should be noted that by combining flood disaster data, foundation reliability, vulnerability analysis, and failure risk assessment, a comprehensive and accurate evaluation of power transmission and transformation facility foundations was achieved. This scheme can quantitatively assess the stability and failure risk of foundations under different flood disaster conditions and identify potential weak points in the foundation structure, thus providing a scientific basis for foundation design and operation and maintenance. Through improvements in design, material selection, and construction quality, the disaster resistance and long-term durability of infrastructure can be effectively enhanced, reducing the risk of failure during disasters and ensuring the safe operation of power transmission and transformation facilities. This method not only improves infrastructure reliability but also provides timely and effective decision support for post-disaster recovery, demonstrating significant practical engineering application value and economic benefits.

[0070] Reference Figure 2As shown in the schematic diagram, the system for analyzing the durability and vulnerability of power transmission and transformation facility foundations provided by this invention includes a database construction module, a data processing module, a sample optimization module, a model construction module, and an analysis module. The modules are interconnected. The database construction module is used to acquire basic environmental data and transmission tower structure data to build a spatial basic database; The data processing module is used to filter out key evaluation parameters from the database and spatially encrypt key evaluation parameters with discontinuous spatial distribution to obtain a continuous set of basic evaluation parameters. The sample optimization module is used to oversample minority class samples using the improved SMOTE algorithm to generate an enhanced sample set from the basic evaluation parameter set. The model building module is used to build a BP neural network model using the augmented sample set as training data, and to globally optimize its initial weights and thresholds to obtain a prediction model. The analysis module is used to establish a basic reliability state equation under flood disaster conditions based on the prediction model, calculate the reliability index, and combine the vulnerability analysis results to quantitatively assess the durability and failure risk under different flood conditions, identify weak links, and output optimization suggestions.

[0071] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0072] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0073] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0074] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0075] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0076] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for analyzing the durability and vulnerability of power transmission and transformation facility foundations, characterized in that, include: Acquire basic environmental data and transmission tower structure data to build a basic database; Parameters are filtered from the database, and spatial densification is performed on parameters with discontinuous spatial distribution to obtain a continuous set of basic evaluation parameters; Oversampling is performed on minority class samples in the basic evaluation parameter set to generate an enhanced sample set; The oversampling process includes: clustering minority class samples, and within each cluster, sampling between the centroid and vertices of a triangle formed by three sample points to generate new samples based on kernel density probability weighting. The neural network is trained with an enhanced sample set, and its initial weights and thresholds are globally optimized to obtain a prediction model; A flood reliability equation is established based on the prediction model, the reliability is calculated, the failure risk is assessed by combining vulnerability analysis, weak links are identified, and optimization suggestions are output.

2. The method according to claim 1, characterized in that, The construction of the basic database includes: A preliminary spatial database was constructed based on the locations of the transmission tower foundations; Establish vertical relationships for each tower foundation; Establish proximity influence association rules based on the spatial proximity relationship between towers; The basic environmental data is classified and environmental gradient association rules are established with the structural data. The vertical correlation, proximity influence correlation, and environmental gradient correlation are integrated into a multi-layer spatial correlation model, and a multi-dimensional spatial index containing vertical, proximity, and environmental gradient is generated for each basic node to obtain a spatial basic database.

3. The method according to claim 1, characterized in that, The step of filtering parameters from the database includes: A mapping relationship is established between the output indicators and the input parameters in the spatial basic database; A high-dimensional input parameter sample set is generated using Sobol sequences to cover the input parameter space; Basic stability analysis calculations were performed on the sample set to obtain simulation data; The first-order sensitivity index and the total sensitivity index of each input parameter were calculated using the Sobol sensitivity analysis method. The parameters are sorted according to the first-order sensitivity index and the total sensitivity index, and the parameters that are higher than the preset threshold are selected as key evaluation parameters.

4. The method according to claim 1, characterized in that, The process of spatially densifying the discontinuous parameters to obtain a continuous set of basic evaluation parameters includes: For key evaluation parameters, spatially discontinuous or sparse data regions are identified based on the density of neighboring points and spatial variation gradient. Trend surface values ​​are obtained by fitting a trend surface to the entire spatial region. Calculate the residuals of key evaluation parameters at the original sampling points; In the spatially discontinuous or sparse data regions, local interpolation is performed using an inverse distance weighted interpolation algorithm based on trend surface values ​​and residuals; The local interpolation results are overlaid with the trend surface values ​​to generate a continuous set of basic evaluation parameters.

5. The method according to claim 1, characterized in that, The process of oversampling minority class samples in the basic evaluation parameter set to generate an enhanced sample set includes: Based on the basic evaluation parameter set and historical damage records, an original sample dataset is constructed. Identify the minority class samples, perform clustering, and obtain the sample distribution within each cluster; New samples are generated in each cluster using a probability density-constrained sampling method. These new samples are then added to the original minority class samples to obtain enhanced evaluation sample data.

6. The method according to claim 5, characterized in that, The generation of new samples in each cluster using a probability density-constrained sampling method includes: Within each cluster, a joint probability density function for samples within the cluster is constructed using kernel density estimation; In each cluster, three sample points are randomly selected to form a triangle, and the centroid of the triangle is calculated. Within the parameter space interval between the centroid and any vertex, weighted random sampling is performed according to the joint probability density function to generate new minority class samples, such that the sampling probability of high-density regions is higher than that of low-density regions.

7. The method according to claim 1, characterized in that, The method of training the neural network with an enhanced sample set includes: Using the augmented sample set as training data, the number of neurons in the input layer and the number of neurons in the output layer are set. The number of hidden layer neurons is determined based on the number of neurons in the input layer and the number of neurons in the output layer. By comparing the mean squared error of each activation function, the activation function with the smallest mean squared error is selected as the optimal function. The learning rate is determined through trial and error to bring the training to convergence. The network was trained using an enhanced sample set to obtain a BP neural network model.

8. The method according to claim 7, characterized in that, The global optimization of its initial weights and thresholds to obtain the prediction model includes: Initialize the weights and thresholds of the BP neural network; Set a fitness function with mean squared error as the optimization objective; Global search is performed by simulating sparrow foraging behavior, which includes a random leap strategy and a local leap strategy, used for global search and local refinement, respectively. The individual positions are sorted and updated according to the fitness values. When the number of iterations reaches a preset value or the mean square error drops to the target accuracy, the optimization is terminated, and the optimized weights and thresholds are obtained. The optimized weights and thresholds are used as initial values ​​to train the BP neural network, thereby obtaining the optimized basic durability and vulnerability prediction model.

9. The method according to claim 1, characterized in that, The process involves establishing a flood reliability equation based on a prediction model, calculating reliability, assessing failure risk through vulnerability analysis, identifying weak points, and outputting optimization suggestions, including: Obtain flood disaster-related data and basic design parameters, and construct the foundation reliability state equation by combining the optimized foundation durability and vulnerability prediction model; The reliability index of the calculation basis is calculated under different flood disaster conditions; Based on the reliability index and the response characteristics of the foundation under flood disaster, a vulnerability analysis is performed, the failure risk index is calculated, and a risk analysis report is output. Based on the risk analysis report, weak points in the infrastructure are identified, and optimization suggestions, including improvements to the infrastructure design, material optimization, or flood-resistant reinforcement measures, are output for these weak points.

10. A system using the method for analyzing the durability and vulnerability of power transmission and transformation facility foundations as described in any one of claims 1-9, characterized in that, include: The database construction module is used to acquire basic environmental data and transmission tower structure data to build a basic database. The data processing module is used to filter parameters from the database and spatially encrypt parameters with discontinuous spatial distribution to obtain a continuous set of basic evaluation parameters. The sample optimization module is used to oversample minority class samples in the basic evaluation parameter set to generate an enhanced sample set. The model building module is used to train the neural network with an augmented sample set and globally optimize its initial weights and thresholds to obtain the prediction model. The analysis module is used to establish flood reliability equations based on prediction models, calculate reliability, assess failure risks by combining vulnerability analysis, identify weak links, and output optimization suggestions.

Images