Method and system for evaluating fault probability of broken wire and pole falling of distribution line in strong wind weather

Abstract

The application discloses a power distribution line breakage and pole overturning fault probability evaluation method and system under strong wind weather, and belongs to the technical field of power distribution line fault risk evaluation.The application firstly calculates key physical characteristic quantities such as conductor cross-section stress and pole root bending moment based on fluid mechanics and structural mechanics, and constructs a fusion feature vector in combination with equipment state evaluation scores and basic parameters;secondly, a wind-line-tower coupling dynamic mechanics simulation model is built and calibrated, and a dynamic fault threshold evaluation model is constructed by using random forest regression;further, multi-scenario simulation fault samples are generated, and a strong wind fault balanced sample library is constructed;finally, a LightGBM binary classification probability model is trained, and a target period fault probability is output.The application effectively solves the sample imbalance problem by combining the advantages of physical mechanism and data driving, improves the evaluation accuracy and generalization ability, and can provide scientific support for power distribution network wind disaster early warning and operation and maintenance prevention and control.

Description

Technical Field

[0001] This invention relates to the field of power system disaster prevention and mitigation technology, specifically to a method and system for assessing the probability of power line breakage and pole collapse during strong winds. Background Technology

[0002] Currently, the risk assessment technology for high wind faults in power distribution lines is mainly divided into two types: data-driven and mechanism-based. However, both have obvious technical limitations and cannot meet the needs of large-scale, refined, and highly reliable risk assessment for power distribution networks.

[0003] Data-driven assessment methods use historical wind data, line log data, and corresponding fault records as training samples. They employ machine learning or deep learning algorithms to fit the mapping relationship between meteorological parameters, equipment parameters, and fault results, directly outputting fault judgment results. This type of method has two inherent drawbacks: First, pure data-driven models are essentially statistical fitting models, lacking clear physical mechanism constraints. The prediction results lack mechanical interpretability, failing to identify the core causes of faults or pinpoint specific weak points in the line. Maintenance personnel can only obtain fault conclusions but cannot formulate targeted maintenance and rectification measures, limiting the reliability and guidance value for engineering applications. Second, poor adaptability to operating conditions leads to insufficient generalization ability. The model's predictive ability highly depends on the coverage of the training samples, making relatively reliable judgments only for historically observed wind levels, equipment types, and terrain scenarios. For extreme strong winds exceeding historical extremes, newly commissioned towers or conductors, wind field distortion areas in special terrains, and scenarios with varying equipment degradation, the model's extrapolation ability is limited, and its generalization performance significantly decreases, making it unsuitable for the complex and ever-changing distribution network operating environment.

[0004] Mechanistic model-based assessment methods are based on fluid mechanics and structural mechanics theories. By building a mechanical calculation model of the conductor-tower system, wind load parameters are input to solve for mechanical response quantities such as conductor cross-sectional stress, tower root bending moment, and component stress. The results are compared with material limit thresholds to determine whether a fault has occurred. The physical logic is clear. However, in engineering applications, the simplified assumptions of the model deviate significantly from actual operating conditions. It ignores the dynamic impact effect of pulsating wind, resulting in a significant deviation between the calculated mechanical response quantities and the actual stress state of the line, leading to insufficient accuracy in the assessment results. Secondly, existing models do not consider the impact of equipment deterioration during operation. Most mechanistic models use the factory-rated parameters of the equipment for calculation, without incorporating the aging and deterioration states of the line during actual operation, such as conductor corrosion cross-sectional area loss, concrete pole cracks, tower foundation settlement, and loose connecting bolts. Equipment deterioration significantly reduces the mechanical ultimate bearing capacity of components and is an important cause of high-wind faults in aging lines. Assessment results using rated parameters are often overly optimistic and will seriously underestimate the actual fault risk of aging lines.

[0005] In summary, the two existing technical approaches each have inherent limitations that are difficult to overcome, and cannot simultaneously meet the comprehensive requirements of high precision, high efficiency, interpretability, and quantifiability for high wind fault assessment of power distribution lines. There is an urgent need to propose a new assessment method that integrates the advantages of physical mechanisms and data-driven approaches. Summary of the Invention

[0006] To address the aforementioned deficiencies in existing technologies, this invention provides a method and system for assessing the probability of power line breakage and pole collapse during windy weather. It constructs a fusion feature vector that combines clear physical meaning with the ability to represent real-world operating conditions, fully incorporating equipment deterioration states. Based on simulation, it expands the sample of rare fault conditions to achieve a quantitative probability output of fault risk, improving the identification accuracy and generalization ability of the fault prediction model. This supports maintenance departments in conducting tiered management based on risk levels, enabling precise allocation of disaster prevention and maintenance resources, and enhancing the engineering practical value of the assessment results.

[0007] The technical solution of this invention is as follows:

[0008] Methods for assessing the probability of power line breakage and pole collapse during strong winds include:

[0009] (1) Constructing the fused feature vector:

[0010] A wind load physical model is established based on the principles of fluid mechanics and structural mechanics to calculate the key physical characteristics of conductors and towers. The state quantities of each component of the distribution line are quantified according to the distribution network equipment condition evaluation guidelines to obtain the equipment condition evaluation score. The key physical characteristics and the equipment condition evaluation score are combined with the equipment foundation parameters to form a fused feature vector. The key physical characteristics include at least conductor cross-sectional stress and tower root bending moment.

[0011] (2) Constructing a dynamic mechanical simulation model:

[0012] A wind-line-tower coupled dynamic mechanical simulation model was built based on the ANSYS platform. Equipment foundation parameters, line equipment status parameters, and wind speed and direction data were loaded under normal operating conditions and high wind weather fault conditions. Simulation data including conductor cross-sectional stress and pole root bending moment were output. The simulation data was verified to calibrate the simulation model.

[0013] (3) Constructing a fault threshold assessment model:

[0014] Fault samples were collected during windy weather. Based on the equipment's basic parameters, equipment condition evaluation scores, and fault mechanical values ​​under windy weather fault conditions, a fault label dataset was constructed. A random forest regression model was trained to obtain a fault threshold assessment model, the output of which is the fault threshold of conductor cross-sectional stress and pole root bending moment. The fault mechanical values ​​include conductor cross-sectional stress and pole root bending moment at the time of the fault.

[0015] (4) Constructing simulated fault samples:

[0016] Based on the probability distribution of historical strong wind weather and equipment basic parameters and line equipment status parameters in the actual power grid, simulation samples of different scenario-line combinations are constructed. A dataset of wind speed and direction data and equipment basic parameters and line equipment status parameters is constructed. The simulation model is input and the simulation data including conductor cross-sectional stress and pole root bending moment is output.

[0017] Based on the simulation samples, obtain their corresponding basic equipment parameters and equipment status evaluation scores, input them into the fault threshold evaluation model, and output the fault thresholds of conductor cross-section stress and pole root bending moment; when the simulation data of conductor cross-section stress and pole root bending moment are less than the fault thresholds of conductor cross-section stress and pole root bending moment, the sample is marked as a normal sample, otherwise it is marked as a simulated fault sample.

[0018] (5) Construct a sample library for high wind fault equilibrium:

[0019] The simulated fault samples are merged with real historical high wind weather fault samples and normal operating condition samples to obtain a high wind fault balanced sample library; the fusion feature vector corresponding to each sample is obtained.

[0020] (6) Constructing a failure probability model:

[0021] The LightGBM binary classification probability prediction model is trained using the fused feature vector of the high wind fault equilibrium sample library and the corresponding fault label to obtain the fault probability prediction model. Its output is the fault probability value of wire breakage and pole collapse. The corresponding fault label refers to the fault state of the sample corresponding to the fused feature vector, that is, whether it belongs to the normal state or the wire breakage and pole collapse state, which can be represented by 0 and 1 respectively.

[0022] (7) Real-time fault assessment and output:

[0023] For the power distribution line segment to be evaluated and real-time and forecast wind speed and direction data, a fused feature vector is generated, which is input into the fault probability prediction model, and the probability values ​​of line breakage and pole collapse faults of the line segment within the target time period are output.

[0024] Furthermore, the calculation of the key physical characteristic quantities includes:

[0025] Obtain structural parameters and wind speed and direction data of the target power distribution line segment, establish a wind load physical model based on the principles of fluid mechanics and structural mechanics, and calculate the key physical characteristic quantities acting on the conductor and tower. The key physical characteristic quantities include at least conductor cross-sectional stress and tower root bending moment. The tower root bending moment includes the tower root bending moment caused by the tower body load and the tower root bending moment caused by the conductor tension load.

[0026] Furthermore, the equipment status evaluation quantification takes the overhead line unit as the evaluation unit, and according to the distribution network equipment status evaluation guidelines, the status of each component included in the overhead line is quantified and the score of the overhead line unit is calculated.

[0027] Furthermore, the basic parameters of the equipment include: conductor outer diameter, conductor span, conductor cross-sectional area, pole / tower type, pole / tower parameters, pole length, burial depth, distance from the tip to the crossarm, root diameter, and tip diameter.

[0028] Furthermore, the line equipment status parameters include: tower mechanical characteristic status parameters, tower tilt appearance status parameters, conductor mechanical characteristic status parameters, conductor sag and other appearance status parameters.

[0029] Furthermore, the aforementioned construction of the wind-line-tower coupled dynamic mechanical simulation model based on the ANSYS platform specifically includes: based on the ANSYS platform, sequentially completing the geometric modeling and material property assignment of the tower, conductor, insulator, hardware and foundation in ANSYS, defining the connection constraints of each component, and establishing the wind-line-tower coupled dynamic mechanical simulation model based on the ANSYS platform.

[0030] Furthermore, it also includes verifying the simulation data described in step (2) and calibrating the simulation model;

[0031] Furthermore, the simulation data described in step (2) is verified, specifically including: using the measured equipment basic parameters, line equipment status parameters, wind speed and direction data, conductor cross-sectional stress, and pole root bending moment as calibration benchmarks, and taking the consistency between the simulation data output by the simulation model and the measured data as the optimization goal, quantitative error index verification, statistical significance test, and fault feature key point matching verification are adopted to determine the degree of matching between the ANSYS simulation model output results and the measured data from three dimensions: numerical accuracy, statistical difference, and fault features.

[0032] Furthermore, the fault feature key point matching and verification specifically includes:

[0033] For fault conditions involving wire breakage and pole collapse during strong winds, the consistency of critical fault parameters is verified, including: matching the maximum conductor cross-sectional stress and maximum pole root bending moment before the fault, verifying whether the simulation model can accurately reproduce the ultimate conductor cross-sectional stress and pole root bending moment response under strong wind impact; matching the timing of abrupt changes in conductor cross-sectional stress and pole root bending moment, verifying whether the simulation model can realistically reproduce the mechanical abrupt changes and structural instability characteristics at the moment of wire breakage and pole collapse; and matching the critical fault wind speed and the location of maximum stress concentration in the structure with the actual fault location.

[0034] Further, the calibration of the simulation model specifically includes:

[0035] Based on the multi-dimensional consistency comparison results between the simulation data and the measured data, differentiated parameter tuning logic and corresponding convergence requirements are set for three types of deviations: overall numerical fitting deviation, systematic statistical deviation, and fault transient characteristic deviation. The simulation model iteration process follows the principle of fixed boundary and single variable to adjust the uncertain parameters until all three convergence criteria of overall numerical fitting deviation, systematic statistical deviation, and fault transient characteristic deviation are met, and the model calibration is completed.

[0036] Furthermore, when training the LightGBM binary classification probability prediction model, the loss function and evaluation metrics are configured as follows: the binary cross-entropy loss function is used as the objective function; the training process monitoring metrics are set as ROC-AUC and LogLoss, to simultaneously monitor the model's discrimination ability and probability accuracy.

[0037] A system for assessing the probability of power line breakage and pole collapse during strong winds includes:

[0038] The fusion feature vector construction module is used to establish a wind load physical model based on the principles of fluid mechanics and structural mechanics, and calculate the key physical feature quantities of conductors and towers; according to the distribution network equipment condition evaluation guidelines, the condition quantities of each component of the distribution line are quantified to obtain the equipment condition evaluation score; the key physical feature quantities, the equipment condition evaluation score and the equipment basic parameters are combined to form the fusion feature vector.

[0039] The wind-line-tower coupled simulation modeling module is used to build a wind-line-tower coupled dynamic mechanical simulation model based on the ANSYS platform. It loads equipment foundation parameters, line equipment status parameters, and wind speed and direction data under normal operating conditions and high wind weather fault conditions, and outputs simulation data including conductor cross-sectional stress and pole root bending moment.

[0040] The fault threshold assessment model construction module is used to collect fault samples in windy weather. Based on the equipment basic parameters, equipment condition evaluation scores, and fault mechanical values ​​under the fault conditions in windy weather, a fault label dataset is constructed. A random forest regression model is trained to obtain the fault threshold assessment model, whose output is the fault threshold of conductor cross-sectional stress and pole root bending moment.

[0041] The simulation fault sample generation module is used to construct simulation sample datasets for different scenario-line combinations based on the probability distribution of historical strong wind weather, equipment basic parameters, and line equipment status parameters in the actual power grid. The calibrated dynamic mechanical simulation model is input, and the simulation data containing conductor cross-sectional stress and pole root bending moment is output. The equipment basic parameters and equipment status evaluation scores of the corresponding simulation samples are input into the fault threshold evaluation model to obtain the corresponding fault threshold. After comparing the simulation data with the fault threshold, normal samples and simulated fault samples are marked.

[0042] The high-wind fault equilibrium sample library construction module is used to merge the simulated fault samples with real historical high-wind weather fault samples and normal operating condition samples to construct a high-wind fault equilibrium sample library and match the fusion feature vector corresponding to each sample.

[0043] The fault probability model training module is used to train the LightGBM binary classification probability prediction model with the fusion feature vector of the strong wind fault equilibrium sample library and the corresponding fault label to obtain the fault probability prediction model of broken line and fallen pole. Its output is the fault probability value of broken line and fallen pole.

[0044] The real-time fault assessment output module is used to generate a fusion feature vector for the power distribution line segment to be assessed by combining real-time and forecast wind speed and direction data, inputting it into the fault probability prediction model, and outputting the probability value of line breakage and pole collapse for the line segment within the target time period.

[0045] This invention transforms heterogeneous data from different sources and in different forms into quantitative features with clear physical meaning and state representation capabilities, solving the problem that traditional features simply pile up raw data and have weak physical correlation.

[0046] This invention constructs a wind load physical model based on fluid mechanics and structural mechanics theories. Wind speed and direction are derived through standardized mechanical formulas and transformed into key physical characteristics such as conductor cross-sectional stress and pole root bending moment, which directly characterize the stress state of the line. This establishes a direct mechanical mapping relationship between meteorological parameters and line fault risk, providing underlying physical constraints for subsequent models and avoiding the logical defects of purely data-driven models.

[0047] Based on the "Guidelines for Status Evaluation of Distribution Network Equipment", this invention transforms qualitative and unstructured equipment status information such as inspection records, test data, family defects, and operating history into calculable equipment status evaluation scores through a rule system of deduction, weighting, and status grading. This enables quantitative characterization of the degree of equipment deterioration and solves the inherent defects of traditional mechanism models that only use factory rated parameters and ignore the actual deterioration of equipment during operation.

[0048] This invention combines mechanical and physical characteristics, equipment status characteristics, and basic equipment parameters into a fused feature vector. It also covers three dimensions: load-side force input, equipment-side attribute status, and wind speed and direction. This comprehensively characterizes the influencing factors of line fault risk, ensuring both the physical interpretability of the features and taking into account the differences in actual equipment operating conditions. This provides a comprehensive and reliable input foundation for subsequent machine learning models.

[0049] This invention first uses multi-dimensional measured data under normal and fault conditions to iteratively calibrate the ANSYS wind-line-tower coupled dynamic mechanical model, ensuring that the model's steady-state mechanical response and transient response under extreme conditions closely match the stress characteristics of the real line. The calibrated simulation model can simulate the line mechanical response under different wind speeds, different equipment states, and different line parameters in batches, breaking through the limitations of the small number of real fault condition samples and insufficient scenario coverage, and generating massive amounts of customizable simulation condition data.

[0050] This invention uses real fault condition samples and mechanical values ​​at the time of the fault to train a random forest regression model to obtain a fault threshold assessment model. Essentially, it fits a nonlinear mapping relationship between "line characteristic parameters" and "critical failure threshold." Unlike traditional methods that use fixed rated failure thresholds, this model can output dynamic fault critical values ​​(conductor cross-sectional stress, pole root bending moment) corresponding to different equipment states and structural parameters. This more closely matches the actual failure boundary after equipment degradation, avoiding the problem of underestimating the risk of aging lines with fixed thresholds.

[0051] This invention provides a method for automatically classifying samples into binary categories (normal / faulty) for any scenario-line combination. This is achieved by using a calibrated mechanical simulation model to obtain the simulated mechanical response under the current wind load, and by using a fault threshold assessment model to determine the critical fault threshold under the current equipment condition. The classification is based on whether the simulated mechanical response exceeds the fault threshold. This labeling method requires no manual intervention, is highly efficient, and its labeling rules have clear mechanical basis, ensuring the accuracy of the labels is supported by physical logic.

[0052] This invention merges simulated fault samples with real historical high wind fault condition samples and normal condition samples, adjusts the distribution ratio of positive and negative samples, and constructs a sample library that conforms to the actual power grid. This eliminates the negative impact of sample imbalance on model training from the data level and avoids the model biasing towards the majority class (normal samples).

[0053] LightGBM is based on a gradient boosting decision tree framework. It excels at processing structured tabular data and can autonomously discover the coupling effects between multi-dimensional features, fit the complex nonlinear mapping relationship between features and fault occurrence probability, and capture the fault risk patterns under the combined effect of multiple factors. The algorithm itself has the characteristics of fast training speed, strong generalization ability, and excellent anti-overfitting ability, and is suitable for the complex working conditions of power distribution lines with multiple parameters and multiple scenarios.

[0054] This invention trains a fault probability model based on a balanced sample library of high-wind faults, effectively correcting the problems of model bias towards the majority class and high false negative rate caused by traditional imbalanced sample training. It significantly improves the model's ability to identify rare fault conditions and ensures the model's sensitivity to high-risk fault scenarios. Attached Figure Description

[0055] Figure 1 This is a flowchart of the method for assessing the probability of power line breakage and pole collapse under strong winds, as described in this invention. Detailed Implementation

[0056] See Figure 1 .

[0057] A method for assessing the probability of power line breakage and pole collapse during strong winds includes:

[0058] 1. Construct the fused feature vector:

[0059] A wind load physical model is established based on the principles of fluid mechanics and structural mechanics to calculate the key physical characteristics of conductors and towers. The status of each component of the distribution line is quantified according to the distribution network equipment status evaluation guidelines to obtain the equipment status evaluation score. The key physical characteristics and the equipment status evaluation score are combined with the equipment foundation parameters to form a fused feature vector. The key physical characteristics include at least conductor cross-sectional stress and tower root bending moment.

[0060] Furthermore, in one embodiment of the present invention, constructing the fused feature vector specifically includes:

[0061] 1.1: Calculation of key physical characteristics: Obtain the structural parameters and wind speed and direction data of the target power distribution line section, establish a wind load physical model based on the principles of fluid mechanics and structural mechanics, and calculate the key physical characteristics acting on the conductor and tower. The key physical characteristics include at least the conductor section stress and the tower root bending moment. The tower root bending moment includes the tower root bending moment caused by the tower body load and the tower root bending moment caused by the conductor tension load.

[0062] The structural parameters include, but are not limited to, conductor outer diameter, conductor span, conductor cross-sectional area, pole shape coefficient, pole length, burial depth, distance from the tip to the crossarm, root diameter, and tip diameter;

[0063] The conductor tension load is calculated using the following formula:

[0064]

[0065] In the formula, The conductor load is in kN. The composite wind speed is expressed in m / s. The angle between the wind direction and the line; This is the wind pressure non-uniformity coefficient; This is the wind pressure height variation coefficient; The conductor shape factor; The outer diameter of the conductor is in mm; The span is the conductor distance, in meters (m).

[0066] The stress in the conductor section is calculated using the following formula:

[0067]

[0068] In the formula The stress at the conductor cross section is in MPa. The cross-sectional area of ​​the conductor is... .

[0069] The pole load is calculated using the following formula:

[0070]

[0071] The bending moment at the base of the pole caused by the load on the pole body is calculated using the following formula:

[0072]

[0073] The bending moment at the pole root caused by the conductor tension load is calculated using the following formula:

[0074]

[0075] In the formula, The load on the pole body is kN; This is the pole shape coefficient; These are the pole length, burial depth, and distance from the tip to the crossarm, respectively, in meters; Root diameter and shoot diameter, mm.

[0076] 1.2: Equipment status evaluation quantification: The overhead line unit is used as the evaluation unit. According to the distribution network equipment status evaluation guidelines, the status of each component included in the overhead line is quantified and the score of the overhead line unit is calculated.

[0077] 1.2.1: The quantification of the status of each component refers to calculating the score of a single component, the score of components of the same type, and the score of an overhead line unit.

[0078] The specific calculation method for the score of a single component is as follows:

[0079] The base score for a single component is 100 points. The maximum deduction value corresponding to the state quantity of the corresponding component is deducted to obtain the initial score of the single component. The initial score is then corrected by considering family defects, service life, family defect coefficient, and life coefficient.

[0080] Furthermore, in one embodiment of the present invention, referring to the "Guidelines for Evaluation of Distribution Network Equipment Status", the evaluation of equipment status is quantified by using the overhead line unit (including components such as poles (foundations), conductors, insulators, iron fittings, guy wires, channels, grounding devices and accessories of the overhead line) as the evaluation unit.

[0081] Table 1: State Quantities of Each Component Included in the Evaluation of Overhead Line Units

[0082] 1.2.2: Obtain the status parameters of each component of the overhead line through methods such as inspection, routine tests, family defects, and operation information. The basic score for each component is 100 points, and the maximum deduction value for each component is 100 points. The status parameters and maximum deduction values ​​for each component are shown in Table 2.

[0083] Table 2: Status Quantities and Maximum Deduction Values ​​of Each Component in an Overhead Line Unit 1 Burial depth P1 / P5 40 2 Inclination P1 40 3 crack P1 40 4 Tower materials, fittings, and iron parts are corroded. P1 / P4 30 5 Protection P1 / P5 20 6 settlement P1 / P5 40 7 Low-pressure rod P1 40 8 Sag P2 20 9 Stock break P2 40 10 Retail Stocks P2 25 11 Insulation damage P2 20 12 temperature P2 40 13 load P2 40 14 Corrosion of wires P2 40 15 foreign body P2 40 16 electrical distance P2 / P5 40 17 Crossing distance P2 40 18 Horizontal distance P2 40 19 filthy P3 40 20 damaged P3 40 21 fixed P3 40 22 temperature P4 40 23 Tighten P4 40 24 curvature P4 40 25 Completeness of attachments P4 40 26 Wire corrosion P5 40 27 tension of the pull cord P5 40 28 Protection distance P6 40 29 Appearance of grounding down conductor P7 40 30 Grounding resistance P7 30 31 Complete signage P8 30 32 Installation of fault indicators, etc. P8 30

[0084] 1.2.3: Calculation of score for a single component:

[0085] Single component score

[0086] Base score for a single component 100—The maximum deduction value in the corresponding component state variables. For components with family defects, the family defect coefficient is used. Components without family defects Lifespan factor (100 - years of operation × 0.3) / 100;

[0087] 1.2.4: Score Calculation for Similar Components:

[0088] Based on the scores of each individual component in the same category, each component is divided into four states: normal state, attention state, abnormal state, and critical state.

[0089] When all components of the same category in the overhead line to be evaluated are in normal condition, the score of that component is the arithmetic mean of the scores of all individual components in that category.

[0090] When one or more components of the same type are in a state that is below normal, the score of that component is the lowest score among all components in that type.

[0091] Furthermore, in one embodiment of the present invention, based on the calculated score of a single component, the evaluation results of a single component are divided into four states: "normal state", "attention state", "abnormal state" and "serious state" with reference to Table 3.

[0092] When all components of a certain type of overhead line to be evaluated are in normal condition, the score of that type of component is the arithmetic mean of the scores of all individual components; when one or more components score below normal condition, the score of that type of component is the same as the component with the lowest score.

[0093] Table 3: Relationship between Evaluation Score and Condition of Overhead Line Components

[0094] 1.2.5: Calculation of score for overhead line unit:

[0095] The scoring rules for overhead line units are as follows:

[0096] When all categories of components are in normal condition, the final score of the overhead line unit is the result of the weighted sum of the scores of each category of components according to the preset weight coefficients.

[0097] When one or more components are in a state below normal, the final score of the overhead line unit is the score corresponding to the component in the worst state.

[0098] Furthermore, in one embodiment of the present invention, when all categories of component scores of the overhead line unit are in a normal state, the state of the overhead line unit is a normal state, and the final score of the overhead line unit is the result of a weighted sum of the scores of all categories of components according to a preset weighting coefficient; the score M of the overhead line unit is... If one or more components score below the normal level, the overhead line unit is classified as the worst-performing component. The final score of the overhead line unit is the score corresponding to the worst-performing component. In the formula, Kp represents the score weight of each component. See Table 4 for reference.

[0099] Table 4: Weights of each component in an overhead line unit

[0100] 1.3: Construction of Feature Vector Fusion

[0101] The key physical characteristics and equipment condition evaluation scores are combined with the equipment's basic parameters to form a fused feature vector. The key physical characteristics include pole root bending moment and conductor cross-sectional stress. Pole root bending moment includes the bending moment caused by pole body load and the bending moment caused by conductor tension load.

[0102] Furthermore, in one embodiment of the present invention, the calculated conductor cross-sectional stress, the pole root bending moment caused by pole load, the pole root bending moment caused by conductor tension load, the equipment condition evaluation score, and the equipment foundation parameters are combined to form a feature vector. .in Key physical characteristics include the bending moment at the pole root caused by pole body load and the bending moment at the pole root caused by conductor tension load. Stress in conductor cross section , It is the equipment status evaluation score. These are the basic parameters of the equipment (including conductor outer diameter, conductor span, conductor cross-sectional area, pole type, pole parameters, pole length, burial depth, distance from the tip to the crossarm, root diameter, tip diameter, etc.).

[0103] 2. Construction of dynamic mechanical simulation model:

[0104] A wind-line-tower coupled dynamic mechanical simulation model was built based on the ANSYS platform. Equipment foundation parameters, line equipment status parameters, and wind speed and direction data were loaded under normal operating conditions and fault conditions in strong winds. The simulation data of line mechanical response, including conductor cross-sectional stress and pole root bending moment, were output. The simulation data was verified to calibrate the simulation model. The line equipment status parameters include: pole mechanical characteristic status quantities, pole tilt appearance status quantities, conductor mechanical characteristic status quantities, conductor sag and other appearance status quantities.

[0105] Furthermore, in one embodiment of the present invention, the collection of equipment basic parameters, line equipment status parameters, and wind speed and direction data under normal operating conditions and fault conditions during strong winds of the power distribution line specifically includes:

[0106] 2.1 Normal Operating Condition Scenario: During a typical weather period characterized by long-term stable operation of the line, no faults, and no abnormal meteorological disturbances, the basic parameters of the equipment are obtained, including conductor outer diameter, conductor span, conductor cross-sectional area, tower type, tower parameters, tower length, burial depth, distance from the top end to the crossarm, root diameter, and top diameter. Wind speed and direction data are also obtained. Line equipment status parameters are collected, including tower mechanical characteristic status quantities, tower tilt appearance status quantities, conductor mechanical characteristic status quantities, conductor sag, and other appearance status quantities. These serve as calibration benchmarks for the normal response of the simulation model, used to verify the accuracy of the steady-state mechanical response of the simulation model under normal loads.

[0107] 2.2 High Wind Weather Fault Conditions: Typical fault cases of distribution line breakage and pole collapse caused by high winds in the history of regional power grids were selected, and complete data of the entire process of the fault period were collected, including wind speed and direction data before and after the fault; the status parameters of line equipment at the time of the fault, and the boundary conditions and load conditions of the high wind weather fault conditions were fully reproduced to verify the dynamic response and fault triggering mechanism of the simulation model under extreme conditions.

[0108] 2.3: Building a Wind-Line-Tower Coupled Dynamic Mechanics Simulation Model Based on the ANSYS Platform: Based on the ANSYS platform, the geometric modeling and material property assignment of the tower, conductor, insulator, hardware, and foundation are completed sequentially in ANSYS. Connection constraints for each component are defined, and a wind-line-tower coupled dynamic mechanics simulation model based on the ANSYS platform is established. Using the ANSYS load timing function, equipment foundation parameters, line equipment status parameters, and wind speed and direction data under normal operating conditions and high wind weather fault conditions are applied to the tower and line structure system across the entire domain. Simulation data including conductor cross-sectional stress and tower root bending moment are output, realistically simulating the transient response of the entire process of dynamic coupled vibration, stress transmission, and tension fluctuation of the line-tower structure under high wind excitation.

[0109] Due to limitations in on-site testing conditions and equipment operating environment, some key mechanical parameters could not be accurately measured, and precise calibration data from manufacturers was unavailable. These parameters are considered uncertain parameters in ANSYS modeling and mainly include: conductor aerodynamic damping coefficient, overall structural damping ratio, and equivalent contact stiffness of hardware connections. These parameters directly affect the response accuracy of ANSYS dynamic simulation.

[0110] 2.4: Verify the simulation data and calibrate the simulation model.

[0111] The simulation data is verified, specifically including: using the measured equipment foundation parameters, line equipment status parameters, wind speed and direction data, conductor cross-sectional stress, and pole root bending moment under normal operating conditions and high wind weather fault conditions as calibration benchmarks; taking the consistency between the simulation data output by the simulation model and the measured data as the optimization goal; and adopting a three-dimensional consistency verification system of quantitative error index verification + statistical significance test + fault feature key point matching verification. From the three dimensions of numerical accuracy, statistical difference, and fault characteristics, the degree of matching between the ANSYS simulation model output results and the measured data is comprehensively determined.

[0112] Furthermore, in one embodiment of the present invention, a three-dimensional consistency verification system is adopted, which combines quantitative error index verification, statistical significance testing, and fault feature key point matching verification. Specifically, it includes:

[0113] 2.4.1: Quantitative Error Index Verification

[0114] The errors between key mechanical parameters (pole root bending moment, conductor section stress) output by ANSYS simulations under normal operating conditions and fault conditions in strong winds, and those measured in the field, are quantified. Evaluation metrics include: Mean Absolute Percentage Error (MAPE) and Coefficient of Determination (R²). MAPE quantifies the relative error between simulated and measured values, evaluating absolute accuracy. R² evaluates the degree of trend fit between simulated and measured curves, reflecting the model's ability to replicate data variation patterns. The calculation formulas are as follows:

[0115] For the k-th sample group, with m sampling points, the formula for Mean Absolute Percentage Error (MAPE) is:

[0116]

[0117] The overall average MAPE of the sample is:

[0118]

[0119] In the formula: , , respectively, are the measured value and the ANSYS simulation value of the i-th sampling point of the k-th sample group, m is the number of sampling points for a single sample, and n is the total number of valid samples.

[0120] For the k-th sample group, with m sampling points, the formula for the coefficient of determination (R²) is:

[0121]

[0122] The overall goodness of fit of the sample is taken as the mean:

[0123]

[0124] In the formula: This represents the mean of the measured data for the k-th sample group; the meanings of the other parameters are the same as above.

[0125] Engineering judgment criteria: Under normal operating conditions, MAPE≤8% and R²≥0.9; under extreme operating conditions with strong winds and dynamic disturbances, the requirements can be appropriately relaxed, and MAPE≤12% and R²≥0.85 can be used to determine that the numerical fitting accuracy of the model meets the requirements.

[0126] 2.4.2: Statistical significance test

[0127] MAPE and R² parameters only reflect the degree of data fit and cannot distinguish whether the error originates from random disturbances in the field or inherent defects in the model. To solve this problem, a statistical hypothesis testing method is used to identify systematic biases and random errors, further improving the rigor of model calibration: ANSYS simulation data and field measured data under the same working conditions are used as paired samples, and significance is determined by calculating the p-value (the p-value is the probability of the current simulation deviation or a more extreme deviation occurring under the premise that the model has no systematic errors and the deviations are all random disturbances).

[0128] When calculating the P-value, it is necessary to distinguish between samples under normal operating conditions and fault conditions caused by strong winds. Under normal operating conditions, the residuals between simulation data and measured data follow a normal distribution, and the paired-samples T-test method is used. However, under extreme operating conditions such as strong wind impact and fault criticality, the stress and displacement response of the line conductor cross-section exhibit abrupt and sudden changes, and the residuals cannot meet the requirements of normal distribution. The traditional T-test results are invalid. In this case, the Wilcoxon signed-rank nonparametric test is used to ensure the effectiveness and rigor of the full-condition model verification.

[0129] (1) Calculation of P-value for paired T-test (normal operating conditions)

[0130] Multiple independent samples of normal operating conditions (without failure) are selected. For the k-th sample group, with m sampling points, the overall average deviation of sample k is calculated. Characterizes the overall simulation deviation level:

[0131]

[0132] Mean of residuals of n independent samples:

[0133]

[0134] Sample bias standard deviation:

[0135]

[0136] T-statistic and degrees of freedom:

[0137]

[0138] Two-tailed test to determine the p-value:

[0139]

[0140]

[0141] In the formula: For degrees of freedom The corresponding t-distribution probability density function, Let n be the gamma function and n be the total number of samples.

[0142] (2) Calculation of P-value for Wilcoxon signed-rank nonparametric test (fault condition in strong winds)

[0143] Multiple independent samples of fault conditions during strong winds are selected. For the k-th sample group with m sampling points, the overall average deviation of sample k is calculated. Characterizes the overall simulation deviation level:

[0144]

[0145] To ensure the validity of the test, [the following is removed] The number of completely unbiased invalid samples is denoted as n', and all subsequent rank statistics are based on the n' groups of valid biases.

[0146] Calculate the absolute value of the bias for all valid samples, eliminate positive and negative interference, retain only the magnitude of the bias, and obtain the effective absolute value of the bias:

[0147] .

[0148] All Sort the samples in ascending order and assign them ranks 1, 2, 3, ... When multiple samples have the same absolute value of bias and the order cannot be distinguished, the average rank method is used to calculate the rank to avoid statistical error. The formula is: .

[0149] In the formula: The initial rank after sorting equal samples. The final rank is used to sort equal samples. This average rank is used uniformly for all equal samples to ensure statistical fairness.

[0150] Based on the original deviation The positive and negative attributes divide all ranks into two categories: positive rank and negative rank.

[0151] when This indicates that the measured value is greater than the simulated value, and the corresponding rank is positive rank.

[0152] when This indicates that the measured value is less than the simulated value, and the corresponding rank is negative.

[0153] The positive rank sum and negative rank sum are obtained by summing them separately: , .

[0154] In the formula: Let be the rank of the k-th valid sample group; The total rank sum of positive deviations reflects the degree to which the overall simulated value is too small. The total rank sum of negative deviations reflects the degree of overall overestimation of the deviation in the simulation values.

[0155] Considering the limited number of valid samples of fault conditions during strong winds, the rank-sum statistic follows a standard Wilcoxon signed-rank exact distribution. By traversing the probability distribution of all rank combinations, the cumulative probability of the current statistic and more extreme cases is calculated to obtain the two-sided P-value, as follows:

[0156]

[0157] In the formula: The rank sum observation statistic, calculated from actual measurements of n' fault condition samples, directly reflects the degree of imbalance in the rank of the sample bias.

[0158] The one-sided cumulative probability (the total probability of the current bias or a more extreme rank bias distribution occurring under the premise that the model has no systematic bias and the bias is only random perturbation) is calculated as follows:

[0159]

[0160] In the formula: T is the theoretical rank sum random variable, which takes the value of a positive integer in the range of [1, n'(n'+1) / 2]. The probability mass of a theoretical rank and random variable taking exactly a specific discrete value t can be obtained by enumerating all equal probability combinations of positive and negative ranks.

[0161] Based on probability symmetry, a correction is made by taking the minimum value of the symmetric tail and amplifying it by a factor of 2 to obtain the final standardized two-sided P-value:

[0162]

[0163] Engineering Judgment Criteria: The preset engineering general significance level α=0.05. When P>0.05, it means that the difference between simulation and actual measurement is not statistically significant, and the error comes entirely from random fluctuations in wind field turbulence, sensor measurement noise and other random factors on site. The physical mechanism and coupling relationship of the model are accurate and reliable. When P≤0.05, it means that the difference is not randomly generated and there is a statistically significant systematic bias. It proves that there are defects in the key parameters or physical mechanisms of the ANSYS model such as damping, stiffness and wind load coupling, and iterative parameter correction is required.

[0164] 2.4.3: Fault Feature Key Point Matching Verification:

[0165] For fault conditions involving wire breakage and pole collapse during strong winds, the focus is on verifying the consistency of critical fault parameters, including: matching the maximum conductor cross-sectional stress and maximum pole root bending moment before the fault to verify whether the simulation model can accurately reproduce the ultimate conductor cross-sectional stress and pole root bending moment response under strong wind impact; matching the timing of abrupt changes in conductor cross-sectional stress and pole root bending moment to verify whether the simulation model can realistically reproduce the mechanical abrupt changes and structural instability characteristics at the moment of wire breakage and pole collapse; and matching the critical fault wind speed and the location of maximum stress concentration in the structure with the actual fault location.

[0166] 2.5: Simulation model calibration, specifically including:

[0167] Based on the multi-dimensional consistency comparison results between the simulation data and the measured data, a closed-loop calibration mechanism of "error source tracing—directional parameter tuning—iterative verification" is established. Differentiated parameter tuning logic and corresponding convergence requirements are set for three types of deviations: overall numerical fitting deviation, systematic statistical deviation, and fault transient characteristic deviation. The simulation model iteration process follows the principle of fixed boundaries and single variables to adjust uncertain parameters until all three convergence criteria—overall numerical fitting deviation, systematic statistical deviation, and fault transient characteristic deviation—are met, thus completing the model calibration. The fault transient characteristic convergence criteria refer to the alignment of stress concentration location, tension mutation law, and failure sequence with engineering reality.

[0168] Furthermore, in one embodiment of the present invention, the overall numerical fitting deviation, systematic statistical deviation, and fault transient characteristic deviation are explained.

[0169] 2.5.1: Overall numerical fitting deviation: If the mean absolute percentage error (MAPE) exceeds the standard and the coefficient of determination (R²) does not meet the threshold, it indicates that there is a deviation between the overall mechanical stiffness and constraint conditions of the model and the actual line.

[0170] Corresponding parameter adjustment logic: Prioritize adjusting inherent structural parameters such as conductor elasticity reduction factor, hardware equivalent stiffness, and tower foundation constraint stiffness;

[0171] The corresponding convergence requirements are: after iteration, the normal operating condition meets MAPE≤8% and R²≥0.9, and the high wind fault condition meets MAPE≤12% and R²≥0.85.

[0172] 2.5.2: Systematic statistical bias: If the paired T test or Wilcoxon signed-rank test yields P≤0.05, it proves that the deviation between simulation and actual measurement shows a significant statistical regularity, which belongs to the systematic defect of the model, rather than random noise.

[0173] Corresponding parameter tuning logic: Combining the positive and negative deviations and the rank distribution characteristics, comprehensively correct the global load coefficient, the overall damping system, and the wind load turbulence correction parameters to eliminate the overall offset trend;

[0174] The corresponding convergence requirement is that after the iteration is completed, both types of test results satisfy P>0.05, proving that the bias is only caused by random disturbances and the model has no systematic error.

[0175] 2.5.3: Deviation in transient characteristics of faults: If the critical conductor section stress, critical root bending moment of the pole, and stress concentration location do not match the actual fault characteristics, it indicates that the model is insufficient in characterizing extreme transient conditions, local contact and wear characteristics.

[0176] Corresponding parameter tuning logic: Targeted fine-tuning of conductor micro-wear reduction coefficient, fitting friction damping, and transient impact constraint parameters;

[0177] The corresponding convergence requirements are: the critical wind speed of the fault, the timing of the tension change, and the location of the failure should be consistent with the actual engineering test case.

[0178] The iterative process strictly adheres to the principles of fixed boundaries and single variables, fine-tuning only one uncertain parameter each time, keeping all input conditions and boundary conditions unchanged, and then re-performing coupled simulations. The three-dimensional verification process is repeated, and the final convergence criteria for the model are set as follows: In terms of quantitative error indicators, R² ≥ 0.90 and MAPE ≤ 8% for normal conditions, and R² ≥ 0.85 and MAPE ≤ 12% for high wind fault conditions; in terms of statistical test indicators, both the paired t-test and Wilcoxon signed-rank test satisfy P > 0.05; in terms of fault characteristics, the stress concentration location, tension mutation law, and failure sequence closely match engineering reality. When all three conditions are met, the model completes iterative calibration.

[0179] 3. Fault Threshold Assessment Model Construction

[0180] Real-world fault condition samples were collected during strong winds. Based on the equipment's basic parameters, equipment condition evaluation scores, and fault mechanical values ​​under these conditions, a random forest regression model was trained to obtain a fault threshold assessment model. The output of this model is the fault threshold of conductor cross-sectional stress and pole root bending moment. The fault mechanical values ​​include conductor cross-sectional stress and pole root bending moment at the moment of the fault.

[0181] Furthermore, in one embodiment of the present invention, constructing the fault threshold assessment model specifically includes:

[0182] 3.1: Data Acquisition of Fault Condition Samples

[0183] Select real-world fault conditions during strong winds and collect the equipment's basic parameters, equipment condition evaluation scores, and fault mechanical values. The fault mechanical values ​​refer to the online measured values ​​of conductor cross-sectional stress and pole root bending moment at the time of the fault.

[0184] 3.2: Sample Size Determination for Streaming

[0185] A preset threshold of 50 valid fault condition samples is set as the critical threshold for modeling. The number of valid fault condition samples is counted. If the number of samples is ≥50, it is determined that the sample is sufficient and enters the machine learning dynamic modeling process; if the number of samples is <50, it is determined that the sample is insufficient and enters the theoretical threshold normal distribution correction process.

[0186] 3.3: Construction of Fault Threshold Assessment Model

[0187] Based on samples of fault conditions during strong winds, a random forest regression model is trained using equipment foundation parameters, equipment condition evaluation scores, and fault mechanical values ​​under these conditions. This results in a fault threshold assessment model, whose output is the fault threshold for conductor cross-sectional stress and pole root bending moment. Specifically, this includes:

[0188] Preprocessing was performed on the equipment basic parameters, equipment condition evaluation scores, and fault mechanical values ​​under the aforementioned high-wind weather fault conditions. This included filling missing values ​​with the mean, removing outliers using the 3σ criterion, and removing duplicate data to ensure the validity of the dataset. The min-max normalization algorithm was used to unify the dimensions of all feature data and map them to the interval [0, 1].

[0189] The preprocessed dataset was divided into training, validation, and test sets in a 7:2:1 ratio to train a random forest regression model, whose output is the fault threshold of conductor cross-sectional stress and pole root bending moment.

[0190] Hyperparameters such as the number of decision trees, maximum tree depth, and subsample sampling ratio are iteratively optimized using the validation set to reduce the risk of model overfitting. Model accuracy is validated using the test set. When the model's coefficient of determination R0 is... 2 When the mean absolute percentage error (MAPE) is ≥0.9 and <3%, the model training is complete, and the fault threshold assessment model is constructed.

[0191] 3.4: Theoretical Threshold Normal Distribution Correction Process

[0192] When the number of valid fault condition samples is insufficient to support machine learning modeling, the benchmark mean values ​​μ of conductor tensile strength and tower bending strength corresponding to the target line type and tower type are obtained with reference to GB50061-2010 "Design Standard for Overhead Power Lines of 66kV and Below" and DL / T5220-2005 "Technical Specification for Design of Overhead Distribution Lines of 10kV and Below". The standard deviation is set as CV·μ (CV is the coefficient of variation, conductor tensile strength is taken as 0.1, and tower bending strength is taken as 0.12) with reference to IEC60826-2017, and a fault threshold normal distribution model is constructed.

[0193] 3.5: Threshold Dynamic Iterative Update

[0194] The system periodically collects newly added operational data, defect and potential data, and new fault condition samples, and iteratively updates the dataset and fault threshold assessment model parameters. If there are sufficient fault condition samples, the random forest regression model is periodically retrained and optimized; if there are insufficient fault condition samples, the system switches to retraining and optimizing the random forest regression model when the number of samples meets the requirements, thus achieving dynamic adaptation of the fault threshold throughout its entire lifecycle.

[0195] 4. Construction of Simulation Fault Samples

[0196] Based on the probability distribution of historical strong wind weather and equipment basic parameters and line equipment status parameters in the actual power grid, simulation samples of different scenario-line combinations are constructed. A dataset of wind speed and direction data and equipment basic parameters and line equipment status parameters is constructed. The simulation model is input and the simulation data including conductor cross-sectional stress and pole root bending moment is output.

[0197] Based on the simulation samples, obtain their corresponding basic equipment parameters and equipment status evaluation scores, input them into the fault threshold evaluation model, and output the fault thresholds of conductor cross-section stress and pole root bending moment; when the simulation data of conductor cross-section stress and pole root bending moment are less than the fault thresholds of conductor cross-section stress and pole root bending moment, the sample is marked as a normal sample, otherwise it is marked as a simulated fault sample.

[0198] 5. Construction of a sample library for high-wind fault balancing

[0199] The simulated fault samples are merged with real historical fault samples during strong winds and normal operating conditions to form a balanced sample library of power line faults during strong winds with balanced label distribution; the fusion feature vector corresponding to each sample is obtained.

[0200] The construction of simulated fault samples focuses on designing corresponding scenario-line combinations for sensitive parameter areas that are prone to failure (such as specific wind angles, high wind speed ranges, and line segments with specific defects), conducting high-density simulations, and generating rare fault samples to solve the problem of sample imbalance.

[0201] 6. Fault probability model training and construction:

[0202] The LightGBM binary classification probability prediction model is trained using the fused feature vectors of the high wind fault equilibrium sample library and the corresponding fault labels to obtain the fault probability prediction model, whose output is the probability value of wire breakage and pole collapse.

[0203] Furthermore, in one embodiment of the present invention, the fault label corresponding to the training of the LightGBM binary classification probability prediction model refers to the fault state of the sample corresponding to the fused feature vector, that is, whether it belongs to the normal state or the broken line / collapsed pole state, which can be represented by 0 and 1 respectively. Specifically, it includes:

[0204] 6.1: Dataset Standardization Preprocessing and Hierarchical Partitioning

[0205] 1. Data Reading: Import the fusion feature vector dataset corresponding to the high wind fault equilibrium sample library. The dataset contains mechanical data such as conductor cross-section stress and pole root bending moment.

[0206] 2. Data cleaning: Remove duplicate samples and extreme abnormal noise samples from the dataset, fill in missing feature values, use the median to fill in continuous features and the mode to fill in categorical features, and ensure the integrity and effectiveness of the dataset.

[0207] 3. Feature standardization processing: All continuous input data are standardized and scaled using the Z-score standardization method to eliminate the impact of feature dimension differences on the training of the LightGBM binary classification probability prediction model. Category features are numerically converted using one-hot encoding.

[0208] 4. Dataset stratification: The dataset is divided into training, validation, and test sets in a stratified random sampling method at a ratio of 7:2:1. The training set is used for learning the parameters of the LightGBM binary classification probability prediction model, the validation set is used for hyperparameter tuning during the training process, and the test set is used for evaluating the model's final generalization ability.

[0209] 6.2: Constructing a Failure Probability Prediction Model

[0210] The LightGBM binary classification probabilistic prediction model is adopted, and the model is set as a binary classification probabilistic prediction task to output the continuous fault probability in the 0-1 interval. The specific model construction operation is as follows:

[0211] 1. Model core mechanism configuration: LightGBM histogram feature discretization and leaf-first growth strategy are enabled to improve the fitting ability and training speed of small-to-medium-dimensional structured power data; a fixed global random seed is used to eliminate random perturbations during training and ensure that the model results are unique and reproducible.

[0212] 2. Loss function and evaluation metric configuration: The binary cross-entropy loss function is used as the objective function; the training process monitoring metrics are set as ROC-AUC and LogLoss to simultaneously monitor the model's discrimination ability and probabilistic accuracy.

[0213] Assuming the model prediction probabilities of all samples are an ordered sequence, and one positive sample and one negative sample are randomly selected, the probability that the prediction probability of the positive sample is greater than that of the negative sample is the ROC-AUC value. The calculation formula for discrete samples is as follows:

[0214]

[0215] In the formula: P is the total number of positive fault samples; N is the total number of negative non-fault samples; Predict the failure probability of the model for the i-th positive sample; Predict the failure probability of the model for the j-th negative sample; This is an indicator function; it takes the value 1 when the condition is true and 0 when the condition is false.

[0216] In a binary fault probability prediction scenario, the standardized LogLoss calculation formula is as follows:

[0217]

[0218] In the formula: N is the total number of samples participating in the evaluation; The true fault label for the i-th sample is 1 for fault and 0 for non-fault. This represents the fault prediction probability of the i-th sample output by the model.

[0219] 3. Model hyperparameter space optimization

[0220] (1) Construction of hyperparameter search space: Based on the core structural parameters and regularization parameters of LightGBM, a parameter optimization space adapted to the fault prediction of power distribution network in windy weather is constructed, which includes: learning rate, maximum tree depth, maximum number of leaf nodes in a single tree, L1 regularization coefficient, and L2 regularization coefficient.

[0221] (2) Cross-validation optimization: Taking the maximization of the validation set ROC-AUC as the core optimization objective, a 5-fold cross-validation combined with a grid search mechanism is adopted to traverse all parameter combinations and select the globally optimal parameter configuration that fits the balanced fault condition sample.

[0222] 4. Iterative Model Training

[0223] Conduct iterative training of the model, combined with dual-index monitoring and single-index early stopping mechanisms, to ensure stable model convergence and optimal generalization performance.

[0224] (1) Iterative training execution: Input the standardized training set into the model to carry out multiple rounds of iterative training. In each round of iteration, the validation set is called to complete the performance verification. The changes of the validation set ROC-AUC and LogLoss indicators are monitored in real time to grasp the convergence status of the model.

[0225] (2) Adaptive early stopping strategy: The training adopts a hierarchical convergence mechanism with dual-index monitoring and single-index judgment, with ROC-AUC as the sole criterion for early stopping, and a 10-round early stopping threshold is set. When the ROC-AUC index on the validation set does not improve effectively for 10 consecutive iterations, the iterative training is automatically terminated, the model weight parameters with the best performance on the validation set during the training process are retained, and invalid iteration results are discarded.

[0226] 6.3: Model Validation

[0227] (1) Independent test set verification: The trained LightGBM binary classification probability prediction model is input into an independent test set that has not participated in training and parameter tuning to complete the model generalization ability test. The values ​​of ROC-AUC and LogLoss in the test set are statistically analyzed to verify the prediction stability of the model in unfamiliar sample scenarios.

[0228] The excellent standard is that ROC-AUC≥0.95 and LogLoss≤0.13. The model has excellent ability to distinguish between faulty and non-faulty working conditions, the predicted probability fits the real working condition very well, and the probability output is accurate and reliable.

[0229] The standard for qualification is that ROC-AUC ≥ 0.90 and LogLoss ≤ 0.15, indicating that the model's probability prediction error is controllable and meets the basic requirements for engineering prediction.

[0230] ROC-AUC < 0.90 or LogLoss > 0.15 is considered unacceptable, indicating insufficient model generalization and discrimination capabilities, distorted risk quantification results, and an inability to meet the requirements for high-precision risk prediction of abnormal weather.

[0231] (2) Overfitting detection: Compare the evaluation metrics of the LightGBM binary classification probability prediction model on the training set and the test set. If the difference between the metrics of the two datasets is less than the preset threshold and there is no obvious cliff drop in the metrics, it is determined that the model has no overfitting or underfitting problems and the model performance is stable and reliable.

[0232] (3) Optimal model solidification

[0233] Save the model weights and hyperparameters for the optimal iteration round, fix the feature input dimensions, input order, and preprocessing rules, and generate an offline reusable model file.

[0234] 7. Real-time fault assessment and output:

[0235] For the power distribution line segment to be evaluated and real-time and forecast wind speed and direction data, a fused feature vector is generated, which is input into the fault probability prediction model, and the probability values ​​of line breakage and pole collapse faults of the line segment within the target time period are output.

[0236] Example 2

[0237] A system for assessing the probability of power line breakage and pole collapse during strong winds includes:

[0238] The fusion feature vector construction module is used to establish a wind load physical model based on the principles of fluid mechanics and structural mechanics, and calculate the key physical feature quantities of conductors and towers; according to the distribution network equipment condition evaluation guidelines, the condition quantities of each component of the distribution line are quantified to obtain the equipment condition evaluation score; the key physical feature quantities, the equipment condition evaluation score and the equipment basic parameters are combined to form the fusion feature vector.

[0239] The wind-line-tower coupled simulation modeling module is used to build a wind-line-tower coupled dynamic mechanical simulation model based on the ANSYS platform. It loads equipment foundation parameters, line equipment status parameters, and wind speed and direction data under normal operating conditions and high wind weather fault conditions, and outputs simulation data including conductor cross-sectional stress and pole root bending moment.

[0240] The fault threshold assessment model construction module is used to collect fault samples in windy weather. Based on the equipment basic parameters, equipment condition evaluation scores, and fault mechanical values ​​under the fault conditions in windy weather, a fault label dataset is constructed. A random forest regression model is trained to obtain the fault threshold assessment model, whose output is the fault threshold of conductor cross-sectional stress and pole root bending moment.

[0241] The simulation fault sample generation module is used to construct simulation sample datasets for different scenario-line combinations based on the probability distribution of historical strong wind weather, equipment basic parameters, and line equipment status parameters in the actual power grid. The calibrated dynamic mechanical simulation model is input, and the simulation data containing conductor cross-sectional stress and pole root bending moment is output. The equipment basic parameters and equipment status evaluation scores of the corresponding simulation samples are input into the fault threshold evaluation model to obtain the corresponding fault threshold. After comparing the simulation data with the fault threshold, normal samples and simulated fault samples are marked.

[0242] The high-wind fault equilibrium sample library construction module is used to merge the simulated fault samples with real historical high-wind weather fault samples and normal operating condition samples to construct a high-wind fault equilibrium sample library and match the fusion feature vector corresponding to each sample.

[0243] The fault probability model training module is used to train the LightGBM binary classification probability prediction model using the fusion feature vector of the strong wind fault equilibrium sample library to obtain the fault probability prediction model of broken line and fallen pole. Its output is the fault probability value of broken line and fallen pole.

[0244] The real-time fault assessment output module is used to generate a fusion feature vector for the power distribution line segment to be assessed by combining real-time and forecast wind speed and direction data, inputting it into the fault probability prediction model, and outputting the probability value of line breakage and pole collapse for the line segment within the target time period.

Claims

1. A method for assessing the probability of power line breakage and pole collapse during strong winds, characterized in that... include: (1) Constructing the fused feature vector: A wind load physical model is established based on the principles of fluid mechanics and structural mechanics to calculate the key physical characteristics of conductors and towers. The state quantities of each component of the distribution line are quantified according to the distribution network equipment condition evaluation guidelines to obtain the equipment condition evaluation score. The key physical characteristics and the equipment condition evaluation score are combined with the equipment foundation parameters to form a fused feature vector. The key physical characteristics include at least conductor cross-sectional stress and tower root bending moment. (2) Constructing a dynamic mechanical simulation model: A wind-line-tower coupled dynamic mechanical simulation model was built based on the ANSYS platform. The equipment foundation parameters, line equipment status parameters, and wind speed and direction data were loaded under normal operating conditions and high wind weather fault conditions. The simulation data included conductor cross-sectional stress and tower root bending moment. (3) Constructing a fault threshold assessment model: Fault samples were collected during windy weather. Based on the equipment's basic parameters, equipment condition evaluation scores, and fault mechanical values ​​under windy weather fault conditions, a fault label dataset was constructed. A random forest regression model was trained to obtain a fault threshold assessment model, the output of which is the fault threshold of conductor cross-sectional stress and pole root bending moment. The fault mechanical values ​​include conductor cross-sectional stress and pole root bending moment at the time of the fault. (4) Constructing simulated fault samples: Based on the probability distribution of historical strong wind weather and equipment basic parameters and line equipment status parameters in the actual power grid, simulation samples of different scenario-line combinations are constructed. A dataset of wind speed and direction data and equipment basic parameters and line equipment status parameters is constructed. The simulation model is input and the simulation data including conductor cross-sectional stress and pole root bending moment is output. Based on the simulation samples, obtain their corresponding basic equipment parameters and equipment status evaluation scores, input them into the fault threshold evaluation model, and output the fault thresholds of conductor cross-section stress and pole root bending moment; when the simulation data of conductor cross-section stress and pole root bending moment are less than the fault thresholds of conductor cross-section stress and pole root bending moment, the sample is marked as a normal sample, otherwise it is marked as a simulated fault sample. (5) Construct a sample library for high wind fault equilibrium: The simulated fault samples are merged with real historical high wind weather fault samples and normal operating condition samples to obtain a high wind fault balanced sample library; the fusion feature vector corresponding to each sample is obtained. (6) Constructing a failure probability model: The LightGBM binary classification probability prediction model is trained using the fused feature vectors of the high wind fault equilibrium sample library and the corresponding fault labels to obtain the fault probability prediction model, whose output is the probability value of wire breakage and pole collapse. (7) Real-time fault assessment and output: For the power distribution line segment to be evaluated and real-time and forecast wind speed and direction data, a fused feature vector is generated, which is input into the fault probability prediction model, and the probability values ​​of line breakage and pole collapse faults of the line segment within the target time period are output.

2. The method for assessing the probability of power line breakage and pole collapse under strong winds as described in claim 1, characterized in that, The calculation of the key physical characteristic quantities includes: Obtain the structural parameters and wind speed and direction data of the target power distribution line section, establish a wind load physical model based on the principles of fluid mechanics and structural mechanics, and calculate the key physical characteristic quantities acting on the conductors and towers. The key physical characteristic quantities include at least the conductor cross-sectional stress and the tower root bending moment. The tower root bending moment includes the tower root bending moment caused by the tower body load and the tower root bending moment caused by the conductor tension load.

3. The method for assessing the probability of power line breakage and pole collapse under strong winds as described in claim 1, characterized in that, The equipment status evaluation quantification takes the overhead line unit as the evaluation unit, and calculates the score of the overhead line unit by quantifying the status of each component included in the overhead line according to the distribution network equipment status evaluation guidelines.

4. The method for assessing the probability of power line breakage and pole collapse under strong winds as described in claim 1, characterized in that, The basic parameters of the equipment include: conductor outer diameter, conductor span, conductor cross-sectional area, pole type, pole parameters, pole length, burial depth, distance from the tip to the crossarm, root diameter, and tip diameter.

5. The method for assessing the probability of power line breakage and pole collapse under strong winds as described in claim 1, characterized in that, The line equipment status parameters include: tower mechanical characteristic status parameters, tower tilt appearance status parameters, conductor mechanical characteristic status parameters, and conductor sag appearance status parameters.

6. The method for assessing the probability of power line breakage and pole collapse under strong winds as described in claim 1, characterized in that, The aforementioned wind-line-tower coupled dynamic mechanical simulation model based on the ANSYS platform specifically includes: based on the ANSYS platform, completing the geometric modeling and material property assignment of the tower, conductor, insulator, hardware and foundation in ANSYS, defining the connection constraints of each component, and establishing the wind-line-tower coupled dynamic mechanical simulation model based on the ANSYS platform.

7. The method for assessing the probability of power line breakage and pole collapse under strong winds as described in claim 1, characterized in that, It also includes verifying the simulation data described in step (2) and calibrating the simulation model.

8. The method for assessing the probability of power line breakage and pole collapse under strong winds as described in claim 7, characterized in that, The simulation data described in step (2) is verified, specifically including: using the measured equipment basic parameters, line equipment status parameters, wind speed and direction data, conductor cross-sectional stress, and pole root bending moment as calibration benchmarks, and taking the consistency between the simulation data output by the simulation model and the measured data as the optimization goal, quantitative error index verification, statistical significance test, and fault feature key point matching verification are adopted to determine the degree of matching between the ANSYS simulation model output results and the measured data from three dimensions: numerical accuracy, statistical difference, and fault features.

9. The method for assessing the probability of power line breakage and pole collapse under strong winds as described in claim 8, characterized in that, The fault feature key point matching and verification specifically includes: For fault conditions involving wire breakage and pole collapse during strong winds, the consistency of critical fault parameters is verified, including: matching the maximum conductor cross-sectional stress and maximum pole root bending moment before the fault, verifying whether the simulation model can accurately reproduce the ultimate conductor cross-sectional stress and pole root bending moment response under strong wind impact; matching the timing of abrupt changes in conductor cross-sectional stress and pole root bending moment, verifying whether the simulation model can realistically reproduce the mechanical abrupt changes and structural instability characteristics at the moment of wire breakage and pole collapse; and matching the critical fault wind speed and the location of maximum stress concentration in the structure with the actual fault location.

10. The method for assessing the probability of power line breakage and pole collapse under strong winds as described in claim 7, characterized in that, The calibration of the simulation model specifically includes: Based on the multi-dimensional consistency comparison results between the simulation data and the measured data, differentiated parameter tuning logic and corresponding convergence requirements are set for three types of deviations: overall numerical fitting deviation, systematic statistical deviation, and fault transient characteristic deviation. The simulation model iteration process follows the principle of fixed boundary and single variable to adjust the uncertain parameters until all three convergence criteria of overall numerical fitting deviation, systematic statistical deviation, and fault transient characteristic deviation are met, and the model calibration is completed.

11. The method for assessing the probability of power line breakage and pole collapse under strong winds as described in claim 1, characterized in that, When training the LightGBM binary classification probability prediction model, the loss function and evaluation metrics are configured as follows: the binary cross-entropy loss function is used as the objective function; the training process monitoring metrics are set as ROC-AUC and LogLoss, and the model's discrimination ability and probability accuracy are monitored simultaneously.

12. A system for assessing the probability of power line breakage and pole collapse during windy weather, characterized in that, include: The feature vector construction module is used to establish a wind load physical model based on the principles of fluid mechanics and structural mechanics, and to calculate the key physical characteristics of conductors and towers. Based on the guidelines for condition evaluation of distribution network equipment, the condition quantities of each component of the distribution line are quantified to obtain the equipment condition evaluation score; the key physical characteristics, the equipment condition evaluation score, and the basic parameters of the equipment are combined to form a fusion feature vector. The wind-line-tower coupled simulation modeling module is used to build a wind-line-tower coupled dynamic mechanical simulation model based on the ANSYS platform. It loads equipment foundation parameters, line equipment status parameters, and wind speed and direction data under normal operating conditions and high wind weather fault conditions, and outputs simulation data including conductor cross-sectional stress and pole root bending moment. The fault threshold assessment model construction module is used to collect fault samples in windy weather. Based on the equipment basic parameters, equipment condition evaluation scores, and fault mechanical values ​​under the fault conditions in windy weather, a fault label dataset is constructed. A random forest regression model is trained to obtain the fault threshold assessment model, whose output is the fault threshold of conductor cross-sectional stress and pole root bending moment. The simulation fault sample generation module is used to construct simulation sample datasets for different scenarios and line combinations based on the probability distribution of historical strong wind weather, equipment basic parameters, and line equipment status parameters in the actual power grid. The input is a calibrated dynamic mechanical simulation model, and the output is simulation data including conductor cross-sectional stress and pole root bending moment. Input the basic parameters of the equipment and the equipment status evaluation score of the corresponding simulation sample into the fault threshold assessment model to obtain the corresponding fault threshold. After comparing the simulation data with the fault threshold, mark the normal sample and the simulated fault sample. The high-wind fault equilibrium sample library construction module is used to merge the simulated fault samples with real historical high-wind weather fault samples and normal operating condition samples to construct a high-wind fault equilibrium sample library and match the fusion feature vector corresponding to each sample. The fault probability model training module is used to train the LightGBM binary classification probability prediction model with the fusion feature vector of the strong wind fault equilibrium sample library and the corresponding fault label to obtain the fault probability prediction model of broken line and fallen pole. Its output is the fault probability value of broken line and fallen pole. The real-time fault assessment output module is used to generate a fusion feature vector for the power distribution line segment to be assessed by combining real-time and forecast wind speed and direction data, inputting it into the fault probability prediction model, and outputting the probability value of line breakage and pole collapse for the line segment within the target time period.

Images