Method for identifying high risk window period of offshore power grid external insulation flashover under approaching typhoon

Abstract

The application belongs to the cross field of atmospheric environment and electric power engineering, and specifically discloses a method for identifying a high-risk window period of flashover of external insulation of a coastal power grid under approaching typhoon, which comprises the following steps: acquiring terrain data, meteorological and atmospheric reanalysis data and field monitoring data of a research area, preprocessing and fusing to form a data set; dividing a four-stage risk link of pollution flashover, establishing mapping to form a candidate feature pool; sequentially adopting physical mechanism matching screening, statistical effectiveness screening, causal effectiveness screening and early warning effectiveness screening to obtain a core feature set, and determining a feature that passes all screening and has a comprehensive core degree greater than a preset threshold as a core feature set for risk assessment of external insulation pollution flashover during a typhoon period; constructing a typhoon terrain dynamic constraint comprehensive index and a typhoon atmospheric terrain coupling risk comprehensive index after the core features are dimensionless, and linearly mapping the two indexes into a probability of a high-risk window period of pollution flashover. The application can quickly identify a short high-risk window period before a typhoon lands.

Description

Technical Field

[0001] This application belongs to the interdisciplinary field of atmospheric environment and power engineering, and more specifically, it relates to a method for identifying the high-risk window period of external insulation flashover in coastal power grids under the approach of a typhoon. Background Technology

[0002] Coastal power transmission and transformation systems are constantly threatened by extreme weather events such as typhoons and tropical cyclones. When a typhoon approaches, the risk of external insulation flashover in the coastal power grid increases significantly. This type of fault is one of the most common fault types in the coastal power grid during typhoons, seriously threatening the safe and stable operation of the power grid.

[0003] Traditional coastal pollution flashover studies often attribute flashover risk solely to the static cumulative effect of sea salt particles, relying mainly on macroscopic wind and rain intensity indicators for risk assessment. They fail to adequately consider the coupling effect of atmospheric conditions and topography during typhoon impact, nor do they take into account the dynamic evolution characteristics of the chemical components of deposited pollution. They lack a systematic understanding of the entire chain mechanism of pollution flashover risk formation, including aerosol transport, pollution deposition and aging, surface wetting and conductivity, and effective precipitation washing, making it difficult to quickly identify high-risk windows before landfall.

[0004] Therefore, how to quickly identify the high-risk window period before a typhoon makes landfall is an urgent problem to be solved. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the purpose of this application is to provide a method for identifying high-risk windows of external insulation flashover in coastal power grids under typhoon approach, which can quickly identify high-risk windows before typhoon landfall.

[0006] To achieve the above objectives, in a first aspect, this application provides a method for identifying high-risk windows for external insulation flashover in coastal power grids under typhoon approach, comprising the following steps: S10: Acquire topographic data, meteorological and atmospheric reanalysis data, and field monitoring data of the study area; preprocess and match the acquired data in time and space; and integrate them to form a standardized and unified analysis dataset with grid correspondence and time synchronization. S20, the flashover of external insulation during the typhoon season is divided into four continuous links: aerosol transport, pollution deposition and aging, surface wetting and conductivity, and rain washing. Based on the standardized and unified analysis dataset, the mapping relationship between each link and candidate features is established to form a full candidate feature pool. S30, based on the standardized unified analysis dataset, the candidate features in the full candidate feature pool are screened sequentially using physical mechanism matching screening, statistical validity screening, causal validity screening and early warning validity screening, and the features that pass all screenings and have a comprehensive core degree greater than a preset threshold are determined as the core feature set for typhoon period external insulation flashover risk assessment. S40, perform dimensionless processing on each feature in the core feature set, and construct a comprehensive typhoon topographic dynamic constraint index based on the processed terrain elevation, horizontal distance between mountains and the coast, and coastal slope. Then, based on the processed TSP mass concentration and Cl... - / Na + The typhoon atmospheric-topographic coupling risk index is constructed by combining mass ratio, electrical conductivity at 25℃, near-surface relative humidity, precipitation, boundary layer height, and the typhoon topographic dynamic constraint comprehensive index. The typhoon atmospheric-topographic coupling risk index is then linearly mapped to the probability of a high-risk window period for pollution flashover.

[0007] As a further preferred embodiment, step S10 includes a key monitoring area delineation step before acquiring data: Using coastal city power supply zones as the basic evaluation unit, the total number of pollution flashover faults during the entire typhoon process was obtained for each evaluation unit within the statistical period T=3. Number of flashover failures during the 24 hours leading up to typhoon landfall. Number of regional power transmission and distribution substations and the average distance from the substation to the coastline Set the distance-to-sea attenuation constant Approximation window magnification factor Smoothing constant Construct a historical high-risk screening index for approaching typhoons. The calculation formula is:

[0008] Risk classification is performed based on index size and threshold conditions. Evaluation units with higher index rankings are identified as core key monitoring areas. Transmission and distribution substations within these areas are designated as core monitoring nodes, and the data acquisition scope for subsequent steps is limited to these core key monitoring areas.

[0009] As a further preferred embodiment, step S10, which involves preprocessing and spatiotemporal matching of the acquired data, specifically includes: Outliers were removed from meteorological and atmospheric reanalysis data using the 3σ Laida criterion, and missing data were filled in using linear interpolation and Kriging. Invalid data were removed from field monitoring data using blank control and parallel sample deviation verification, and background ion concentration was deducted. Projection transformation, invalid value filling, and core parameter standardization extraction were performed on topographic data. The spatial dimension uniformly adopts the WGS84 geographic coordinate system, with a uniform resolution of 1km×1km grid, and interpolation is performed using the inverse distance weighting method with terrain correction factor; the temporal dimension is uniformly set to hourly resolution, and each timestamp is matched with a typhoon process stage label.

[0010] As a further preferred embodiment, in step S20, the full candidate feature pool Candidate feature subset of aerosol transport links Candidate feature subset of fouling deposition and aging process Candidate feature subset of surface wetting conductivity Candidate feature subsets for precipitation and rain washing process Composition, expression is ; in, Corresponding to the process of salt spray aerosol transport and topographic convergence driven by the outer circulation of the typhoon, the characteristics include near-surface wind speed, boundary layer height, straight distance from the coastline, horizontal distance between the mountains and the coast, elevation, topographic relief, slope aspect, and coastal topographic slope. Corresponding to the deposition and chemical aging processes of aerosol particles, characteristics include TSP mass concentration, Cl... - / Na + Mass ratio, SO4² - Mass concentration, conductivity at 25℃; Corresponding to the conductive process of contaminated insulating surfaces under high humidity conditions, the characteristics include near-surface relative humidity, air temperature, and conductivity at 25°C; The corresponding effective precipitation washes away the pollutants, characterized by hourly precipitation.

[0011] As a further preferred option, in step S30, the physical mechanism matching screening uses the physical mechanism anchoring coefficient P(f), which is a binary rigid admission coefficient. P(f)∈{0,1}. If the candidate feature directly participates in any of the four consecutive physical links and has a clear risk-causing or risk-controlling logic, then P(f)=1 is assigned; otherwise, P(f)=0 is assigned and directly eliminated. The statistical validity screening uses the weighted nonlinear association strength S(f), with S(f) ranging from [0,1]. An effective threshold S(f) ≥ 0.45 is set, and features that do not meet the threshold are directly removed. The causal validity screening uses the independent risk contribution C(f), with C(f) ranging from [0,1]. An effective threshold is set: C(f) ≥ 0.10 and the 95% highest posterior density interval of C(f) does not contain 0. Features that do not meet the threshold are directly removed. The effectiveness screening of early warning adopts the high-risk window period sensitivity index W(f), with the value range of [0,1]. An effective threshold of W(f) ≥ 0.20 is set, and features that do not meet the threshold are directly eliminated. Candidate features selected through the fourth-order screening must also satisfy a comprehensive core score (Core(f)) greater than the comprehensive judgment threshold. .

[0012] As a further preferred embodiment, the formula for calculating the weighted nonlinear correlation strength S(f) is:

[0013] In the formula, w(f) is the weighted Kendall rank correlation coefficient between candidate feature f and pollution flashover risk label, which is used to characterize the monotonic correlation strength in time series samples; MIC(f) is the maximum information coefficient between candidate feature f and pollution flashover risk label, which is used to characterize the nonlinear correlation strength.

[0014] As a further preferred embodiment, the independent risk contribution C(f) is calculated based on a hierarchical Bayesian logistic regression model. ATE(f) represents the average treatment effect of candidate feature f after controlling for other candidate features, differences in evaluation units, and differences in typhoon processes. It is used to characterize the marginal causal contribution of this feature to the probability of pollution flashover risk, and the 95% highest posterior density interval of ATE(f) does not include 0. The high-risk window period sensitivity index W(f) is calculated based on the time-series SHAP contribution, and the calculation formula is as follows: ,in denoted as the total number of samples within the window period, win is the sample set within the high-risk warning window period 24 hours before typhoon landfall, and SHAP(f,t) is the SHAP contribution value of candidate feature f to the risk probability output by the model at time t.

[0015] As a further preferred embodiment, in step S40, the min-max normalization method is used to perform dimensionless processing on each feature in the core feature set. The normalization formula is: , where x The standardized value. The original value of the indicator. , These are the minimum and maximum values ​​of the indicator in the entire dataset, respectively.

[0016] As a further optimization, in step S40, a comprehensive index of typhoon topographic dynamic constraints is constructed. The specific formula is:

[0017] in The processed, standardized values ​​of terrain elevation. The standardized values ​​for the horizontal distance between the mountain range and the coastline after processing. This represents the standardized value of the coastal terrain slope after processing. , , These are the weighting coefficients for terrain factors; , , It is a non-linear power; This is the terrain coupling correction factor. ; Revision for the minimum value; The value range is [0, 1.5]; Constructing a comprehensive typhoon-atmospheric-topographic coupled risk index The specific formula is:

[0018] in The normalized value of the TSP mass concentration after treatment. For the treated Cl - / Na + Quality ratio standardized value The normalized conductivity value at 25℃ after treatment. The normalized value of near-surface relative humidity after processing. The processed precipitation standardized value, This is the standardized value of the processed boundary layer height; , , These are the weighting coefficients for pollutant items; , , The pollutant term is a nonlinear power; , , The meteorological term is a nonlinear power; The value range is [0, 2.5]; The typhoon atmospheric topography coupled with the comprehensive risk index The linear mapping represents the probability P of the high-risk window period for pollution flashover, and the basic calculation formula is: ,in , ;like If ≥2.0, then directly set P=1. If P ≤ 0, then set P = 0 directly; when P ≥ 0.8, it is determined to be the high-risk window period for pollution flashover.

[0019] Secondly, this application provides a system for identifying high-risk windows of external insulation flashover in coastal power grids under typhoon approach, including a processor, a storage medium, and a signal transmitting module. The storage medium stores instructions, and the processor loads and executes the instructions to implement the method for identifying high-risk windows of external insulation flashover in coastal power grids under typhoon approach as described in any of the above claims. The signal transmitting module is connected to the processor and is used to send a warning signal to the power grid dispatching platform, operation and maintenance terminal, SMS platform, or audible and visual alarm device after the processor determines that the target area has entered a high-risk window of external insulation flashover.

[0020] The beneficial effects of this application are as follows: (1) Rapid identification of high-risk windows, breaking through the blind spots of traditional early warning. The system sorts out the whole chain of risk mechanism in the four stages of aerosol transport before typhoon landfall - pollution deposition and aging - surface wetting and conductivity - effective precipitation and rain washing. It can quickly capture the short-term high-risk window within 24 hours before typhoon landfall, overcome the industry pain point that traditional methods cannot identify the critical risk stage before landfall, and the early warning time resolution reaches 1 hour.

[0021] (2) The accuracy of early warning has been greatly improved, and the rates of missed and false alarms have been significantly reduced. A four-dimensional progressive feature screening system of physical mechanism anchoring, statistical validity, causal validity and early warning validity was constructed. Nine core risk-causing features were selected and combined with a three-order coupling model of topographic dynamic constraints, meteorological phases and pollution chemical activity. Compared with the traditional single-indicator early warning method, the accuracy of early warning has been improved, the rate of missed and false alarms has been reduced, and the risk quantification is more in line with the actual working conditions.

[0022] (3) Deep integration of multi-source data enables more refined spatial risk characterization. By integrating high-resolution topographic, meteorological, atmospheric reanalysis, field monitoring and fault labeling data, and adopting a spatiotemporal matching method with topographic correction, the distortion that may be caused by traditional spatial interpolation is reduced, and a 1km×1km grid-level refined risk assessment is achieved, reflecting the modulation effect of the complex coastal topography on salt spray transport and pollution flashover risk.

[0023] (4) The risk levels are clear and operable, supporting the hierarchical operation and maintenance of the power grid. The comprehensive risk index is mapped to a probability value of 0 to 1, and the warning thresholds of high risk and ultra-high risk are clearly defined. The window period duration, scope of impact and core driving factors are automatically output, forming a closed loop of risk identification, level determination, warning issuance and hierarchical handling. This can provide a reference for the power grid to deploy in advance, adjust load and carry out emergency repairs, and greatly improve the efficiency of power grid operation and maintenance during typhoons.

[0024] (5) It has strong universality and engineering applicability, and is suitable for the entire coastal power grid. The method has a certain degree of adaptability to different coastal areas and voltage levels. It can be widely applied to typhoon-affected areas such as the southeast coast of my country, the Yangtze River Delta, and the Bohai Rim. It covers the main grid of transmission, transformation and distribution substations and is backward compatible with distribution networks. Under the condition of having basic meteorological, environmental and power grid monitoring data, it can reduce the need for new monitoring equipment and reuse existing meteorological, environmental and power grid monitoring data. It has low deployment cost and strong engineering feasibility. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of the overall process for identifying and issuing early warning of high-risk windows for external insulation flashover in coastal power grids during typhoon approach, as provided in this application. Figure 2 This is a schematic diagram of the four-stage, full-chain risk mechanism of flashover during the approach of a typhoon, provided in this application. Figure 3This is a schematic diagram of the hardware device provided in this application. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0027] This application proposes a method for constructing a model to identify high-risk short-term window periods of external insulation flashover in coastal power grids under typhoon approach. The aim is to analyze the entire chain of risk mechanisms of pollution flashover during typhoons and establish a four-dimensional progressive core risk feature screening system. By integrating multi-source data for quantitative analysis, a third-order coupled risk assessment model is constructed to achieve rapid identification, classification, and early warning of high-risk window periods, providing technical support for the safe operation and maintenance of power grids during typhoons.

[0028] like Figure 1 As shown, the method for constructing a short-term window period identification model for high-risk external insulation flashover of coastal power grids under typhoon approach, provided in this application, includes the following steps: S1: Identification and monitoring scope delineation of key substation areas based on historical typhoon impacts Taking my country's coastal areas as the research scope, this study combines historical typhoon pollution flashover fault data, power grid asset distribution (power grid asset distribution refers to the layout and spatial distribution characteristics of core production assets in the power system that undertake transmission, distribution, power supply, and dispatch functions in four dimensions: geographic space, voltage level, functional zoning, ownership, and operation and maintenance, and is the core basic data for power grid planning, operation and maintenance management, and disaster risk prevention and control) and geospatial information. It adopts a historical high-risk screening index model for approaching typhoons to identify high-risk key substation areas and delineate graded monitoring ranges, providing targeted monitoring targets for subsequent risk warnings.

[0029] 1. Establish standardized evaluation units and basic datasets. Coastal city power supply zones are used as the basic evaluation units, ensuring that the number of transmission, distribution, and substations within each unit meets the requirements for statistical validity. The statistical period T=3 years is used to uniformly collect the total number of pollution flashover faults during the entire typhoon process in each unit within a 3-year statistical period. Number of flashover failures during the 24 hours leading up to typhoon landfall. Number of regional power transmission and distribution substations and the average distance from the substation to the coastline And set the model parameters: distance-to-sea attenuation constant. Approximation window magnification factor Smoothing constant .

[0030] 2. Construct and calculate historical high-risk screening index for approaching typhoons. The calculation formula is: This index comprehensively reflects the risk intensity of a unit site, the degree of coastal salt spray exposure, and the concentration of pollution flashover faults during the typhoon approach window. The higher the index, the higher the risk of pollution flashover in the region during the typhoon approach phase.

[0031] 3. Risk classification and monitoring scope delineation are based on index magnitude and threshold conditions. All evaluation units are classified according to... The regions are sorted in descending order, with high index rankings, high proportion of faults during the window period, and prominent unit risk intensity as the core key monitoring areas, and the transmission, transformation, and distribution substations within the region are the core monitoring nodes.

[0032] S2: Acquisition of Multi-Source Basic Data and Construction of Unified Analysis Dataset Based on the key substation areas identified by S1, and in accordance with the requirements for identifying the risk of flashover pollution in coastal insulation during typhoon seasons, the system acquires multi-source basic data, completes preprocessing and fusion, and constructs a unified analysis dataset. Specific implementation details are as follows: 1. Acquisition of multi-source basic data High-resolution topographic data: Acquire 30m resolution DEM data of the study area, extract core parameters such as elevation, slope, and relative position of the shoreline, characterize the constraint effect of topography on the transport of sea salt aerosols, and provide a spatial basis for the simulation of salt fog spatial transport.

[0033] Meteorological and atmospheric reanalysis data: Acquire hourly typhoon track, intensity, precipitation, temperature, humidity and wind field, and other refined meteorological elements from CMA-STI typhoon track dataset, ground meteorological stations and radar networks; Simultaneously acquire core parameters such as sea salt aerosol deposition flux, three-dimensional meteorological field, and boundary layer height from NASA MERRA-2 reanalysis data to fully characterize the entire physical process of sea salt aerosol generation, transport and deposition.

[0034] On-site monitoring data: TSP sampling and monitoring points were set up at key substations and their surrounding areas to carry out continuous sampling throughout the entire cycle before, during and after the typhoon's impact. Filter membrane conductivity testing and water-soluble ion detection were completed to form a time-synchronized continuous observation dataset, providing ground-based true value verification for the quantification of pollution flashover risk.

[0035] 2. Integration of Data Preprocessing and Standardization Differentiated preprocessing by type: Adaptive schemes are adopted for different data characteristics. For time-series meteorological and reanalysis data, outliers are removed using the 3σ Laida criterion, and missing data are filled in by linear interpolation and Kriging. For field monitoring data, invalid data are removed by blank control and parallel sample deviation verification, and background ion concentration is deducted. For topographic data, projection transformation, invalid value filling and core parameter standardization extraction are performed.

[0036] Two-dimensional spatiotemporal matching: Construct a two-dimensional matching framework of "space-time". The spatial dimension adopts the WGS84 geographic coordinate system and the resolution is uniformly 1km×1km grid. The inverse distance weighting interpolation method with terrain correction factor is used to solve the spatial distortion problem of traditional interpolation that does not consider the influence of terrain. The temporal dimension is uniformly set to hourly resolution. Each timestamp is matched with the typhoon process stage label to achieve accurate alignment of data with key processes of pollution flashover.

[0037] Construction of standardized analysis dataset: After preprocessing and spatiotemporal matching, the dataset integrates all elements of terrain parameters, meteorological elements, aerosol parameters, field monitoring data, and historical pollution flashover fault labels to form a standardized analysis dataset with grid correspondence and time synchronization, providing a unified and high-quality data foundation for subsequent risk identification and model training.

[0038] S3: Analysis of the entire chain of risk mechanisms for flashover and construction of candidate feature pool 1. The four-stage, full-chain risk mechanism of flashover caused by pollution during typhoon approach, see... Figure 2 ; 2. Construction of the full candidate feature pool.

[0039] The flashover hazard process of external insulation during typhoons is divided into four continuous stages: aerosol transport, pollution deposition and aging, surface wetting and conductivity, and rain washing. A mapping relationship is established between these hazard-causing stages and candidate features, thus forming a full pool of candidate features for subsequent screening. Its expression is:

[0040]

[0041] In the formula, This is the full candidate feature pool; , , , These are candidate feature subsets corresponding to aerosol transport, fouling deposition and aging, surface wetting and conductivity, and precipitation washing. To unify the analysis of the first in the dataset One original variable or derived variable; For the first One risky link; For the mechanism mapping function, when the feature When the corresponding risk-causing link can be directly characterized, take Otherwise take Through the above mapping, multi-source input data are categorized according to the entire chain of pollution flashover risk mechanisms, and a full candidate feature pool is output, providing a physical basis and selection objects for subsequent core feature screening.

[0042] Furthermore, the unified analysis dataset can be represented as:

[0043] in, For meteorological element data, Data related to aerosols and particulate matter. Data on the chemical composition and conductive activity of the contaminant. For topographic and geomorphological data, This is historical flashover fault tagging data. Based on the input dataset. Based on the above mechanism mapping relationship, the mechanism classification of candidate features and the construction of the feature pool are completed. The specific mapping is as follows: Aerosol transport links ( ): Corresponds to the process of salt spray aerosol transport and topographic convergence driven by the outer circulation of the typhoon. The characteristics include: near-surface wind speed, boundary layer height, straight distance from the coastline, horizontal distance between the mountain and the coast, elevation, topographic relief, slope aspect, and coastal topographic slope.

[0044] Fouling and aging process ( ): Corresponding to the deposition and chemical aging process of aerosol particles, characterized by: TSP mass concentration, mass ratio Mass concentration, conductivity at 25℃.

[0045] Surface wetting conductive links ( ): Corresponds to the process of conductive conduction due to contamination on the insulating surface under high humidity conditions, characterized by: near-surface relative humidity, air temperature, and conductivity at 25℃.

[0046] Rainfall washing process ( ): This corresponds to the effective precipitation process that washes away pollutants, and its characteristics include: hourly precipitation.

[0047] Based on the above mapping relationship, a full candidate feature pool is constructed for subsequent screening. Specifically, it includes: Atmospheric dimension candidate features: TSP mass concentration, Cl - / Na + Mass ratio, electrical conductivity at 25℃, near-surface relative humidity, hourly precipitation, boundary layer height, near-surface wind speed, air temperature, SO4² - mass concentration; Candidate features for topographic dimensions: elevation, horizontal distance between mountains and the coast, coastal slope, straight distance from the coastline, topographic relief, and slope aspect.

[0048] S4: Multi-stage progressive core feature selection driven by both formula quantification and empirical data Based on the S3 full-chain risk mechanism and candidate feature pool, a four-dimensional quantitative screening formula system is constructed. Using the unified analysis dataset of the Fujian-Guangdong coastal areas as the computational basis, each round of screening simultaneously completes the rigid determination of formula quantification and empirical data statistical verification, ultimately determining the core risk feature set of pollution flashover. Specific implementation details are as follows: 1. Construct a comprehensive judgment formula system based on core features A comprehensive formula for determining the core features across all dimensions is established to standardize the determination of the risk-causing coreness of candidate features. The formula expression is as follows:

[0049] In the formula, f is the candidate feature index related to the risk of flashover during the typhoon period to be determined, with a value range of [0, +∞). The higher the value, the stronger the comprehensive characterization ability of the feature to the risk of flashover. 2. P(f), S(f), C(f), and W(f) are four-dimensional progressive screening indicators, which must be judged sequentially. Features that fail to pass any of the dimensions are directly eliminated. Figure 2 As shown.

[0050] First-order screening: Physical mechanism anchoring coefficient P(f) — mechanism matching screening.

[0051] P(f) is a binary rigid admission coefficient, P(f)∈{0,1}, which represents the matching degree between the candidate feature and the risk mechanism of the entire pollution flashover chain determined by S3. If the feature directly participates in any physical link and has a clear risk-causing or risk-controlling logic, P(f)=1 is assigned; otherwise, P(f)=0 is assigned and it is directly eliminated.

[0052] Second-order screening: Weighted nonlinear correlation strength S(f) – statistical validity screening.

[0053] S(f) is a continuous statistical validity index, ranging from [0,1]. It comprehensively characterizes the degree of nonlinear correlation between candidate features and pollution flashover risk, and uses a weighted Kendall rank correlation coefficient. With the maximum information coefficient Equal-weighted fusion calculation, i.e.:

[0054] Set effective threshold Features that do not meet the threshold are directly removed.

[0055] Third-order screening: Independent risk contribution C(f) – causal effectiveness screening.

[0056] C(f) is a continuous causal validity index, with a value range of [0,1], representing the independent causal contribution of candidate features to pollution flashover risk; based on a hierarchical Bayesian Logistic regression model, taking... Set effective threshold Furthermore, the 95% highest posterior density interval (HPDI) of ATE(f) does not contain 0, and features that do not meet the requirements are directly removed.

[0057] Fourth-stage screening: High-risk window period sensitivity index W(f) – screening for early warning effectiveness.

[0058] W(f) is a continuous early warning effectiveness index, with a value range of [0,1]. It characterizes the sensitivity of candidate features to pollution flashover risk response during the high-risk early warning window period of 24 hours before typhoon landfall. The average risk contribution of features within the window period is calculated based on the contribution of time-series SHAP (SHapley Additive ex Planations), using the following formula:

[0059] in The total number of samples within the window period; setting an effective threshold. Features that do not meet the threshold are directly removed.

[0060] 3. Final determination and comprehensive assessment of the core feature set Set a comprehensive judgment threshold The coreity of the nine features that passed the fourth-order screening was calculated. The results show that the coreity of the following nine features is: The calculated results range from [0.73, 0.88], all of which are greater than the comprehensive judgment threshold. The final set of features for assessing the risk of flashover pollution in external insulation of power transmission lines during typhoons was determined, and it is divided into two main dimensions: Key characteristic indicators of atmospheric conditions: Standardized value of TSP mass concentration ( ), Cl - / Na + Standardized value of quality ratio ( ), Standardized values ​​( ), near-surface relative humidity standardized value ( Standardized values ​​of precipitation ( ) ), boundary layer height normalized value ( ); Core topographic feature indicators: Standardized values ​​of elevation ( Standardized values ​​of horizontal distance between mountains and coastlines ( Standardized values ​​of coastal topographic slope ( ).

[0061] S5: Construction and Early Warning of Multi-Dimensional Coupled External Insulation Flashover Risk Assessment Model Based on the core feature set output by S4, a third-order coupled risk assessment model integrating topographic dynamic constraints, meteorological phases, and pollution chemical activity is constructed to quantify the comprehensive risk of external insulation flashover during typhoons. The specific implementation is as follows: 1. First, the min-max normalization method is used to make the data dimensionless, transforming atmospheric condition indicators and topographic indicators of different dimensions to the [0,1] interval, thus eliminating the interference of dimension differences on model training. The normalization formula is:

[0062] In the formula, x The standardized value. The original value of the indicator. , These are the minimum and maximum values ​​of the indicator in the entire dataset, respectively.

[0063] Subsequently, a comprehensive index of typhoon topographic and dynamic constraints was constructed. The synergistic constraint effect of topography on the outer circulation of typhoons and the transport of salt spray particulate matter is quantified by the following formula:

[0064] The parameters in the formula are calibrated based on engineering experience in the typhoon area of ​​southeastern coastal China: , , For terrain factor weighting coefficients: ; Non-linear powers: , , [1.2, 2.0] represents the risk amplification index of the terrain factor (when the power = 1, the terrain term is linear and cannot characterize the accelerated amplification effect of risk; when the power > 1, the function is convex and will produce a nonlinear amplification effect on highly risky terrain conditions, which is consistent with actual physical laws: the influence of terrain on typhoon circulation and salt spray particulate transport is not linear. In areas with low terrain, near mountains, and gentle slopes, salt spray will accumulate faster, and the risk of flashover will show a nonlinear characteristic of "the closer to the threshold, the faster it increases". The lower limit of 1.2 ensures the most basic nonlinear amplification effect and avoids the model from degenerating into a linear model, which cannot capture the risk amplification effect of terrain; the upper limit of 2.0 avoids over-amplifying the influence of extreme terrain, prevents a few extreme values ​​from causing exponential distortion and a decrease in the model's generalization ability, and ensures the stability of the model within the study area). =1.5, =1.3, =1.4; Terrain coupling correction factor: The difference between the topography and slope represents the synergistic coupling effect of topography and slope. The greater the difference between the two, the more complex the modulation of the typhoon dynamic field by the topography and the higher the risk correction. =0.01, [0, 1.5], the larger the value, the stronger the dynamic constraint of the terrain on the typhoon, and the higher the risk of flashover. 2. Typhoon Atmospheric Topographic Coupling Risk Index

[0065] To quantify the synergistic risk-causing effects of multiple factors, a hierarchical coupling structure of "pollution-meteorology-topography" is constructed as the core output of the risk assessment model. This structure breaks down atmospheric factors into pollutants and meteorological sub-items, respectively characterizing the formation and diffusion processes of salt spray pollution, and then combines them with the topographic sub-item. Multiplication embodies a third-order collaborative risk mechanism; nonlinear power scaling is used instead of linear weighting, and a boundary layer height inverse constraint term is applied. This characterizes the dynamic characteristics of pollutant retention and accumulation, thereby constructing a comprehensive typhoon-atmospheric coupled risk index. The formula is as follows:

[0066] The parameters in the formula are calibrated based on engineering experience in the typhoon area of ​​southeastern coastal China: For pollutant items, weighting coefficients

[0067] For the nonlinear power of the pollutant term,

[0068] For meteorological terms, nonlinear powers.

[0069] The minimum value was revised to ε = 0.01. ∈[0,2.5], the larger the value, the higher the risk of flashover of external insulation during typhoons; 3. Calculate the probability P of the high-risk window period for pollution flashover. Will The linear mapping is a probability value between 0 and 1, which aligns with the quantitative requirements of engineering early warning. The formula and correction rules are as follows: Basic calculation formula: ; Fixed parameters: (No risk factors) (Extreme risk threshold); Special case correction: like Let P=1 directly (the probability of flashover is 100%). like We directly set P=0 (the model determines that the risk is low or the warning threshold has not been reached).

[0070] The model calculates and outputs the probability P of the high-risk window period for pollution flashover in real time. It classifies the risk level into two dimensions according to preset thresholds: high risk level (0.8≤P<0.95) and extremely high risk level (P≥0.95). When P≥0.8, the system automatically determines that it has entered the high-risk window period for pollution flashover and automatically extracts the expected duration of the window period, the scope of affected substations and lines, and core risk driving indicators. It compiles standardized early warning information and pushes it to dispatching personnel and on-site operation and maintenance managers through multiple channels such as the power grid operation and maintenance platform, SMS, and voice dispatch notification.

[0071] The beneficial effects of this application are as follows: (1) Rapid identification of high-risk windows, breaking through the blind spots of traditional early warning. The system sorts out the whole chain of risk mechanism in the four stages of aerosol transport before typhoon landfall - pollution deposition and aging - surface wetting and conductivity - effective precipitation and rain washing. It can quickly capture the short-term high-risk window within 24 hours before typhoon landfall, overcome the industry pain point that traditional methods cannot identify the critical risk stage before landfall, and the early warning time resolution reaches 1 hour.

[0072] (2) The accuracy of early warning has been greatly improved, and the rates of missed and false alarms have been significantly reduced. A four-dimensional progressive feature screening system of physical mechanism anchoring, statistical validity, causal validity and early warning validity was constructed. Nine core risk-causing features were selected and combined with a three-order coupling model of topographic dynamic constraints, meteorological phases and pollution chemical activity. Compared with the traditional single-indicator early warning method, the accuracy of early warning has been improved, the rate of missed and false alarms has been reduced, and the risk quantification is more in line with the actual working conditions.

[0073] (3) Deep integration of multi-source data enables more refined spatial risk characterization. By integrating high-resolution topographic, meteorological, atmospheric reanalysis, field monitoring and fault labeling data, and adopting a spatiotemporal matching method with topographic correction, the distortion that may be caused by traditional spatial interpolation is reduced, and a 1km×1km grid-level refined risk assessment is achieved, reflecting the modulation effect of the complex coastal topography on salt spray transport and pollution flashover risk.

[0074] (4) The risk levels are clear and operable, supporting the hierarchical operation and maintenance of the power grid. The comprehensive risk index is mapped to a probability value of 0 to 1, and the warning thresholds of high risk and ultra-high risk are clearly defined. The window period duration, scope of impact and core driving factors are automatically output, forming a closed loop of risk identification, level determination, warning issuance and hierarchical handling. This can provide a reference for the power grid to deploy in advance, adjust load and carry out emergency repairs, and greatly improve the efficiency of power grid operation and maintenance during typhoons.

[0075] (5) It has strong universality and engineering applicability, and is suitable for the entire coastal power grid. The method has a certain degree of adaptability to different coastal areas and voltage levels. It can be widely applied to typhoon-affected areas such as the southeast coast of my country, the Yangtze River Delta, and the Bohai Rim. It covers the main grid of transmission, transformation and distribution substations and is backward compatible with distribution networks. Under the condition of having basic meteorological, environmental and power grid monitoring data, it can reduce the need for new monitoring equipment and reuse existing meteorological, environmental and power grid monitoring data. It has low deployment cost and strong engineering feasibility.

[0076] The following are specific embodiments provided in this application: Example 1 This embodiment provides a method for identifying and issuing early warnings of high-risk windows for external insulation flashover in coastal power systems during typhoon approach. Taking the coastal power supply area of ​​Quanzhou City, Fujian Province as the implementation target, the method first screens key monitoring targets for coastal power transmission and distribution stations and their respective power supply zones around Quanzhou Bay, Weitou Bay, Anhai Bay, and Chongwu Peninsula. Then, it identifies and issues early warnings for high-risk windows for external insulation flashover during the typhoon approach phase for the selected key power transmission and distribution stations.

[0077] 1. Screening of key monitoring areas and power transmission and distribution stations The data collected included external insulation flashover fault records, power grid asset distribution data, geographical coordinates of each transmission and distribution substation, distance from the substation to the coastline, total number of pollution flashover faults during the typhoon process, and number of pollution flashover faults within the 24-hour window before the typhoon landfall in the coastal power supply area of ​​Quanzhou City from 2023 to 2025.

[0078] Using the power supply zones or substation clusters of coastal counties (cities, districts) in Quanzhou City as evaluation units, the number of substations, the total number of flashover faults during the entire typhoon process, the number of flashover faults in the 24-hour approach window before typhoon landfall, and the average distance from the coastline were statistically analyzed within each evaluation unit. The statistical period T=3 years, the distance-to-sea attenuation constant d0=20 km, the approach window magnification factor β=1.5, and the smoothing constant α=1. Based on the historical high-risk screening index model established in this application, the historical high-risk screening index H for each evaluation unit was calculated, and the evaluation units were ranked.

[0079] In this embodiment, the screening results of each evaluation unit are shown in Table 1.

[0080] Table 1

[0081] As shown in Table 1, the high-risk screening index of the Jinjiang-Shenhu area in Jinjiang City and the Xiangzhi-Hanjiang area in Shishi City is higher than that of other areas. Both areas also exhibit characteristics such as a high proportion of flashover failures in the 24 hours prior to landfall, high failure intensity per site, and proximity to the coastline. Therefore, the Jinjiang-Shenhu area in Jinjiang City and the Xiangzhi-Hanjiang area in Shishi City are identified as key monitoring areas.

[0082] After further comparing the distribution of power transmission and distribution stations, historical fault records, and coastal salt spray exposure conditions in the two key monitoring areas, the 10 kV coastal power distribution station (No. JJ-WT-01, hereinafter referred to as Jinjing Weitou Bay Coastal Station) located along the Weitou Bay coast in Jinjing Town, Jinjiang City, within the Jinjing-Shenhu area of ​​Jinjiang City, was selected as the core monitoring object for this embodiment. This station is approximately 5.6 km from the coastline, surrounded by low-altitude coastal plains transitioning to hilly terrain to the northwest, exhibiting typical topographical conditions for the transport, retention, and deposition of sea salt aerosols driven by the outer circulation of typhoons. Centered on the Jinjing Weitou Bay Coastal Station, a 5 km radius area was designated as the core monitoring area, a 5-20 km radius area as the extended monitoring area, and a 20-30 km radius area as the background analysis area.

[0083] 2. Acquisition of multi-source basic data and construction of a unified analysis dataset After identifying the Jinjingweitou Bay coastal station as the key monitoring target, the impact process of Typhoon Gaemi (No. 3 of 2024) was selected as the analysis object for the example. The analysis period was from 00:00 on July 23, 2024 to 24:00 on July 27, 2024, with a uniform time resolution of 1 hour.

[0084] The data obtained includes: Topographic data: Obtain 30 m resolution DEM data of the study area and extract parameters such as elevation, slope, topographic relief, distance from the coastline, and horizontal distance between the mountains and the coastline.

[0085] Meteorological data: Obtain hourly observation data from ground meteorological stations around the target power transmission and distribution substation, including near-surface relative humidity, air temperature, wind speed, wind direction, and hourly precipitation.

[0086] Typhoon path data: Obtain hourly or 3-hourly data on the typhoon's center location, central pressure, maximum wind speed, direction of movement, and speed of movement, and interpolate to a 1-hour time resolution.

[0087] Atmospheric reanalysis data: Obtain data such as boundary layer height, near-surface wind field, sea salt aerosol mass concentration, and sea salt aerosol deposition flux from MERRA-2 reanalysis data.

[0088] On-site monitoring data: TSP sampling points were set up along the Jinjing Weitou Bay coastline and its surrounding areas. Continuous sampling was conducted before, during, and after the typhoon's impact to obtain TSP mass concentration, water-soluble ion concentration, and Cl concentration. - / Na + Mass ratio and conductivity data at 25 °C.

[0089] The above data underwent preprocessing. For meteorological and reanalysis data, outliers were removed using the 3σ criterion, and short-term missing data were filled using linear interpolation. For field-sampled data, quality control was performed using blank filter subtraction, parallel sample deviation verification, and outlier removal. For topographic data, projection transformation, invalid value filling, and spatial resampling were completed. All data were unified to the WGS84 coordinate system, with a spatial resolution of 1 km × 1 km and a temporal resolution of 1 h, forming a standardized unified analysis dataset for the Jinjingweitou Bay coastal station.

[0090] 3. Extraction of core risk-causing features Based on the core risk-causing feature set identified in this application, nine core features are extracted from the unified analysis dataset, namely: TSP mass concentration, Cl... - / Na + Mass ratio, conductivity at 25 °C, relative humidity near the ground, hourly precipitation, boundary layer height, elevation and depression values, horizontal distance between mountains and coastline, and coastal topographic slope.

[0091] The static topographic features of the Jinjingweitou Bay coastal station are shown in Table 2.

[0092] Table 2

[0093] Table 3 shows some hourly dynamic characteristics during the approaching typhoon.

[0094] Table 3

[0095] The data above shows that from 08:00 to 15:00 on July 25th, the relative humidity in the target area along Weitou Bay in Jinjing Town remained above 90%, the TSP concentration remained at a high level, the conductivity at 25℃ increased significantly, and the boundary layer height decreased, indicating that the near-surface pollutant diffusion capacity was weakened, and the surface contaminants on the external insulation surface exhibited strong conductivity under high humidity conditions. Simultaneously, the hourly rainfall during this period was relatively low, failing to provide sufficient and effective flushing of contaminants. Therefore, environmental conditions were present for a high-risk window for external insulation flashover.

[0096] 4. Risk Index Calculation and High-Risk Window Identification The nine core risk-causing characteristics mentioned above were dimensionless by using the min-max normalization method, transforming each indicator to the 0-1 range. Subsequently, calculations were performed according to the typhoon topographic dynamic constraint comprehensive index and the typhoon atmospheric topographic coupled risk-causing comprehensive index constructed in this application.

[0097] In this embodiment, the weights for topographic factors are: elevation (0.40), horizontal distance between mountains and coastline (0.35), and coastal slope (0.25); the topographic nonlinear powers are 1.5, 1.3, and 1.4, respectively; and the minimum correction term ε is 0.01. Among the pollutants, TSP mass concentration is weighted at 0.35, and Cl... - / Na + The weight of mass ratio is 0.25, and the weight of conductivity at 25 ℃ is 0.40; in the meteorological items, the weight of relative humidity is 0.40, the weight of precipitation erosion inhibition is 0.30, and the weight of boundary layer height reverse constraint is 0.30.

[0098] The model calculates the combined typhoon-atmospheric-topographical risk index at each moment in real time and maps it to the probability P of a high-risk window period for external insulation flashover. A high-risk window period is defined as P ≥ 0.80, and an extremely high-risk window period is defined as P ≥ 0.95.

[0099] In this embodiment, the risk probability calculation results of the Jinjingweitou Bay coastal stations during the typhoon approach phase are shown in Table 4.

[0100] Table 4

[0101] Calculation results show that from 08:00 to 17:00 on July 25, 2024, the risk probability P of the Jinjing Weitou Bay coastal station and surrounding line area was consistently greater than 0.80, and the system determined this continuous period to be a high-risk window for external insulation flashover. Specifically, the risk probability reached 0.91 at 12:00 on July 25, representing the highest risk moment during this typhoon. The main driving factors for this window were: continuously increasing near-surface relative humidity, increased TSP mass concentration, increased conductivity at 25 ℃, decreased boundary layer height, and insufficient effective precipitation erosion.

[0102] After 18:00 on July 25, as the hourly rainfall increased, the TSP mass concentration and conductivity at 25 ℃ decreased, and the pollutants were partially washed away, reducing the risk probability to 0.59. The system determined that the target area along Weitou Bay in Jinjing Town had exited the high-risk window period.

[0103] 5. Early warning information generation and operation and maintenance response When the model calculates a risk probability P ≥ 0.80 for the Jinjingweitou Bay coastal station, the system automatically generates a high-risk warning for external insulation flashover. The warning information includes the warning level, expected duration, affected area, main driving factors, and recommended maintenance measures.

[0104] In this embodiment, the warning information generated by the system is as follows: "Based on model calculations, the Jinjingweitou Bay coastal substation and its surrounding 10 kV outgoing line area entered a high-risk window for external insulation flashover at 08:00 on July 25, 2024, which is expected to last until 17:00 on July 25, 2024. The current risk level is high, with a maximum risk probability of 0.91. The main risk drivers are increased relative humidity near the ground, increased TSP mass concentration, increased conductivity at 25 ℃, decreased boundary layer height, and insufficient precipitation erosion. It is recommended that the dispatching and maintenance departments strengthen the monitoring of key equipment status, conduct infrared or ultraviolet detection of external insulation in transmission and distribution substations, make emergency preparations for pollution flashover prevention, and conduct targeted inspections of key lines and equipment." After receiving the early warning, the dispatching personnel will initiate tiered operation and maintenance measures according to the warning level. For high-risk levels, the focus will be on inspecting the external insulation condition, detecting partial discharge and ultraviolet discharge in equipment, patrolling key lines and corridors, and prepositioning emergency repair teams. For extremely high-risk levels, further measures will be taken, such as adjusting operating modes, ensuring the protection of important loads, stationing repair teams on-site, and prepositioning emergency supplies.

[0105] Therefore, this embodiment realizes a complete process from screening key sites based on historical faults and geographical exposure conditions, to multi-source data fusion, extraction of core risk-causing features, calculation of risk probability, identification of high-risk windows, and issuance of early warnings. It can identify short-term high-risk stages before typhoon landfall, where "high humidity has formed but effective precipitation has not yet fully occurred," providing a basis for the prevention and control of external insulation flashover risks in coastal power systems during typhoons.

[0106] Example 2: Please see Figure 3 , Figure 3 This is a schematic diagram of the hardware device in an embodiment of this application. The hardware device specifically includes: a high-risk window period identification system 401 for external insulation flashover of coastal power systems during typhoon approach, a processor 402, a storage medium 403, and a signal transmission module 404.

[0107] A system 401 for identifying high-risk windows of external insulation flashover in coastal power systems during typhoon approach: System 401 is used to implement a method for identifying and warning of high-risk windows of external insulation flashover in coastal power systems during typhoon approach, and to complete the screening of key monitoring stations, acquisition of multi-source data, calculation of risk probability and identification of high-risk windows.

[0108] Processor 402: Processor 402 loads and executes instructions and data in storage medium 403 to implement a method for identifying and warning of high-risk windows for external insulation flashover in coastal power systems during typhoon approach.

[0109] Storage medium 403: Storage medium 403 stores instructions and data; storage medium 403 is used to implement a method for identifying and warning of high-risk windows of external insulation flashover in coastal power systems during typhoon approach.

[0110] Signal transmitting module 404: The signal transmitting module 404 is connected to the processor 402 and is used to send a warning signal to the power grid dispatching platform, operation and maintenance terminal, SMS platform or audible and visual alarm device after the processor 402 determines that the target area has entered the high risk window period of external insulation flashover, so as to realize the timely push and response of risk information.

[0111] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for identifying high-risk windows for external insulation flashover in coastal power grids under typhoon approach, characterized in that, Includes the following steps: S10: Acquire topographic data, meteorological and atmospheric reanalysis data, and field monitoring data of the study area; preprocess and match the acquired data in time and space; and integrate them to form a standardized and unified analysis dataset with grid correspondence and time synchronization. S20, the flashover of external insulation during the typhoon season is divided into four continuous links: aerosol transport, pollution deposition and aging, surface wetting and conductivity, and rain washing. Based on the standardized and unified analysis dataset, the mapping relationship between each link and candidate features is established to form a full candidate feature pool. S30, based on the standardized unified analysis dataset, the candidate features in the full candidate feature pool are screened sequentially using physical mechanism matching screening, statistical validity screening, causal validity screening and early warning validity screening, and the features that pass all screenings and have a comprehensive core degree greater than a preset threshold are determined as the core feature set for typhoon period external insulation flashover risk assessment. S40, perform dimensionless processing on each feature in the core feature set, and construct a comprehensive typhoon topographic dynamic constraint index based on the processed terrain elevation, horizontal distance between mountains and the coast, and coastal slope. Then, based on the processed TSP mass concentration and Cl... - / Na + The typhoon atmospheric-topographic coupling risk index is constructed by combining mass ratio, electrical conductivity at 25℃, near-surface relative humidity, precipitation, boundary layer height, and the typhoon topographic dynamic constraint comprehensive index. The typhoon atmospheric-topographic coupling risk index is then linearly mapped to the probability of a high-risk window period for pollution flashover.

2. The method for identifying the high-risk window period of external insulation flashover in coastal power grids under typhoon approach as described in claim 1, characterized in that, Step S10 includes a key monitoring area delineation step before data acquisition: Using coastal city power supply zones as the basic evaluation unit, the total number of pollution flashover faults during the entire typhoon process was obtained for each evaluation unit within the statistical period T=3. Number of flashover failures during the 24 hours leading up to typhoon landfall. Number of regional power transmission and distribution substations and the average distance from the substation to the coastline Set the distance-to-sea attenuation constant Approximation window magnification factor Smoothing constant Construct a historical high-risk screening index for approaching typhoons. The calculation formula is: Risk classification is performed based on index size and threshold conditions. Evaluation units with higher index rankings are identified as core key monitoring areas. Transmission and distribution substations within these areas are designated as core monitoring nodes, and the data acquisition scope for subsequent steps is limited to these core key monitoring areas.

3. The method for identifying the high-risk window period of external insulation flashover in coastal power grids under typhoon approach, as described in claim 1, is characterized in that... In step S10, the preprocessing and spatiotemporal matching of the acquired data specifically include: Outliers were removed from meteorological and atmospheric reanalysis data using the 3σ Laida criterion, and missing data were filled in using linear interpolation and Kriging. Invalid data were removed from field monitoring data using blank control and parallel sample deviation verification, and background ion concentration was deducted. Projection transformation, invalid value filling, and core parameter standardization extraction were performed on topographic data. The spatial dimension uniformly adopts the WGS84 geographic coordinate system, with a uniform resolution of 1km×1km grid, and interpolation is performed using the inverse distance weighting method with terrain correction factor; the temporal dimension is uniformly set to hourly resolution, and each timestamp is matched with a typhoon process stage label.

4. The method for identifying the high-risk window period of external insulation flashover in coastal power grids under typhoon approach as described in claim 1, characterized in that, In step S20, the full candidate feature pool Candidate feature subset of aerosol transport links Candidate feature subset of fouling deposition and aging process Candidate feature subset of surface wetting conductivity Candidate feature subsets for precipitation and rain washing process Composition, expression is ; in, Corresponding to the process of salt spray aerosol transport and topographic convergence driven by the outer circulation of the typhoon, the characteristics include near-surface wind speed, boundary layer height, straight distance from the coastline, horizontal distance between the mountains and the coast, elevation, topographic relief, slope aspect, and coastal topographic slope. Corresponding to the deposition and chemical aging processes of aerosol particles, characteristics include TSP mass concentration, Cl... - / Na + Mass ratio, SO4² - Mass concentration, conductivity at 25℃; Corresponding to the conductive process of contaminated insulating surfaces under high humidity conditions, the characteristics include near-surface relative humidity, air temperature, and conductivity at 25°C; The corresponding effective precipitation washes away the pollutants, characterized by hourly precipitation.

5. The method for identifying the high-risk window period of external insulation flashover in coastal power grids under typhoon approach as described in claim 1, characterized in that, In step S30, the physical mechanism matching screening uses the physical mechanism anchoring coefficient P(f), which is a binary rigid admission coefficient. P(f)∈{0,1}. If the candidate feature directly participates in any of the four consecutive physical links and has a clear risk-causing or risk-controlling logic, then P(f)=1 is assigned; otherwise, P(f)=0 is assigned and it is directly eliminated. The statistical validity screening uses the weighted nonlinear association strength S(f), with S(f) ranging from [0,1]. An effective threshold S(f) ≥ 0.45 is set, and features that do not meet the threshold are directly removed. The causal validity screening uses the independent risk contribution C(f), with C(f) ranging from [0,1]. An effective threshold is set: C(f) ≥ 0.10 and the 95% highest posterior density interval of C(f) does not contain 0. Features that do not meet the threshold are directly removed. The effectiveness screening of early warning adopts the high-risk window period sensitivity index W(f), with the value range of [0,1]. An effective threshold of W(f) ≥ 0.20 is set, and features that do not meet the threshold are directly eliminated. Candidate features selected through the fourth-order screening must also satisfy a comprehensive core score (Core(f)) greater than the comprehensive judgment threshold. .

6. The method for identifying the high-risk window period of external insulation flashover in coastal power grids under typhoon approach, as described in claim 5, is characterized in that... The formula for calculating the weighted nonlinear correlation strength S(f) is as follows: In the formula, is the weighted Kendall rank correlation coefficient between candidate feature f and pollution flashover risk label, used to characterize the monotonic association strength in time series samples; MIC(f) is the maximum information coefficient between candidate feature f and pollution flashover risk label, used to characterize the nonlinear association strength.

7. The method for identifying the high-risk window period of external insulation flashover in coastal power grids under typhoon approach as described in claim 5, characterized in that, The independent risk contribution C(f) is calculated based on a hierarchical Bayesian logistic regression model. ATE(f) represents the average treatment effect of candidate feature f after controlling for other candidate features, differences in evaluation units, and differences in typhoon processes. It is used to characterize the marginal causal contribution of this feature to the probability of pollution flashover risk, and the 95% highest posterior density interval of ATE(f) does not include 0. The high-risk window period sensitivity index W(f) is calculated based on the time-series SHAP contribution, and the calculation formula is as follows: ,in denoted as the total number of samples within the window period, win is the sample set within the high-risk warning window period 24 hours before typhoon landfall, and SHAP(f,t) is the SHAP contribution value of candidate feature f to the risk probability output by the model at time t.

8. The method for identifying the high-risk window period of external insulation flashover in coastal power grids under typhoon approach, as described in claim 1, is characterized in that... In step S40, the min-max normalization method is used to perform dimensionless processing on each feature in the core feature set. The normalization formula is as follows: ,in The standardized value. The original value of the indicator. , These are the minimum and maximum values ​​of the indicator in the entire dataset, respectively.

9. The method for identifying the high-risk window period of external insulation flashover in coastal power grids under typhoon approach as described in claim 1, characterized in that, In step S40, a comprehensive index of typhoon topographic dynamic constraints is constructed. The specific formula is: in The processed, standardized values ​​of terrain elevation. The standardized values ​​for the horizontal distance between the mountain range and the coastline after processing. This represents the standardized value of the coastal terrain slope after processing. , , These are the weighting coefficients for terrain factors; , , It is a non-linear power; This is the terrain coupling correction factor. ; Correction for the minimum value; The value range is [0, 1.5]; Constructing a comprehensive typhoon-atmospheric-topographic coupled risk index The specific formula is: in The normalized value of the TSP mass concentration after treatment. For the treated Cl - / Na + Quality ratio standardized value The normalized conductivity value at 25℃ after treatment. The normalized value of near-surface relative humidity after processing. The processed precipitation standardized value, This is the standardized value of the processed boundary layer height; , , These are the weighting coefficients for pollutant items; , , The pollutant term is a nonlinear power; , , The meteorological term is a nonlinear power; The value range is [0, 2.5]; The typhoon atmospheric topography coupled with the comprehensive risk index The linear mapping represents the probability P of the high-risk window period for pollution flashover, and the basic calculation formula is: ,in , ;like If ≥2.0, then directly set P=1. If P ≤ 0, then set P = 0 directly; when P ≥ 0.8, it is determined to be the high-risk window period for pollution flashover.

10. A system for identifying high-risk windows of external insulation flashover in coastal power grids during typhoon approach, characterized in that, The system includes a processor, a storage medium, and a signal transmitting module. The storage medium stores instructions, and the processor loads and executes the instructions to implement the method for identifying the high-risk window period of external insulation flashover in coastal power grids under typhoon approach as described in any one of claims 1 to 9. The signal transmitting module is connected to the processor and is used to send a warning signal to the power grid dispatching platform, operation and maintenance terminal, SMS platform, or audible and visual alarm device after the processor determines that the target area has entered the high-risk window period of external insulation flashover.

Images