A power grid outage influence evaluation method based on severe convective weather index

Abstract

The present application relates to the technical field of power grid weather disaster influence evaluation, in particular to a power grid outage influence evaluation method based on severe convective weather index. The present application obtains data corresponding to historical severe convective processes and to-be-evaluated severe convective processes, constructs a historical sample set and a to-be-evaluated sample according to time and regional correlation; extracts basic disaster features, comprehensive warning features and regional exposure degree features, constructs compound disaster interaction features, forms a severe convective weather influence feature set; takes the actual outage user number corresponding to the historical sample set as a training target to train a machine learning regression model, inversely transforms the prediction result, obtains a predicted outage user number and determines it as a severe convective weather index; and determines a power grid outage influence index according to its percentile in a historical severe convective weather index set. The present application comprehensively reflects the superposition of multiple disaster species and the regional exposure degree, realizes quantitative evaluation and historical comparison of power grid outage influence of severe convective weather in different time and different regions.

Description

Technical Field

[0001] This invention relates to the field of power grid meteorological disaster impact assessment technology, specifically a power grid outage impact assessment method based on severe convective weather index. Background Technology

[0002] Severe convective weather is characterized by its sudden onset, rapid changes in its impact area, and the simultaneous occurrence of multiple disaster factors. Heavy rain, strong winds, lightning, and hail can cause line tripping, transformer outages, substation blackouts, and user power outages within a short period. Assessing the impact of power grid outages caused by severe convective weather is a crucial foundation for conducting power grid operation risk assessments and emergency response.

[0003] Current assessments of the impact of severe convective weather on power grids primarily rely on meteorological warning levels, single meteorological thresholds, or human experience to determine the level of risk based on whether meteorological elements meet preset conditions. While these methods can reflect the intensity of severe convective weather, meteorological warning levels and meteorological element values ​​cannot directly correlate with the actual number of users experiencing power outages. The same warning level may result in different scales of power outages in different regions.

[0004] Furthermore, existing methods do not adequately consider the superimposed effects of heavy rain, strong winds, lightning, and hail, making it difficult to reflect the combined impact of multiple disasters occurring simultaneously. They also lack comprehensive consideration of regional exposure factors such as county area, total number of users in a county, and user density. The assessment results are usually expressed in terms of warning level or risk level, lacking a unified power grid outage impact index that can characterize the scale of users predicting power outages and support historical comparisons, making it difficult to quantitatively compare the power grid outage impacts of different times, regions, and different severe convective processes. Summary of the Invention

[0005] This invention provides a method for assessing the impact of power grid outages based on a severe convective weather index. It integrates real-time severe convective weather data, meteorological warning signals, regional exposure, and historical outage data to quantitatively assess the degree of impact of power grid outages.

[0006] To achieve the above objectives, the present invention provides the following technical solution: This invention provides a method for assessing the impact of power grid outages based on a severe convective weather index, comprising: S100: Acquire real-time data of severe convective weather, meteorological warning signal data, regional exposure data and actual number of users experiencing power outages corresponding to historical severe convective processes, as well as real-time data of severe convective weather, meteorological warning signal data and regional exposure data corresponding to the severe convective process to be evaluated, and preprocess the acquired data; S200: Correlate the preprocessed data according to time and region to construct a historical sample set and samples to be evaluated; S300: Extract basic disaster features, comprehensive early warning features, and regional exposure features from the historical sample set and the sample to be evaluated. Construct composite disaster interaction features based on the combination relationship between the features corresponding to different disaster types in the basic disaster features. Combine the basic disaster features, the comprehensive early warning features, the regional exposure features, and the composite disaster interaction features to form a set of severe convective weather impact features. S400: Use the set of severe convective weather impact features corresponding to the historical sample set as the model input, and use the logarithmic transformation of the actual number of power outage users corresponding to the historical sample set as the training target to train the machine learning regression model. S500: Input the historical sample set and the set of severe convective weather impact features corresponding to the sample to be evaluated into the trained machine learning regression model, perform inverse transformation on the obtained prediction results to obtain the predicted number of power outage users, and determine the predicted number of power outage users as the severe convective weather index. S600: Construct a historical severe convective weather index set based on the severe convective weather index corresponding to the historical sample set, calculate the percentile of the severe convective weather index corresponding to the sample to be evaluated in the historical severe convective weather index set, and determine the percentile as the power grid outage impact index.

[0007] As a preferred embodiment of the present invention, in S100, the specific steps of the preprocessing include: Delete completely empty records and standardize the format of the time field, administrative region name, and data field name; Convert data fields recorded as percentages to numerical values ​​and fill in missing warning counts in the weather warning signal data with zeros; Replace the infinite values ​​in the remaining numeric fields with the missing values, and fill in the missing values ​​with the median of the corresponding numeric field.

[0008] As a preferred embodiment of the present invention, in S200, the specific steps for constructing the historical sample set and the samples to be evaluated include: The combination of time, round, city, and district / county is used as the primary key of the sample; Historical samples are formed by associating the real-time data of severe convective weather, meteorological warning signal data, regional exposure data, and actual number of users experiencing power outages with the primary key of the samples, and these historical samples are then aggregated to form a historical sample set. The samples to be evaluated are formed by associating the real-time severe convective weather data, meteorological warning signal data, and regional exposure data corresponding to the severe convective process to be evaluated with the sample primary key.

[0009] As a preferred embodiment of the present invention, in S300, the basic disaster characteristics, the comprehensive early warning characteristics, and the regional exposure characteristics include: The basic disaster characteristics include maximum hourly rainfall intensity, maximum cumulative rainfall, rainstorm coverage, coverage of stations with maximum winds of level 8 or above, number of unique stations with maximum winds of level 8 or above, maximum maximum wind speed, number of lightning records, lightning density, maximum hourly lightning frequency, lightning duration, and number of different levels of hail warnings. The comprehensive early warning features include the total number of wind, rain, thunder, and hail warnings, the total number of orange-red warnings, and a red warning indicator to indicate whether a red warning exists. The regional exposure characteristics include the area of ​​the district / county, the total number of users in the county / county, and the user density. The user density is determined based on the ratio of the total number of users in the county / county to the area of ​​the district / county.

[0010] As a preferred embodiment of the present invention, in S300, the comprehensive early warning features further include rainstorm intensity, strong wind intensity, lightning intensity, and hail intensity, and the steps for constructing the rainstorm intensity, the strong wind intensity, the lightning intensity, and the hail intensity include: The number of yellow, orange, and red alerts corresponding to heavy rain, strong winds, thunderstorms, and hail were obtained respectively. For each type of disaster, the sum of the number of yellow alerts, twice the number of orange alerts, and three times the number of red alerts is determined as the warning intensity for that type of disaster. The warning intensities corresponding to heavy rain, strong wind, lightning, and hail are respectively defined as the heavy rain intensity, the strong wind intensity, the lightning intensity, and the hail intensity.

[0011] As a preferred embodiment of the present invention, in S300, the composite disaster interaction features include rain-thunder interaction features, wind-thunder interaction features, and rain-wind interaction features, and the construction steps of the composite disaster interaction features include: The maximum cumulative rainfall and the number of lightning records are each incremented by one and then logarithmed. The product of the two logarithmic values ​​is determined as the rain-lightning interaction feature. The logarithm of the number of lightning records is taken by adding one to the number of records. The product of the logarithm and the maximum wind speed is determined as the wind-lightning interaction feature. The maximum hourly rainfall intensity is increased by one and then the logarithm is taken. The product of the logarithm and the maximum wind speed is determined as the rain-wind interaction feature.

[0012] As a preferred embodiment of the present invention, in step S400, the specific steps for training the machine learning regression model include: Arrange the set of severe convective weather impact features corresponding to the historical sample set to form the model input matrix; The training target vector is formed by adding one to the number of actual power outage users corresponding to the historical sample set and taking the natural logarithm. The random forest regression model is used to fit the model input matrix and the training target vector to obtain the trained machine learning regression model.

[0013] As a preferred embodiment of the present invention, in S500, the specific steps of performing an inverse transformation on the obtained prediction result to obtain the predicted number of users experiencing power outages, and determining the predicted number of users experiencing power outages as a severe convective weather index, include: Input the set of severe convective weather impact features corresponding to the historical sample set and the sample to be evaluated into the trained machine learning regression model to obtain the prediction results on a logarithmic scale. The predicted number of users experiencing power outages is obtained by performing an exponential operation on the predicted result and then subtracting one from it. Any predicted number of users experiencing power outages that is less than zero is then adjusted to zero. The predicted number of users experiencing power outages corresponding to the historical sample set and the predicted number of users experiencing power outages corresponding to the sample to be evaluated are respectively determined as the severe convective weather index of the corresponding sample.

[0014] As a preferred embodiment of the present invention, in S600, the calculation step of the percentile specifically includes: Arrange the severe convective weather indices corresponding to the historical sample set according to their numerical values ​​to form a historical severe convective weather index set. Determine the ranking position of the severe convective weather index corresponding to the sample to be evaluated in the set of historical severe convective weather indices; Calculate the percentile value ranging from 0 to 100 based on the sorting position, and determine the percentile as the power grid outage impact index.

[0015] As a preferred embodiment of the present invention, in S600, the verification step before outputting the power grid outage impact index includes: Calculate the Pearson correlation coefficient between the severe convective weather index corresponding to the historical sample set and the actual number of users experiencing power outages; The Pearson correlation coefficient is compared with a preset correlation threshold. When the Pearson correlation coefficient is lower than the preset correlation threshold, the power grid outage impact index is stopped from being output; when the Pearson correlation coefficient reaches the preset correlation threshold, the power grid outage impact index is output.

[0016] The beneficial effects of this invention are: 1. This invention correlates real-time data of severe convective weather, meteorological warning signal data, regional exposure data, and actual number of users experiencing power outages according to time and region, so that the evaluation results simultaneously consider severe convective weather conditions and the exposure level of regional power grid users.

[0017] 2. This invention constructs rain-thunder interaction features, wind-thunder interaction features, and rain-wind interaction features. By using composite disaster interaction features, it characterizes the combination relationship when different disaster types occur together, thus avoiding the problem that it is difficult to reflect the superimposed effects of multiple disasters by using only a single disaster feature.

[0018] 3. This invention establishes a mapping relationship between the set of impact features of severe convective weather and the actual number of users experiencing power outages through a machine learning regression model, and determines the predicted number of users experiencing power outages as a severe convective weather index, so that the assessment results have a clear meaning of the scale of users experiencing power outages.

[0019] 4. This invention determines the power outage impact index based on the percentile of the severe convective weather index in the historical severe convective weather index set, and converts the prediction results of different scales into a unified historical relative position, which facilitates horizontal comparison of different times, different regions and different severe convective processes. Attached Figure Description

[0020] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart illustrating a method for assessing the impact of power outages on the grid based on a severe convective weather index, according to the present invention. Figure 2 This is a schematic diagram illustrating the multi-source data association and the construction of historical sample sets and samples to be evaluated in this invention; Figure 3 This is a flowchart of the machine learning regression model training, severe convective weather index calculation, and power outage impact index calculation of the present invention. Figure 4 This is a schematic diagram illustrating the conversion of the severe convective weather index into the power grid outage impact index according to the present invention. Figure 5 This is a comparison chart of the severe convective weather index at the district / county level and the actual number of users experiencing power outages in this invention; Figure 6 This is a schematic diagram illustrating the weighting of the impact characteristics of severe convective weather at the district / county level according to the present invention. Figure 7 This is a comparison chart of the severe convective weather index and the actual number of users experiencing power outages for a representative sample of districts and counties in this invention. Figure 8 This is a comparison chart of the predicted number of users experiencing power outages at the provincial level and the actual number of users experiencing power outages, as presented in this invention. Detailed Implementation

[0021] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0022] Example 1: As Figure 1 As shown, the present invention provides a method for assessing the impact of power grid outages based on a severe convective weather index, comprising: S100: Acquire real-time data of severe convective weather, meteorological warning signal data, regional exposure data and actual number of users experiencing power outages corresponding to historical severe convective processes, as well as real-time data of severe convective weather, meteorological warning signal data and regional exposure data corresponding to the severe convective process to be evaluated, and preprocess the acquired data; Specifically, the historical severe convective process refers to a severe convective process that has already occurred and has corresponding data on the number of users experiencing power outages, while the severe convective process to be evaluated refers to a severe convective weather index and a power grid outage impact index that need to be calculated.

[0023] The real-time data on severe convective weather includes data related to rainfall, strong winds, lightning, and hail. Rainfall data includes the maximum hourly rainfall intensity, the maximum cumulative rainfall, and the coverage rate of heavy rainfall. Strong wind data includes the coverage rate of stations with maximum winds of force 8 or above, the number of unique stations with maximum winds of force 8 or above, the total number of unique stations in the region, and the maximum maximum wind speed. Lightning data includes the number of lightning records, the density of ground flashes during the event, the maximum hourly lightning strike frequency, and the duration of lightning strikes. Hail data includes records of hail events.

[0024] The meteorological warning signal data includes the warning levels and number of warnings corresponding to heavy rain, strong winds, thunderstorms, and hail. The warning levels include blue, yellow, orange, and red warnings. The regional exposure data includes the area of ​​the district / county and the total number of users in the county. The actual number of users experiencing power outages includes the actual number of users experiencing power outages at different times and in different areas during historical severe convective weather events.

[0025] Historical severe convective weather data, meteorological warning signal data, regional exposure data, and actual power outage data for corresponding severe convective events all contain corresponding time and regional information, which are used to construct a historical sample set; the severe convective weather data, meteorological warning signal data, and regional exposure data for the severe convective events to be evaluated all contain corresponding time and regional information, which are used to construct the samples to be evaluated.

[0026] Furthermore, the specific steps of the preprocessing include: Delete records where all data fields are empty, and unify the format of time fields, administrative region names, and data field names from different data sources so that data corresponding to the same time and the same region adopts a consistent record format; Data fields recorded as percentages will be converted to numerical values. These percentage-based data fields include heavy rain coverage and coverage of stations experiencing maximum winds of force 8 or higher. The converted data will be represented using a standardized numerical format. Check the number of warnings in the meteorological warning signal data. When a certain time and area lacks the number of warnings corresponding to the disaster type or warning level, fill the missing number of warnings to zero. Examine the remaining numeric fields, replace infinity values ​​with missing values, calculate the median of the non-missing data in each numeric field, and fill in the missing values ​​using the median of the corresponding numeric field.

[0027] The data corresponding to historical severe convective events and the data corresponding to the severe convective events to be evaluated are processed using the same preprocessing rules. The preprocessed data retains the original time and regional information for subsequent correlation based on time and region.

[0028] S200: Correlate the preprocessed data according to time and region to construct a historical sample set and samples to be evaluated; Specifically, the primary key of the sample is a combination of time, round, city, and district / county. Time is used to determine the period during which the data corresponds to a severe convective event; round is used to distinguish different severe convective event records within the same time range; and city and district / county are used to determine the region corresponding to the data. Data with the same primary key are grouped into the same sample.

[0029] Furthermore, the specific steps for constructing the historical sample set and the samples to be evaluated include: Based on the primary key of the samples, historical samples are formed by associating real-time severe convective weather data, meteorological warning signal data, regional exposure data, and actual number of power outage users with the same time, round, city, and county from the preprocessed historical severe convective weather data. Each historical sample represents the severe convective weather conditions, meteorological warning conditions, regional exposure conditions, and actual number of power outage users in the corresponding time and region.

[0030] Historical samples corresponding to different sample primary keys are collected to form a historical sample set. The real-time data of severe convective weather, meteorological warning signal data, and regional exposure data in the historical sample set are used to form a set of severe convective weather impact characteristics, and the data on the actual number of users experiencing power outages are used to train a machine learning regression model.

[0031] Based on the primary key of the sample, data on actual severe convective weather, meteorological warning signals, and regional exposure are associated with the same time, round, city, and county from the preprocessed data corresponding to the severe convective process to be evaluated, forming a sample to be evaluated. This sample does not include actual power outage data; the corresponding predicted power outage number, severe convective weather index, and power grid outage impact index are determined in subsequent steps.

[0032] Historical samples and samples to be evaluated use the same time field format, administrative region name, and data field name, so that the historical sample set and the similar data in the samples to be evaluated have consistent data structure and data meaning.

[0033] like Figure 2 As shown, real-time severe convective weather data, meteorological warning signal data, regional exposure data, and actual power outage data are correlated according to the sample primary key. Specifically, the four types of data corresponding to historical severe convective events are correlated to form historical samples, and multiple historical samples are aggregated to form a historical sample set. The real-time severe convective weather data, meteorological warning signal data, and regional exposure data corresponding to the severe convective event to be evaluated are correlated to form the sample to be evaluated. The historical sample set and the sample to be evaluated serve as the data foundation for subsequent feature extraction and severe convective weather impact assessment.

[0034] S300: Extract basic disaster features, comprehensive early warning features, and regional exposure features from the historical sample set and the sample to be evaluated. Construct composite disaster interaction features based on the combination relationship between the features corresponding to different disaster types in the basic disaster features. Combine the basic disaster features, the comprehensive early warning features, the regional exposure features, and the composite disaster interaction features to form a set of severe convective weather impact features. Specifically, feature extraction and feature construction are performed on each historical sample in the historical sample set and the sample to be evaluated. Each historical sample and the sample to be evaluated use the same feature name, calculation method, and arrangement order to ensure that the set of severe convective weather impact features used to train the machine learning regression model is consistent with the set of severe convective weather impact features used to calculate the severe convective weather index.

[0035] Furthermore, the basic disaster characteristics, the comprehensive early warning characteristics, and the regional exposure characteristics include: The basic disaster characteristics include maximum hourly rainfall intensity, maximum cumulative rainfall, rainstorm coverage, coverage of stations with maximum winds of level 8 or above, number of unique stations with maximum winds of level 8 or above, total number of unique stations in the region, maximum maximum wind speed, number of lightning records, lightning density, maximum hourly lightning frequency, lightning duration, and number of hail warnings at different levels. Specifically, maximum hourly rainfall intensity, maximum cumulative rainfall, and rainstorm coverage are extracted from rainfall-related data; coverage of stations with maximum winds of level 8 or above, number of unique stations with maximum winds of level 8 or above, total number of unique stations in the region, and maximum maximum wind speed are extracted from wind-related data; number of lightning records, lightning density, maximum hourly lightning frequency, and lightning duration are extracted from lightning-related data; and the number of hail warnings at different levels is extracted from meteorological warning signal data corresponding to hail.

[0036] The comprehensive early warning features include the total number of wind, rain, thunder, and hail warnings, the total number of orange-red warnings, and a red-warning indicator to show whether a red warning exists. The total number of warnings for heavy rain, strong winds, thunder, and hail is obtained by summing the number of warnings for each of these three conditions; the total number of orange and red warnings for heavy rain, strong winds, thunder, and hail is obtained by summing the number of orange and red warnings for each of these three conditions. In one implementation, when a red warning exists in the corresponding sample, the red-warning indicator is set to 1; when no red warning exists in the corresponding sample, the red-warning indicator is set to 0.

[0037] The regional exposure features include county / district area, total number of users in the county, and user density. The county / district area and total number of users in the county are extracted from the regional exposure data, and the user density is determined based on the ratio of the total number of users in the county to the county / district area, i.e.: User density = Total number of users in the county / County / district area.

[0038] Furthermore, the comprehensive early warning features also include rainstorm intensity, gale intensity, lightning intensity, and hail intensity. The steps for constructing the rainstorm intensity, gale intensity, lightning intensity, and hail intensity include: The number of yellow, orange, and red alerts corresponding to heavy rain, strong winds, thunderstorms, and hail were obtained respectively. For each type of disaster, the sum of the number of yellow alerts, twice the number of orange alerts, and three times the number of red alerts is determined as the warning intensity for that type of disaster. The warning intensities corresponding to heavy rain, strong winds, lightning, and hail are respectively defined as heavy rain intensity, strong wind intensity, lightning intensity, and hail intensity.

[0039] Through the above processing, the intensity of rainstorms, strong winds, lightning, and hail can simultaneously reflect the number of warnings and the warning level for the corresponding disaster type, thus avoiding the problem of not being able to distinguish different warning levels when only the number of warnings is used.

[0040] Furthermore, the composite disaster interaction features include rain-thunder interaction features, wind-thunder interaction features, and rain-wind interaction features. The steps for constructing the composite disaster interaction features include: The maximum cumulative rainfall and the number of lightning records are each increased by one and then the logarithm is taken. The product of the two logarithmic values ​​is determined as the rain-lightning interaction feature, that is: Rain-lightning interaction feature = log(1 + maximum cumulative rainfall) × log(1 + number of lightning records); Add one to the number of lightning records and take the logarithm. Multiply the logarithm by the maximum wind speed and determine the wind-lightning interaction feature, i.e.: wind-lightning interaction feature = maximum wind speed × log(1 + number of lightning records); Add one to the maximum hourly rainfall intensity and take the logarithm. Multiply the resulting logarithmic value by the maximum wind speed and determine the rain-wind interaction characteristic, i.e.: Rain-wind interaction characteristic = log(1 + maximum hourly rainfall intensity) × maximum wind speed.

[0041] Here, log represents the natural logarithm; one is added to the maximum cumulative rainfall, the number of lightning records, and the maximum hourly rainfall intensity to ensure that logarithmic calculations can still be performed when the corresponding values ​​are zero. Rain-thunder interaction characteristics, wind-thunder interaction characteristics, and rain-wind interaction characteristics respectively characterize the combined relationships between rainfall and lightning, strong wind and lightning, and rainfall and strong wind in the same severe convective process.

[0042] Finally, according to the preset feature arrangement order, the basic disaster features, comprehensive early warning features, regional exposure features and complex disaster interaction features corresponding to each historical sample and the sample to be evaluated are combined to form the severe convective weather impact feature set of the corresponding sample, which is used to train the machine learning regression model and calculate the severe convective weather index.

[0043] S400: Use the set of severe convective weather impact features corresponding to the historical sample set as the model input, and use the logarithmic transformation of the actual number of power outage users corresponding to the historical sample set as the training target to train the machine learning regression model. Specifically, according to the feature arrangement order determined in S300, the basic disaster features, comprehensive early warning features, regional exposure features and compound disaster interaction features corresponding to each historical sample in the historical sample set are arranged in sequence, so that the set of severe convective weather impact features corresponding to each historical sample has the same feature dimension and arrangement order.

[0044] Furthermore, the specific steps for training the machine learning regression model include: Following the order of the historical samples, the sets of severe convective weather impact features corresponding to each historical sample are arranged row by row to form the model input matrix. Each row in the model input matrix corresponds to a historical sample, and each column corresponds to a feature from the set of severe convective weather impact features.

[0045] Obtain the actual number of users experiencing power outages corresponding to each historical sample. Increment this number by one and take its natural logarithm to obtain the corresponding training target value: Training target value = log(1 + actual number of users experiencing power outages). Arrange the training target values ​​into a training target vector according to the same historical sample arrangement order as the model input matrix, ensuring a one-to-one correspondence between each row of the model input matrix and a training target value in the training target vector. Here, log represents the natural logarithm.

[0046] A logarithmic transformation is applied to the actual number of users experiencing power outages to reduce the impact of differences in the number of actual users experiencing power outages between different historical samples on model training, and to make the distribution of the training target value smoother.

[0047] A random forest regression model is used to fit the model input matrix and the training target vector. During model training, sampling with replacement is performed from the historical sample set to form multiple sets of training samples, and regression decision trees are constructed based on each set of training samples. In the node partitioning process of each regression decision tree, candidate features are selected from the set of severe convective weather impact features, and the node partitioning method is determined according to the regression error corresponding to the candidate partitioning results, until the preset tree depth or node sample number condition is reached. The output results of multiple regression decision trees are averaged to establish a mapping relationship between the set of severe convective weather impact features and the logarithmic value of the actual number of power outage users, thus obtaining the trained machine learning regression model.

[0048] In one implementation, the parameters of the random forest regression model are set as follows: , , , , .in, Indicates the number of regression decision trees; Indicates the maximum depth of a single regression decision tree; This represents the minimum number of samples required for an internal node to continue splitting. This represents the minimum number of samples contained in a leaf node; Used to set a random seed during model training.

[0049] The trained machine learning regression model takes the set of features affecting severe convective weather as input and outputs the prediction results of the corresponding samples on the logarithmic scale of the actual number of power outage users, which are then used for inverse transformation to determine the predicted number of power outage users and the severe convective weather index.

[0050] S500: Input the historical sample set and the set of severe convective weather impact features corresponding to the sample to be evaluated into the trained machine learning regression model, perform inverse transformation on the obtained prediction results to obtain the predicted number of power outage users, and determine the predicted number of power outage users as the severe convective weather index. Specifically, according to the feature names and arrangement order used when training the machine learning regression model, the sets of severe convective weather impact features corresponding to each historical sample and the sets of severe convective weather impact features corresponding to the sample to be evaluated are input into the trained machine learning regression model. Based on the mapping relationship established in S400, the trained machine learning regression model outputs the prediction results at the logarithmic scale corresponding to each historical sample and the sample to be evaluated.

[0051] Furthermore, the specific steps of performing an inverse transformation on the obtained prediction results to obtain the predicted number of users experiencing power outages, and determining the predicted number of users experiencing power outages as a severe convective weather index, include: The predicted number of power outage users is obtained by subtracting one from the exponentially calculated prediction results for each historical sample and the sample to be evaluated on a logarithmic scale, i.e.: Predicted number of power outage users = exp(prediction result on a logarithmic scale) - 1. Here, exp represents exponential operation with the natural constant as the base. This inverse transformation corresponds to the process in S400 of adding one to the actual number of power outage users and then taking the natural logarithm, and is used to restore the prediction results on a logarithmic scale to the scale of the number of power outage users.

[0052] When the predicted number of users experiencing power outages obtained from the inverse transformation is less than zero, the corresponding predicted number of users experiencing power outages is adjusted to zero to avoid negative values ​​that do not reflect the actual meaning of the number of users experiencing power outages.

[0053] The predicted number of users experiencing power outages corresponding to each historical sample is determined as the severe convective weather index corresponding to each historical sample, and the predicted number of users experiencing power outages corresponding to the sample to be evaluated is determined as the severe convective weather index corresponding to the sample to be evaluated. Therefore, the severe convective weather index and the predicted number of users experiencing power outages use the same numerical scale to represent the scale of users experiencing power outages under corresponding severe convective weather conditions.

[0054] like Figure 3 As shown, the severe convective weather impact feature set comprises basic disaster characteristics, comprehensive early warning characteristics, regional exposure characteristics, and complex disaster interaction characteristics. This severe convective weather impact feature set serves as the input to a machine learning regression model, with the logarithmic value of the actual number of users experiencing power outages serving as the training objective. The trained machine learning regression model outputs prediction results on a logarithmic scale, which are then inversely transformed to obtain the severe convective weather index. The severe convective weather index undergoes percentile mapping in subsequent steps to obtain the power grid outage impact index.

[0055] S600: Construct a historical severe convective weather index set based on the severe convective weather index corresponding to the historical sample set, calculate the percentile of the severe convective weather index corresponding to the sample to be evaluated in the historical severe convective weather index set, and determine the percentile as the power grid outage impact index.

[0056] Specifically, the severe convective weather indices corresponding to each historical sample obtained from S500 are acquired and arranged in ascending order of value to form a historical severe convective weather index set. This historical severe convective weather index set is used to characterize the distribution of the predicted power outage scale during historical severe convective events.

[0057] Furthermore, the calculation steps for the percentile specifically include: Determine the ranking position of the severe convective weather index corresponding to the sample to be evaluated in the historical severe convective weather index set, and calculate the percentile value ranging from 0 to 100 based on the ranking position.

[0058] In one implementation, the number of severe convective weather indices in the historical severe convective weather index set that are less than or equal to the severe convective weather index corresponding to the sample to be evaluated is counted. The ratio of the number of severe convective weather indices to the total number of severe convective weather indices in the historical severe convective weather index set is multiplied by 100 to obtain the percentile, i.e.: Power outage impact index = Number of severe convective weather indices less than or equal to the severe convective weather index corresponding to the sample to be evaluated / Total number of historical severe convective weather indices × 100.

[0059] The percentile is determined as the power grid outage impact index corresponding to the sample to be evaluated. When the power grid outage impact index is closer to 100, it indicates that the severe convective weather index corresponding to the sample to be evaluated has a higher ranking position in the historical severe convective weather index set, and the corresponding predicted outage user scale is larger relative to the historical sample; when the power grid outage impact index is closer to 0, it indicates that the predicted outage user scale corresponding to the sample to be evaluated is smaller relative to the historical sample.

[0060] Furthermore, before outputting the power outage impact index, the trained machine learning regression model is validated. Following the same historical sample arrangement order, the severe convective weather index corresponding to each historical sample is mapped one-to-one with the actual number of users experiencing power outages, and the Pearson correlation coefficient between the severe convective weather index corresponding to the historical sample set and the actual number of users experiencing power outages is calculated.

[0061] The Pearson correlation coefficient is compared with a preset correlation threshold. When the Pearson correlation coefficient is lower than the preset correlation threshold, it indicates that the correlation between the severe convective weather index corresponding to the historical sample set and the actual number of users experiencing power outages has not met the output conditions, and the output of the power grid outage impact index is stopped; when the Pearson correlation coefficient reaches the preset correlation threshold, the power grid outage impact index corresponding to the sample to be evaluated is output.

[0062] The preset correlation threshold is pre-set according to the evaluation scale and business requirements. In the county-level evaluation scenario, the preset correlation threshold is set to 0.8. When the Pearson correlation coefficient is greater than or equal to 0.8, the predicted number of power outage users, the severe convective weather index, and the power grid outage impact index corresponding to the sample to be evaluated are output.

[0063] For historical samples with actual power outage data, residuals, absolute errors, relative errors, outage rates, and predicted outage rates can be calculated based on the predicted and actual outage numbers. The residual is the difference between the predicted and actual outage numbers, and the absolute error is the absolute value of the residual. When the actual outage number is greater than 0, the relative error is the ratio of the residual to the actual outage number multiplied by 100%; when the actual outage number is 0, the relative error is not calculated. The outage rate is the ratio of the actual outage number to the total number of users in the county multiplied by 100%, and the predicted outage rate is the ratio of the predicted outage number to the total number of users in the county multiplied by 100%.

[0064] like Figure 4 As shown, the historical severe convective weather index set forms a historical severe convective weather index distribution, and the severe convective weather index corresponding to the sample to be evaluated is located at the corresponding position in the historical severe convective weather index distribution. Based on this position, the historical percentile corresponding to the sample to be evaluated is determined, and the historical percentile is mapped to a power grid outage impact index with a value range of 0 to 100.

[0065] In summary, this embodiment obtains the predicted number of users experiencing power outages, the severe convective weather index, and the power grid outage impact index through multi-source data association, feature construction, model training, and index conversion, thereby achieving a quantitative assessment of the degree of power grid outage impact of the severe convective process to be evaluated.

[0066] Example 2: This example uses the power grid outage impact assessment method based on the severe convective weather index described in Example 1 to assess and verify severe convective weather processes at the county and district level.

[0067] Specifically, district and county-level data corresponding to multiple historical severe convective events were selected as historical samples. The primary keys of the samples were constructed according to time, round, city, and district / county. The real-time data of severe convective weather, meteorological warning signal data, regional exposure data, and actual power outage data were correlated to form a district / county-level historical sample set.

[0068] Based on the historical sample set, basic disaster features, comprehensive early warning features, and regional exposure features are extracted. Then, based on the combination relationships between features corresponding to different disaster types within the basic disaster features, composite disaster interaction features are constructed, forming a set of severe convective weather impact features corresponding to each historical sample. The set of severe convective weather impact features corresponding to each historical sample is input into a trained machine learning regression model. The obtained prediction results are inversely transformed to obtain the predicted number of users experiencing power outages for each historical sample. This predicted number of users experiencing power outages is then determined as the severe convective weather index for the corresponding historical sample.

[0069] To verify the ability of the severe convective weather index to characterize the actual impact of power outages, the severe convective weather index corresponding to each historical sample was compared and analyzed with the actual number of users experiencing power outages.

[0070] like Figure 5 As shown, the horizontal axis represents the logarithmic value of the actual number of users experiencing power outages plus one, and the vertical axis represents the logarithmic value of the severe convective weather index plus one. The scatter plots in the graph represent the correspondence between the actual number of users experiencing power outages and the severe convective weather index for different historical samples. Figure 5 It can be seen that the severe convective weather index has a good correlation with the actual number of users experiencing power outages on a logarithmic scale, indicating that the severe convective weather index obtained based on the set of severe convective weather impact characteristics can reflect the scale of power outages caused by different severe convective processes.

[0071] Furthermore, a correlation analysis was conducted between the severe convective weather index corresponding to historical samples at the district and county levels and the actual number of users experiencing power outages. The Pearson correlation coefficient, log-scaled Pearson correlation coefficient, Spearman correlation coefficient, coefficient of determination, mean absolute error, and root mean square error were calculated. Specifically, the Pearson correlation coefficient was used to evaluate the degree of linear correlation between the severe convective weather index and the actual number of users experiencing power outages; the Spearman correlation coefficient was used to evaluate the ranking relationship between the two; the coefficient of determination was used to evaluate the explanatory power of the prediction results for changes in the actual number of users experiencing power outages; and the mean absolute error and root mean square error were used to evaluate the level of prediction error.

[0072] In this embodiment, the Pearson correlation coefficient between the severe convective weather index corresponding to the historical samples at the district / county level and the actual number of users experiencing power outages is 0.872, which reaches the preset correlation threshold, indicating that the severe convective weather index can well reflect the actual scale of users experiencing power outages corresponding to severe convective weather processes at the district / county level.

[0073] Furthermore, feature contribution analysis is performed on the output of the trained machine learning regression model.

[0074] like Figure 6 As shown, different severe convective weather impact characteristics contribute to the prediction results to varying degrees. Among them, the characteristics of rain-thunder interaction, total number of users in the county, maximum 1-hour rainfall intensity, maximum cumulative rainfall, county area, and user density have a high contribution, indicating that the combined impact of disaster factors such as rainfall and lightning, as well as the degree of regional user exposure, all affect the power outage impact assessment results.

[0075] Furthermore, typical strong convective processes from different years were selected for comparative analysis.

[0076] like Figure 7As shown, there is a corresponding relationship between the severe convective weather index corresponding to severe convective events in different years and the actual number of users experiencing power outages. Samples with higher severe convective weather indices usually correspond to a larger number of actual power outage users, indicating that the proposed method can quantitatively distinguish the degree of power outage impact of different severe convective events by utilizing the correlation between basic disaster characteristics, comprehensive early warning characteristics, regional exposure characteristics, and complex disaster interaction characteristics in historical samples.

[0077] In summary, this embodiment verifies the effectiveness of the power grid outage impact assessment method based on the severe convective weather index using historical samples at the district and county levels. The verification results show that the method can comprehensively consider the impact of multiple disasters and the degree of regional exposure, and quantitatively assess the power grid outage impact caused by severe convective weather.

[0078] Example 3: This example verifies the extended application of the method described in Example 1 at the provincial level.

[0079] Specifically, in the provincial-level assessment scenario, the severe convective process corresponding to the date is used as the basis for sample division. This is combined with provincial-level data on actual severe convective weather, meteorological warning signals, and the actual number of users experiencing power outages to form a provincial historical sample set. Each historical sample represents the severe convective weather conditions, meteorological warnings, and the actual number of users experiencing power outages on the corresponding date or during the severe convective process.

[0080] Basic disaster characteristics and comprehensive early warning characteristics are extracted from the aforementioned provincial historical sample sets to form a provincial severe convective weather characteristic set corresponding to each provincial historical sample. The basic disaster characteristics include the maximum 1-hour rainfall intensity, the maximum cumulative rainfall, the rainstorm coverage rate, the coverage rate of stations with maximum winds of level 8 or above, the number of unique stations with maximum winds of level 8 or above, the number of lightning records, and the maximum 1-hour lightning strike frequency; the comprehensive early warning characteristics include the number of warnings corresponding to rainstorms, strong winds, lightning, and hail.

[0081] Following the model training method described in Example 1, the provincial severe convective weather feature sets corresponding to each provincial historical sample are arranged to form a model input matrix. The actual number of power outage users corresponding to each provincial historical sample is incremented by one, and the natural logarithm is taken to form a training target vector, which is then used to train a machine learning regression model. The provincial severe convective weather feature sets corresponding to each provincial historical sample are input into the trained machine learning regression model, and the obtained prediction results are inversely transformed to obtain the predicted number of power outage users corresponding to each provincial historical sample. The predicted number of power outage users is then determined as the severe convective weather index for the corresponding historical sample.

[0082] A set of historical severe convective weather indices is constructed based on the severe convective weather indices corresponding to historical samples in each province. The percentile of the severe convective weather index corresponding to the sample to be evaluated in the set of historical severe convective weather indices is calculated, and the percentile is determined as the power grid outage impact index corresponding to the sample to be evaluated.

[0083] like Figure 8 As shown, the horizontal axis represents the actual number of users experiencing power outages, and the vertical axis represents the predicted number of users experiencing power outages. The dashed line in the figure represents the reference relationship when the predicted number of users experiencing power outages equals the actual number of users experiencing power outages. Each scatter point represents the correspondence between the actual number of users experiencing power outages and the predicted number of users experiencing power outages corresponding to different provincial historical samples. The closer the scatter point is to the dashed line, the smaller the prediction error of the corresponding historical sample.

[0084] In this embodiment, the number of provincial-level historical samples is 68, and the correlation coefficient between the severe convective weather index corresponding to each provincial-level historical sample and the actual number of users experiencing power outages is 0.7487. The date corresponding to the highest power grid outage impact index at the provincial level is April 12, 2025, and the date corresponding to the highest number of actual power outage users is April 21, 2025, with a maximum of 698,583 users experiencing power outages.

[0085] Feature contribution analysis was performed on the trained machine learning regression model. The results showed that the number of unique stations with maximum winds of level 8 or above had the highest contribution, followed by the number of lightning records. This indicates that the influence range of strong winds and the intensity of lightning activity at the provincial scale have a high impact on the predicted number of users experiencing power outages.

[0086] In summary, this embodiment demonstrates that the method can calculate the severe convective weather index and the power grid outage impact index at the provincial level, and quantitatively compare the impact of power grid outages on the entire province caused by different dates or severe convective processes.

[0087] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for assessing the impact of power grid outages based on a severe convective weather index, characterized in that, include: S100: Acquire real-time data of severe convective weather, meteorological warning signal data, regional exposure data and actual number of users experiencing power outages corresponding to historical severe convective processes, as well as real-time data of severe convective weather, meteorological warning signal data and regional exposure data corresponding to the severe convective process to be evaluated, and preprocess the acquired data; S200: Correlate the preprocessed data according to time and region to construct a historical sample set and samples to be evaluated; S300: Extract basic disaster features, comprehensive early warning features, and regional exposure features from the historical sample set and the sample to be evaluated. Construct composite disaster interaction features based on the combination relationship between the features corresponding to different disaster types in the basic disaster features. Combine the basic disaster features, the comprehensive early warning features, the regional exposure features, and the composite disaster interaction features to form a set of severe convective weather impact features. S400: Use the set of severe convective weather impact features corresponding to the historical sample set as the model input, and use the logarithmic transformation of the actual number of power outage users corresponding to the historical sample set as the training target to train the machine learning regression model. S500: Input the historical sample set and the set of severe convective weather impact features corresponding to the sample to be evaluated into the trained machine learning regression model, perform inverse transformation on the obtained prediction results to obtain the predicted number of power outage users, and determine the predicted number of power outage users as the severe convective weather index. S600: Construct a historical severe convective weather index set based on the severe convective weather index corresponding to the historical sample set, calculate the percentile of the severe convective weather index corresponding to the sample to be evaluated in the historical severe convective weather index set, and determine the percentile as the power grid outage impact index.

2. The method for assessing the impact of power grid outages based on a severe convective weather index as described in claim 1, characterized in that, In S100, the specific steps of the preprocessing include: Delete completely empty records and standardize the format of the time field, administrative region name, and data field name; Convert data fields recorded as percentages to numerical values ​​and fill missing warning counts in weather warning signal data with zeros; Replace the infinite values ​​in the remaining numeric fields with the missing values, and fill in the missing values ​​with the median of the corresponding numeric field.

3. The method for assessing the impact of power grid outages based on a severe convective weather index as described in claim 1, characterized in that, In S200, the specific steps for constructing the historical sample set and the samples to be evaluated include: The combination of time, round, city, and district / county is used as the primary key of the sample; Historical samples are formed by associating the real-time data of severe convective weather, meteorological warning signal data, regional exposure data, and actual number of users experiencing power outages with the primary key of the samples, and these historical samples are then aggregated to form a historical sample set. The samples to be evaluated are formed by associating the real-time severe convective weather data, meteorological warning signal data, and regional exposure data corresponding to the severe convective process to be evaluated with the sample primary key.

4. The method for assessing the impact of power grid outages based on a severe convective weather index as described in claim 1, characterized in that, In S300, the basic disaster characteristics, the comprehensive early warning characteristics, and the regional exposure characteristics include: The basic disaster characteristics include maximum hourly rainfall intensity, maximum cumulative rainfall, rainstorm coverage, coverage of stations with maximum winds of level 8 or above, number of unique stations with maximum winds of level 8 or above, maximum maximum wind speed, number of lightning records, lightning density, maximum hourly lightning frequency, lightning duration, and number of different levels of hail warnings. The comprehensive early warning features include the total number of wind, rain, thunder, and hail warnings, the total number of orange-red warnings, and a red warning indicator to indicate whether a red warning exists. The regional exposure characteristics include the area of ​​the district / county, the total number of users in the county / county, and the user density. The user density is determined based on the ratio of the total number of users in the county / county to the area of ​​the district / county.

5. The method for assessing the impact of power grid outages based on a severe convective weather index according to claim 4, characterized in that, In S300, the comprehensive early warning features also include rainstorm intensity, strong wind intensity, lightning intensity, and hail intensity. The steps for constructing the rainstorm intensity, strong wind intensity, lightning intensity, and hail intensity include: The number of yellow, orange, and red alerts corresponding to heavy rain, strong winds, thunderstorms, and hail were obtained respectively. For each type of disaster, the sum of the number of yellow alerts, twice the number of orange alerts, and three times the number of red alerts is determined as the warning intensity for that type of disaster. The warning intensities corresponding to heavy rain, strong wind, lightning, and hail are respectively defined as the heavy rain intensity, the strong wind intensity, the lightning intensity, and the hail intensity.

6. The method for assessing the impact of power grid outages based on a severe convective weather index according to claim 4, characterized in that, In S300, the composite disaster interaction features include rain-thunder interaction features, wind-thunder interaction features, and rain-wind interaction features. The steps for constructing the composite disaster interaction features include: The maximum cumulative rainfall and the number of lightning records are each incremented by one and then logarithmed. The product of the two logarithmic values ​​is determined as the rain-lightning interaction feature. The logarithm of the number of lightning records is taken by adding one to the number of records. The product of the logarithm and the maximum wind speed is determined as the wind-lightning interaction feature. The maximum hourly rainfall intensity is increased by one and then the logarithm is taken. The product of the logarithm and the maximum wind speed is determined as the rain-wind interaction feature.

7. The method for assessing the impact of power grid outages based on a severe convective weather index according to claim 1, characterized in that, In S400, the specific steps for training the machine learning regression model include: Arrange the set of severe convective weather impact features corresponding to the historical sample set to form the model input matrix; The training target vector is formed by adding one to the number of actual power outage users corresponding to the historical sample set and taking the natural logarithm. The random forest regression model is used to fit the model input matrix and the training target vector to obtain the trained machine learning regression model.

8. The method for assessing the impact of power grid outages based on a severe convective weather index according to claim 7, characterized in that, In S500, the specific steps of performing an inverse transformation on the obtained prediction results to obtain the predicted number of users experiencing power outages, and determining the predicted number of users experiencing power outages as a severe convective weather index, include: Input the set of severe convective weather impact features corresponding to the historical sample set and the sample to be evaluated into the trained machine learning regression model to obtain the prediction results on a logarithmic scale. The predicted number of users is obtained by performing an exponential operation on the predicted result and then subtracting one. Any predicted number of users with a power outage that is less than zero is then adjusted to zero. The predicted number of users experiencing power outages corresponding to the historical sample set and the predicted number of users experiencing power outages corresponding to the sample to be evaluated are respectively determined as the severe convective weather index of the corresponding sample.

9. The method for assessing the impact of power grid outages based on a severe convective weather index according to claim 1, characterized in that, In S600, the calculation steps for the percentile specifically include: Arrange the severe convective weather indices corresponding to the historical sample set according to their numerical values ​​to form a historical severe convective weather index set. Determine the ranking position of the severe convective weather index corresponding to the sample to be evaluated in the set of historical severe convective weather indices; The percentile, ranging from 0 to 100, is calculated based on the sorting position, and the percentile is determined as the power grid outage impact index.

10. The method for assessing the impact of power grid outages based on a severe convective weather index according to claim 1, characterized in that, In S600, the verification steps before outputting the power outage impact index include: Calculate the Pearson correlation coefficient between the severe convective weather index corresponding to the historical sample set and the actual number of users experiencing power outages; The Pearson correlation coefficient is compared with a preset correlation threshold. When the Pearson correlation coefficient is lower than the preset correlation threshold, the power grid outage impact index is stopped from being output; when the Pearson correlation coefficient reaches the preset correlation threshold, the power grid outage impact index is output.

Images