Identification method and system based on short-time heavy rainfall precursor of electric power facility

Abstract

The invention discloses an identification method and system based on a short-time heavy rainfall precursor of an electric power facility, and particularly relates to the technical field of identification of the short-time heavy rainfall precursor, and the method comprises the steps: obtaining original multi-source observation data related to the short-time heavy rainfall in the electric power facility in an electric power corridor region of a southwest region, and carrying out the time alignment; outputting a multi-source observation data set covering the power corridor area in the southwest region; performing quality control processing on the multi-source observation data set, checking the consistency and validity of the multi-source observation data set, removing missing values and abnormal values, and outputting a multi-source quality control data set; and constructing a precursor fragment population by fusing the enhancement factor and the attenuation static factor, executing coevolution grouping and frequency similarity modeling, and outputting a short-time heavy rainfall precursor identification result with a stable structure and comprehensive expression.

Description

Technical Field

[0001] This invention relates to the field of short-term heavy rainfall precursor identification technology, and more specifically, to a method and system for identifying short-term heavy rainfall precursors based on power facilities. Background Technology

[0002] The Southwest region has significant topographic differences and a large latitudinal and longitude span. It encompasses various landform types, including plateaus, basins, and hills. These different landforms exhibit distinct climatic characteristics, which are particularly pronounced under extreme high-impact weather conditions.

[0003] Short-duration heavy precipitation, as a typical high-impact weather event, involves the interaction of atmospheric systems of different properties and scales under the complex terrain of the plateau, resulting in drastic weather changes that are difficult to predict accurately.

[0004] While existing research has made progress in data collection, reanalysis field applications, and radar monitoring, its application in power facility disaster scenarios is still significantly insufficient, especially in the identification of precursor factors of short-term heavy precipitation. Most existing methods rely on enhanced signals such as significant increase in humidity, enhanced radar echo, and sudden increase in ground wind speed, while lacking the identification and utilization of attenuated or stationary signals that may appear in the short period of time before the disaster is triggered.

[0005] This shows that the precursor factor dataset formed after data collection and quality control processing has missing data, and the risk assessment results show systematic bias, making it difficult to fully characterize the potential impact of short-term heavy rainfall on power facilities. Summary of the Invention

[0006] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a method and system for identifying short-term heavy rainfall precursors based on power facilities. By fusing enhancement factors and attenuation quiescent factors to construct a precursor fragment population, performing co-evolutionary grouping and frequency similarity modeling, the system outputs a short-term heavy rainfall precursor identification result that is structurally stable and comprehensively expressed.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a method for identifying precursors of short-term heavy rainfall to power facilities, comprising:

[0008] S1. Obtain raw multi-source observation data related to short-term heavy rainfall in power facilities in the power corridor area of ​​Southwest China, perform time alignment, and output a multi-source observation dataset covering the power corridor area of ​​Southwest China.

[0009] S2. Perform quality control processing on the multi-source observation dataset to verify the consistency and validity of the multi-source observation dataset, remove missing values ​​and outliers, and output the multi-source quality control dataset.

[0010] S3. Perform annotation operations on the multi-source quality control dataset. The enhanced signals that show increased humidity, enhanced echo, and sudden increase in wind speed are labeled as enhancement factors. The attenuated or stationary signals that show decreased humidity, weakened echo, and stationary wind speed are labeled as attenuation stationary factors. These are used as different fragment units of the precursor factor mosaic and the precursor factor mosaic set is output.

[0011] S4. Based on the co-evolutionary clustering method, the enhanced factor fragments and decaying stationary factor fragments in the precursor factor puzzle set are used as input. According to the frequency of occurrence in the power facility disaster records, the enhanced factor fragments and decaying stationary factor fragments that appear frequently in multiple power facility disaster records are retained as stable precursor fragments, and the stable precursor fragment set is output.

[0012] S5. Combine the enhanced factor fragments and attenuated static factor fragments in the stable precursor fragment set in pairs to form candidate splicing sequences. Calculate the early warning lead time and false alarm rate of each candidate splicing sequence in the power facility disaster record and compare them with the preset threshold. Retain the candidate splicing sequences that meet the preset threshold to form a precursor factor dataset.

[0013] In a preferred embodiment, in S1, raw multi-source observation data related to short-term heavy rainfall are acquired from power facilities in the power corridor area of ​​Southwest China. The raw multi-source observation data includes data from ground meteorological observation stations, weather radar data, reanalysis data, and records of power facility disasters and proprietary monitoring data.

[0014] Extract unified timestamp information from the original multi-source observation data, divide the time window according to the preset time interval, and perform interpolation to fill in the record points that do not completely coincide with the window time.

[0015] When data from a certain source contains observation records in two adjacent windows, the time difference and value difference of the corresponding observation records are taken. The value of the previous observation record is added to the product of the corresponding proportion and the value difference to obtain the observation value at the corresponding time in the window.

[0016] When a data source is missing adjacent records, the observation record closest to the window is used as a substitute to obtain the complete set of window observations;

[0017] After interpolating and completing the data from each source, the observations in each window are merged according to the preset time order, and a multi-source observation dataset aligned to a unified standard time series and covering the power corridor area in Southwest China is output.

[0018] In a preferred embodiment, in S2, in the aligned multi-source observation dataset, it is detected whether there are records in each time window that cannot be filled in by interpolation or the nearest value. If there are, the corresponding record is marked as missing; otherwise, it is marked as complete.

[0019] The system detects whether the difference between observations in a multi-source observation dataset within the same spatial neighborhood exceeds a preset threshold. Observations exceeding the threshold are marked as abnormal, while those not exceeding the threshold are marked as normal.

[0020] Numerical validity checks are performed on the observations in the multi-source observation dataset. Observations that exceed a preset reasonable range are marked as invalid, and those that do not are marked as valid.

[0021] Perform a removal operation on observation records marked as missing, abnormal, or invalid, and output a multi-source quality control dataset that meets the conditions of completeness, consistency, and validity.

[0022] In a preferred embodiment, in S3, the variation characteristics of humidity, radar echo intensity, and wind speed variables in the multi-source quality control dataset are determined:

[0023] At the same spatial location, the difference between the current humidity value in the current time window and the humidity value in the adjacent previous time window is taken. When the difference is greater than the humidity threshold, it is determined that the humidity is rising; when the difference is less than or equal to the humidity threshold, it is determined that the humidity is falling.

[0024] At the same spatial location, calculate the difference between the representative value of radar echo intensity in the current time window and the mean value of radar echo intensity in multiple adjacent windows. When the difference is greater than the echo threshold, it is determined to be an enhanced echo; when the difference is less than or equal to the echo threshold, it is determined to be a weakened echo.

[0025] In the same spatial location, the ratio of the wind speed representative value of the current time window to the wind speed representative value of the adjacent previous time window is calculated. When the ratio is greater than the wind speed ratio threshold, it is determined that the wind speed has suddenly increased. When the ratio is less than or equal to the wind speed ratio threshold, it is determined that the wind speed is stationary.

[0026] When any one of the following conditions is met—increased humidity, enhanced echo, or sudden increase in wind speed—the corresponding observation record will be marked as an enhancement factor.

[0027] When any one of the following conditions is met—decreased humidity, weakened echo, or stationary wind speed—the corresponding observation record will be marked as the attenuation stationary factor.

[0028] The labeled enhancement factor and decay quiescent factor are used as different fragment units of the precursor factor mosaic, and a set of enhancement factor fragments and a set of decay quiescent factor fragments are established in each time window. The sets of enhancement factor fragments and decay quiescent factor fragments in each time window are combined in sequence to output the precursor factor mosaic set.

[0029] In a preferred embodiment, in S4, the enhancement factor fragments and decay quiescent factor fragments in the precursor factor puzzle set are respectively used as the initial enhancement cluster and the initial decay quiescent cluster to form the initial cluster population.

[0030] Co-evolutionary iterative operation is performed on the initial population. During the iteration, the number of times any two precursor factor fragments co-occur in the power facility disaster records is counted, and the ratio of the number of co-occurrences to the total number of power facility disaster records containing at least one fragment is calculated to obtain the similarity between the two. When the similarity is greater than or equal to a preset threshold, the corresponding fragments are assigned to the same population.

[0031] When the similarity is less than a preset threshold, the precursor factor fragments from different clusters are cross-recombined to generate new candidate pairings, and the similarity is recalculated until the similarity result no longer changes or the number of iterations reaches the preset upper limit.

[0032] In a preferred embodiment, S4 further includes determining the corresponding cluster as a stable cluster structure when the similarity between any two fragments within a cluster remains above a preset threshold in two consecutive iterations.

[0033] The frequency of occurrence of each fragment in the stable cluster structure in the power facility disaster records is calculated by comparing it with the total number of disaster records. Fragments with a frequency greater than a preset ratio threshold are retained as stable precursor fragments, and a stable precursor fragment set is output.

[0034] In a preferred embodiment, in S5, the enhancement factor fragments and decaying stationary factor fragments in the stable precursor fragment set are spliced ​​in pairs to form a candidate splicing sequence.

[0035] For each candidate splicing sequence, extract the disaster start time from the power facility disaster record and the first trigger time of the candidate splicing sequence, calculate the time difference between the two, and obtain the early warning lead time of the candidate splicing sequence;

[0036] For each candidate splicing sequence, the number of times the candidate splicing sequence is triggered in records without disasters is counted, and the total number of triggers in all records is counted. The ratio of the two is used as the false alarm rate of the candidate splicing sequence.

[0037] In a preferred embodiment, S5 further includes comparing the warning lead time and false alarm rate of each candidate splicing sequence with preset thresholds, retaining candidate splicing sequences whose warning lead time and false alarm rate both meet the preset thresholds, and forming a precursor factor dataset for monitoring and risk assessment of short-term heavy rainfall precursors in power facilities.

[0038] A system for identifying precursors of short-term heavy rainfall to power facilities includes a data alignment module, a quality control rejection module, a labeling module, a clustering and screening module, and a splicing and evaluation module.

[0039] The data alignment module is used to acquire raw multi-source observation data related to short-term heavy precipitation in power facilities in the power corridor area of ​​Southwest China, perform time alignment, and output a multi-source observation dataset covering the power corridor area of ​​Southwest China.

[0040] The quality control and elimination module is used to perform quality control processing on the multi-source observation dataset, verify the consistency and validity of the multi-source observation dataset, eliminate missing values ​​and outliers, and output a multi-source quality control dataset for the identification of precursors to short-term heavy rainfall in power facilities and risk assessment.

[0041] The annotation module is used to perform annotation operations on the multi-source quality control dataset. Enhanced signals that are characterized by increased humidity, enhanced echo, and sudden increase in wind speed are annotated as enhancement factors, and attenuated or stationary signals that are characterized by decreased humidity, weakened echo, and stationary wind speed are annotated as attenuation and stationary factors. These are used as different fragment units of the precursor factor mosaic and the precursor factor mosaic set is output.

[0042] The clustering and filtering module is based on the co-evolutionary clustering method. It takes the enhancing factor fragments and decaying stationary factor fragments in the precursor factor puzzle set as input, and performs statistics based on their frequency of occurrence in power facility disaster records. It retains the enhancing factor fragments and decaying stationary factor fragments that appear frequently in multiple power facility disaster records as stable precursor fragments and outputs a stable precursor fragment set.

[0043] The splicing evaluation module is used to splice and combine the enhanced factor fragments and attenuated stationary factor fragments in the stable precursor fragment set in pairs to form candidate splicing sequences. The warning lead time and false alarm rate of each candidate splicing sequence in the power facility disaster record are calculated and compared with the preset threshold. The candidate splicing sequences that both the warning lead time and the false alarm rate meet the preset threshold are retained to form a precursor factor dataset for monitoring and risk assessment of short-term heavy precipitation precursors in power facilities.

[0044] The technical effects and advantages of this invention are as follows:

[0045] 1. This scheme constructs a dual-structure precursor factor mosaic set containing enhanced fragments and attenuated static fragments, and introduces a cluster similarity judgment and collaborative recombination mechanism to expand the coverage of precursor factor types. It solves the problem that existing methods only rely on enhanced signals and ignore attenuated static precursors, thereby improving the accuracy of early identification and the forward-looking response to short-term heavy precipitation disasters in complex terrain backgrounds.

[0046] 2. By constructing clusters based on the co-occurrence frequency of fragments and performing similarity determination and collaborative recombination, a stable population structure is formed, thereby improving the structural consistency and characteristic representativeness of precursor factors in the spatiotemporal dimensions.

[0047] 3. By detecting the convergence of fragment similarity changes in cross-cycle iterations, we can identify and lock onto fragment groups with stable frequency responses, thereby improving the structural convergence and population stability representation of the final precursor recognition results.

[0048] 4. A nested computational structure is proposed to express the frequency of fragment cluster occurrence, avoiding reliance on rule settings, realizing mathematical modeling of the fusion relationship of precursor clusters, and improving the model's universality and structural derivability;

[0049] 5. When the similarity is insufficient, a cross-group fragment recombination and recalculation mechanism is implemented to support the adaptive updating of precursor structures under highly dynamic meteorological backgrounds and improve the robustness of the model in non-steady-state weather. Attached Figure Description

[0050] Figure 1 This is a flowchart outlining the method steps of the present invention;

[0051] Figure 2 This is a schematic diagram of the system module structure of the present invention;

[0052] Figure 3 This is a flowchart of the multi-source data normalization process of the present invention;

[0053] Figure 4 This is a flowchart of the quality control screening process of the present invention;

[0054] Figure 5 This is a flowchart of the precursor factor annotation process of the present invention;

[0055] Figure 6 This is a flowchart of the cluster evolution screening process of the present invention;

[0056] Figure 7 This is a flowchart of the precursor splicing and screening process of the present invention. Detailed Implementation

[0057] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0058] Refer to the instruction manual appendix Figure 1-7 An embodiment of the present invention provides a method for identifying precursors of short-term heavy rainfall to power facilities, comprising:

[0059] S1. Obtain raw multi-source observation data related to short-term heavy rainfall in power facilities in the power corridor area of ​​Southwest China, perform time alignment, and output a multi-source observation dataset covering the power corridor area of ​​Southwest China. The raw multi-source observation data includes ground meteorological observation station data, weather radar data, reanalysis data, and power facility disaster records and proprietary monitoring data.

[0060] S2. Perform quality control processing on the multi-source observation dataset to verify the consistency and validity of the multi-source observation dataset, remove its missing values ​​and outliers, and output a multi-source quality control dataset for the identification and risk assessment of short-term heavy rainfall precursors in power facilities.

[0061] S3. Perform annotation operations on the multi-source quality control dataset. The enhanced signals that show increased humidity, enhanced echo, and sudden increase in wind speed are labeled as enhancement factors. The attenuated or stationary signals that show decreased humidity, weakened echo, and stationary wind speed are labeled as attenuation and stationary factors. These are used as different fragment units of the precursor factor mosaic, and the precursor factor mosaic set is output. The precursor factor mosaic set is used for combined modeling of short-term heavy precipitation precursors of power facilities.

[0062] S4. Based on the co-evolutionary clustering method, the enhanced factor fragments and decaying stationary factor fragments in the precursor factor mosaic set are used as inputs. According to their frequency of occurrence in the power facility disaster records, the enhanced factor fragments and decaying stationary factor fragments that appear frequently in multiple power facility disaster records are retained as stable precursor fragments. The stable precursor fragment set is output. The stable precursor fragment set is used to support the identification, modeling and disaster risk assessment of short-term heavy rainfall precursors in power facilities.

[0063] S5. The enhanced factor fragments and attenuated static factor fragments in the stable precursor fragment set are spliced ​​together in pairs to form candidate splicing sequences. The warning lead time and false alarm rate of each candidate splicing sequence in the power facility disaster record are calculated and compared with the preset threshold. The candidate splicing sequences that both the warning lead time and the false alarm rate meet the preset threshold are retained to form a precursor factor dataset for monitoring and risk assessment of short-term heavy precipitation precursors in power facilities.

[0064] In S1, raw multi-source observation data related to short-term heavy rainfall are acquired from power facilities in the power corridor area of ​​Southwest China. The raw multi-source observation data includes data from ground meteorological observation stations, weather radar data, reanalysis data, and records of power facility disasters and proprietary monitoring data.

[0065] The unified timestamp information of the original multi-source observation data is extracted separately, and the time window is divided according to the preset time interval. Interpolation is performed to fill in the record points that do not completely coincide with the window time. The record point refers to a single observation record with a timestamp in the original multi-source observation data.

[0066] When there are observation records in two adjacent windows for data from a certain source, the time difference and value difference of the corresponding observation records are taken. The value of the previous observation record is added to the product of the corresponding proportion and the value difference to obtain the observation value at the corresponding time of the window. Here, the observation record refers to a single observation value with a timestamp generated by each data source at a specific time in the original multi-source observation data.

[0067] When a data source is missing adjacent records, the observation record closest to the window is used as a substitute to obtain the complete set of window observations;

[0068] After completing the interpolation and completion of data from various sources, the observations in each window are merged according to the preset time order, and a multi-source observation dataset aligned to a unified standard time series and covering the power corridor area in Southwest China is output. The multi-source observation dataset is used to construct a benchmark sample system for short-term heavy rainfall precursors of power facilities and serves as a training and verification basis for disaster risk assessment.

[0069] In S2, in the aligned multi-source observation dataset, it is detected whether there are records in each time window that cannot be filled in by interpolation or the nearest value. If there are, the corresponding record is marked as missing; otherwise, it is marked as complete.

[0070] The system detects whether the difference between observations in a multi-source observation dataset within the same spatial neighborhood exceeds a preset threshold. Observations exceeding the threshold are marked as abnormal, while those not exceeding the threshold are marked as normal.

[0071] Numerical validity checks are performed on the observations in the multi-source observation dataset. Observations that exceed a preset reasonable range are marked as invalid, and those that do not are marked as valid.

[0072] Perform a removal operation on observation records marked as missing, abnormal, or invalid, and output a multi-source quality control dataset that meets the conditions of completeness, consistency, and validity.

[0073] In S3, the variation characteristics of humidity, radar echo intensity, and wind speed variables in the multi-source quality control dataset are determined:

[0074] At the same spatial location, the difference between the current humidity value in the current time window and the humidity value in the adjacent previous time window is taken. When the difference is greater than the humidity threshold, it is determined that the humidity is rising; when the difference is less than or equal to the humidity threshold, it is determined that the humidity is falling.

[0075] At the same spatial location, calculate the difference between the representative value of radar echo intensity in the current time window and the mean value of radar echo intensity in multiple adjacent windows. When the difference is greater than the echo threshold, it is determined to be an enhanced echo; when the difference is less than or equal to the echo threshold, it is determined to be a weakened echo.

[0076] At the same spatial location, the ratio of the representative wind speed value of the current time window to the representative wind speed value of the adjacent previous time window is calculated. When the ratio is greater than the wind speed ratio threshold, it is determined that the wind speed has suddenly increased; when the ratio is less than or equal to the wind speed ratio threshold, it is determined that the wind speed has stopped. The representative value refers to the observation value obtained at the corresponding time of the window after time alignment and quality control processing.

[0077] When any one of the following conditions is met—increased humidity, enhanced echo, or sudden increase in wind speed—the corresponding observation record will be marked as an enhancement factor.

[0078] When any one of the following conditions is met—decreased humidity, weakened echo, or stationary wind speed—the corresponding observation record will be marked as the attenuation stationary factor.

[0079] The labeled enhancement factor and decay quiescent factor are used as different fragment units of the precursor factor mosaic, and a set of enhancement factor fragments and a set of decay quiescent factor fragments are established in each time window. The sets of enhancement factor fragments and decay quiescent factor fragments in each time window are combined in sequence to output the precursor factor mosaic set.

[0080] In S4, the enhancement factor fragments and decay quiescent factor fragments in the precursor factor puzzle set are used as the initial enhancement cluster and the initial decay quiescent cluster, respectively, to form the initial cluster population.

[0081] Co-evolutionary iterative operation is performed on the initial population. During the iteration, the number of times any two precursor factor fragments co-occur in the power facility disaster record is counted, and the ratio of the number of co-occurrences to the total number of power facility disaster records containing at least one of the fragments is calculated to obtain the similarity between the two. When the similarity is greater than or equal to a preset threshold, the corresponding fragments are assigned to the same population.

[0082] When the similarity is less than a preset threshold, the precursor factor fragments from different clusters are cross-recombined to generate new candidate pairings, and the similarity is recalculated until the similarity result no longer changes or the number of iterations reaches the preset upper limit.

[0083] It should be noted that in the formula structure involved in this scheme, dimensionless terms can be used as proportional or structural adjustment factors. When combined with quantities with units, they only play a role in numerical scaling and do not introduce new physical dimensions. Therefore, they will not change or confuse the overall unit system. This combination of "dimensionless terms and terms with units" can be understood as a composite structural expression commonly used in mathematical physics modeling. It conforms to the principle of dimensional consistency and has a clear physical interpretation basis.

[0084] Secondly, in the formula structure of this scheme, if multiple variables with different physical units are involved, including but not limited to time, mass or energy variables, their joint appearance is to express the collaborative modeling relationship of multiple physical mechanisms. Each variable can form a unified structure through function mapping, ratio combination or normalization adjustment, with clear units and clear meaning. The overall expression conforms to the principle of dimensional consistency and the conventional formula of engineering modeling.

[0085] In this solution, constants, weights, adjustment factors, threshold parameters, proportional coefficients, etc., are all adjustable control parameters for different application environments. Their values ​​depend on the target equipment configuration, data input characteristics, and performance optimization goals. During the implementation phase, they are set to converge within a reasonable range through model verification, performance constraints, or engineering calibration. Although these parameters do not have a unique preset value, they have clear adjustment logic and calculation paths. They belong to the deterministic setting process in engineering implementation. The purpose of this setting is to ensure that the solution is both universally adaptable and reproducible and operable, without affecting its technical clarity and feasibility.

[0086] S4 also includes determining the corresponding cluster as a stable cluster structure when the similarity between any two fragments within a cluster remains above a preset threshold in two consecutive iterations.

[0087] The frequency of occurrence of each fragment in the stable cluster structure in the power facility disaster record is statistically analyzed, and the ratio of the frequency to the total number of disaster records is calculated to obtain the frequency of occurrence. Fragments with a frequency of occurrence greater than a preset ratio threshold are retained as stable precursor fragments, and a stable precursor fragment set is output.

[0088] Define frequency of occurrence

[0089]

[0090] in Describing the group G qThe set of all precursor factor fragment pairs; (u,v) denotes the cluster G. q Any two precursor factor fragments; This represents the path similarity of the precursor factor fragment pairs in the nth round; express Path similarity of the previous round of precursor factor fragment pairs; This represents the dynamic similarity fusion function, which combines the current similarity with the previous similarity and incorporates stability adjustment. This represents the trend enhancement function, which is used to non-linearly amplify the mean similarity value. G represents the nth round of clustering. q The weight function; dn represents the integration variable, which is used to transform the discrete iterative process of evolution round n into a continuous integral; N q Describing the group G q Maximum number of iteration rounds; (·) represents the frequency normalization function, which is used to convert the structural integral result into the frequency value.

[0091] Furthermore, in In the formula, the path similarity of each pair of precursor factor fragments is... Dynamic fusion function based on similarity After stability adjustment, the mean is taken and then the trend enhancement function is applied. Nonlinear amplification, and weighting function Multiplication is based on the frequency normalization function Obtain the final occurrence frequency

[0092]

[0093] Where exp(·) represents the exponential function; This indicates the difference in similarity between current precursor factor fragment pairs in two consecutive evolutionary rounds. Used to construct stability adjustment factors;

[0094] Furthermore, in In the formula, based on the exponential function exp(·), the similarity difference of the current precursor factor fragment pair in two consecutive rounds of evolution is expressed. With current precursor factor fragments Multiply to obtain the dynamic similarity fusion function.

[0095]

[0096] Where α represents the nonlinear amplification structure constant, and α is used to naturally highlight the expression position of high input value within the function structure without changing the judgment threshold or logical condition; x represents the weighted average of the similarity of the precursor factor fragments in the current round; log(·) represents the logarithmic function;

[0097] Furthermore, in In the formula, the trend enhancement function is obtained by multiplying the weighted mean x of the similarity of the precursor factor fragments in the current round with the nonlinear amplification structure constant α, based on the logarithmic function log(·). (x);

[0098]

[0099] in Describing the group G q The similarity variance in the nth round;

[0100]

[0101] Where p represents the stability enhancement value of the integral structure output; β represents the frequency normalization adjustment parameter, β is used to prevent exceeding the maximum frequency of 1; min(·) represents the selection function, which is used to directly take the smaller of two known inputs as the output. In this formula, it is achieved through... The frequency calculation result is limited to a range of no more than 1, ensuring that the numerical output always remains within a logically reasonable range.

[0102] In S5, the enhancement factor fragments and decaying quiescent factor fragments in the stable precursor fragment set are spliced ​​in pairs to form a candidate splicing sequence.

[0103] For each candidate splicing sequence, extract the disaster start time from the power facility disaster record and the first trigger time of the candidate splicing sequence, calculate the time difference between the two, and obtain the early warning lead time of the candidate splicing sequence;

[0104] For each candidate splicing sequence, the number of times the candidate splicing sequence is triggered in records without disasters is counted, and the total number of times it is triggered in all records is counted. The ratio of the two is used as the false alarm rate of the candidate splicing sequence.

[0105] S5 also includes comparing the warning lead time and false alarm rate of each candidate splicing sequence with preset thresholds, retaining candidate splicing sequences whose warning lead time and false alarm rate both meet the preset thresholds, and forming a precursor factor dataset for monitoring and risk assessment of short-term heavy rainfall precursors in power facilities.

[0106] A system for identifying precursors of short-term heavy rainfall to power facilities includes a data alignment module, a quality control rejection module, a labeling module, a clustering and screening module, and a splicing and evaluation module.

[0107] The data alignment module is used to acquire raw multi-source observation data related to short-term heavy rainfall in power facilities in the power corridor area of ​​Southwest China, perform time alignment, and output a multi-source observation dataset covering the power corridor area of ​​Southwest China. The raw multi-source observation data includes ground meteorological observation station data, weather radar data, reanalysis data, and power facility disaster records and proprietary monitoring data.

[0108] The quality control and elimination module is used to perform quality control processing on the multi-source observation dataset, verify the consistency and validity of the multi-source observation dataset, and eliminate its missing values ​​and outliers, outputting a multi-source quality control dataset for the identification and risk assessment of short-term heavy rainfall precursors in power facilities.

[0109] The annotation module is used to perform annotation operations on the multi-source quality control dataset. Enhanced signals that manifest as increased humidity, enhanced echo, and sudden increase in wind speed are labeled as enhancement factors, and attenuated or stationary signals that manifest as decreased humidity, weakened echo, and stationary wind speed are labeled as attenuation and stationary factors. These are then used as different fragment units of the precursor factor mosaic, and the precursor factor mosaic set is output. The precursor factor mosaic set is used for combined modeling of short-term heavy precipitation precursors to power facilities.

[0110] The clustering and screening module is based on the co-evolutionary clustering method. It takes the enhanced factor fragments and decaying static factor fragments in the precursor factor mosaic set as input, and performs statistical analysis based on their frequency of occurrence in power facility disaster records. It retains the enhanced factor fragments and decaying static factor fragments that appear frequently in multiple power facility disaster records as stable precursor fragments, and outputs a stable precursor fragment set. The stable precursor fragment set is used to support the identification, modeling and disaster risk assessment of short-term heavy rainfall precursors for power facilities.

[0111] The splicing evaluation module is used to splice and combine the enhanced factor fragments and attenuated stationary factor fragments in the stable precursor fragment set in pairs to form candidate splicing sequences. The warning lead time and false alarm rate of each candidate splicing sequence in the power facility disaster record are calculated and compared with the preset threshold. The candidate splicing sequences that both the warning lead time and the false alarm rate meet the preset threshold are retained to form a precursor factor dataset for monitoring and risk assessment of short-term heavy precipitation precursors in power facilities.

[0112] It should be noted that, including but not limited to: this solution was developed based on an analysis of the technical challenges of accurately identifying precursory fragments related to short-term heavy rainfall in traditional power facility disaster early warning systems. Existing methods often rely on rules such as meteorological factor mutations, structural damage events, or fuzzy semantic annotations for judgment, but lack a deep understanding of the combination structure of fragment factors in the disaster evolution chain. This makes it difficult to uncover high-risk signals with recurrence, group co-occurrence, and structural stability. In particular, when there are discrete descriptions across scenarios, devices, and times in disaster texts, traditional methods have omissions and misjudgments in the identification of high-frequency fragment groups.

[0113] To address the aforementioned issues, the solution constructs an identification path based on fragment group evolution and frequency extraction, integrating precursor factor extraction, initial group generation, similarity-driven recombination iteration, stability determination, and frequency calculation mechanisms to achieve structural identification and quantifiable representation of high-risk fragment groups of short-duration heavy precipitation.

[0114] The implementation path consists of six steps:

[0115] The first step is to use a structured parsing tool to decompose the historical disaster data of power facilities, extract all environmental descriptions, facility status and event semantics associated with short-term heavy rainfall, and refine them into the smallest granular set of precursor factor fragments, which serve as the basic input for the fragment identification process.

[0116] The second step is to construct an initial group population based on the co-occurrence relationship of fragments. The initial group is composed of enhancing factor fragments and the initial group is composed of static or decaying factor fragments. For each pair of fragments, the similarity index is calculated based on the ratio of the number of times they co-occur in the disaster record to the number of times they appear alone. The current group structure is then used to determine whether they are classified into the same group.

[0117] The third step is to perform a cross-recombination operation on fragment pairs with similarity below the threshold, construct candidate fragment combinations, recalculate the similarity after combination, add the new combination to the current group if it meets the similarity criteria, and remove it if it does not meet the criteria. This process is repeated for multiple iterations until the similarity between any two fragments remains stable in two consecutive iterations, and is then marked as a stable fragment group structure.

[0118] The fourth step is to count the actual number of times each stable fragment cluster appears in the power facility disaster sample, and divide it by the total number of samples to calculate the actual frequency of the fragment cluster.

[0119] The fifth step, in order to enhance the dynamic adaptability of frequency expression, is to construct a scoring function for the degree of state change, structural position transfer and fluctuation amplitude among fragments within the fragment group, forming a unified evaluation model for fusion trend smoothness, structural consistency and frequency intensity, and finally to obtain the comprehensive occurrence score of each stable fragment group.

[0120] The sixth step is to select fragment clusters that score above a preset threshold based on the scoring results. These fragment clusters are then output as reproducible and stable precursor fragment clusters, which are used for the construction of subsequent disaster early warning models or as risk warning inputs for actual operation and maintenance systems.

[0121] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for identifying a short-time heavy rain precursor of a power facility, characterized in that, Comprise: S1, obtain the original multi-source observation data related to short-time heavy rain in the power corridor region of Southwest China, perform time alignment, and output the multi-source observation data set covering the power corridor region of Southwest China; S2, perform quality control processing on the multi-source observation data set, test the consistency and effectiveness of the multi-source observation data set, and eliminate missing values and abnormal values, output the multi-source quality control data set; S3, perform labeling operation on the multi-source quality control data set, label the enhancement signal showing humidity rise, echo enhancement and wind speed sudden increase as enhancement factor, label the attenuation or static signal showing humidity drop, echo attenuation and wind speed static as attenuation static factor, and respectively as different fragment units of precursor factor jigsaw, output the precursor factor jigsaw set; S4, perform labeling operation on the multi-source quality control data set, label the enhancement signal showing humidity rise, echo enhancement and wind speed sudden increase as enhancement factor, label the attenuation or static signal showing humidity drop, echo attenuation and wind speed static as attenuation static factor, and respectively as different fragment units of precursor factor jigsaw, output the precursor factor jigsaw set; S5, combine the enhancement factor fragments and attenuation static factor fragments in the stable precursor fragment set two by two to form candidate splicing sequences, calculate the warning lead time and false alarm rate of each candidate splicing sequence in the power facility disaster record, and compare with the preset threshold, retain the candidate splicing sequence that meets the preset threshold, and form the precursor factor data set.

2. The identification method of short-time heavy rain precursor of power facilities according to claim 1, wherein: In S1, the original multi-source observation data related to short-time heavy rain in the power corridor region of Southwest China is obtained, and the original multi-source observation data includes ground meteorological observation station data, weather radar data, reanalysis data, and power facility disaster record and self-monitoring data; The uniform timestamp information of the original multi-source observation data is extracted respectively, and time window division is performed according to the preset time interval, and interpolation is performed on the record points that do not completely coincide with the window time; When there are observation records of a certain source in adjacent two windows, the time difference and value difference of the corresponding observation records are taken, the time interval between the corresponding observation record and the previous observation record is taken as the proportion of the time difference, the value of the previous observation record is added to the product of the corresponding proportion and the value difference, and the observation value at the corresponding time of the window is obtained; When a certain source of data lacks adjacent records, the closest observation record to the window is used to replace it, and the supplemented window observation value set is obtained; After completing the interpolation and supplement of each source data, the observation values in each window are merged according to the preset time sequence, and the multi-source observation data set aligned to the uniform standard time sequence and covering the power corridor region of Southwest China is output.

3. The identification method of short-time heavy rain precursor of power facilities according to claim 2, wherein: In S2, in the aligned multi-source observation data set, it is detected whether there are records that cannot be supplemented by interpolation or nearest value replacement in each time window, if there are, the corresponding records are marked as missing, otherwise they are marked as complete; Detecting whether the difference between the observation values in the multi-source observation data set in the same spatial neighborhood exceeds the preset threshold, marking the observation records exceeding the threshold as abnormal, and otherwise marking them as normal; Performing numerical validity test on the observation values in the multi-source observation data set, marking the observation values exceeding the preset reasonable interval as invalid, and otherwise marking them as valid; Performing elimination operation on the observation records marked as missing, abnormal or invalid, and outputting the multi-source quality control data set meeting the conditions of integrity, consistency and validity.

4. The method according to claim 3, wherein: In S3, in the multi-source quality control data set, the change characteristics of humidity, radar echo intensity and wind speed variables are determined: In the same spatial position, the difference between the humidity representative value of the current time window and the humidity representative value of the adjacent previous time window is taken, and when the difference is greater than the humidity threshold, it is determined that the humidity rises, and when the difference is less than or equal to the humidity threshold, it is determined that the humidity falls; In the same spatial position, the difference between the radar echo intensity representative value in the current time window and the average of the radar echo intensity representative values in the adjacent multiple windows is calculated, and when the difference is greater than the echo threshold, it is determined that the echo is enhanced, and when the difference is less than or equal to the echo threshold, it is determined that the echo is weakened; In the same spatial position, the ratio of the wind speed representative value in the current time window to the wind speed representative value in the adjacent previous time window is calculated, and when the ratio is greater than the wind speed ratio threshold, it is determined that the wind speed suddenly increases, and when the ratio is less than or equal to the wind speed ratio threshold, it is determined that the wind speed is stationary; When any one of the conditions of humidity rising, echo enhancement and wind speed sudden increase is determined, the corresponding observation record is marked as an enhancement factor; When any one of the conditions of humidity falling, echo weakening and wind speed stationary is determined, the corresponding observation record is marked as an attenuation stationary factor; The marked enhancement factor and attenuation stationary factor are respectively taken as different fragment units of the precursor factor jigsaw puzzle, and an enhancement factor fragment set and an attenuation stationary factor fragment set are established in each time window, and the enhancement factor fragment set and the attenuation stationary factor fragment set of each time window are combined in turn, and the precursor factor jigsaw puzzle set is output.

5. The method according to claim 4, wherein: In S4, the enhancement factor fragments and the attenuation stationary factor fragments in the precursor factor jigsaw puzzle set are respectively taken as enhancement initial clusters and attenuation stationary initial clusters, forming an initial cluster population; In the iteration process, the number of common occurrences of any two precursor factor fragments in the power facility disaster record is counted, and the similarity is calculated by taking the ratio of the number of common occurrences to the total number of power facility disaster records containing at least one fragment, and when the similarity is greater than or equal to the preset threshold, the corresponding fragments are divided into the same cluster. When the similarity is less than the preset threshold, cross-recombination is performed on the precursor factor fragments from different clusters to generate new candidate pairing relationships, and the similarity is recalculated until the similarity result no longer changes or the number of iterations reaches the preset upper limit.

6. The method according to claim 5, wherein: S4 further comprises determining that the corresponding cluster is a stable cluster structure when the similarity between any two fragments in the cluster remains higher than the preset threshold in two consecutive iterations; The number of occurrences of each fragment in the stable cluster structure in the power facility disaster record is counted, and a ratio calculation is performed with the total number of disaster records to obtain the occurrence frequency. Fragments with an occurrence frequency greater than a preset proportion threshold are retained as stable precursor fragments, and a stable precursor fragment set is output.

7. The method according to claim 6, wherein: In S5, the enhancement factor fragments in the stable precursor fragment set are spliced with the decay and static factor fragments one by one to form candidate splicing sequences; For each candidate splicing sequence, the disaster starting time in the power facility disaster record and the first trigger time of the candidate splicing sequence are extracted, the time difference between the two is calculated, and the warning lead time of the candidate splicing sequence is obtained; For each candidate splicing sequence, the number of triggers in the candidate splicing sequence in the non-disaster record is counted, and the total number of triggers in all records is counted. The ratio of the two is taken as the false alarm rate of the candidate splicing sequence.

8. The method according to claim 7, wherein: S5 further comprises comparing the warning lead time and the false alarm rate of each candidate splicing sequence with the preset threshold, respectively, retaining the candidate splicing sequence that meets the preset threshold for both the warning lead time and the false alarm rate, and forming a precursor factor data set for power facility short-time heavy rain precursor monitoring and risk assessment.

9. A system for identifying short-time heavy rain precursors of power facilities, comprising a method for identifying short-time heavy rain precursors of power facilities according to claim 8, comprising a data alignment module, a quality control rejection module, a labeling module, a cluster screening module, and a splicing evaluation module, wherein: The data alignment module is used to obtain original multi-source observation data related to short-time heavy rain in the power corridor region of the southwest region, perform time alignment, and output a multi-source observation data set covering the power corridor region of the southwest region; The quality control rejection module is used to perform quality control processing on the multi-source observation data set, verify the consistency and effectiveness of the multi-source observation data set, and reject missing values and outliers, output a multi-source quality control data set for power facility short-time heavy rain precursor identification and risk assessment; The labeling module is used to perform labeling operations on the multi-source quality control data set, label the enhancement signals that appear as humidity rise, echo enhancement, and wind speed surge as enhancement factors, label the decay or static signals that appear as humidity drop, echo attenuation, and wind speed static as decay and static factors, and output a precursor factor puzzle set. The grouping screening module takes the enhanced factor fragments and the attenuated stationary factor fragments in the precursor factor puzzle set as inputs based on the co-evolution grouping method, and performs statistics according to the frequency of occurrence in the power facility disaster records, retains the enhanced factor fragments and the attenuated stationary factor fragments with high frequency of occurrence in multiple power facility disaster records as stable precursor fragments, and outputs a stable precursor fragment set; The splicing evaluation module is used for splicing and combining the enhanced factor fragments and the attenuated stationary factor fragments in the stable precursor fragment set two by two to form candidate splicing sequences, calculating the early warning lead time and the false alarm rate of each candidate splicing sequence in the power facility disaster records, and comparing with the preset threshold, retaining the candidate splicing sequences with the early warning lead time and the false alarm rate meeting the preset threshold to form a precursor factor data set for monitoring and risk assessment of short-time heavy rain precursors of power facilities.

Images