Construction method and system of simplified model for lightning risk assessment of complex hybrid power grid
By constructing a simplified model for lightning damage risk assessment in complex hybrid power grids, and using coding processing and cluster analysis to extract disaster-causing factors, the problem of insufficient lightning damage risk assessment in the power grid is solved, the evaluation efficiency and accuracy are improved, and the calculation cost is reduced.
Patent Information
- Application Number
- CN202311788814.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-12-25
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2043-12-25
AI Technical Summary
The lack of effective lightning damage risk assessment methods in the prior art makes it difficult to avoid lightning damage risks during power grid operation, and there are safety hazards and property losses.
The construction method of a simplified model for lightning damage risk assessment of complex hybrid power grids is adopted. By collecting line lightning fault feature vectors, coding processing, correlation analysis, regression model training and clustering analysis, disaster-causing factors are extracted, and used to characterize lightning damage are constructed. Mathematical model of risk.
It improves the efficiency and effectiveness of lightning damage risk assessment, simplifies the lightning damage risk assessment model in different typical scenarios, reduces the calculation cost, and improves the credibility and accuracy of the evaluation results.
Smart Images

Figure CN117973180B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of power grids, and in particular relates to a method and system for constructing a simplified model for lightning risk assessment of a complex hybrid power grid. Background Art
[0002] The existing technology lacks methods for assessing lightning damage risks, which results in the inability to effectively avoid lightning damage risks during power grid operation, which can easily cause safety hazards and property losses. Summary of the invention
[0003] The purpose of the present invention is to address the deficiencies of the above-mentioned background technology and to provide a method and system for constructing a simplified model for lightning damage risk assessment in a complex hybrid power grid, which can effectively assess the lightning damage risk.
[0004] The technical solution adopted by the present invention is: a method for constructing a simplified model for lightning risk assessment of a complex hybrid power grid, comprising the following steps:
[0005] Based on the historical data of the power grid to be evaluated, the line lightning fault feature vector is collected and the line lightning fault feature vector is encoded;
[0006] Perform correlation analysis on the line lightning fault feature vectors after coding, and take the line lightning fault feature vectors with correlation rankings ranging from the first to the set rankings as preliminary factors;
[0007] The preliminary factors are used as input variables of the regression model, the regression model is trained, and the rationality of the preliminary factors is verified based on the trained regression model;
[0008] Perform cluster analysis on the line lightning fault feature vector after encoding, and select the disaster-causing factor from the preliminary factors that have passed the rationality verification according to the cluster analysis results;
[0009] Based on the historical data of the power grid to be evaluated, a simplified model is constructed to characterize the mathematical relationship between disaster factors and lightning damage risks.
[0010] In the above technical scheme, the line lightning fault characteristic vector includes: province, voltage level, tripping month, flashover form, number of times on the same tower, lightning current amplitude, lightning protection measures, tower type, tower height, grounding resistance, protection angle, large side shielding distance, small side shielding distance, topography, ground inclination, altitude, ground lightning density, thunderstorm day and number of discharge traces.
[0011] In the above technical solution, the encoding process includes: for variables that do not have sequence characteristics in the line lightning fault feature vector, using a virtual coding method to encode them into virtual variables; directly performing sequence coding processing on the number of discharge traces; assigning digital codes to categorical variables according to categories; using proportional normalization processing for protection angles and ground inclination angles; and using minimum and maximum normalization processing for quantitative variables.
[0012] In the above technical solution, during the correlation analysis, Cramer's V coefficient is used to calculate the correlation between categorical variables; Spearman rank correlation is used to calculate the correlation between ordinal variables and quantitative variables; Somer's D is used to calculate the correlation between ordinal variables; eta coefficient is used to calculate the correlation between categorical variables and quantitative variables; ordinal variables are regarded as categorical variables, and Cramer's V coefficient is used to calculate the correlation between categorical variables and ordinal variables.
[0013] In the above technical scheme, province, tripping month, flashover form, number of times on the same tower, lightning protection measures, tower type, topography and landform are categorical variables; voltage level and number of discharge traces are ordinal variables; lightning current amplitude, tower height, grounding resistance, protection angle, large side span, small side span, ground inclination, altitude, ground flash density and thunderstorm days are quantitative variables.
[0014] In the above technical solution, the regression model takes the number of discharge traces as the dependent variable.
[0015] In the above technical solution, the regression model adopts the Logistic regression model; when the number of discharge traces is less than or equal to the set value, it is marked as category 0, and when the number of discharge traces is greater than the set value, it is marked as category 1.
[0016] In the above technical solution, based on the trained regression model, the significance of each input variable is calculated, and the input variables whose significance is less than the set value are determined to pass the rationality verification.
[0017] In the above technical solution, the classification of the dependent variable in the cluster analysis is determined according to the number of discharge traces.
[0018] In the above technical solution, the F value of the line lightning fault feature vector is obtained by cluster analysis and sorted from large to small, and the line lightning fault feature vector ranked in the range from the first to the set ranking and passed the rationality verification is selected as the disaster factor.
[0019] In the above technical solution, the lightning damage risk is characterized by flashover; the flashover forms include shielding and counterattack.
[0020] In the above technical solution, the mathematical expression of the simplified model is:
[0021]
[0022] Where b is a constant; x n , n=1,2,...,7 are disaster factors; w n is the corresponding weight of the nth hazard factor, and f(X) represents the lightning damage risk.
[0023] In the above technical solution, a simplified model is fitted and calculated based on the historical data of the power grid to be evaluated, and the corresponding weight value of each disaster-causing factor is obtained by solving it.
[0024] The present invention also provides a system for constructing a simplified model for lightning risk assessment of a complex hybrid power grid, which is characterized by comprising a data collection module, a coding processing module, a correlation analysis module, a regression analysis module, a cluster analysis module and a simplified model construction module; wherein,
[0025] The data collection module is used to collect line lightning fault feature vectors based on historical data of the power grid to be evaluated;
[0026] The encoding processing module is used to encode the line lightning fault feature vector;
[0027] The correlation analysis module is used to perform correlation analysis on the line lightning fault feature vector after coding, and take the line lightning fault feature vector with correlation ranking in a set ranking interval as the preliminary factor;
[0028] The regression analysis module is used to use the preliminary factors as input variables of the regression model, train the regression model, and verify the rationality of the preliminary factors based on the trained regression model;
[0029] The cluster analysis module is used to perform cluster analysis on the line lightning fault feature vector after encoding processing, and select the disaster-causing factor from the preliminary factors that have passed the rationality verification according to the cluster analysis results;
[0030] The simplified model building module is used to build a simplified model for characterizing the mathematical relationship between disaster factors and lightning damage risks based on the historical data of the power grid to be evaluated.
[0031] In the above technical solution, the coding processing module uses a dummy coding method to encode variables that do not have sequence characteristics in the line lightning fault feature vector into dummy variables; directly performs sequence coding processing on the number of discharge traces; assigns digital codes to categorical variables according to categories; uses proportional normalization processing for protection angles and ground inclination angles; and uses minimum and maximum normalization processing for quantitative variables.
[0032] In the above technical solution, during the correlation analysis process, the correlation analysis module uses Cramer's V coefficient to calculate the correlation between categorical variables; uses Spearman rank correlation to calculate the correlation between ordinal variables and quantitative variables; uses Somer's D to calculate the correlation between ordinal variables; uses eta coefficient to calculate the correlation between categorical variables and quantitative variables; regards ordinal variables as categorical variables, and uses Cramer's V coefficient to calculate the correlation between categorical variables and ordinal variables.
[0033] In the above technical solution, the regression model in the regression analysis module adopts a Logistic regression model; when the number of discharge traces is less than or equal to a set value, it is marked as category 0, and when the number of discharge traces is greater than the set value, it is marked as category 1.
[0034] In the above technical solution, the regression analysis module calculates the significance of each input variable based on the trained regression model, and determines that the input variables with significance less than the set value pass the rationality verification.
[0035] In the above technical solution, the cluster analysis module obtains the F value of the line lightning fault feature vector through cluster analysis and sorts it from large to small, and selects the line lightning fault feature vector that is ranked between the first and set rankings and passes the rationality verification as the disaster factor.
[0036] In the above technical solution, the mathematical expression of the simplified model in the simplified model construction module is:
[0037]
[0038] Where b is a constant; x n , n=1,2,...,7 are disaster factors; w n is the corresponding weight of the nth hazard factor, and f(X) represents the lightning damage risk.
[0039] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for constructing a simplified model for lightning risk assessment of a complex hybrid power grid as described in the above technical solution.
[0040] The beneficial effects of the present invention are as follows: the present invention extracts lightning disaster factors by performing correlation analysis, regression model analysis and cluster analysis on multiple groups of line lightning fault data, simplifies the lightning damage risk assessment model under different typical scenarios, and improves the efficiency of lightning damage assessment and prevention in the power system.
[0041] Furthermore, the line lightning fault characteristics selected by the present invention can fully characterize the lightning strike status of the line and improve the effectiveness of risk assessment.
[0042] Furthermore, another virtual coding method proposed in the present invention can express plain text data using numbers to provide data support for subsequent correlation analysis, effectively saving computing costs and improving computing accuracy.
[0043] Furthermore, the present invention adopts different methods to calculate the correlation between different types of variables, which fits the characteristics of different variables and effectively improves the credibility of the calculation results.
[0044] Furthermore, the present invention divides the data into categorical variables, ordinal variables and quantitative variables according to the data attributes of different lightning fault characteristics, so as to facilitate the calculation of the correlation between different types of variables and further improve the credibility of the calculation results.
[0045] Furthermore, when measuring lightning damage and its distribution pattern, it is considered that the current manifestation of line lightning tripping is relatively simple, and the data only have records after tripping. If there is no tripping after being struck by lightning, such data is difficult to search. Therefore, the present invention takes the number of discharge traces as the key observation object based on the line patrol and accident analysis reports of on-site operation and maintenance personnel, applies regression analysis and cluster analysis, effectively obtains lightning strike records, and improves the credibility of subsequent data analysis and model application.
[0046] Furthermore, the number of categories of the discharge trace quantity set in the present invention is greater than 2, and a Logistic regression model is used to complete nonlinear and multi-classification tasks.
[0047] Furthermore, the present invention can determine whether the variables involved in the model calculation have contributed to the model through significance. The contributing variables are considered to have passed the rationality verification and can be used as the selection objects of disaster-causing factors, which effectively ensures the validity of the selection of disaster-causing factors and can improve the credibility of the simplified model.
[0048] Furthermore, the present invention determines the importance of factors by the influence of factors on clusters, thereby ensuring the credibility of the selection of disaster-causing factors.
[0049] Furthermore, since the forms of lightning faults in complex hybrid power grids are mainly divided into three types: induced lightning faults, counter-strike overvoltage, and shielding failure overvoltage, and considering that the amount of induced lightning fault data is too small and the characteristics are not obvious, it is difficult to train a credible model. Therefore, the present invention selects shielding failure counter-strike to characterize the lightning damage risk and ensure the effectiveness of the output results of the simplified model.
[0050] Furthermore, the present invention simplifies the complex association mechanism of lightning damage forms into a mathematical model, and can obtain the lightning damage risk with extremely high accuracy according to given parameters, thereby ensuring the accuracy of the evaluation results and saving calculation costs.
[0051] Furthermore, the present invention uses a large amount of real historical data to fit a simplified model, which can effectively improve the credibility of the model and the accuracy of the results. BRIEF DESCRIPTION OF THE DRAWINGS
[0052] Figure 1 It is a schematic diagram of the method flow of the present invention;
[0053] Figure 2 It is a schematic diagram of the correlation analysis results of a specific embodiment. DETAILED DESCRIPTION
[0054] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments to facilitate a clear understanding of the present invention, but they do not constitute a limitation on the present invention.
[0055] like Figure 1 As shown, the present invention provides a method for constructing a simplified model for lightning risk assessment of a complex hybrid power grid, comprising the following steps:
[0056] Based on the historical data of the power grid to be evaluated, the line lightning fault feature vector is collected and the line lightning fault feature vector is encoded;
[0057] Conduct correlation analysis on the line lightning fault feature vectors after coding, and take the line lightning fault feature vectors with correlation ranking in the set ranking interval as preliminary factors;
[0058] The preliminary factors are used as input variables of the regression model, the regression model is trained, and the rationality of the preliminary factors is verified based on the trained regression model;
[0059] Perform cluster analysis on the line lightning fault feature vector after encoding, and select the disaster-causing factor from the preliminary factors that have passed the rationality verification according to the cluster analysis results;
[0060] Based on the historical data of the power grid to be evaluated, a simplified model is constructed to characterize the mathematical relationship between disaster factors and lightning damage risks.
[0061] The present invention also provides a system for constructing a simplified model for lightning risk assessment of a complex hybrid power grid, which is characterized by comprising a data collection module, a coding processing module, a correlation analysis module, a regression analysis module, a cluster analysis module and a simplified model construction module; wherein,
[0062] The data collection module is used to collect line lightning fault feature vectors based on historical data of the power grid to be evaluated;
[0063] The encoding processing module is used to encode the line lightning fault feature vector;
[0064] The correlation analysis module is used to perform correlation analysis on the line lightning fault feature vector after coding, and take the line lightning fault feature vector with correlation ranking in a set ranking interval as the preliminary factor;
[0065] The regression analysis module is used to use the preliminary factors as input variables of the regression model, train the regression model, and verify the rationality of the preliminary factors based on the trained regression model;
[0066] The cluster analysis module is used to perform cluster analysis on the line lightning fault feature vector after encoding processing, and select the disaster-causing factor from the preliminary factors that have passed the rationality verification according to the cluster analysis results;
[0067] The simplified model building module is used to build a simplified model for characterizing the mathematical relationship between disaster factors and lightning damage risks based on the historical data of the power grid to be evaluated.
[0068] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for constructing a simplified model for lightning risk assessment of a complex hybrid power grid as described in the above technical solution.
[0069] The principle of the present invention is further explained below with reference to specific embodiments.
[0070] This specific embodiment proposes a method for establishing a simplified model for transmission line lightning risk assessment, which mainly includes the following steps:
[0071] Step S1: Based on the historical data of the line to be evaluated, count the lightning accidents, collect the line lightning fault feature vectors, and pre-process the data.
[0072] Summarize the lines where lightning accidents occurred, collect the characteristic variables of line lightning faults: province, voltage level, tripping month, flashover form, number of times on the same tower, lightning current amplitude, lightning protection measures, tower type, tower height, ground resistance, protection angle, large side shielding distance, small side shielding distance, topography, ground inclination, altitude, ground lightning density, thunderstorm days and number of discharge traces, and preprocess the data, which includes removing null values, wrong values, normalization, etc.
[0073] The variable data description is shown in Table 1.
[0074] Table 1 Lightning fault characteristics and correlation analysis data description
[0075]
[0076]
[0077] Step S2: Encode the original text data and standardize the encoded data. Since the original collected data is pure text data and cannot be analyzed for correlation, this specific embodiment proposes a text data encoding method. Specifically, for variables that do not have sequence characteristics such as tower type and lightning protection measures, the dummy encoding method DummyEncoding is used to encode them into dummy variables Dummy Variable; the number of discharge traces is directly sequence encoded; other categorical variables are assigned digital codes according to categories; the protection angle and ground inclination angle are normalized using proportional normalization, and the quantitative variables are normalized using min-max normalization.
[0078] The coding process of lightning fault characteristic variables is shown in Table 2.
[0079] Table 2 Lightning fault characteristics and correlation analysis coding table
[0080]
[0081]
[0082] Step S3: Based on historical data, correlation analysis, regression analysis and cluster analysis are performed on the encoded line lightning fault feature vector to extract lightning disaster factors. The specific analysis method flow is as follows:
[0083] (1) First, a correlation analysis was conducted: province, tripping month, flashover form, number of times on the same tower, lightning protection measures, tower type, and topography were categorical variables; voltage level and number of discharge traces were ordinal variables; lightning current amplitude, tower height, ground resistance, protection angle, large side span, small side span, ground inclination, altitude, ground-to-ground lightning density (main discharge), ground-to-ground lightning density (return stroke), and thunderstorm day were quantitative variables.
[0084] The correlation between each two categorical variables was calculated using Cramer's V coefficient.
[0085]
[0086]
[0087] Among them, i is the observed frequency of level i in the contingency table between the variables currently involved in the calculation of Cramer's V coefficient, p i is the expected frequency of level i, n is the total sample size, class_A and class_B are the number of categories of the variables currently involved in the calculation of Cramer's V coefficient, and k = class_A × class_B.
[0088] If the items involved in the calculation are the tripping month and the flashover type, then class_A is the number of categories of the tripping month (1 - 12, 12), and class_B is the number of categories of the flashover type (backflash, shielding failure, 2).
[0089] If the items involved in the calculation are the terrain and the tower type, then class_A is the number of categories of the terrain, and class_B is the number of categories of the tower type.
[0090] Spearman rank correlation is used to calculate the correlation between each ordinal variable and each quantitative variable.
[0091]
[0092] where d j is the difference in the ranking positions between the two variables currently involved in the Spearman calculation in the j-th sample. For example, for the variable voltage level 1 - 5, if the voltage level of a certain piece of data is 4, then the ranking position of its voltage level is 2.
[0093] Regarding the explanation of the ranking positions, for variables A and B, if the j-th sample ranks 3 in variable A and 5 in variable B, then the difference in the ranking positions of the j-th sample between the two variables is 2.
[0094] Somer's D is used to calculate the correlation between every two ordinal variables.
[0095]
[0096] where N C is the number of concordant pairs, N D is the number of discordant pairs, and N T is the number of tied pairs.
[0097] For example: For a pair of numbers (x1, y1) and (x2, y2), x and y represent the ranking positions: x1 represents the position of data 1 in ordinal variable 1, y1 represents the position of data 1 in ordinal variable 2; x2 represents the position of data 2 in ordinal variable 1, and y2 represents the position of data 2 in ordinal variable 2.
[0098] If (x1 > x2 && y1 > y2) || (x1 < x2 && y1 < y2), then (x1, y1) and (x2, y2) are called concordant;
[0099] If (x1 > x2 && y1 < y2) || (x1 < x2 && y1 > y2), then (x1, y1) and (x2, y2) are called discordant;
[0100] If (x1 = x2 && y1 = y2), then (x1, y1) and (x2, y2) are called tied.
[0101] The eta coefficient was used to calculate the correlation between categorical and quantitative variables.
[0102]
[0103] Where h is the number of categories of the categorical variable between the two variables currently involved in the eta coefficient calculation, is the average value of the quantitative variable under each category of the categorical variable, is the mean value of the quantitative variable.
[0104] Since there is no correlation calculation method for categorical variables and ordinal variables, the ordinal variables are regarded as categorical variables, and Cramer's V coefficient is used to calculate the correlation between the two.
[0105] The final calculation results of the correlation analysis are as follows: Figure 2 As shown, the calculation results of all the above correlations are sorted, and the variables with the highest ranking are used as factors with greater impact on lightning damage, and the factors with high correlation can be used as input variables of the regression model. In this specific embodiment, the factors with high correlation include: province, month, tower type, topography, altitude, ground-to-ground lightning density (main discharge), ground-to-ground lightning density (return stroke) and thunderstorm day.
[0106] (2) Then, regression analysis was performed: the dependent variable was set as the number of discharge traces, and the province, month, tower type, topography, altitude, ground-to-ground lightning density (main discharge), ground-to-ground lightning density (return stroke) and thunderstorm day were set as independent variables. The significance level p < 0.05 indicated that at least one of the independent variables included in the regression analysis equation group was valid, which further indicated that the establishment of the entire regression equation was meaningful.
[0107] Given a data set after preprocessing and related analysis S contains the top N correlations in the correlation analysis L variables; for each X n , there is X n ={m 1 ,m 2 ,…,m Q}, X n It includes Q dimensions, corresponding to the variables in the specific embodiment.
[0108] Step 1: Establish the model equation of ordered logistic regression:
[0109]
[0110] where P(Y≤r) is the probability that the dependent variable is less than or equal to r, α r is the intercept of the rth level, β 1 ,β2 ,...,β Q is the characteristic coefficient.
[0111] Step 2: The dependent variable is divided into two categories: 0 and 1, where Y≤r is represented as category 0, and Y>r is represented as 1. That is, when the number of discharge traces is less than or equal to the set value, it is marked as category 0, and when the number of discharge traces is greater than the set value, it is marked as category 1.
[0112] Step 3: For each binary logistic regression model, use the Log-Likelihood function as the loss function for parameter estimation and model training.
[0113]
[0114]
[0115] Where p(Y=1) is the probability that the dependent variable is 1, p n is the predicted probability of the nth sample calculated based on the model and parameter β, y n is the actual category of the nth sample. The parameter β is estimated by minimizing the negative Log-Likelihood function via gradient descent.
[0116] Step 4: Let r = r + 1, and repeat steps 2 and 3 R times, where R is the number of categories of the ordered logistic regression dependent variable minus 1.
[0117] The regression model is trained using the model equation and loss function constructed above. The trained regression model can be used to verify whether the disaster-causing factors extracted by the correlation analysis are reasonable: by calculating the significance of each input variable, the input variables whose significance is less than the set value pass the rationality verification.
[0118] In this specific embodiment, the significance of whether the conductor is damaged and whether the insulator is damaged in the regression analysis is less than 0.05, indicating that these two variables are effective for the regression equation of the model. The final calculation results of the regression analysis are shown in Table 3.
[0119] Table 3 Regression analysis results (parameter estimates)
[0120]
[0121]
[0122] (3) Cluster analysis:
[0123] This specific embodiment uses K-means cluster analysis to cluster 19 attributes including province, voltage level, tripping month, flashover form, number of times on the same tower, lightning current amplitude, lightning protection measures, tower type, tower height, ground resistance, protection angle, large side shielding distance, small side shielding distance, topography, ground inclination, altitude, ground lightning density (main discharge), ground lightning density (return stroke) and thunderstorm day.
[0124] The classified data include: province, voltage level, tripping month, flashover form, number of times on the same tower, lightning protection measures, tower type and topography; the sequenced data include lightning current amplitude, tower height, grounding resistance, protection angle, large side shield distance, small side shield distance, ground inclination, altitude, ground lightning density (main discharge), ground lightning density (return stroke) and thunderstorm days.
[0125] Cluster analysis was performed using SPSS software, and the calculation steps are as follows:
[0126] Given a preprocessed data set D = {X 1 ,X 2 ,…,X N}, D contains N variables; for each X n , there is X n ={m 1 ,m 2 ,…,m Q}, X n It includes Q dimensions, corresponding to the variables in the specific embodiment.
[0127] The first step is to randomly initialize K cluster centers {C 1 ,C 2 ,…,C K},1<K≤N, the specific value of K is determined by the elbow method based on the analysis, C k ,1≤k≤K represents the center of the kth cluster.
[0128] The second step is to calculate the Euclidean distance from each object to each cluster center. The calculation formula is: where dis(X n ,C k ) represents object X n to C k The Euclidean distance, X n,q Represents X n The attribute of the qth dimension in C n,q Represents C n The properties of the qth dimension in .
[0129] The third step is to n , according to dis(X n ,C k ) assign it to C with the smallest Euclidean distancek In the above example, we get the updated K clusters {S 1 ,S 2 ,…,S K}. And re-determine the cluster center C of each cluster k , S k,num Represents S k Cluster X n There are num.
[0130] Step 4: Repeat steps 2 and 3 until C k No change occurs.
[0131] Finally, the variance analysis table obtained through cluster analysis is used to obtain the F value of the line lightning fault feature vector and sort it from large to small, and the line lightning fault feature vector with a high ranking and passed the rationality verification is selected as the disaster factor.
[0132] In cluster analysis, Euclidean distance is used to measure the distance between data and cluster center. Since the dependent variable is the number of discharge traces, there are 5 categories: {0, 1, 2, 3, 4}, the number of clusters is set to 5. The final calculation results of cluster analysis in this specific embodiment are shown in Table 4.
[0133] Table 4 Cluster analysis results (ANOVA table)
[0134]
[0135] Step S4: Based on the disaster-causing factors extracted by correlation analysis, regression analysis and cluster analysis, seven factors are comprehensively extracted to construct a simplified model for lightning damage risk assessment.
[0136] This specific embodiment extracts the disaster-causing factors required for the simplified model based on the correlation analysis results of the correlation analysis variables, the parameter estimation value table in the regression analysis and the variance analysis table in the cluster analysis, according to the high or low F value corresponding to each variable.
[0137] The lightning damage risk assessment is simplified to the determination of the flashover form, specifically, the determination of whether the flashover form is a shielding strike or a counterstrike. The model is simplified into the formula:
[0138]
[0139] Where b is a constant, x n ,n=1,2,...,7 are independent variables, which are the disaster factors extracted by the three correlation analysis methods. n is the corresponding weight of the nth hazard factor, and f(X) represents the lightning damage risk.
[0140] The simplified model is fitted and calculated based on the historical data of the power grid to be evaluated, and the corresponding weight value of each disaster-causing factor is obtained.
[0141] The simplified model was fitted using the LevenBerg-Marquardt algorithm, and the calculation method is as follows.
[0142]
[0143] Where J represents the Jacobian determinant, I represents the identity matrix, λ is the damping factor, and a n are the parameters of the model.
[0144] The loss function uses the residual sum of squares (RSS), which is calculated as follows.
[0145]
[0146] where f i Indicates the actual value, Represents the predicted value.
[0147] In this specific embodiment, 70% of the data is used for training and 30% of the data is used for testing. The final simplified model is:
[0148]
[0149] The simplified model has an accuracy of 96.00% on the training set and 93.53% on the test set.
[0150] The contents not described in detail in this specification belong to the prior art known to professional and technical personnel in this field.
Claims
1. A method for constructing a simplified model for lightning risk assessment of complex hybrid power grids, characterized by: The following steps are involved: Based on the historical data of the power grid to be evaluated, the line lightning fault feature vector is collected and the line lightning fault feature vector is encoded; Perform correlation analysis on the line lightning fault feature vectors after coding, and take the line lightning fault feature vectors with correlation rankings ranging from the first to the set rankings as preliminary factors; The preliminary factors are used as input variables of the regression model, the regression model is trained, and the rationality of the preliminary factors is verified based on the trained regression model; Perform cluster analysis on the line lightning fault feature vector after encoding, and select the disaster-causing factor from the preliminary factors that have passed the rationality verification according to the cluster analysis results; Based on the historical data of the power grid to be evaluated, a simplified model is constructed to characterize the mathematical relationship between the disaster factors and the lightning damage risk. The mathematical expression of the simplified model is: ; Where b is a constant; is the disaster factor; n is the corresponding weight of the nth hazard factor, and f(X) represents the lightning damage risk.
2. A method according to claim 1, characterized in that: The line lightning fault characteristic vector includes: province, voltage level, tripping month, flashover form, number of times on the same tower, lightning current amplitude, lightning protection measures, tower type, tower height, grounding resistance, protection angle, large side shielding distance, small side shielding distance, topography, ground inclination, altitude, ground lightning density, thunderstorm day and number of discharge traces.
3. A method according to claim 1, characterized in that: The encoding process includes: for variables that do not have sequence characteristics in the line lightning fault feature vector, using a dummy encoding method to encode them into dummy variables; directly performing sequence encoding processing on the number of discharge traces; assigning digital codes to categorical variables according to categories; using proportional normalization processing on protection angles and ground inclination angles; and using minimum and maximum normalization processing on quantitative variables.
4. A method according to claim 1, characterized in that: In the process of correlation analysis, Cramer's V coefficient is used to calculate the correlation between categorical variables; Spearman rank correlation is used to calculate the correlation between ordinal variables and quantitative variables; Somer's D is used to calculate the correlation between ordinal variables; eta coefficient is used to calculate the correlation between categorical variables and quantitative variables; ordinal variables are regarded as categorical variables, and Cramer's V coefficient is used to calculate the correlation between categorical variables and ordinal variables.
5. A method according to claim 4, characterized in that: Province, tripping month, flashover form, number of times on the same tower, lightning protection measures, tower type, and topography are categorical variables; voltage level and number of discharge traces are ordinal variables; lightning current amplitude, tower height, grounding resistance, protection angle, large side span, small side span, ground inclination, altitude, ground flash density, and thunderstorm days are quantitative variables.
6. A method according to claim 2, characterized in that: The regression model included the number of discharge traces as the dependent variable.
7. A method according to claim 6, characterized in that: The regression model adopts the Logistic regression model; when the number of discharge traces is less than or equal to the set value, it is marked as category 0, and when the number of discharge traces is greater than the set value, it is marked as category 1.
8. A method according to claim 6, characterized in that: Based on the trained regression model, the significance of each input variable is calculated, and the input variables whose significance is less than the set value are judged to pass the rationality verification.
9. A method according to claim 3, characterized in that: The classification of the dependent variable in the cluster analysis was determined based on the number of discharge traces.
10. A method according to claim 9, characterized in that: The F value of the line lightning fault feature vector is obtained through cluster analysis and sorted from large to small, and the line lightning fault feature vector that is ranked between the first and set rankings and passes the rationality verification is selected as the disaster factor.
11. A method according to claim 2, characterized in that: The lightning damage risk is characterized by flashover forms; the flashover forms include shielding stroke and backstroke.
12. A method according to claim 11, characterized in that: The simplified model is fitted and calculated based on the historical data of the power grid to be evaluated, and the corresponding weight value of each disaster-causing factor is obtained.
13. A system for constructing a simplified model for lightning risk assessment of complex hybrid power grids, characterized by: It includes data collection module, coding processing module, correlation analysis module, regression analysis module, cluster analysis module and simplified model building module; among them, The data collection module is used to collect line lightning fault feature vectors based on historical data of the power grid to be evaluated; The encoding processing module is used to encode the line lightning fault feature vector; The correlation analysis module is used to perform correlation analysis on the line lightning fault feature vector after coding, and take the line lightning fault feature vector with correlation ranking in a set ranking interval as the preliminary factor; The regression analysis module is used to use the preliminary factors as input variables of the regression model, train the regression model, and verify the rationality of the preliminary factors based on the trained regression model; The cluster analysis module is used to perform cluster analysis on the line lightning fault feature vector after encoding processing, and select the disaster-causing factor from the preliminary factors that have passed the rationality verification according to the cluster analysis results; The simplified model building module is used to build a simplified model for characterizing the mathematical relationship between disaster factors and lightning damage risks based on the historical data of the power grid to be evaluated; The mathematical expression of the simplified model is: ; Where b is a constant; is the disaster factor; n is the corresponding weight of the nth hazard factor, and f(X) represents the lightning damage risk.
14. The system for constructing a simplified model for lightning risk assessment of a complex hybrid power grid according to claim 13, characterized in that: The coding processing module uses a dummy coding method to encode variables that do not have sequence characteristics in the line lightning fault feature vector into dummy variables; directly performs sequence coding processing on the number of discharge traces; assigns digital codes to categorical variables according to categories; uses proportional normalization processing on protection angles and ground inclination angles; and uses minimum and maximum normalization processing on quantitative variables.
15. The system for constructing a simplified model for lightning risk assessment of a complex hybrid power grid according to claim 13, characterized in that: In the process of correlation analysis, the correlation analysis module uses Cramer's V coefficient to calculate the correlation between categorical variables; uses Spearman rank correlation to calculate the correlation between ordinal variables and quantitative variables; uses Somer's D to calculate the correlation between ordinal variables; uses eta coefficient to calculate the correlation between categorical variables and quantitative variables; regards ordinal variables as categorical variables, and uses Cramer's V coefficient to calculate the correlation between categorical variables and ordinal variables.
16. The system for constructing a simplified model for lightning risk assessment of a complex hybrid power grid according to claim 13, characterized in that: The regression model in the regression analysis module adopts the Logistic regression model; when the number of discharge traces is less than or equal to the set value, it is marked as category 0, and when the number of discharge traces is greater than the set value, it is marked as category 1.
17. The system for constructing a simplified model for lightning risk assessment of a complex hybrid power grid according to claim 13, characterized in that: In the regression analysis module, the significance of each input variable is calculated based on the trained regression model, and the input variables whose significance is less than the set value are judged to pass the rationality verification.
18. The system for constructing a simplified model for lightning risk assessment of a complex hybrid power grid according to claim 13, characterized in that: The cluster analysis module obtains the F value of the line lightning fault feature vector through cluster analysis and sorts it from large to small, and selects the line lightning fault feature vector that is ranked between the first and set rankings and passes the rationality verification as the disaster factor.
19. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for constructing a simplified model for lightning risk assessment of a complex hybrid power grid according to any one of claims 1 to 12 is implemented.
Citation Information
Patent Citations
Power grid lightning damage risk early warning method
CN112686544A
Urban power distribution network fault prediction method and system in extreme weather
CN117057480A
Multi-dimensional analysis method for tripping risk of whole transmission line due to lightning shielding failure
US20230243883A1
