A method and system for constructing a parametric insurance trigger index for lightning strike risk
By constructing a parameterized insurance trigger index based on lightning monitoring data, the problem of low efficiency in manual loss assessment in existing lightning insurance has been solved. This enables accurate assessment and rapid claims processing of lightning strike risks, improves the timeliness and fairness of insurance services, and reduces operating costs.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NANJING UNIV OF INFORMATION SCI & TECH
- Filing Date
- 2026-06-22
- Publication Date
- 2026-07-24
AI Technical Summary
Existing lightning disaster insurance relies on manual on-site investigations, resulting in slow response, high costs, and results that are easily influenced by subjectivity. This fails to meet the demand for rapid claims settlement and lacks transparent and quantifiable criteria for assessing the severity of disasters.
By constructing a parameterized insurance trigger index based on lightning monitoring data, using data from the Fengyun-4 meteorological satellite to obtain lightning strike characteristics, and employing principal component analysis, random forest method, and linear regression method to calculate the trigger index, a comprehensive lightning strike risk trigger index is constructed in conjunction with disaster data to achieve automatic claims settlement.
It enables objective and quantitative assessment of lightning strike risk, shortens the claims process, improves the timeliness and fairness of insurance services, reduces operating costs, and achieves a technological leap from traditional loss assessment to data-driven assessment, supporting real-time early warning and rapid compensation.
Smart Images
Figure CN122453540A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of machine learning and big data analysis, and in particular relates to a method and system for constructing a parameterized insurance trigger index for lightning strike risk. Background Technology
[0002] Faced with the high frequency and significant losses associated with natural risks like lightning, insurance has become one of the core measures for people to obtain financial compensation. However, in existing lightning disaster insurance businesses, loss assessment mainly relies on manual on-site investigations after the disaster. This method is not only slow to respond and costly in terms of manpower, but also difficult to implement when disasters cause traffic disruptions or regional lockdowns, failing to meet the urgent needs of policyholders for rapid financial compensation, resulting in low market efficiency. Furthermore, the results of manual loss assessments are easily influenced by the subjective experience of the investigators, which can easily lead to disputes regarding the definition of the scope of loss and the assessment of the degree of loss, resulting in lengthy claims processes and a poor customer experience.
[0003] The above limitations reflect the lack of a transparent, quantifiable, and tamper-proof basis for measuring the severity of disasters in existing insurance products, especially lightning insurance products. Summary of the Invention
[0004] The purpose of this invention is to objectively and in real time measure the severity of lightning disasters suffered by a specific area within a specific time period, thereby providing a reliable and efficient standardized basis for the design, pricing, claims triggering, and loss assessment of insurance products, and improving the scale of intelligent and digital development of lightning insurance.
[0005] This invention provides a method for constructing a parameterized insurance trigger index for lightning strike risk, the specific steps of which are as follows: Acquire sequential lightning strike data and derived lightning characteristics of a sample area, wherein the derived lightning characteristics are obtained based on the sequential lightning strike data; the sample area includes insured and uninsured areas. The derived lightning features are filtered to obtain the filtered derived lightning features; Calculate the trigger index of the insured area based on the screened derivative lightning characteristics; A comprehensive lightning strike risk trigger index is constructed based on the aforementioned trigger index.
[0006] Furthermore, the sequence lightning strike data is L2 level lightning group data generated hourly by the lightning imager of Fengyun-4 meteorological satellite A, with a time range of N days before the insurance application date. After preprocessing, a feature analysis database is formed. The preprocessing refers to retaining only lightning strike data with complete information and where the latitude and longitude of the lightning occurrence are located within the sample area. In this invention, the time range is 20 days before the insurance application date. The sequence lightning strike data includes the latitude and longitude coordinates of the lightning group occurrence, the precise time of the lightning group occurrence (accurate to milliseconds), the flash type of the lightning group, the radiation intensity of the lightning group, the number of lightning clusters contained in the lightning group, the spatial coverage area of the lightning group, and the recording quality of individual lightning events.
[0007] Furthermore, the derived lightning characteristics include ten features: the number of lightning clusters within a lightning group, the mean radiation of the lightning group, the standard deviation of the lightning group radiation, the proportion of high-intensity lightning, the proportion of afternoon lightning, the mean longitude, the mean latitude, the standard deviation of longitude, the standard deviation of latitude, and the coefficient of variation of the lightning group coverage area. Among these, the proportion of high-intensity lightning is defined as the proportion of lightning groups whose radiation intensity is higher than the 75th percentile of the region; the proportion of afternoon lightning is defined as the proportion of lightning groups occurring between 12:00 and 18:00 each day out of all lightning groups in the local area.
[0008] Furthermore, the derived lightning features are screened, specifically by: calculating the correlation coefficient between each pair of derived lightning features to form a feature correlation matrix, setting a correlation coefficient threshold, and determining that the two derived lightning features are highly correlated if the absolute value of the correlation coefficient between two derived lightning features is greater than or equal to the correlation coefficient threshold. Calculate the variance of the two highly correlated derived lightning features separately, and delete the derived lightning features with smaller variances.
[0009] Furthermore, the comprehensive lightning strike risk trigger index of each sample area during the benchmark period (such as N days before the insurance purchase or a fixed historical period) is used as the benchmark index, and the benchmark ranking of each sample area is determined. During the insurance period, calculate the current comprehensive lightning strike risk trigger index for each insured area and determine the current ranking of each insured area; If the ranking of the insured region improves by a certain number of places compared to the benchmark ranking, or if the percentage increase of the comprehensive lightning strike risk trigger index compared to the benchmark index reaches a certain threshold, then a claim will be triggered.
[0010] Furthermore, based on the screened derived lightning characteristics, the triggering index of each sample region was calculated using principal component analysis, random forest method and linear regression method respectively; The triggering index for each sample region is calculated using principal component analysis (PCA). Specifically, after standardizing the sequential lightning strike data and derived lightning characteristics, a standardized feature matrix is obtained, and the covariance matrix is calculated. The covariance matrix is then decomposed into eigenvalues and corresponding eigenvectors. The eigenvalues are further decomposed into weight vectors using principal component directions. Multiplying the standardized feature matrix by the weight vectors yields the weights of each eigenvalue under PCA, with the sum of the squares of all weights being 1. After normalizing the eigenvalues, they are multiplied by their corresponding weights and summed to obtain the triggering index. The random forest method is used to calculate the trigger index for each sample region. Specifically, from the filtered derived lightning features, features representing the risk level are selected as the target variable, or the number of lightning clusters contained in a lightning group is used as the default target variable to establish N. tree N decision trees are used, with each tree sampling samples using a bootstrap method. A subset of features is randomly selected when a node splits. The Gini coefficient is used to calculate the feature importance of each tree for N. tree The importance of each decision tree is averaged to obtain the average importance score of each feature in the target variable, which is then used as the weight of each feature in this method. The weights are normalized to the interval [0,1] and sum to 1. The trigger index is obtained by multiplying each feature in the target variable by its corresponding weight and summing the results. In this invention, N... tree Select 100; The linear regression method calculates the triggering index for each sample region. Specifically, it uses the number of lightning clusters in a lightning group as the dependent variable by default, or selects a feature representing the risk level as the dependent variable and the remaining features as independent variables from the filtered derived lightning features. A linear relationship model is established between the dependent and independent variables, and the regression coefficients are used as the weights of each feature. The ordinary least squares method is used to estimate the regression coefficients. The regression coefficients of each independent variable without the intercept term are extracted from the fitting results and used as the weights of the remaining features in the calculation of the trigger index. The trigger index is obtained by normalizing the feature values of the remaining features, multiplying them by the corresponding weights, and summing them.
[0011] Furthermore, a comprehensive lightning strike risk triggering index is constructed based on the aforementioned triggering index. Specifically, this involves: combining relevant disaster data to determine the quantitative relationship between the regional comprehensive lightning strike risk triggering index and the triggering indices obtained by each independent method; if the disaster data does not support obtaining other quantitative relationships, the average of the three independent indices is used, that is, the average of the triggering indices calculated based on principal component analysis, random forest method and linear regression method is taken to obtain the comprehensive lightning strike risk triggering index. In the development and application of lightning strike disaster index insurance products: using the obtained comprehensive lightning strike risk trigger index as a benchmark, relevant information such as historical insurance premium rates, rates of return, and commission rates is collected to form specific insurance and claims plans. After a claim is triggered, the premium is calculated as follows: , The compensation can be calculated using the limited expected compensation method, and the formula is as follows: , in, Let X represent the finite expected compensation, be a random variable of loss, and follow a continuous distribution; its distribution function is: The density function is , u represents the limit of compensation. Calculate compensation below the limit. Calculate compensation exceeding the limit, i.e., calculate according to the limit.
[0012] This invention also provides a parameterized insurance trigger index construction system for lightning strike risk, comprising: The data acquisition and database construction module is used to acquire sequential lightning strike data of the sample area, calculate derived lightning characteristics based on the sequential lightning strike data, and construct a feature analysis database. The feature filtering module is used to filter the derived lightning features in the feature analysis database to obtain the filtered derived lightning features. The trigger index calculation module calculates the trigger index of the insured area based on the screened derivative lightning characteristics, using at least one of the principal component analysis, random forest, and linear regression methods. The comprehensive index construction module constructs a comprehensive lightning strike risk trigger index based on the trigger index.
[0013] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the above-described method for constructing a parameterized insurance trigger index for lightning strike risk.
[0014] A computer storage medium storing a computer program that, when executed by a processor, implements the aforementioned method for constructing a parameterized insurance trigger index for lightning strike risk.
[0015] Compared with the prior art, the present invention has the following advantages: 1. This invention solves the problems of low efficiency and strong subjectivity in existing technologies that rely on manual on-site loss assessment, as well as the technical difficulties of traditional agricultural insurance in accurately assessing risks and quickly settling claims for sudden and localized meteorological disasters such as lightning strikes. By constructing a parameterized insurance trigger index based on lightning monitoring data, this invention achieves an objective quantitative assessment of lightning strike risks, shortening the claims settlement cycle from the traditional months to a few days or even real-time triggering, significantly improving the timeliness and fairness of insurance services. 2. It realizes dynamic insurance index assessment of physical parameters, which changes the traditional static and extensive pricing model of insurance products and achieves the technical effect of refined pricing of lightning strike risk. This enables the scientific formulation of premium levels, making the premiums and risk levels accurately matched, and avoiding the insurance operation risks caused by risk mismatch. 3. By establishing a quantitative relationship model between lightning parameters and disaster losses, the technical problem of establishing a direct correlation between lightning activity intensity and economic losses in existing technologies has been solved; the influence of different lightning parameters on the loss rate of the target object has been revealed, enabling accurate prediction of the degree of loss and automatic compensation when the monitoring data reaches the preset threshold, realizing a technological leap from "loss assessment based on experience" to "data-triggered"; 4. Significantly reduces the operating costs of insurance companies and the waiting costs of policyholders, achieving the technical effect of optimizing the allocation of insurance resources. Under the traditional claims settlement model, each lightning strike requires a significant investment of manpower for on-site investigation and loss assessment, with operating costs accounting for approximately 15%-30% of the premium. This invention employs a parameterized triggering mechanism, enabling claims settlement without on-site investigation, saving the vast majority of the operating costs associated with this traditional model, while allowing policyholders to quickly receive compensation for post-disaster recovery. 5. It realizes dynamic monitoring and early warning response to lightning strike risks, overcoming the limitation of existing insurance products that only respond passively after a disaster. The system of this invention can access lightning monitoring data from meteorological departments in real time. When the monitored radiation intensity or ground flash density approaches the trigger threshold, it can issue an early warning to the insured and prompt them to take temporary protective measures. This realizes the innovation of the insurance service model from "post-disaster compensation" to "pre-disaster prevention", minimizing actual disaster losses. Attached Figure Description
[0016] Figure 1 This is a flowchart illustrating a method for constructing a parameterized insurance trigger index for lightning strike risk.
[0017] Figure 2 It is a related heat map of various derived lightning characteristics drawn based on the collected lightning group data in Example 1. Detailed Implementation
[0018] The technical solution of the present invention will be further described below with reference to the accompanying drawings and tables.
[0019] Example 1:
[0020] like Figure 1 As shown, a method for constructing a parameterized insurance trigger index for lightning strike risk is presented. This method is applied to a lightning strike disaster insurance compensation scheme formulation system and includes the following steps:
[0021] S1. Obtaining lightning strike record data: Based on publicly available or provided data from meteorological stations, collect sequential lightning strike data from sample areas and perform data preprocessing to form a feature analysis database;
[0022] The specific operation process of the trigger index and disaster assessment method for data extraction from China's Fengyun-4 meteorological satellite provided in this embodiment of the invention is as follows.
[0023] First, we obtain the main information about lightning disasters in this embodiment. According to incomplete statistics, a total of 449 lightning disasters occurred nationwide in 2019, of which 8 caused fires or explosions.
[0024] This case study collects hourly L2-level lightning group data generated by the lightning imager on the Fengyun-4A meteorological satellite from May 1st to May 20th, 2019. Specifically, this case study collects the occurrence time of lightning groups, the number of lightning events included in each lightning group, the latitude and longitude range covered by the lightning groups, and the radiation intensity of the lightning groups within the range of 0°~159.07°E and 0°~60.72°N from May 1st to May 20th, 2019.
[0025] This case study preprocesses the collected lightning data. It filters out all data with a quality flag of 0 and complete information, selects several provinces, autonomous regions, and municipalities in China as sample areas, and filters out all data whose lightning occurrence latitude and longitude locations meet the requirements.
[0026] After preprocessing, this case study collected data on 263,172 lightning clusters from 00:02:45 on May 1, 2019 to 23:28:59 on May 20, 2019. The data covers longitudes of 74.67°E to 134.77°E and latitudes of 21.18°N to 53.47°N, with an average radiation intensity of 149.69 μJ / m². 2 / ster.
[0027] S2. Feature Engineering and Lightning Disaster Feature Assessment: Based on the lightning strike data in S1, the derived lightning characteristics of each sample area are calculated, and feature selection is performed in the feature analysis database formed in S1 through feature engineering.
[0028] Based on the original features, derived lightning characteristics of each sample area are calculated, including the number of lightning clusters in a lightning group, the mean radiation of the lightning group, the standard deviation of the lightning group radiation, the proportion of high-intensity lightning, the proportion of afternoon lightning, the mean longitude, the mean latitude, the standard deviation of longitude, the standard deviation of latitude, and the coefficient of variation of the lightning group coverage area. The correlation coefficients between all derived lightning characteristics are calculated, a feature correlation matrix is constructed, and a correlation heatmap of each derived lightning characteristic is plotted, such as... Figure 2 As shown.
[0029] Calculations show that the longitude standard deviation is highly correlated with the latitude standard deviation, with a correlation coefficient of 0.810; the mean of lightning group radiation is highly correlated with the standard deviation of lightning group radiation, with a correlation coefficient of 0.866; the variance of the longitude standard deviation is higher than that of the latitude standard deviation, and the variance of the lightning group radiation standard deviation is higher than that of the latitude standard deviation. Therefore, the features of latitude standard deviation and mean of lightning group radiation are deleted.
[0030] S3. Establish an independent lightning risk triggering index construction model: Calculate the importance of each feature as a weight based on the original and derived lightning features retained in S1 and S2, obtain the triggering index of the sample area under the independent method, and establish a sample area triggering index calculation model.
[0031] In this case, the importance of each feature was calculated as a weight using principal component analysis, random forest method, and linear regression. The results are shown in Tables 1, 2, and 3.
[0032] Table 1. Characteristic weights based on principal component analysis.
[0033]
[0034] Table 2 Feature weights based on random forest method
[0035]
[0036] Table 3 Feature weights based on linear regression method
[0037]
[0038] S4. Construct a comprehensive lightning strike risk triggering index calculation model: Based on the lightning risk triggering index calculated by each method in S3, and referring to the lightning strike and related disaster data collected in S1, determine the quantitative relationship between the comprehensive lightning strike risk triggering index of the target area and the triggering index of each independent method;
[0039] Based on the weights calculated above, the comprehensive lightning strike risk triggering index for each sample area from 00:02:45 on May 1, 2019 to May 20, 2019 was obtained. Taking the top ten sample areas in terms of comprehensive lightning strike risk triggering index as an example, their comprehensive lightning strike risk triggering index is calculated as shown in Table 4.
[0040] Table 4. Comprehensive Lightning Strike Risk Trigger Index of the Top Ten Sample Areas
[0041]
[0042] Tables 1, 2, and 3 show the weights and methods for calculating the trigger index during the insurance period. Table 4 shows the comprehensive lightning strike risk trigger index for each sample area, which serves as the data benchmark for triggering claims. During the insurance period, if the ranking of the comprehensive trigger index for the insured area improves beyond the agreed-upon ranking or the comprehensive trigger index improves by a specified percentage, a claim will be triggered.
[0043] S5. Development and Application of Lightning Strike Disaster Index Insurance Products: Using the comprehensive lightning strike risk trigger index obtained in S4 as a benchmark, collect relevant information such as insurance premium rates, yield rates, and commission rates over the years to form specific insurance and claims solutions.
[0044] This case assumes the insurance company's expected rate of return is 4%. If the compensation amount calculated based on historical loss data using the technical solution is P, then the premium is... .
[0045] By applying this method for constructing a parameterized insurance trigger index for lightning strike risk, this case effectively and timely measures the severity of lightning disasters suffered by different specific sample areas within a specific time period. It also provides an objective and reliable reference for the triggering conditions of parameterized insurance, thereby reducing labor costs and improving the operational efficiency of insurance products.
[0046] Example 2:
[0047] Based on the parameterized insurance trigger index construction method for lightning strike risk in Specific Example 1, this invention also provides a parameterized insurance trigger index construction system for lightning strike risk, comprising:
[0048] The data acquisition and database construction module is used to acquire sequential lightning strike data of the sample area and calculate derived lightning characteristics based on the sequential lightning strike data;
[0049] The feature filtering module is used to filter the derived lightning features to obtain the filtered derived lightning features;
[0050] The trigger index calculation module calculates the trigger index of the insured area based on the screened derivative lightning characteristics, using at least one of the principal component analysis, random forest and linear regression methods in Example 1.
[0051] The comprehensive index construction module constructs a comprehensive lightning strike risk trigger index based on the trigger index.
[0052] Example 3:
[0053] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the parameterized insurance trigger index construction method for lightning strike risk described in Embodiment 1 above.
[0054] Example 4:
[0055] A computer storage medium storing a computer program that, when executed by a processor, implements the parameterized insurance trigger index construction method for lightning strike risk described in Embodiment 1 above.
Claims
1. A method for constructing a parameterized insurance trigger index for lightning strike risk, characterized in that, Includes the following steps: Acquire sequential lightning strike data and derived lightning characteristics of a sample area, wherein the derived lightning characteristics are obtained based on the sequential lightning strike data; the sample area includes insured and uninsured areas. The derived lightning features are filtered to obtain the filtered derived lightning features; Calculate the trigger index of the insured area based on the screened derivative lightning characteristics; A comprehensive lightning strike risk trigger index is constructed based on the aforementioned trigger index.
2. The method for constructing a parameterized insurance trigger index for lightning strike risk according to claim 1, characterized in that, The sequence lightning strike data is L2 level lightning group data generated hourly by the lightning imager of Fengyun-4 meteorological satellite A. The sequence lightning strike data includes the latitude and longitude coordinates of the lightning group, the precise time of the lightning group's occurrence, the flash type of the lightning group, the radiation intensity of the lightning group, the number of lightning clusters contained in the lightning group, the spatial coverage area of the lightning group, and the recording quality of individual lightning events.
3. The method for constructing a parameterized insurance trigger index for lightning strike risk according to claim 1 or 2, characterized in that, The derived lightning characteristics include the number of lightning clusters in the lightning group, the mean radiation of the lightning group, the standard deviation of the lightning group radiation, the proportion of high-intensity lightning, the proportion of afternoon lightning, the mean longitude, the mean latitude, the standard deviation of longitude, the standard deviation of latitude, and the coefficient of variation of the lightning group coverage area.
4. The method for constructing a parameterized insurance trigger index for lightning strike risk according to claim 1, characterized in that, The derivative lightning features are screened by: calculating the correlation coefficient between each pair of derivative lightning features to form a feature correlation matrix, setting a correlation coefficient threshold, and determining that the two derivative lightning features are highly correlated if the absolute value of the correlation coefficient between two derivative lightning features is greater than or equal to the correlation coefficient threshold. Calculate the variance of the two highly correlated derived lightning features separately, and delete the derived lightning features with smaller variances.
5. The method for constructing a parameterized insurance trigger index for lightning strike risk according to claim 1, characterized in that, The comprehensive lightning strike risk triggering index of each sample area in the baseline period is used as the benchmark index, and the benchmark ranking of each sample area is determined. During the insurance period, calculate the current comprehensive lightning strike risk trigger index for each insured area and determine the current ranking of each insured area; If the ranking of the insured region improves by a certain number of places compared to the benchmark ranking, or if the percentage increase of the comprehensive lightning strike risk trigger index compared to the benchmark index reaches a certain threshold, then a claim will be triggered.
6. The method for constructing a parameterized insurance trigger index for lightning strike risk according to claim 1 or 5, characterized in that, Based on the screened derivative lightning characteristics, the triggering index of each sample region was calculated using principal component analysis, random forest method and linear regression method respectively. The triggering index for each sample region is calculated using principal component analysis (PCA). Specifically, after standardizing the sequential lightning strike data and derived lightning characteristics, a standardized feature matrix is obtained, and the covariance matrix is calculated. The covariance matrix is then decomposed into eigenvalues and corresponding eigenvectors. The eigenvalues are further decomposed into weight vectors using principal component directions. Multiplying the standardized feature matrix by the weight vectors yields the weights of each eigenvalue under PCA. The eigenvalues are then normalized, multiplied by their corresponding weights, and summed to obtain the triggering index. The trigger index for each sample region is calculated using the random forest method. Specifically, from the filtered derived lightning features, features representing the risk level are selected as the target variable, and an N... tree Each decision tree uses a bootstrap sampling method to extract samples, and a subset of features is randomly selected when a node splits. The Gini coefficient is used to calculate the feature importance of each tree for N. tree The average importance score of each feature in the target variable is obtained by averaging the importance of each decision tree, and this average score is used as the weight of each feature in the target variable under this method. The trigger index is obtained by multiplying each feature in the target variable by its corresponding weight and summing the results. The trigger index for each sample region is calculated using linear regression. Specifically, from the screened derivative lightning features, a feature representing the risk level is selected as the dependent variable, and the remaining features are independent variables. A linear relationship model between the dependent and independent variables is established, and the regression coefficients of each independent variable are estimated using the least squares method. The regression coefficients of each independent variable without the intercept term are extracted, and the regression coefficients without the intercept term are used as the weights of the remaining features in the calculation of the trigger index. The feature values of the remaining features are normalized, multiplied by the corresponding weights, and summed to obtain the trigger index.
7. The method for constructing a parameterized insurance trigger index for lightning strike risk according to claim 1, characterized in that, A comprehensive lightning strike risk triggering index is constructed based on the aforementioned triggering index. Specifically, the comprehensive lightning strike risk triggering index is obtained by averaging the triggering indices calculated using principal component analysis, random forest, and linear regression methods.
8. A parameterized insurance trigger index construction system for lightning strike risk, characterized in that, include: The data acquisition and database construction module is used to acquire sequential lightning strike data of the sample area, calculate derived lightning characteristics based on the sequential lightning strike data, and construct a feature analysis database. The feature filtering module is used to filter the derived lightning features in the feature analysis database to obtain the filtered derived lightning features. The trigger index calculation module calculates the trigger index of the insured area based on the screened derivative lightning characteristics, using at least one of the principal component analysis, random forest, and linear regression methods. The comprehensive index construction module constructs a comprehensive lightning strike risk trigger index based on the trigger index.
9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements a method for constructing a parameterized insurance trigger index for lightning strike risk, as described in any one of claims 1 to 7.
10. A computer storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements a method for constructing a parameterized insurance trigger index for lightning strike risk, as described in any one of claims 1 to 7.