A method for dynamic assessment and estimation of typhoon disaster for marine ranching

Abstract

This application relates to a method for dynamic assessment and prediction of typhoon disasters for marine ranching. It involves acquiring meteorological element forecast data and optimal tropical cyclone track data during the impact period of a target typhoon, and obtaining historical typhoon damage data and historical meteorological data to form a historical sample set. Based on the forecast data and optimal track data, candidate disaster-causing factors are constructed. The grey relational degree, linear correlation, and monotonic relationship between the candidate disaster-causing factors and historical disaster losses are calculated. After comprehensive evaluation, key disaster-causing factors are selected. The weights of key disaster-causing factors are determined through pairwise comparisons, and the dimensionality of the key disaster-causing factors and their weights is reduced to obtain low-dimensional features. The meteorological element forecast data is matched to marine ranching raster cells to generate rasterized input data. Based on the weights and low-dimensional features, the disaster risk value of each raster cell is calculated, forming a target typhoon disaster risk distribution field with configurable spatial resolution. The data is then visualized and updated synchronously with the forecast data.

Description

Technical Field

[0001] This application relates to the field of typhoon disaster assessment technology, and in particular to a method for dynamic assessment and prediction of typhoon disasters for marine ranches. Background Technology

[0002] Typhoons, as high-energy extreme weather systems forming over tropical oceans, often bring destructive processes such as strong winds, torrential rains, and storm surges when they approach or pass through sea areas, posing a serious threat to marine ranches. Marine ranches, as typical integrated marine production and ecological utilization systems, have widely distributed facilities, scattered operations, and limited disaster resistance capabilities, making them highly sensitive to weather changes. Once affected by typhoons, they are prone to damage to aquaculture facilities, failure of anchoring systems, and large-scale escape of farmed organisms, resulting in significant economic losses. Therefore, conducting typhoon disaster risk assessments and dynamic forecasts is an important technical requirement for ensuring the safe production of marine ranches and improving disaster prevention and control capabilities.

[0003] Currently, existing typhoon disaster assessment research mainly focuses on several key areas, such as utilizing advanced technologies like Geographic Information Systems (GIS) and remote sensing to assess typhoon disaster risks through GIS, open data, and fuzzy comprehensive assessment theory; or proposing more comprehensive disaster system concepts to establish corresponding fuzzy risk assessment models. With the rise of big data technology, its application in disaster risk assessment is constantly expanding. By integrating historical meteorological data, geographic information, and socioeconomic data, researchers utilize big data analytics algorithms to achieve efficient data processing and analysis. Furthermore, the development of artificial intelligence has further promoted the refinement and precision of disaster assessment methods. For example, methods such as random forests, combined weighting, K-means clustering, and deep learning are gradually being introduced into the construction of typhoon disaster level assessment models.

[0004] While disaster risk assessment technology has made significant progress in typhoon prevention and control, existing research largely focuses on regional assessments at the provincial or larger scales, resulting in relatively coarse spatial resolution and difficulty in depicting the characteristics of disaster changes at local scales. For operational scenarios such as marine ranching, which are highly sensitive to meteorological disturbances and have small spatial scales, it is difficult to obtain sufficiently detailed risk information from existing medium- and large-scale assessment methods. Furthermore, existing methods generally lack a dynamic risk prediction framework that can be updated in real time as the typhoon evolves, limiting the timeliness of disaster response. Summary of the Invention

[0005] To at least partially overcome the problem in related technologies that the spatial resolution of typhoon disaster assessment in operational scenarios that are highly sensitive to meteorological disturbances and have a small spatial scale is relatively coarse and difficult to characterize the disaster change characteristics at the local scale, this application provides a dynamic assessment and prediction method for typhoon disasters for marine ranches.

[0006] The proposed solution is as follows:

[0007] A method for dynamic assessment and prediction of typhoon disasters for marine ranching includes: Acquire multi-source meteorological data during the period of impact of the target typhoon. The multi-source meteorological data should include at least meteorological element forecast data and optimal tropical cyclone track data. Obtain disaster loss data and historical meteorological data during the impact of historical typhoons, and associate the disaster loss data with the historical meteorological data according to the spatiotemporal matching rules to form a historical sample set; A set of candidate disaster-causing factors was constructed based on meteorological element forecast data and optimal tropical cyclone track data; The grey relational degree, linear correlation and monotonic relationship between each candidate disaster-causing factor and historical disaster losses were calculated based on the historical sample set. Based on the calculation results, a comprehensive evaluation of the candidate disaster-causing factors is conducted, and key disaster-causing factors are selected based on the evaluation results. The key disaster-causing factors are compared pairwise to determine the weight of each key disaster-causing factor. The key disaster-causing factors and their weights are subjected to dimensionality reduction processing to generate low-dimensional features of the key disaster-causing factors. Meteorological element forecast data is matched to the raster cells of the marine ranch to generate rasterized input data; Based on the rasterized input data, the disaster risk value of each raster unit during the impact of the target typhoon is calculated based on the weights and low-dimensional features of each key disaster-causing factor, forming a target typhoon disaster risk distribution field with configurable spatial resolution. The target typhoon disaster risk field is visualized and updated synchronously with meteorological element forecast data.

[0008] Preferably, the multi-source meteorological data further includes meteorological reanalysis data; The meteorological element forecast data is gridded forecast data, including at least one or more of the following: wind speed, wind direction, precipitation, air pressure, and temperature; the optimal track data of the tropical cyclone includes at least one or more of the following: typhoon center location, typhoon intensity characterization information, movement speed, and radius of the strong wind circle; the meteorological reanalysis data includes at least one or more of the following: wind speed, wind direction, precipitation, air pressure, and temperature. The candidate disaster-causing factors set includes at least one or more of the following: typhoon impact duration factor, distance from typhoon center to marine ranch factor, maximum wind speed level factor, average wind speed level factor, average precipitation factor, and total precipitation factor; The method further includes: Within a preset historical time window, the meteorological reanalysis data and the meteorological element forecast data are aligned at the same scale, and the deviation correction amount is obtained by statistically analyzing the difference between the two at their corresponding spatiotemporal locations. The deviation correction amount is then applied to the meteorological element forecast data during the period of influence of the target typhoon to obtain the corrected meteorological element forecast data.

[0009] Preferably, the method further includes: Perform dimensional impact elimination processing on candidate hazard-causing factors in the set, including: For each candidate disaster-causing factor, determine the maximum and minimum values ​​within the historical sample set, and then scale the candidate disaster-causing factors to a preset numerical range according to the ratio of their values ​​to the maximum and minimum values. Alternatively, the mean and standard deviation of each candidate disaster-causing factor can be determined within the historical sample set, and the candidate disaster-causing factors can be standardized according to their values ​​relative to the mean and standard deviation.

[0010] Preferably, the linear correlation and monotonic relationship between each candidate disaster-causing factor and historical disaster losses are calculated based on a historical sample set, including: The Pearson correlation coefficient between each candidate disaster-causing factor and historical disaster losses was calculated as a linear correlation. The Spearman rank correlation coefficient between each candidate disaster-causing factor and historical disaster losses was calculated as a monotonic relationship.

[0011] Preferably, the candidate disaster-causing factors are comprehensively evaluated based on the calculation results, and key disaster-causing factors are selected based on the evaluation results, including: Robustness tests were performed on grey relational analysis, and significance tests were performed on Pearson correlation coefficient and Spearman rank correlation coefficient. After the robustness test of grey relational degree is passed, the candidate disaster-causing factors with grey relational degree not lower than the preset relational threshold and passing at least one significance test are selected as key disaster-causing factors. Alternatively, candidate disaster-causing factors with a gray correlation degree not lower than a preset correlation threshold and that pass two significance tests can be used as key disaster-causing factors.

[0012] Preferably, pairwise comparisons are performed on key disaster-causing factors to determine the weight of each key disaster-causing factor, including: A judgment matrix is ​​constructed based on the pairwise importance comparison results of key disaster-causing factors; the matrix elements of the judgment matrix are used to characterize the relative importance between corresponding two key disaster-causing factors. The maximum eigenvalue and the corresponding eigenvector are calculated based on the judgment matrix, and the eigenvector is normalized. A consistency check is performed on the judgment matrix. When the consistency meets the preset threshold, each component of the normalized feature vector is determined as the weight of the corresponding key disaster-causing factor. Otherwise, the pairwise comparison results are adjusted and recalculated.

[0013] Preferably, the key disaster-causing factors and their weights are subjected to dimensionality reduction processing to generate low-dimensional features of the key disaster-causing factors, including: A high-dimensional feature space composed of the values ​​of key disaster-causing factors is constructed based on the historical sample set, and the values ​​of each key disaster-causing factor are weighted according to the weight of the key disaster-causing factors. Construct a projection evaluation index for evaluating dimensionality reduction results, wherein the projection evaluation index is used to characterize the dispersion and clustering characteristics of the projected samples; The projection evaluation index is optimized by searching in the projection direction space through iterative search or intelligent optimization. The weighted key disaster-causing factor feature vectors are projected along the optimal projection direction, and the projection result is output as the low-dimensional feature of the key disaster-causing factor.

[0014] Preferably, meteorological element forecast data is matched to the raster cells of the marine ranch to generate rasterized input data, including: The meteorological element forecast data is processed for time synchronization and spatial matching, and outlier detection and processing are performed to obtain processed meteorological grid data. Based on the spatial distribution of the grid cells of the marine ranch, the meteorological grid data is mapped to the corresponding grid cells through interpolation or neighborhood selection, and the rasterized input data is generated by organizing it according to a uniform time step.

[0015] Preferably, based on the rasterized input data, the disaster risk value of each raster cell during the impact of the target typhoon is calculated according to the weights and low-dimensional features of each key disaster-causing factor, including: Based on the rasterized input data, extract the key disaster-causing factor values ​​corresponding to each raster unit during the impact of the target typhoon, and construct the key disaster-causing factor vector for each raster unit; The vector of key disaster-causing factors is weighted according to the weight of each key disaster-causing factor, and the weighted vector of key disaster-causing factors is mapped to the low-dimensional feature space of the key disaster-causing factors to obtain a comprehensive disaster-causing index for characterizing the comprehensive disaster-causing intensity of typhoons. Based on the disaster loss data of the historical sample set, a vulnerability characterization is constructed. The vulnerability characterization includes at least: performing dimensionless processing on the disaster loss data according to the loss elements, and summarizing it in the dimension of historical typhoon samples to obtain the average vulnerability of the grid cells. The comprehensive disaster-causing index and the average vulnerability are fused and calculated according to a preset fusion rule to obtain the disaster risk value of each grid cell. The disaster risk values ​​are then combined according to the spatial location of the grid cells to form a target typhoon disaster risk distribution field with configurable spatial resolution. When the meteorological element forecast data is updated, the rasterized input data is updated and the disaster risk value and the target typhoon disaster risk distribution field are updated simultaneously.

[0016] Preferably, the target typhoon disaster risk field is visualized and output, including: A risk level classification map is generated based on the disaster risk value of each grid cell according to a preset classification rule; Based on a set of grid cells whose disaster risk value is not lower than a preset threshold, the risk impact range is extracted and an impact range map is generated. The disaster risk value is rendered in a time series based on the continuous time step during the impact of the target typhoon, generating a dynamic risk field that changes over time.

[0017] The technical solution provided in this application may include the following beneficial effects: This technical solution integrates multi-source meteorological data during the impact of a target typhoon and performs high-resolution gridding processing. It matches meteorological element forecast data with optimal tropical cyclone path data to marine ranch grid cells, constructing gridded input data and forming a disaster risk distribution field with configurable spatial resolution. This allows for accurate identification of the typhoon's impact characteristics on small-scale marine ranches, with significantly higher assessment precision than traditional regional-scale assessment models. Simultaneously, based on historical sample sets, it conducts hierarchical diagnostic analysis of candidate disaster-causing factors using grey relational analysis, linear correlation, and monotonic relationships. Combined with pairwise comparisons of key disaster-causing factors to determine weights, it forms a structured and quantifiable disaster-causing factor system, clarifying the relative importance of different meteorological factors and providing good physical interpretability for risk calculation results. Furthermore, by dimensionality reduction of key disaster-causing factors and their weights to generate low-dimensional features, it achieves information extraction and noise reduction. Disaster risk values ​​and visualized risk fields are updated synchronously when meteorological element forecast data is updated, effectively reflecting the rapid changes of typhoons, improving the timeliness and stability of disaster risk prediction, and providing reliable spatial decision support for disaster prevention, mitigation, and production scheduling in marine ranches.

[0018] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0019] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0020] Figure 1 This is a flowchart illustrating a method for dynamic assessment and prediction of typhoon disasters for marine ranches, provided in one embodiment of this application. Detailed Implementation

[0021] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0022] Example 1 Traditional typhoon disaster assessment studies mostly focus on regional assessments at the provincial or larger scale, resulting in relatively coarse spatial resolution. This makes it difficult to characterize the disaster changes at local scales, such as marine ranches, leading to insufficiently detailed risk information. Furthermore, existing methods generally lack a dynamic risk prediction framework that can be updated in real time as the typhoon evolves, limiting the timeliness of disaster response.

[0023] This application proposes a dynamic assessment and prediction method for typhoon disasters in marine ranching, referring to... Figure 1 ,include: S1. Obtain multi-source meteorological data during the period of impact of the target typhoon. The multi-source meteorological data shall include at least meteorological element forecast data and tropical cyclone optimal track data. S2. Obtain disaster loss data and historical meteorological data during the impact of historical typhoons, and associate the disaster loss data with the historical meteorological data according to the spatiotemporal matching rules to form a historical sample set; S3. Construct a set of candidate disaster-causing factors based on meteorological element forecast data and optimal tropical cyclone path data; S4. Calculate the grey relational degree, linear correlation and monotonic relationship between each candidate disaster-causing factor and historical disaster losses based on the historical sample set; S5. Based on the calculation results, conduct a comprehensive evaluation of the candidate disaster-causing factors, and select the key disaster-causing factors based on the evaluation results; S6. Perform pairwise comparisons on key disaster-causing factors to determine the weight of each key disaster-causing factor. S7. Perform dimensionality reduction processing on key disaster-causing factors and their weights to generate low-dimensional features of key disaster-causing factors. S8. Match meteorological element forecast data to the raster cells of the marine ranch to generate rasterized input data; S9. Based on the rasterized input data, calculate the disaster risk value of each raster unit during the period of impact of the target typhoon based on the weight and low-dimensional features of each key disaster-causing factor, and form a target typhoon disaster risk distribution field with configurable spatial resolution. S10. Visualize the target typhoon disaster risk field and update it synchronously with meteorological element forecast data.

[0024] For ease of understanding, the following explains some key terms in this embodiment: Multi-source meteorological data refers to a collection of data acquired from different sources to describe the meteorological conditions of typhoons and their affected areas. This data collection may include meteorological element forecast data output from numerical weather prediction models, actual typhoon observation data, satellite remote sensing data, and meteorological reanalysis data, aiming to provide comprehensive and accurate meteorological information.

[0025] Meteorological element forecast data refers to data that uses numerical weather prediction models to predict one or more meteorological elements (such as wind speed, wind direction, precipitation, air pressure, and temperature) for a specific region over a future period. This data is usually provided in a gridded format and forms the basis for typhoon disaster forecasting.

[0026] Optimal track data for tropical cyclones refers to authoritative data released by meteorological agencies after comprehensive analysis and correction. This data describes key characteristics of tropical cyclones (such as typhoons), including their historical movement paths, center locations, intensity, speed, and the radius of their strong wind circle. It includes one or more of the following: typhoon center location, typhoon intensity information, speed, and radius of the strong wind circle. This data reflects the actual evolution of typhoons and is of great significance for historical disaster analysis and model training.

[0027] A historical sample set refers to a dataset created by collecting disaster loss data and corresponding historical meteorological data during historical typhoon impacts, and then integrating them according to specific spatiotemporal matching rules. This sample set is used to reveal the intrinsic relationship between meteorological factors and disaster losses, and serves as the basis for constructing assessment models and screening disaster-causing factors.

[0028] The candidate hazard-causing factor set refers to a series of physical quantities or indicators that may lead to disasters in marine ranching, extracted or calculated from meteorological element forecast data and optimal tropical cyclone track data. These factors are considered potential hazard-causing factors and require further analysis to determine their correlation with disaster losses. The candidate hazard-causing factor set includes at least one or more of the following: typhoon impact duration factor, typhoon center to marine ranching distance factor, maximum wind speed level factor, average wind speed level factor, average precipitation factor, and total precipitation factor.

[0029] Key disaster-causing factors refer to those factors that, after comprehensive evaluation and screening of a set of candidate disaster-causing factors, are identified as having a significant impact on typhoon disaster losses in marine ranches. These factors are given higher importance in disaster risk assessment models.

[0030] Low-dimensional features refer to simplified feature vectors that effectively characterize the overall disaster-causing intensity of typhoons, obtained by reducing the dimensionality of high-dimensional key disaster-causing factors and their weights. These feature vectors retain the main information of the original data while reducing the complexity of the model.

[0031] A grid cell refers to dividing a marine ranching area into a series of grid units with uniform size and shape. Each grid cell represents a specific geographical area within the marine ranching area and serves as the basic spatial unit for conducting refined disaster risk assessments.

[0032] Rasterized input data refers to data that is generated by mapping meteorological forecast data and other input information onto various raster cells of a marine ranch through spatial matching and interpolation, and then organizing the data according to a uniform time step. This data provides the basis for calculating the disaster risk value of each raster cell.

[0033] The disaster risk value is a quantitative indicator that calculates the probability and severity of disaster losses suffered by each grid cell during the impact of a target typhoon by comprehensively considering the disaster-causing intensity of the typhoon and the vulnerability of the marine ranch.

[0034] The target typhoon disaster risk distribution field refers to a map that combines the disaster risk values ​​of each grid cell according to their spatial location, forming a visual representation of the spatial distribution of disaster risk in a marine ranching area under the influence of a target typhoon. This distribution field has configurable spatial resolution and can be dynamically updated.

[0035] When implementing this method, the first step is to acquire multi-source meteorological data during the period of impact of the target typhoon. This multi-source meteorological data can include meteorological element forecast data and optimal tropical cyclone track data. For example, meteorological element forecast data can be obtained manually by downloading it from the meteorological department's website or received through a subscription service. Optimal tropical cyclone track data can be manually retrieved from historical databases or obtained through a meteorological information platform. This data provides the foundational information for subsequent disaster assessment and forecasting.

[0036] Subsequently, it is necessary to obtain disaster loss data and historical meteorological data during historical typhoon impact periods. Historical disaster loss data can be obtained from disaster reports from marine ranches, insurance claim records, or expert assessments. Historical meteorological data can be obtained from meteorological observation stations, satellite remote sensing, or historical meteorological databases. After acquisition, simple spatiotemporal matching rules can be used, such as establishing a one-to-one association between historical meteorological data and disaster loss data within the same time period and geographically close locations, thereby forming a historical sample set.

[0037] Furthermore, based on meteorological element forecast data and optimal tropical cyclone track data, a set of candidate disaster-causing factors can be constructed. These factors can be selected based on experience or preliminary analysis, aiming to cover the various impacts that typhoons may cause.

[0038] Based on this, and using historical sample sets, it is necessary to calculate the degree of correlation between each candidate disaster-causing factor and historical disaster losses. Specifically, grey relational analysis can be calculated to measure the geometric similarity between factors and losses; linear correlation can be calculated to assess whether there is a linear relationship between factors and losses; and monotonicity can be calculated to determine whether there is a consistent trend in the changes between factors and losses.

[0039] Subsequently, based on the above calculation results, a comprehensive evaluation of the candidate disaster-causing factors can be conducted. Then, based on the evaluation results, key disaster-causing factors with a significant impact on disaster losses are selected from the set of candidate disaster-causing factors.

[0040] After identifying the key disaster-causing factors, it is necessary to compare these factors pairwise to determine their relative importance in influencing disaster losses, thereby obtaining the weight of each key disaster-causing factor. For example, an expert scoring method can be used, where multiple experts rank or assign values ​​to each pair of key disaster-causing factors based on their importance, and then these scoring results are simply averaged or summed to obtain a preliminary weight allocation.

[0041] To simplify the model and improve computational efficiency, the key disaster-causing factors and their weights can be dimensionality-reduced to generate low-dimensional features of these factors. The high-dimensional data of key disaster-causing factors is then projected into a lower-dimensional space, thereby extracting a few comprehensive features that represent the main information of the original data.

[0042] Next, the meteorological forecast data needs to be matched to the raster cells of the marine ranch to generate rasterized input data. For example, the meteorological forecast grid point data closest to the center of each grid cell can be directly assigned to that grid cell, or a simple average of several adjacent meteorological forecast grid point data points can be performed before assignment. In this way, discrete meteorological forecast data can be converted into rasterized data corresponding to the spatial distribution of the marine ranch.

[0043] Based on rasterized input data, the disaster risk value of each raster cell during the impact of a target typhoon can be calculated according to the weights and low-dimensional features of each key disaster-causing factor. For example, a preliminary risk index can be obtained by simply linearly weighting and summing the values ​​of the key disaster-causing factors of each raster cell with their corresponding weights. Then, combined with the vulnerability information of the marine ranch itself (e.g., facility type, aquaculture density, etc.), the disaster risk value of each raster cell can be calculated using a simple multiplication or addition model. These risk values ​​can be combined according to the spatial location of the raster cells to form a target typhoon disaster risk distribution field with configurable spatial resolution.

[0044] Finally, the target typhoon disaster risk field is visualized and output. Simultaneously, when meteorological forecast data is updated, the above calculation process can be re-executed, and the disaster risk value and risk distribution field can be updated synchronously to provide real-time risk information.

[0045] This technical solution integrates multi-source meteorological data during the impact of a target typhoon and performs high-resolution gridding processing. It matches meteorological element forecast data with optimal tropical cyclone path data to marine ranch grid cells, constructing gridded input data and forming a disaster risk distribution field with configurable spatial resolution. This allows for accurate identification of the typhoon's impact characteristics on small-scale marine ranches, with significantly higher assessment precision than traditional regional-scale assessment models. Simultaneously, based on historical sample sets, it conducts hierarchical diagnostic analysis of candidate disaster-causing factors using grey relational analysis, linear correlation, and monotonic relationships. Combined with pairwise comparisons of key disaster-causing factors to determine weights, it forms a structured and quantifiable disaster-causing factor system, clarifying the relative importance of different meteorological factors and providing good physical interpretability for risk calculation results. Furthermore, by dimensionality reduction of key disaster-causing factors and their weights to generate low-dimensional features, it achieves information extraction and noise reduction. Disaster risk values ​​and visualized risk fields are updated synchronously when meteorological element forecast data is updated, effectively reflecting the rapid changes of typhoons, improving the timeliness and stability of disaster risk prediction, and providing reliable spatial decision support for disaster prevention, mitigation, and production scheduling in marine ranches.

[0046] Example 2 It should be noted that multi-source meteorological data also includes meteorological reanalysis data; The method also includes: Within a preset historical time window, meteorological reanalysis data and meteorological element forecast data are aligned at the same scale, and the deviation correction amount is obtained by statistically analyzing the difference between the two at their corresponding spatiotemporal locations. The deviation correction amount is then applied to the meteorological element forecast data during the period of influence of the target typhoon to obtain the corrected meteorological element forecast data.

[0047] This application introduces meteorological reanalysis data and aligns it with meteorological element forecast data at the same scale and corrects for biases, effectively correcting systematic biases and uncertainties in the meteorological element forecast data. This makes the meteorological input data used to construct the candidate disaster-causing factor set more accurate, thereby improving the accuracy and reliability of subsequent disaster-causing factor screening, weight determination, and final disaster risk value calculation.

[0048] Example 3 It should be noted that the method also includes: Perform dimensional impact elimination processing on candidate hazard-causing factors in the set, including: For each candidate disaster-causing factor, determine the maximum and minimum values ​​within the historical sample set, and scale the candidate disaster-causing factors to a preset numerical range according to the ratio of their values ​​to the maximum and minimum values. Alternatively, the mean and standard deviation of each candidate disaster-causing factor can be determined within the historical sample set, and the candidate disaster-causing factors can be standardized according to their values ​​relative to the mean and standard deviation.

[0049] The elimination of dimensional effects can include one of the following two methods: One approach is to determine the maximum and minimum values ​​of each candidate disaster-causing factor within a historical sample set, and then scale the candidate disaster-causing factors to a preset numerical range according to their values ​​relative to these maximum and minimum values. This processing method, known as Min-Max normalization, identifies the maximum and minimum values ​​of each candidate disaster-causing factor in the historical sample set, and then linearly maps all values ​​of that factor to a preset numerical range, such as [0, 1] or [-1, 1]. Specifically, for any candidate disaster-causing factor, its original value is converted into a new value within the preset range, thereby unifying all factors to the same numerical range and eliminating dimensional differences.

[0050] Another approach is to determine the mean and standard deviation of each candidate disaster-causing factor within the historical sample set, and then standardize the candidate factors according to their values ​​relative to these mean and standard deviation. This process, known as Z-score standardization, calculates the mean and standard deviation of each candidate disaster-causing factor within the historical sample set, and then converts all values ​​of that factor into a dimensionless numerical value. After standardization, the mean of each candidate disaster-causing factor will approach 0, and the standard deviation will approach 1, resulting in a data distribution with similar center and dispersion, thereby eliminating the influence of dimensions and units.

[0051] By performing dimensionality elimination processing on candidate disaster-causing factors in the dataset, whether using Min-Max normalization or Z-score standardization, the influence caused by differences in dimensions and numerical ranges among different disaster-causing factors can be effectively eliminated. This ensures that all factors are on a uniform and comparable scale when subsequently calculating the grey relational degree, linear correlation, and monotonic relationship between each candidate disaster-causing factor and historical disaster losses. This avoids the dominance of certain factors with large numerical ranges in the analysis results, thereby improving the accuracy and fairness of the correlation analysis. Ultimately, this helps to more accurately screen key disaster-causing factors and determine their weights, laying a solid foundation for subsequent disaster risk value calculations, and thus improving the reliability and accuracy of the entire typhoon disaster dynamic assessment and prediction method.

[0052] After performing dimensionality elimination processing on the candidate disaster-causing factors in the set, the grey relational degree, linear correlation and monotonic relationship between each candidate disaster-causing factor and historical disaster losses are calculated based on the historical sample set.

[0053] Preferably, the Pearson correlation coefficient between each candidate disaster-causing factor and historical disaster losses is calculated as a linear correlation. Specifically, the formula for calculating the Pearson correlation coefficient is as follows: ; in, and Indicates the first The values ​​of candidate disaster-causing factors and historical disaster losses in a number of historical sample cases. and Its sample case mean This represents the Pearson correlation coefficient; Calculate the Spearman rank correlation coefficient between each candidate disaster-causing factor and historical disaster losses as a monotonic relationship; Specifically, the formula for calculating the Spearman rank correlation coefficient is as follows: ; in, Indicates the first The difference between the candidate disaster-causing factor values ​​and the historical disaster loss values ​​in a historical sample case, where n represents the number of samples. This represents the Spearman rank correlation coefficient.

[0054] Based on this, a comprehensive evaluation of the candidate disaster-causing factors is conducted according to the calculation results, and key disaster-causing factors are selected based on the evaluation results, including: Robustness tests were performed on grey relational analysis, and significance tests were performed on Pearson correlation coefficient and Spearman rank correlation coefficient. The formula for the significance test of the Pearson correlation coefficient is: ; in, This indicates the significance level of the Pearson correlation coefficient; The significance test formula for Spearman's rank correlation coefficient is as follows: ; in, This indicates the significance level of the Spearman rank correlation coefficient.

[0055] After the robustness test of grey relational degree is passed, the candidate disaster-causing factors with grey relational degree not lower than the preset relational threshold and passing at least one significance test are selected as key disaster-causing factors. Alternatively, candidate disaster-causing factors with a gray correlation degree not lower than a preset correlation threshold and that pass two significance tests can be used as key disaster-causing factors.

[0056] Specifically, the robustness test of grey relational degree aims to evaluate the stability and reliability of the grey relational degree calculation results in order to avoid distortion of the relational degree results due to outliers in the sample data or small changes in the data distribution.

[0057] Meanwhile, the significance test is used to determine whether the calculated Pearson correlation coefficient and Spearman rank correlation coefficient are statistically significant, that is, to determine whether the observed correlation is real and not caused by random sampling error.

[0058] After passing the robustness test of grey relational analysis, this application selects candidate disaster-causing factors with a grey relational degree not lower than a preset correlation threshold and that pass at least one significance test as key disaster-causing factors. This screening criterion combines the systematic advantages of grey relational analysis with the rigor of statistical significance testing, ensuring that there is a sufficiently strong correlation between disaster-causing factors and historical disaster losses, and that this correlation is statistically reliable, excluding cases of accidental or weak correlations.

[0059] Alternatively, to further enhance the rigor of the screening, this application may also select candidate disaster-causing factors that have a grey relational degree not lower than a preset correlation threshold and pass both significance tests as key disaster-causing factors. This more stringent screening condition ensures that there is not only a significant linear relationship between disaster-causing factors and disaster losses, but also a significant monotonic relationship, thus demonstrating statistical significance in different types of correlation measures.

[0060] By introducing robustness tests on grey relational analysis and significance tests on Pearson correlation coefficient and Spearman rank correlation coefficient, this application can effectively identify candidate disaster-causing factors with stable and statistically significant associations with historical disaster losses. This multi-dimensional and multi-level testing mechanism avoids misjudgments that may arise from relying solely on a single correlation metric, such as incorrectly identifying non-critical factors as critical factors due to data noise or randomness, or overlooking truly important disaster-causing factors. By combining grey relational analysis thresholds with at least one or two significance tests, this application can more accurately and reliably screen key disaster-causing factors, ensuring that the selected factors are not only sufficiently strong in terms of correlation but also statistically reliable. This significantly improves the input quality of subsequent disaster-causing factor weight determination and low-dimensional feature generation, resulting in more accurate typhoon disaster risk values ​​for marine ranches and a more instructive disaster risk distribution field, providing stronger technical support for disaster prevention and mitigation decisions in marine ranches.

[0061] Example 4 It should be noted that pairwise comparisons are performed on key disaster-causing factors to determine the weight of each key disaster-causing factor, including: A judgment matrix is ​​constructed based on the pairwise importance comparison results of key disaster-causing factors; the matrix elements of the judgment matrix are used to characterize the relative importance between corresponding two key disaster-causing factors. Calculate the largest eigenvalue and the corresponding eigenvector based on the judgment matrix, and then normalize the eigenvector. A consistency check is performed on the judgment matrix. When the consistency meets the preset threshold, each component of the normalized feature vector is determined as the weight of the corresponding key disaster-causing factor; otherwise, the pairwise comparison results are adjusted and recalculated.

[0062] This embodiment provides a systematic and objective method for determining the weights of key disaster-causing factors. By constructing a judgment matrix and performing pairwise importance comparisons, expert experience or domain knowledge can be transformed into quantitative data, avoiding the arbitrariness of simple subjective assignment. Furthermore, by calculating the largest eigenvalue and its corresponding eigenvector and performing normalization, the relative importance of each disaster-causing factor can be scientifically extracted from these comparisons, ensuring the mathematical rationality of the weights. Crucially, the introduction of a consistency check mechanism effectively identifies and corrects logical contradictions or inconsistencies in the judgment matrix, thereby ensuring the high reliability and stability of the determined weights. When consistency requirements are not met, the comparison results are adjusted and recalculated, forming a feedback loop that ensures the final weights accurately reflect the true impact of each key disaster-causing factor on typhoon disaster losses. This provides a more accurate and reliable input for subsequent disaster risk value calculations, significantly improving the accuracy and scientific rigor of dynamic typhoon disaster assessment and prediction.

[0063] Example 5 It should be noted that the key disaster-causing factors and their weights undergo dimensionality reduction processing to generate low-dimensional features of the key disaster-causing factors, including: A high-dimensional feature space composed of the values ​​of key disaster-causing factors is constructed based on the historical sample set, and the values ​​of each key disaster-causing factor are weighted according to their respective weights. Construct a projection evaluation index to evaluate the dimensionality reduction results. The projection evaluation index is used to characterize the dispersion and clustering characteristics of the projected samples. The projection direction that achieves the optimal projection evaluation index is searched in the projection direction space through iterative search or intelligent optimization. The weighted key disaster-causing factor feature vectors are projected along the optimal projection direction, and the projection result is used as the low-dimensional feature of the key disaster-causing factor.

[0064] Specifically, constructing a high-dimensional feature space aims to integrate the values ​​of multiple key disaster-causing factors for each historical sample into a unified mathematical representation. For each sample in the historical sample set, the values ​​of its corresponding key disaster-causing factors are considered as coordinate components of that sample in the high-dimensional space. For example, if there are N key disaster-causing factors, each historical sample will correspond to a point in N-dimensional space. Weighting the values ​​of each key disaster-causing factor is to fully reflect the relative importance of different key disaster-causing factors in the disaster-causing process when constructing the high-dimensional feature space. These weights are determined based on pairwise comparisons of the key disaster-causing factors in the early stages, reflecting their contribution to disaster losses. The weighting process can be achieved by multiplying the original value of each key disaster-causing factor by its corresponding weight, thereby allowing factors with higher importance to have a greater influence in the feature space.

[0065] Building upon this foundation, constructing projection evaluation metrics is a key quantitative standard for measuring the effectiveness of dimensionality reduction. Its core function is to assess whether, after dimensionality reduction, the data in the lower-dimensional space can still effectively retain important information from the original high-dimensional data, particularly the relative relationships between samples. "Dispersion" typically refers to the degree of dispersion of data points along the projection direction; for example, maximizing the variance of the projected data ensures that the projected data still has good discriminative power. "Clustering characteristics" focus on whether data points can maintain their original clustering structure in the lower-dimensional space; that is, similar sample points remain close together after projection, while dissimilar sample points remain separated. For example, metrics can be designed to measure the compactness of intra-class distances and the dispersivity of inter-class distances. By comprehensively considering these characteristics, this metric can guide the dimensionality reduction process, ensuring that while reducing dimensionality, the intrinsic structure and information of the data are preserved to the greatest extent possible.

[0066] Because the number and complexity of possible projection directions in high-dimensional space are enormous, direct exhaustive search is impractical. Therefore, iterative search or intelligent optimization methods are effective ways to find the optimal projection direction. Iterative search methods typically approach the optimal solution by gradually adjusting the projection direction and using feedback from the projection evaluation index after each adjustment. For example, gradient descent or its variants can be used to search along the direction in which the evaluation index gradient increases. Intelligent optimization methods, such as genetic algorithms and particle swarm optimization, simulate evolution or group behavior in nature to perform a global search in the projection direction space, avoiding getting trapped in local optima and thus more effectively finding the optimal projection direction that maximizes or minimizes the projection evaluation index (depending on the index design). This process ensures that the dimensionality reduction result retains the most effective information from the original data.

[0067] Once the optimal projection direction is determined, projection mapping is the core operation for transforming high-dimensional data into a low-dimensional space. Each weighted key hazard causative factor feature vector (i.e., a point in the high-dimensional space) is projected along this direction. Specifically, this typically involves calculating the dot product of the feature vector and the optimal projection direction vector, resulting in a scalar value or a low-dimensional vector, which represents the new coordinates of the sample in the low-dimensional space. The output projection result is the low-dimensional feature of the key hazard causative factor.

[0068] Example 6 It should be noted that the meteorological element forecast data is matched to the raster cells of the marine ranch to generate rasterized input data, including: The meteorological element forecast data is processed for time synchronization and spatial matching, and outlier detection and processing are performed to obtain processed meteorological grid data. Based on the spatial distribution of the grid cells of the marine ranch, meteorological grid data is mapped to the corresponding grid cells through interpolation or neighborhood selection, and rasterized input data is generated by organizing the data at a uniform time step.

[0069] Through the above technical solution, this application can effectively solve the problem of inconsistency in time and space of meteorological data from different sources or with different time steps, and ensure the basic quality and uniformity of the data.

[0070] Furthermore, through outlier detection and processing, errors or unreasonable values ​​in the data can be identified and corrected in a timely manner, significantly improving the reliability and accuracy of meteorological data and avoiding interference from outlier data with subsequent disaster risk assessment results.

[0071] Based on this, according to the spatial distribution of the grid cells of the marine ranch, the processed meteorological grid data is accurately mapped to the corresponding grid cells by interpolation or neighborhood selection, and organized according to a uniform time step, thereby ensuring that each marine ranch grid cell can obtain accurate, continuous meteorological element forecast values ​​that are highly matched with its own spatial location.

[0072] This provides high-quality and refined input data for subsequent calculation of the disaster risk value of each grid cell based on the weights and low-dimensional features of each key disaster-causing factor, greatly improving the spatial resolution and assessment accuracy of the target typhoon disaster risk distribution field, and making the risk assessment results more instructive and practical.

[0073] Example 7 It should be noted that, based on the rasterized input data, the disaster risk value of each raster cell during the impact of the target typhoon is calculated according to the weights and low-dimensional features of each key disaster-causing factor, including: Based on the rasterized input data, the key disaster-causing factor values ​​of each raster cell during the impact of the target typhoon are extracted, and the key disaster-causing factor vector of each raster cell is constructed. The key disaster-causing factor vectors are weighted according to the weights of each key disaster-causing factor, and the weighted key disaster-causing factor vectors are mapped to the low-dimensional feature space of the key disaster-causing factors to obtain a comprehensive disaster-causing index for characterizing the comprehensive disaster-causing intensity of typhoons. Vulnerability characterization is constructed based on disaster loss data from historical sample sets. Vulnerability characterization includes at least the following: performing dimensionless processing on disaster loss data according to loss elements, and summarizing the data across historical typhoon sample dimensions to obtain the average vulnerability of the raster cells. The comprehensive disaster-causing index and the average vulnerability are fused and calculated according to the preset fusion rules to obtain the disaster risk value of each grid cell. The disaster risk values ​​are then combined according to the spatial location of the grid cells to form a target typhoon disaster risk distribution field with configurable spatial resolution. When updating meteorological element forecast data, the rasterized input data is updated and the disaster risk value and the target typhoon disaster risk distribution field are updated simultaneously.

[0074] Specifically, based on the rasterized input data, the key disaster-causing factor values ​​corresponding to each raster cell during the impact of the target typhoon are extracted, and a key disaster-causing factor vector for each raster cell is constructed, aiming to prepare input data for subsequent risk calculations. This process identifies and extracts data values ​​corresponding to pre-selected key disaster-causing factors from the rasterized input data already matched to the marine ranching raster cells, for each raster cell and the entire duration of the target typhoon's impact. For example, if "maximum wind speed" and "total precipitation" are identified as key disaster-causing factors, the maximum wind speed and total precipitation values ​​for each raster cell during the typhoon's impact are obtained from the rasterized input data. These extracted values ​​are then organized into a vector, where each component of the vector represents the value of that raster cell on a specific key disaster-causing factor, thus forming a localized feature representation reflecting the potential impact of the typhoon.

[0075] The key disaster-causing factor vectors are weighted according to the weights of each key disaster-causing factor, and then mapped to the low-dimensional feature space of the key disaster-causing factors to obtain a comprehensive disaster-causing index for characterizing the overall disaster-causing intensity of a typhoon. The purpose of this step is to integrate multi-dimensional key disaster-causing factors into a single index that comprehensively reflects the potential impact intensity of a typhoon, while considering the relative importance and inherent structure of each factor. Specifically, firstly, the key disaster-causing factor vectors of each grid cell are weighted using previously determined weights (e.g., obtained through methods such as the analytic hierarchy process) to highlight those factors with a greater impact on the disaster. Then, the weighted vectors are projected onto the low-dimensional feature space obtained through pre-dimensionality reduction. This mapping process can capture the main variations and correlation patterns among key disaster-causing factors, and the final projection result is the comprehensive disaster-causing index, which can effectively characterize the overall disaster-causing intensity of a typhoon in a specific grid cell.

[0076] A vulnerability characterization is constructed based on disaster loss data from the historical sample set. This vulnerability characterization includes at least the following steps: performing dimensionless processing on the disaster loss data according to loss factors, and summarizing the data across historical typhoon samples to obtain the average vulnerability of the raster cells. This step aims to quantify the inherent vulnerability of marine ranches (or their components within the raster cells) to typhoon disasters, independent of the typhoon's intensity. Specifically, the disaster loss data from the historical sample set is first processed. Since the disaster loss data may contain multiple loss factors with different dimensions, it is necessary to perform dimensionless processing on these data according to loss factors, such as through normalization (scaling the values ​​to the 0-1 range) or standardization (converting to a distribution with a mean of 0 and a standard deviation of 1) to eliminate dimensional differences and ensure comparability between different loss factors. Then, across the historical typhoon sample dimension, the dimensionless loss data for each raster cell is summarized, for example, by calculating its mean or median. This mean represents the average vulnerability of the raster cell, reflecting its inherent tendency to be damaged in past typhoon events.

[0077] The comprehensive disaster-causing index and the average vulnerability are fused and calculated according to a preset fusion rule to obtain the disaster risk value of each grid cell. The disaster risk values ​​are then combined according to the spatial location of the grid cells to form a target typhoon disaster risk distribution field with configurable spatial resolution.

[0078] This step aims to combine the potential impact intensity of a typhoon (comprehensive disaster index) with the inherent vulnerability of the marine ranch (average vulnerability) to derive a comprehensive disaster risk value for each location. The fusion calculation employs a pre-defined fusion rule, which can be a multiplicative model (e.g., risk = index × vulnerability) or a more complex nonlinear function, depending on the characteristics of the marine ranch and the type of disaster.

[0079] The result of the fusion calculation is the disaster risk value for each grid cell. By arranging and combining these risk values ​​according to their corresponding grid cell spatial locations, a target typhoon disaster risk distribution field is formed. This distribution field can intuitively show the spatial distribution of risks to marine ranches under the influence of typhoons, and its spatial resolution can be configured according to needs.

[0080] When meteorological element forecast data is updated, the rasterized input data is updated, and the disaster risk value and the target typhoon disaster risk distribution field are updated synchronously. This step aims to ensure the real-time and timely nature of disaster risk assessment results. When new meteorological element forecast data (e.g., hourly or daily updates from meteorological models) becomes available, the system immediately updates the rasterized input data generated based on this forecast data. Subsequently, the entire risk calculation process, from extracting key disaster-causing factor values, constructing vectors, weighted mapping, calculating comprehensive disaster-causing indicators, to fusing with average vulnerability to calculate disaster risk values, is re-executed for all affected raster cells. This synchronous update mechanism ensures that the disaster risk value of each raster cell and the entire target typhoon disaster risk distribution field always reflect the latest forecast information, thereby providing dynamic and responsive risk assessment capabilities.

[0081] Example 8 It should be noted that the visualization output of the target typhoon disaster risk field includes: A risk level classification map is generated based on the disaster risk value of each grid cell according to a preset classification rule; Based on a set of grid cells whose disaster risk value is not lower than a preset threshold, the risk impact range is extracted and an impact range map is generated. The disaster risk value is rendered in a time series based on the continuous time step during the impact of the target typhoon, generating a dynamic risk field that changes over time.

[0082] Visualizing the target typhoon disaster risk field aims to transform abstract, numerical disaster risk data into intuitive and easy-to-understand graphical or image formats, enabling users to quickly and accurately obtain information, thereby improving data readability and assisting decision-makers in risk assessment and emergency response. This can be achieved through Geographic Information System (GIS) platforms, professional data visualization software, or custom-developed user interfaces. Output formats can include static images, dynamic videos, or interactive map interfaces.

[0083] When generating a risk level grading map, the system processes the disaster risk values ​​of each grid cell according to preset grading rules. The risk level grading map discretizes continuous disaster risk values ​​into several risk levels, distinguishing them on the map using different colors, textures, or symbols, allowing users to easily identify the risk levels of different areas. Preset grading rules can employ various methods such as equal-interval grading, equal-quantity grading, or custom grading. For example, risk values ​​can be divided into several levels such as "low risk," "medium risk," "high risk," and "extremely high risk," and assigned corresponding colors such as green, yellow, orange, and red for rendering. In practice, the system iterates through all grid cells, determines their risk level based on their calculated disaster risk values, compares them with the preset grading rules, and then draws them on the map using the corresponding visual elements.

[0084] To clearly define the boundaries of areas potentially significantly impacted by typhoon disasters, this application also extracts the risk impact range and generates an impact range map based on a set of raster cells with disaster risk values ​​not lower than a preset threshold. The risk impact range map helps users focus on the areas requiring the most attention, thereby optimizing resource allocation and emergency deployment. First, one or more risk thresholds are set; for example, areas with risk values ​​higher than a specific value are defined as "affected areas." The system filters out all raster cells with disaster risk values ​​reaching or exceeding the threshold, aggregating these raster cells to form one or more connected regions. Then, boundary extraction, smoothing, or buffer analysis can be performed on these regions to generate clear risk impact range boundaries, which are then displayed on the map using highlights, outlines, or fill colors.

[0085] Furthermore, to demonstrate the changing trend of typhoon disaster risk over time, this application performs time-series rendering of disaster risk values ​​based on continuous time steps during the impact period of the target typhoon, generating a dynamic risk field that changes over time. The dynamic risk field can show the generation, development, movement, and dissipation of risks, which is crucial for predicting disaster development trends and formulating dynamic emergency plans. Since meteorological element forecast data is updated over time, disaster risk values ​​are also calculated at different time steps. The system calculates and generates a corresponding disaster risk distribution field for each continuous time step during the impact period of the target typhoon (e.g., every hour or every three hours). Then, these risk distribution fields generated at different time steps are sequentially played or animated to form a dynamic, time-varying risk evolution process. This can be achieved by providing interactive controls such as a timeline slider and play / pause buttons on the visualization interface, allowing users to observe the dynamic changes in the risk field.

[0086] Specific implementation examples: Disaster Risk Assessment and Forecast for Marine Ranches in Beibu Gulf, Guangxi Based on Typhoon Wipha in 2019 We obtained the optimal track data (center location, maximum wind speed, etc.) of Tropical Cyclone Wipha, along with corresponding reanalysis data and numerical forecast data for the corresponding time period, including elements such as surface wind speed, sea level pressure, and precipitation. We then performed time alignment, spatial interpolation, and cleaning on the above data to construct a multi-source meteorological input set suitable for algorithmic analysis.

[0087] Based on a sample database of historical typhoon events, a stratified diagnostic analysis was conducted on six indicators, including wind speed, precipitation, and distance from the typhoon center, to identify the key disaster-causing factors most relevant to the disaster situation. The weight of each factor was determined using the analytic hierarchy process (AHP), and the multidimensional factors were projected into a comprehensive disaster index using the projection pursuit method to characterize the overall disaster-causing intensity of the typhoon.

[0088] By combining the above comprehensive indicators with hourly meteorological data during the impact of Typhoon Wipha, a disaster risk field with a resolution of 0.25° was constructed for the Beibu Gulf. The results show that during the period of 09:00–11:00 UTC, the typhoon center was still far from the pasture area (>80 km), and the real-time output risk level remained at "low to moderate", with a relatively stable risk distribution. As the typhoon center moved into the hinterland of the Beibu Gulf and continued to approach the pasture (<50 km) during the period of 12:00–15:00 UTC, the risk field updated by the model showed a rapid upward trend, and the high-risk area ("severe and above") gradually expanded towards the pasture in the continuous output time series.

[0089] Continuous hourly forecasts demonstrate that this method can simultaneously reflect the real-time evolution of typhoon location, wind field intensity, and precipitation changes. The morphology and location of high-risk areas continuously adjust with the typhoon's path, achieving dynamic characterization and real-time prediction of typhoon impacts. This hourly update mechanism allows for the timely capture of changing trends in disaster risk before the actual impact of a typhoon, providing marine ranching with more timely early warning information.

[0090] It is understood that the same or similar parts in the above embodiments can be referred to each other, and the contents not described in detail in some embodiments can be referred to the same or similar contents in other embodiments.

[0091] It should be noted that in the description of this application, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance. Furthermore, in the description of this application, unless otherwise stated, "a plurality of" means at least two.

[0092] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the function involved, as will be understood by those skilled in the art to which embodiments of this application pertain.

[0093] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0094] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0095] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0096] The storage media mentioned above can be read-only memory, disk, or optical disk, etc.

[0097] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0098] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for dynamic assessment and prediction of typhoon disasters for marine ranches, characterized in that, include: Acquire multi-source meteorological data during the period of impact of the target typhoon. The multi-source meteorological data should include at least meteorological element forecast data and optimal tropical cyclone track data. Obtain disaster loss data and historical meteorological data during the impact of historical typhoons, and associate the disaster loss data with the historical meteorological data according to the spatiotemporal matching rules to form a historical sample set; A set of candidate disaster-causing factors was constructed based on meteorological element forecast data and optimal tropical cyclone track data; The grey relational degree, linear correlation and monotonic relationship between each candidate disaster-causing factor and historical disaster losses were calculated based on the historical sample set. Based on the calculation results, a comprehensive evaluation of the candidate disaster-causing factors is conducted, and key disaster-causing factors are selected based on the evaluation results. The key disaster-causing factors are compared pairwise to determine the weight of each key disaster-causing factor. The key disaster-causing factors and their weights are subjected to dimensionality reduction processing to generate low-dimensional features of the key disaster-causing factors. Meteorological element forecast data is matched to the raster cells of the marine ranch to generate rasterized input data; Based on the rasterized input data, the disaster risk value of each raster unit during the impact of the target typhoon is calculated based on the weights and low-dimensional features of each key disaster-causing factor, forming a target typhoon disaster risk distribution field with configurable spatial resolution. The target typhoon disaster risk field is visualized and updated synchronously with meteorological element forecast data.

2. The method according to claim 1, characterized in that, The multi-source meteorological data also includes meteorological reanalysis data; The meteorological element forecast data is gridded forecast data, including at least one or more of the following: wind speed, wind direction, precipitation, air pressure, and temperature; the optimal track data of the tropical cyclone includes at least one or more of the following: typhoon center location, typhoon intensity characterization information, movement speed, and radius of the strong wind circle; the meteorological reanalysis data includes at least one or more of the following: wind speed, wind direction, precipitation, air pressure, and temperature. The candidate disaster-causing factors set includes at least one or more of the following: typhoon impact duration factor, distance from typhoon center to marine ranch factor, maximum wind speed level factor, average wind speed level factor, average precipitation factor, and total precipitation factor; The method further includes: Within a preset historical time window, the meteorological reanalysis data and the meteorological element forecast data are aligned at the same scale, and the deviation correction amount is obtained by statistically analyzing the difference between the two at their corresponding spatiotemporal locations. The deviation correction amount is then applied to the meteorological element forecast data during the period of influence of the target typhoon to obtain the corrected meteorological element forecast data.

3. The method according to claim 1, characterized in that, The method further includes: Perform dimensional impact elimination processing on candidate hazard-causing factors in the set, including: For each candidate disaster-causing factor, determine the maximum and minimum values ​​within the historical sample set, and then scale the candidate disaster-causing factors to a preset numerical range according to the ratio of their values ​​to the maximum and minimum values. Alternatively, the mean and standard deviation of each candidate disaster-causing factor can be determined within the historical sample set, and the candidate disaster-causing factors can be standardized according to their values ​​relative to the mean and standard deviation.

4. The method according to claim 3, characterized in that, Based on historical sample sets, the linear correlation and monotonic relationship between each candidate disaster-causing factor and historical disaster losses were calculated, including: The Pearson correlation coefficient between each candidate disaster-causing factor and historical disaster losses was calculated as a linear correlation. The Spearman rank correlation coefficient between each candidate disaster-causing factor and historical disaster losses was calculated as a monotonic relationship.

5. The method according to claim 4, characterized in that, Based on the calculation results, a comprehensive evaluation of the candidate disaster-causing factors is conducted, and key disaster-causing factors are selected based on the evaluation results, including: Robustness tests were performed on grey relational analysis, and significance tests were performed on Pearson correlation coefficient and Spearman rank correlation coefficient. After the robustness test of grey relational degree is passed, the candidate disaster-causing factors with grey relational degree not lower than the preset relational threshold and passing at least one significance test are selected as key disaster-causing factors. Alternatively, candidate disaster-causing factors with a gray correlation degree not lower than a preset correlation threshold and that pass two significance tests can be used as key disaster-causing factors.

6. The method according to claim 1, characterized in that, The key disaster-causing factors are compared pairwise to determine the weight of each key disaster-causing factor, including: A judgment matrix is ​​constructed based on the pairwise importance comparison results of key disaster-causing factors; the matrix elements of the judgment matrix are used to characterize the relative importance between corresponding two key disaster-causing factors. The maximum eigenvalue and the corresponding eigenvector are calculated based on the judgment matrix, and the eigenvector is normalized. A consistency check is performed on the judgment matrix. When the consistency meets the preset threshold, each component of the normalized feature vector is determined as the weight of the corresponding key disaster-causing factor. Otherwise, the pairwise comparison results are adjusted and recalculated.

7. The method according to claim 1, characterized in that, Dimensionality reduction is performed on key disaster-causing factors and their weights to generate low-dimensional features of the key disaster-causing factors, including: A high-dimensional feature space composed of the values ​​of key disaster-causing factors is constructed based on the historical sample set, and the values ​​of each key disaster-causing factor are weighted according to the weight of the key disaster-causing factors. Construct a projection evaluation index for evaluating dimensionality reduction results, wherein the projection evaluation index is used to characterize the dispersion and clustering characteristics of the projected samples; The projection evaluation index is optimized by searching in the projection direction space through iterative search or intelligent optimization. The weighted key disaster-causing factor feature vectors are projected along the optimal projection direction, and the projection result is output as the low-dimensional feature of the key disaster-causing factor.

8. The method according to claim 1, characterized in that, Meteorological element forecast data is matched to the raster cells of the marine ranch to generate rasterized input data, including: The meteorological element forecast data is processed for time synchronization and spatial matching, and outlier detection and processing are performed to obtain processed meteorological grid data. Based on the spatial distribution of the grid cells of the marine ranch, the meteorological grid data is mapped to the corresponding grid cells through interpolation or neighborhood selection, and the rasterized input data is generated by organizing it according to a uniform time step.

9. The method according to claim 1, characterized in that, Based on the rasterized input data, the disaster risk value of each raster cell during the impact of the target typhoon is calculated according to the weights and low-dimensional features of each key disaster-causing factor, including: Based on the rasterized input data, extract the key disaster-causing factor values ​​of each raster unit during the period of impact of the target typhoon, and construct the key disaster-causing factor vector of each raster unit; The vector of key disaster-causing factors is weighted according to the weight of each key disaster-causing factor, and the weighted vector of key disaster-causing factors is mapped to the low-dimensional feature space of the key disaster-causing factors to obtain a comprehensive disaster-causing index for characterizing the comprehensive disaster-causing intensity of typhoons. Based on the disaster loss data of the historical sample set, a vulnerability characterization is constructed. The vulnerability characterization includes at least: performing dimensionless processing on the disaster loss data according to the loss elements, and summarizing it in the dimension of historical typhoon samples to obtain the average vulnerability of the grid cells. The comprehensive disaster-causing index and the average vulnerability are fused and calculated according to a preset fusion rule to obtain the disaster risk value of each grid cell. The disaster risk values ​​are then combined according to the spatial location of the grid cells to form a target typhoon disaster risk distribution field with configurable spatial resolution. When the meteorological element forecast data is updated, the rasterized input data is updated and the disaster risk value and the target typhoon disaster risk distribution field are updated simultaneously.

10. The method according to claim 1, characterized in that, Visualize and output the target typhoon disaster risk field, including: A risk level classification map is generated based on the disaster risk value of each grid cell according to a preset classification rule; Based on a set of grid cells whose disaster risk value is not lower than a preset threshold, the risk impact range is extracted and an impact range map is generated. The disaster risk value is rendered in a time series based on the continuous time step during the impact of the target typhoon, generating a dynamic risk field that changes over time.

Images