Method and system for predicting casualties caused by earthquake landslide in consideration of population dynamic characteristics and environmental heterogeneity

Abstract

The invention discloses a method and a system for predicting casualties caused by earthquake landslide in consideration of population dynamic characteristics and environmental heterogeneity. The method comprises the following steps: step 1, collecting all environmental factor data and multi-source population distribution data in a target earthquake region according to a preset space unit; 2, inputting the collected environmental factor data into a preset logistic regression model to obtain an earthquake landslide probability; 3, processing the multi-source population distribution data through a data fusion method, and obtaining the dynamic population distribution of the earthquake occurrence time period; 4, processing the environmental factor data based on a random forest algorithm model and an SHAP method to obtain an environmental heterogeneity parameter and an environmental factor contribution degree; and 5, according to the probability prediction model of the casualties caused by the earthquake landslide, obtaining probability distribution of the casualties caused by the earthquake landslide. According to the method, the precision of evaluating casualties caused by earthquake landslide is improved, and a reliable basis for formulating an action scheme is provided for earthquake rescue.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field, and in particular to a method and system for predicting casualties caused by an earthquake landslide by considering population dynamic characteristics and environmental heterogeneity. BACKGROUND

[0002] Geological secondary disasters triggered by an earthquake are manifested in various forms of slope material movement, and the mechanical mechanism mainly includes various typical types: rock collapse phenomenon, mainly characterized by the instantaneous disintegration of steep rock mass under the action of seismic load; rockfall disaster in the form of discrete body movement, which is manifested in the bouncing and rolling process of a single or multiple rock blocks along the slope surface; soil landslide in the form of continuous deformation, the mechanical essence of which is shear failure on the potential sliding surface; and mudflow movement with the characteristics of solid-liquid two-phase flow. These disaster types cause significant casualties in multiple strong earthquakes due to their suddenness and chain effect.

[0003] The existing earthquake landslide population loss evaluation system mainly adopts a statistical method based on field investigation. In the implementation process, a multidisciplinary joint investigation team needs to be organized to collect disaster data through point-by-point checking. This mode exposes obvious time lag defects in earthquake emergency response. It takes a long time from the occurrence of an earthquake to the completion of all landslide point checking, which seriously lags behind the golden rescue period. At the same time, due to the interference of factors such as landslide burial depth and rescue progress, there are significant uncontrollable underreporting and late reporting phenomena in field statistics. Although the observation technology provides a new technical means for landslide identification, it is limited by the revisit period of optical remote sensing satellites, weather influence, and data accuracy, so that remote sensing monitoring cannot meet the real-time requirements of emergency response.

[0004] In recent years, in order to meet the timeliness and accuracy requirements of earthquake disaster emergency, quantitative evaluation means of casualties caused by earthquake landslide have been proposed. At present, the existing technology provides a population loss rapid evaluation method based on earthquake landslide occurrence probability. This method first constructs an earthquake landslide probability field based on the Newmark displacement model or machine learning algorithm, and estimates the population loss caused by an earthquake landslide in combination with the population density distribution. This technology compresses the evaluation time of earthquake landslide casualties from several days to several hours, which provides key decision support for emergency rescue and resource allocation.

[0005] However, there is no perfect technical solution for earthquake landslide personnel casualty assessment. In existing earthquake landslide personnel casualty assessment research, on the one hand, static population assumption is generally used, and the dynamic characteristics of population are ignored, resulting in systematic deviation between the results of personnel casualty risk assessment and spatial distribution prediction and the actual disaster situation; on the other hand, the spatial characteristics of environmental factors and their complex correlation mechanism are often simplified as single factor or deterministic mapping relationship, without considering the spatial heterogeneity of environmental factors and the spatio-temporal non-stationarity of the relationship between human activities. The static assumption of population distribution and the heterogeneity of environmental impact factors are the key problems affecting the prediction accuracy. The gap between the personnel casualty assessment results and the actual needs of emergency disposal and rescue is still large. SUMMARY

[0006] In order to solve the problem of low personnel casualty prediction accuracy caused by ignoring the dynamic characteristics of population and not considering the heterogeneity of environmental impact factors in the prior art, the present application provides a personnel casualty prediction method and system caused by earthquake landslide considering the dynamic characteristics of population and the heterogeneity of environment. The dynamic characteristics of population at the time of earthquake are obtained by data fusion, and the environmental heterogeneity parameters are obtained based on the random forest algorithm model. Finally, the probability prediction model of personnel casualty caused by earthquake landslide is constructed to realize accurate prediction of personnel casualty caused by earthquake landslide, and the prediction accuracy of personnel casualty is improved.

[0007] In order to achieve the above purpose, the technical scheme of the present application is: The first aspect of the present application provides a personnel casualty prediction method caused by earthquake landslide considering the dynamic characteristics of population and the heterogeneity of environment, comprising: Step 1: Collect all environmental factor data and multi-source population distribution data in the target earthquake area according to the preset spatial unit; wherein the environmental factor data includes terrain factor, address factor and basic geographic environment factor, and the multi-source population distribution data includes population kilometer grid data and mobile positioning data; Step 2: input the collected environmental factor data into the preset logistic regression model to obtain the probability of earthquake landslide, so as to obtain the probability of landslide and facilitate subsequent prediction of personnel casualty caused by landslide; Step 3: process the multi-source population distribution data by using data fusion method to obtain the dynamic population distribution at the earthquake period, so as to obtain the dynamic characteristics of population and improve the accuracy of casualty prediction; Step 4: process the environmental factor data based on the random forest algorithm model and SHAP method to obtain the environmental heterogeneity parameters and the contribution degree of environmental factors, so as to integrate the influence of environment on personnel casualty into the prediction model and improve the prediction accuracy; Step five: constructing a probability prediction model of casualties caused by earthquake landslides according to the environmental heterogeneity parameters and the contribution degrees of environmental factors, inputting the environmental heterogeneity parameters, the probability of earthquake landslides and the dynamic population distribution in the earthquake occurrence period into the probability prediction model of casualties caused by earthquake landslides to obtain the probability distribution of casualties caused by earthquake landslides.

[0008] Further, the step two specifically comprises: The correlation between any two environmental factors in all environmental factor data is obtained by using the Spearman rank correlation formula, the environmental factors are screened according to the correlation, and the screened environmental factors are obtained for eliminating the impact factors with weak correlation; The screened environmental factors are input into a preset logistic regression model to obtain the probability of earthquake landslides.

[0009] Further, the step three specifically comprises: The population kilometer grid data and the mobile positioning data are respectively subjected to two-dimensional discrete Fourier transform to obtain frequency domain images; wherein the mobile positioning data includes an anonymous user equipment spatial position data set collected by an operator communication network within a preset time window before and after the earthquake occurrence moment; The amplitude information is obtained based on the frequency domain image corresponding to the population kilometer grid data, and the phase information is obtained based on the frequency domain image corresponding to the mobile positioning data; The amplitude information and the phase information are processed according to two-dimensional discrete inverse Fourier transform to obtain the dynamic population distribution in the earthquake occurrence period, so as to improve the prediction accuracy.

[0010] Further, the step four specifically comprises: The environmental element data are input into a preset random forest algorithm model to obtain the environmental heterogeneity parameters; The contribution degrees of the environmental factors are obtained based on the SHAP method and the Gini value index method on the basis of the random forest algorithm model.

[0011] Further, the contribution degrees of the environmental factors obtained based on the SHAP method and the Gini value index method on the basis of the random forest algorithm model specifically comprise: The feature contribution value is calculated according to the environmental heterogeneity parameters, and the n x m feature contribution value matrix is constructed according to the feature contribution value, wherein n is the sample size and m is the feature number; The Gini value drop of each feature in the random forest algorithm model is calculated, the feature contribution value matrix and the Gini drop are weighted and fused to obtain the importance score of the environmental factor; The importance score of the environmental factor is normalized to obtain the contribution degree of the environmental factor.

[0012] Further, the probability prediction model of casualties caused by the earthquake landslide is constructed according to the environmental heterogeneity parameter and the environmental factor contribution degree, and specifically comprises the following steps. The environmental factors are sorted according to the environmental factor contribution degree, and the top two environmental factors are obtained; and the environmental correction coefficient is obtained according to the top two environmental factors and the environmental heterogeneity parameter. The probability prediction model of casualties caused by the earthquake landslide is constructed according to the environmental correction coefficient.

[0013] Further, the environmental correction coefficient is expressed by the following formula: wherein, the environmental correction coefficient, the environmental heterogeneity parameter, and different normalized environmental factors, the coefficient, the comprehensive function, and y is the historical real earthquake case casualty distribution.

[0014] Further, the probability prediction model of casualties caused by the earthquake landslide is expressed by the following formula: wherein, R pop the probability distribution of casualties caused by the earthquake landslide, H landslide the probability of the earthquake landslide, P dynamic the dynamic population distribution in the earthquake period, E uncertainty the environmental heterogeneity parameter.

[0015] The second aspect of the present application proposes a system for predicting casualties caused by an earthquake landslide considering the dynamic characteristics of the population and the environmental heterogeneity, comprising: A data collection module is configured to collect all environmental factor data and multi-source population distribution data in a target earthquake area according to a preset spatial unit; wherein the environmental factor data includes terrain factors, address factors and basic geographic environmental factors, and the multi-source population distribution data includes population kilometer grid data and mobile positioning data. A landslide prediction module is configured to input the collected environmental factor data into a preset logistic regression model to obtain the probability of the earthquake landslide, so as to facilitate the prediction of casualties caused by the landslide. The dynamic population distribution module is used to process multi-source population distribution data using data fusion methods to obtain the dynamic population distribution during the earthquake period, which facilitates the acquisition of population dynamic characteristics and improves the accuracy of casualty prediction. The element extraction module is used to process environmental factor data based on the random forest algorithm model and SHAP method to obtain environmental heterogeneity parameters and environmental factor contribution, which makes it easier to integrate the impact of the environment on human casualties into the prediction model to improve prediction accuracy. The prediction module is used to construct a probability prediction model for casualties caused by earthquakes and landslides based on environmental heterogeneity parameters and the contribution of environmental factors. The environmental heterogeneity parameters, the probability of earthquakes and landslides, and the dynamic population distribution during the earthquake occurrence period are input into the probability prediction model for casualties caused by earthquakes and landslides to obtain the probability distribution of casualties caused by earthquakes and landslides.

[0016] The beneficial effects of this invention are: This invention improves the accuracy of casualty assessment caused by earthquakes and landslides, providing a reliable basis for earthquake rescue action planning. Regarding prediction accuracy, it integrates the spatial accuracy of kilometer-grid population data with the dynamic real-time nature of mobile positioning data, overcoming the shortcomings of insufficient spatiotemporal resolution in traditional static population models. In terms of timeliness, it determines device locations using mobile positioning data, which can meet the critical 72-hour emergency response requirement with 5-minute updates to operator signaling data. Regarding environmental heterogeneity parameters, it employs the nonlinear coupling effect of random forest quantification of multidimensional factors, significantly improving the spatial prediction accuracy of landslide burial casualty rates. Attached Figure Description

[0017] Figure 1 A flowchart illustrating a method for predicting casualties caused by earthquake-induced landslides, taking into account population dynamics and environmental heterogeneity, is provided for embodiments of the present invention.

[0018] Figure 2 This is a flowchart illustrating the training of a method for predicting casualties caused by earthquake landslides, taking into account population dynamics and environmental heterogeneity, as provided in an embodiment of the present invention.

[0019] Figure 3 This is an architecture diagram of an earthquake landslide casualty prediction system that takes into account population dynamics and environmental heterogeneity, provided as an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of the embodiments of this invention will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0021] Example 1 like Figure 1 As shown, a method for predicting casualties caused by earthquake-induced landslides, considering population dynamics and environmental heterogeneity, includes: S101: Collect all environmental factor data and multi-source population distribution data within the target seismic area according to the preset spatial units; among which, the environmental factor data includes topographic factors, geographical factors and basic geographic environmental factors, and the multi-source population distribution data includes population kilometer grid data and mobile positioning data.

[0022] S102: Input the collected environmental factor data into the preset logistic regression model to obtain the earthquake landslide probability.

[0023] S103: Use data fusion methods to process multi-source population distribution data to obtain the dynamic population distribution during the earthquake period.

[0024] S104: Based on the random forest algorithm model and SHAP method, environmental factor data are processed to obtain environmental heterogeneity parameters and environmental factor contribution.

[0025] S105: Construct a probability prediction model for casualties caused by earthquake landslides based on environmental heterogeneity parameters and the contribution of environmental factors. Input environmental heterogeneity parameters, earthquake landslide probability, and dynamic population distribution during the earthquake occurrence period into the probability prediction model for casualties caused by earthquake landslides to obtain the probability distribution of casualties caused by earthquake landslides.

[0026] This invention first obtains the probability of earthquake landslides through environmental factors, then obtains the dynamic distribution of the population using population kilometer grid data and mobile positioning data, which can more accurately determine the distribution of people during an earthquake. Finally, the environmental factors are processed to obtain environmental heterogeneity parameters and the contribution of environmental factors. Based on the environmental heterogeneity parameters, a probability prediction model for earthquake landslide-related casualties is constructed, and the probability distribution of earthquake landslide-related casualties is obtained from the model. This invention combines dynamic population distribution and environmental heterogeneity to improve the accuracy of prediction.

[0027] Example 2 Based on the above embodiments, such as Figure 2As shown, this invention proposes a specific process for a method to predict casualties caused by earthquake-induced landslides, considering population dynamics and environmental heterogeneity, including: S201: Collect earthquake case data, including earthquake intensity, earthquake casualty records, co-seismic landslide data, multi-source population distribution data in earthquake zones, and basic geographical, topographical, geological, meteorological and other environmental data within earthquake-affected areas.

[0028] Specifically, data on earthquake cases and their consequences will be collected, including earthquake intensity distribution data and earthquake casualty records. Data on coseismic landslides will be collected, primarily focusing on the spatial distribution of coseismic landslides occurring during earthquake events, including their location and spatial geometry. Environmental factor data will be collected, including at least topographic factors (such as elevation, slope, aspect, surface relief, surface curvature, and topographic humidity index), geological factors (such as stratigraphic lithology and fault distribution), and basic geographical environmental factors (such as soil properties, vegetation characteristics, water system distribution, road distribution, and building distribution). Multi-source population data for the earthquake zone will be collected, including at least kilometer-grid data of population distribution in the earthquake zone and mobile location data during the earthquake's occurrence period. Multi-source population distribution data includes both kilometer-grid population data and mobile location data.

[0029] S202: Using the logistic regression algorithm and the modeling approach of earthquake landslide probability hazard assessment, an earthquake landslide probability hazard assessment model is constructed based on earthquake landslide samples and multi-factor environmental factors to obtain earthquake landslide probability distribution data.

[0030] Specifically, using coseismic landslide data, we selected environmental factors that influence earthquake-triggered landslides, and used the landslide / non-landslide area ratio to select negative samples to create a sample dataset. The training set accounted for 70% to 80% of the sample data, and the validation set accounted for 20% to 30%.

[0031] The Spearman rank correlation test is used to examine the correlation between various environmental factors, quantify the degree of correlation between them, and select environmental factors for modeling. The Spearman rank correlation is calculated as follows: in, Let i be the i-th element in variable x. Let be the i-th element in variable y, and N be the total number of samples. Let y be the average value of the variable. Let x be the average value of the variable.

[0032] Using the selected environmental factors and landslide sample dataset as input variables, a hazard assessment model was established by training the model using the training dataset. The hazard assessment model was a logistic regression model. The trained model was validated using a validation set, and the predicted results were compared with those of geological hazard samples. Validation metrics included accuracy, recall, F1 score, ROC curve, and AUC value. Using the constructed and trained assessment model, the probability of a landslide occurring at any spatial location under seismic intensity was calculated, and earthquake landslide probability distribution data were obtained.

[0033] S203: Based on the data fusion method, extract multi-dimensional population information from the population kilometer grid data of the earthquake zone and the mobile positioning data during the earthquake occurrence period, including high-precision spatial distribution and high-time dynamic features, to obtain dynamic population feature data that can characterize the real-time population distribution during the earthquake occurrence period.

[0034] Specifically, through two-dimensional discrete Fourier transform, the population kilometer grid data and mobile positioning data are transformed into a frequency domain image represented by amplitude spectrum and phase spectrum. The two-dimensional discrete Fourier transform can be expressed as: in, F(u, v) For frequency domain images, u =0, 1, 2, …, v =0, 1, 2, …, M represents the data in the row direction of the population kilometer grid data, N represents the data in the column direction of the population kilometer grid data, and f(x,y) is the original population kilometer grid data.

[0035] The amplitude information of the population distribution to be merged is obtained by calculating the frequency domain image corresponding to the population kilometer grid data, and the phase information of the population distribution to be merged is obtained by calculating the frequency domain image corresponding to the mobile positioning data.

[0036] Dynamic population distribution characteristics data with high reliability and timeliness are obtained by using the two-dimensional discrete Fourier inverse transform. The inverse Fourier transform can be expressed as: in, DFT -1 This represents the inverse Fourier transform. F h (u, v) A frequency domain image representing the population distribution to be merged. A h (u, v) and Ψ h (u, v) These represent the amplitude spectrum and phase spectrum of the frequency domain image, respectively. This represents a dynamic population distribution.

[0037] S204: Using the actual distribution of casualties in earthquake landslides and combining various environmental factors that may affect the distribution of casualties, a random forest algorithm is used to construct an environmental heterogeneity evaluation model for casualties in earthquake landslides and obtain environmental heterogeneity parameters.

[0038] Specifically, environmental factors are constructed by selecting data on elevation and its topographic parameters, strata lithology, faults, soil types, land use types, vegetation cover, water systems, roads, buildings, and other environmental factors.

[0039] By using correlation analysis and environmental factor elimination, we can ensure that there is little spatial autocorrelation among the environmental factors.

[0040] For negative sample selection, negative samples will be randomly generated in areas more than 100 meters away from positive samples within the earthquake zone at a ratio of 1:10 to construct a training sample set for the random forest algorithm. Using the training dataset, an environmental heterogeneity assessment model affecting the distribution of casualties will be established to obtain the assessment results characterizing the environmental heterogeneity of casualties in earthquake-induced landslides. The machine learning interpretability SHAP method will be used, employing the Gini index method to give the importance score (VIM) of environmental factors and obtain their contribution.

[0041] Specifically, feature contribution values ​​are calculated based on environmental heterogeneity parameters. An n×m feature contribution value matrix is ​​then constructed, where n is the sample size and m is the number of features. The Gini value decrease for each feature at the split node in the random forest algorithm model is calculated. The feature contribution value matrix and the Gini value decrease are then weighted and fused to obtain the importance score of the environmental factor. The importance score of the environmental factor is then normalized to obtain the contribution degree of the environmental factor.

[0042] S205: Construct a probability prediction model for casualties caused by earthquake landslides based on environmental heterogeneity parameters and the contribution of environmental factors. Input environmental heterogeneity parameters, earthquake landslide probability, and dynamic population distribution during the earthquake occurrence period into the probability prediction model for casualties caused by earthquake landslides to obtain the probability distribution of casualties caused by earthquake landslides.

[0043] Specifically, based on modeling case data and following the assessment approach of "earthquake landslide hazard - dynamic population exposure - environmental factor heterogeneity," a probability prediction model for casualties caused by earthquake landslides is constructed, which can be expressed as: in, R pop The probability distribution of casualties in earthquake-induced landslides. H landslide This represents the probability of an earthquake-induced landslide. P dynamicThis represents the dynamic population distribution during the period in which the earthquake occurred. E uncertainty This is a parameter representing environmental heterogeneity.

[0044] Based on the ranking of environmental factors by their contribution, the two dominant environmental factors were identified. x a , x b Based on the environmental heterogeneity assessment results, the environmental correction coefficient for predicting the probability of casualties in earthquake-induced landslides is calculated. k , can be represented as: in, The environmental correction coefficient is spatially distributed data, representing the correction coefficient on each grid. For environmental heterogeneity parameters, and These are different normalized environmental factors. cov Describing covariance, The coefficient can be obtained from real historical earthquake cases. Let y be a comprehensive function, representing a linear or nonlinear function that comprehensively considers the environmental impact of multiple factors, where y represents the distribution of casualties in real historical earthquake cases.

[0045] Calculate the number of casualties and the spatial probability distribution caused by earthquake-induced landslides.

[0046] Example 3 Based on the above embodiments, the present invention proposes a specific process for a method to predict casualties caused by earthquake landslides, considering population dynamics and environmental heterogeneity, including: S301: Collect all environmental factor data and multi-source population distribution data within the target seismic area according to the preset spatial units; among which, environmental factor data includes topographic factors, geographical factors and basic geographic environmental factors, and multi-source population distribution data includes population kilometer grid data and mobile positioning data.

[0047] S302: Input the collected environmental factor data into the preset logistic regression model to obtain the earthquake landslide probability.

[0048] Specifically, the Spearman rank correlation formula is used to obtain the correlation between any two environmental factors in all environmental factor data. The environmental factors are then screened based on the correlation to obtain the screened environmental factors.

[0049] The selected environmental factors are input into a preset logistic regression model to obtain the probability of earthquake landslides.

[0050] S303: Use data fusion methods to process multi-source population distribution data to obtain the dynamic population distribution during the earthquake period.

[0051] Specifically, two-dimensional discrete Fourier transforms were performed on the population kilometer grid data and the mobile location data to obtain a frequency domain image; among which, the mobile location data included an anonymous user equipment spatial location dataset collected by the operator's communication network within a preset time window before and after the earthquake.

[0052] Amplitude information is obtained from the frequency domain image corresponding to the population kilometer grid data, and phase information is obtained from the frequency domain image corresponding to the mobile positioning data.

[0053] The amplitude and phase information are processed by the two-dimensional discrete Fourier inverse transform to obtain the dynamic population distribution during the earthquake period.

[0054] S304: Based on the random forest algorithm model and SHAP method, environmental factor data are processed to obtain environmental heterogeneity parameters and environmental factor contributions.

[0055] Specifically, environmental element data is input into a pre-defined random forest algorithm model to obtain environmental heterogeneity parameters. Environmental heterogeneity parameters represent the non-uniform (heterogeneous) distribution characteristics exhibited in the spatial dimension after the combined effects of all environmental elements.

[0056] Environmental factor contributions are derived using the SHAP and Gini index methods based on a random forest algorithm model. Specifically, feature contribution values ​​are calculated based on environmental heterogeneity parameters, and an n×m feature contribution value matrix is ​​constructed, where n is the sample size and m is the number of features. The Gini value decrease for each feature at the split node in the random forest algorithm model is calculated, and the feature contribution value matrix and the Gini value decrease are weighted and fused to obtain the importance score of the environmental factor. The importance score of the environmental factor is then normalized to obtain the environmental factor contribution.

[0057] S305: Construct a probability prediction model for casualties caused by earthquake landslides based on environmental heterogeneity parameters and the contribution of environmental factors. Input environmental heterogeneity parameters, earthquake landslide probability, and dynamic population distribution during the earthquake occurrence period into the probability prediction model for casualties caused by earthquake landslides to obtain the probability distribution of casualties caused by earthquake landslides.

[0058] Specifically, the probability prediction model for casualties caused by earthquake-induced landslides is expressed by the following formula: in, R pop The probability distribution of casualties in earthquake-induced landslides.H landslide This represents the probability of an earthquake-induced landslide. P dynamic This represents the dynamic population distribution during the period in which the earthquake occurred. E uncertainty This is a parameter representing environmental heterogeneity.

[0059] Based on the ranking of environmental factors by their contribution, the two dominant environmental factors were identified. x a , x b Based on the environmental heterogeneity assessment results, the environmental correction coefficient for predicting the probability of casualties in earthquake-induced landslides is calculated. k , can be represented as: in, The environmental correction coefficient is spatially distributed data, representing the correction coefficient on each grid. For environmental heterogeneity parameters, and These are different normalized environmental factors. cov Describing covariance, The coefficient can be obtained from real historical earthquake cases. Let y be a comprehensive function, representing a linear or nonlinear function that comprehensively considers the environmental impact of multiple factors, where y represents the distribution of casualties in real historical earthquake cases.

[0060] Example 4 Based on the above embodiments, such as Figure 3 As shown, this invention proposes a system for predicting casualties caused by earthquake-induced landslides, considering population dynamics and environmental heterogeneity, comprising: The data collection module is used to collect all environmental element data and multi-source population distribution data in the target earthquake area according to the preset spatial units; among which, the environmental element data includes earthquake landslide influencing factors, and the multi-source population distribution data includes population kilometer grid data and mobile positioning data; The landslide prediction module is used to input the collected environmental factor data into a preset logistic regression model to obtain the probability of earthquake landslides. The dynamic population distribution module is used to process multi-source population distribution data using data fusion methods to obtain the dynamic population distribution during the earthquake period. The element extraction module is used to process environmental element data based on the random forest algorithm model and SHAP method to obtain environmental heterogeneity parameters and environmental factor contribution. The prediction module is used to construct a probability prediction model for casualties caused by earthquakes and landslides based on environmental heterogeneity parameters and the contribution of environmental factors. The environmental heterogeneity parameters, the probability of earthquakes and landslides, and the dynamic population distribution during the earthquake occurrence period are input into the probability prediction model for casualties caused by earthquakes and landslides to obtain the probability distribution of casualties caused by earthquakes and landslides.

[0061] It should be noted that the earthquake landslide casualty prediction system provided in this embodiment of the invention, which considers population dynamics and environmental heterogeneity, is for the purpose of implementing the above-mentioned earthquake landslide casualty prediction method that considers population dynamics and environmental heterogeneity. Its specific functions can be referred to in the above-mentioned method embodiments, and will not be repeated here.

[0062] In summary, this invention improves the accuracy of assessing casualties caused by earthquakes and landslides, providing a reliable basis for developing action plans for earthquake relief. Regarding prediction accuracy, the integration of spatial accuracy from kilometer-grid population data with the dynamic real-time nature of mobile positioning data overcomes the shortcomings of insufficient spatiotemporal resolution in traditional static population models. In terms of timeliness, determining device locations using mobile positioning data, even with 5-minute updates to operator signaling data, meets the critical 72-hour window for emergency response. Regarding environmental heterogeneity parameters, the use of random forest quantification of the nonlinear coupling effect of multidimensional factors significantly improves the spatial prediction accuracy of landslide burial casualty rates.

[0063] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1.A method for predicting casualties caused by an earthquake-induced landslide, taking into account population dynamics and environmental heterogeneity, characterized in that, The application relates to a method for predicting the probability of casualties caused by a landslide during an earthquake. The method comprises the following steps: Step 1: collecting all environmental factor data and multi-source population distribution data in a target earthquake area according to a preset spatial unit; wherein the environmental factor data comprises terrain factors, address factors and basic geographical environmental factors, and the multi-source population distribution data comprises population kilometer grid data and mobile positioning data; Step 2: inputting the collected environmental factor data into a preset logistic regression model to obtain a landslide probability during an earthquake; Step 3: processing the multi-source population distribution data by using a data fusion method to obtain a dynamic population distribution during an earthquake; Step 4: processing the environmental factor data based on a random forest algorithm model and a SHAP method to obtain environmental heterogeneity parameters and environmental factor contribution degrees; 2.The method of claim 1, wherein, Step 5: constructing a probability prediction model for casualties caused by a landslide during an earthquake according to the environmental heterogeneity parameters and the environmental factor contribution degrees, inputting the environmental heterogeneity parameters, the landslide probability during an earthquake and the dynamic population distribution during an earthquake into the probability prediction model for casualties caused by a landslide during an earthquake, and obtaining a probability distribution of casualties caused by a landslide during an earthquake. The step 2 specifically comprises the following steps: obtaining the correlation between any two environmental factors in all environmental factor data by using a Spearman rank correlation formula, screening the environmental factors according to the correlation, and obtaining screened environmental factors; 3.The method of claim 1, wherein the method further comprises: inputting the screened environmental factors into a preset logistic regression model to obtain a landslide probability during an earthquake. The step 3 specifically comprises the following steps: respectively performing two-dimensional discrete Fourier transform on the population kilometer grid data and the mobile positioning data to obtain frequency domain images; wherein the mobile positioning data comprises a set of anonymous user equipment spatial position data collected by an operator communication network within a preset time window before and after an earthquake; obtaining amplitude information based on the frequency domain image corresponding to the population kilometer grid data and obtaining phase information based on the frequency domain image corresponding to the mobile positioning data; 4.The method of claim 1, wherein the method further comprises, processing the amplitude information and the phase information according to two-dimensional discrete inverse Fourier transform to obtain a dynamic population distribution during an earthquake. The step 4 specifically comprises the following steps: inputting the environmental factor data into a preset random forest algorithm model to obtain environmental heterogeneity parameters; 5.The method of claim 4, wherein the method is characterized by, obtaining environmental factor contribution degrees based on a SHAP method and a Gini value index method on the basis of the random forest algorithm model. The step of obtaining environmental factor contribution degrees based on a SHAP method and a Gini value index method on the basis of the random forest algorithm model specifically comprises the following steps: calculating feature contribution values according to the environmental heterogeneity parameters, and constructing a feature contribution value matrix of n*m according to the feature contribution values, wherein n is a sample size and m is a feature number; calculating the Gini value reduction amount of each feature when a split node in the random forest algorithm model, weighting and fusing the feature contribution value matrix and the Gini value reduction amount to obtain an importance score of the environmental factor; 6.The method of claim 4, wherein the method is characterized by, normalizing the importance score of the environmental factor to obtain the environmental factor contribution degree. The step of constructing a probability prediction model for casualties caused by a landslide during an earthquake according to the environmental heterogeneity parameters and the environmental factor contribution degrees specifically comprises the following steps: sorting the environmental factors according to the environmental factor contribution degrees to obtain the top two environmental factors, and obtaining an environmental correction coefficient according to the top two environmental factors and the environmental heterogeneity parameters. A probability prediction model for casualties caused by a seismic landslide is constructed according to an environmental correction coefficient. 7.The method of claim 6, wherein the method is characterized by, The environmental correction coefficient is expressed by the following formula: wherein, is an environmental correction coefficient, is an environmental heterogeneity parameter, and are different normalized environmental factors, respectively, is a coefficient, is a comprehensive function, y is a historical real earthquake case personnel casualty distribution. 8.The method of claim 7, wherein the method further comprises, The probability prediction model for casualties caused by a seismic landslide is expressed by the following formula: wherein, R pop is a probability distribution of casualties for a seismic landslide, H landslide is a probability of occurrence of a seismic landslide, P dynamic is a dynamic population distribution for a time period of a seismic event, E uncertainty is an environmental heterogeneity parameter. 9.A system for predicting casualties caused by an earthquake landslide, considering population dynamics and environmental heterogeneity, characterized by, The probability prediction model for casualties caused by a seismic landslide is constructed according to an environmental correction coefficient. The probability prediction model for casualties caused by a seismic landslide is constructed according to an environmental correction coefficient. The data collection module is configured to collect all environmental factor data and multi-source population distribution data in a target seismic area according to a preset spatial unit; the environmental factor data includes seismic landslide impact factors, and the multi-source population distribution data includes population kilometer grid data and mobile positioning data; The landslide prediction module is configured to input the collected environmental factor data into a preset logistic regression model to obtain a seismic landslide probability; The dynamic population distribution module is configured to process the multi-source population distribution data by using a data fusion method to obtain a dynamic population distribution during a seismic occurrence period; The factor extraction module is configured to process the environmental factor data based on a random forest algorithm model and a SHAP method to obtain environmental heterogeneity parameters and environmental factor contribution degrees; The prediction module is configured to construct a probability prediction model for casualties caused by a seismic landslide according to the environmental heterogeneity parameters and the environmental factor contribution degrees, input the environmental heterogeneity parameters, the seismic landslide probability, and the dynamic population distribution during the seismic occurrence period into the probability prediction model for casualties caused by a seismic landslide, and obtain a probability distribution of casualties caused by a seismic landslide.

Images