Geological disaster identifying, monitoring and early warning system and method based on AI technology

Abstract

The invention discloses a geological disaster identification monitoring early warning system and method based on an AI technology, and relates to the technical field of geological disaster monitoring. The method comprises the following steps: collecting a multi-source data set in a target area, preprocessing the multi-source data set, and carrying out hidden danger labeling to obtain a plurality of mark points; the technical key points are as follows: intelligent benchmarking of a current high-risk area and a historical disaster case is realized by adopting a technical scheme of comprehensive similarity calculation and a KNN algorithm, an effect of intelligent grade mapping is achieved, scientificity and reliability of disaster grade judgment are improved, and by comparing the similarity of a current characteristic value and a historical disaster sample, the judgment accuracy of the disaster grade is improved. According to the method, the closest historical case can be quickly positioned, the risk level of the current area is determined accordingly, the adaptive capacity and the response speed of an early warning system are further enhanced, and meanwhile, the method can effectively balance the accuracy and the response speed of risk judgment.

Description

Technical Field

[0001] The present invention relates to the field of geological disaster monitoring technology, and specifically to a geological disaster identification, monitoring and early warning system and method based on AI technology. Background Art

[0002] In recent years, geological disaster monitoring technology has made significant progress with the help of AI technology, especially in the identification and early warning of geological disasters such as landslides and mudslides. Traditional methods mainly rely on on-site surveys and single sensor data (such as seismometers and rain gauges). This method has problems such as limited coverage, poor real-time performance and insufficient accuracy. Geological disaster identification based on AI technology achieves high-precision prediction or early warning of geological disasters through data fusion and machine learning algorithms.

[0003] Traditional geological hazard monitoring technology relies primarily on field surveys, single-sensor data (such as seismometers and rain gauges), and static models, which present significant technical limitations. First, the data source is single, and relying solely on limited ground-based observation equipment makes it difficult to fully cover large areas, resulting in inaccurate risk assessments. For example, in landslide monitoring, traditional methods may overlook small but persistent surface changes that are often key to early warning. Second, due to insufficient real-time and dynamic adjustment capabilities, traditional fixed-parameter models cannot promptly adapt to environmental changes, such as the impact of changing meteorological conditions on the probability of landslides. This makes the early warning system slow to respond to sudden extreme weather events, reducing the effectiveness of warnings. Third, the threshold setting method lacks scientific basis and cannot effectively balance the accuracy of risk assessment and response speed. It also lacks an assessment of the chain reaction of disasters. For example, if a landslide with a higher warning level occurs in any adjacent area, whether the resulting vibration or other factors will affect adjacent areas, and whether the degree of impact will change the original risk level, resulting in inaccurate warning levels, are all technical issues that need to be addressed. Summary of the Invention

[0004] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0005] A method for identifying, monitoring and warning geological disasters based on AI technology, comprising the following steps:

[0006] Collect multi-source datasets in the target area, pre-process them, and mark hidden dangers to obtain a number of marked points. The multi-source datasets include at least SAR data, drone aerial images, geographic information, and meteorological data.

[0007] Within the target area, areas with slopes greater than 45° are identified as high-risk areas based on geographic information. Based on multi-source datasets and hidden danger annotations, feature sets are screened within any high-risk area, and the primary risk index corresponding to each high-risk area is calculated based on the feature sets.

[0008] The GBDT model is used, with meteorological data as input and the output as the probability of landslide occurrence;

[0009] Construct a risk index correction model to correct the primary risk index based on the probability of landslide occurrence, and derive the secondary risk index corresponding to the high-risk area;

[0010] Compare the multi-source data set corresponding to any secondary risk index with similar data from the early stages of different geological disasters in history to calculate the comprehensive similarity; set a benchmarking threshold, traverse all geological disaster event samples in the historical database, screen several similar samples, use the KNN algorithm for level mapping, and output the preliminary disaster risk level;

[0011] Analyze the interval between two adjacent excessive secondary risk indices and determine whether the interval and the two adjacent excessive secondary risk indices meet the set constraints. If so, trigger the linkage risk assessment mechanism to generate a joint risk index; determine whether the joint risk index exceeds the standard index threshold;

[0012] If not, the preliminary disaster risk level will be used as the final disaster risk level;

[0013] If so, the disaster level correction action is performed and the corresponding warning level is sent based on the final disaster risk level;

[0014] Under the condition that the secondary risk index R_sy continues to accumulate, the AI ​​model is used to generate the probability distribution interval of the secondary risk index R_sy; the database used in the KNN algorithm is regularly updated, and the AI ​​model is retrained with new samples; wherein the AI ​​model is: any one of GPR, BNN and random forest.

[0015] Furthermore, the process of preprocessing the multi-source datasets includes at least: denoising, registration, and spatiotemporal alignment in combination with meteorological data;

[0016] SAR data at least include: cumulative deformation and deformation rate;

[0017] UAV aerial images include at least: several surface images;

[0018] Geographic information shall include at least: terrain data and slope layers;

[0019] Meteorological data include at least: rainfall and temperature.

[0020] Furthermore, the feature set includes at least: cumulative deformation, deformation rate, and marker density;

[0021] The primary risk index R_py corresponding to each high-risk area is calculated based on the following method:

[0022] ;

[0023] Where x i : the i-th eigenvalue, f i (): normalization function, w i : The weight of each eigenvalue, i=1, 2, ..., n, where n is the total number of eigenvalues.

[0024] Furthermore, the secondary risk index R_sy corresponding to the high-risk area is generated according to the following method:

[0025] ;

[0026] Where, α: correction coefficient, P_landslide: probability of landslide occurrence;

[0027] Collect the number of all identified high-risk areas N_r and their corresponding primary risk index R_py in the target area j Where j = 1, 2, ..., N_r, calculate the primary risk index R_py j The mean R_py_avg and standard deviation σ_py;

[0028] The correction factor α is adjusted as follows:

[0029] ;

[0030] Where α0 is the initial weight; q1 and q2 are adjustment coefficients, both ranging from [0, 1]; N_total is the number of grids in the target area.

[0031] Furthermore, the comprehensive similarity S_siy is calculated and the conditions for screening several similar samples are:

[0032] The comprehensive similarity S_siy ≥ the benchmark threshold S_th is satisfied.

[0033] Furthermore, the constraints are:

[0034] Condition 1: Are the geological conditions similar or identical?

[0035] Condition 2: Whether the maximum distance between two high-risk areas with excessive secondary risk indices is lower than the calibration value.

[0036] Furthermore, the process of triggering the linkage risk assessment mechanism is as follows:

[0037] Construct a linkage risk index model, input two exceeded secondary risk indices and the geographical distance between the center points of the high-risk areas corresponding to the two exceeded secondary risk indices, and output the joint risk index R_cd;

[0038] When running the linkage risk index model, the method is as follows:

[0039] ;

[0040] Where R_sy A and Résy B : The secondary risk index of exceeding the standard for two adjacent high-risk areas, β: linkage enhancement coefficient, ranging from [0, 1], γ: distance attenuation coefficient, ranging from [0, 0.1]; d: geographical distance value.

[0041] Furthermore, the process of performing the disaster level correction action is: marking two adjacent high-risk areas as an overall area, and raising the area with the lower preliminary disaster risk level by one level to obtain the final disaster risk level.

[0042] A geological disaster identification, monitoring and early warning system based on AI technology, the system includes:

[0043] Data acquisition module: collects multi-source data sets in the target area, pre-processes the multi-source data sets, and annotates hidden dangers to obtain a number of marked points. The multi-source data sets include at least: SAR data, drone aerial images, geographic information, and meteorological data;

[0044] Risk Identification Module: Within the target area, areas with slopes greater than 45° are identified as high-risk areas based on geographic information. Based on multi-source datasets and hidden danger annotations, a feature set is screened for any high-risk area, and the primary risk index corresponding to each high-risk area is calculated based on the feature set.

[0045] Probability prediction module: adopts GBDT model, with meteorological data as input and landslide occurrence probability as output;

[0046] Risk correction module: Constructs a risk index correction model to correct the primary risk index based on the probability of landslide occurrence, and derives the secondary risk index corresponding to the high-risk area;

[0047] Comprehensive benchmarking module: Compare the multi-source data set corresponding to any secondary risk index with similar data from the early stages of different geological disasters in history, and calculate the comprehensive similarity; set the benchmarking threshold, traverse all geological disaster event samples in the historical database, screen several similar samples, use the KNN algorithm for level mapping, and output the preliminary disaster risk level;

[0048] Linkage analysis module: analyze the interval region between the two adjacent exceeding secondary risk indexes, judge whether the interval region and the two adjacent exceeding secondary risk indexes meet the set constraint condition, and trigger the linkage risk assessment mechanism when meeting, for generating a joint risk index; judge whether the joint risk index exceeds the standard index threshold;

[0049] If not, the preliminary disaster risk level is taken as the final disaster risk level;

[0050] If yes, perform a disaster level correction action, and send the corresponding warning level based on the final disaster risk level;

[0051] Adaptive optimization module: under the condition of continuous accumulation of secondary risk index R_sy, generate the probability distribution interval of secondary risk index R_sy using an AI model; regularly update the database used in KNN algorithm and retrain the AI model using new samples; wherein the AI model is any one of GPR, BNN and random forest.

[0052] The application provides a geological disaster identification monitoring and early warning system and method based on AI technology, which has the following beneficial effects:

[0053] (1) The scheme realizes comprehensive and effective risk assessment of geological disaster hidden danger points by adopting multi-source data set fusion, achieves high-precision risk identification effect, which not only improves the detection capability of potential landslide area, but also solves the problem of inaccurate risk assessment caused by single data source in traditional method;

[0054] (2) The scheme adopts the technical scheme of predicting landslide probability by means of GBDT model and correcting primary risk index, realizes dynamic adjustment from primary risk index to secondary risk index, and completes further accurate risk quantification effect, the linkage mechanism not only considers geological conditions, but also takes into account the influence of real-time meteorological changes, effectively solves the problem that the traditional fixed parameter model cannot adapt to complex environmental changes, and ensures the timeliness and accuracy of early warning action;

[0055] (3) The primary risk index generated in this scheme can, on the one hand, provide the basic input for risk modeling of subsequent models, so that it can deeply identify whether there is high geological instability in high-risk areas while considering the influence of meteorological or linkage effects; on the other hand, it can also serve as an important basis for dynamic parameter adjustment, used to guide the adaptive adjustment of the correction coefficient α. By counting the number of high-risk areas and their corresponding primary risk index distribution, the risk concentration and difference in the entire target area can be judged, thereby determining the weight of the landslide probability in the calculation of the secondary risk index. This mechanism realizes the transition from "static assessment" to "dynamic response", improving the sensitivity of the secondary risk index to local high-risk events and the robustness of the overall system.

[0056] (4) This scheme adopts the technical scheme of comprehensive similarity calculation and KNN algorithm to realize the intelligent matching of current high-risk areas with historical disaster cases, achieve the effect of intelligent level mapping, and improve the scientificity and reliability of disaster level determination. By comparing the similarity between the current feature value and the historical disaster sample, the closest historical case can be quickly located, and the risk level of the current area can be determined accordingly, further enhancing the adaptive ability and response speed of the early warning system. Different from the traditional manual setting of thresholds that emphasizes response speed, and the existing scheme that emphasizes accuracy by comprehensively calculating the required threshold based on historical data, this method can effectively balance the accuracy and response speed of risk determination and meet actual needs.

[0057] (5) This plan adopts the relevant technologies based on the joint risk index model and the modified disaster level to achieve the quantitative assessment of potential chain reactions between adjacent high-risk areas, thus achieving the effect of regional risk prevention and control closed loop. It solves the problem of traditional isolated analysis of high-risk points ignoring the surrounding impacts, making the early warning plan more comprehensive and flexible. By effectively predicting and responding to possible chain disaster events, the original risk level can be changed, and corresponding warnings can be sent to attract higher levels of attention, thereby improving the robustness and predictability of the overall plan. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 This is a flowchart of the overall steps of a geological disaster identification, monitoring and early warning method based on AI technology in the present invention. DETAILED DESCRIPTION

[0059] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0060] Example 1:

[0061] See also Figure 1 This embodiment provides a method for identifying, monitoring, and warning geological hazards based on AI technology. This method implements a full-process description of local geological hazard identification and warning, covering key steps from data collection, risk modeling, landslide probability prediction, historical similarity analysis, regional linkage effect assessment, and AI adaptive optimization. This method is suitable for high-precision geological hazard risk warning scenarios based on the fusion of multi-source remote sensing, meteorological, and geological data, and can provide strong technical support for disaster prevention and mitigation under different conditions.

[0062] The specific steps of this method are as follows:

[0063] S1. Data acquisition and preprocessing:

[0064] S1.1 Data Collection:

[0065] Collect multi-source data sets in the target area; the multi-source data sets include at least: SAR data, drone aerial imagery, geographic information, and meteorological data; form a grid or cell structure in the target area to facilitate subsequent monitoring;

[0066] Remote sensing (SAR) data include: cumulative deformation and deformation rate in different sub-areas (which can be high-risk areas);

[0067] Drone aerial images include: several surface images;

[0068] Geographic information includes: terrain data (DEM) and slope layers;

[0069] Meteorological data include: rainfall and temperature in the target area;

[0070] It should be noted that SBAS-InSAR technology is used to analyze changes in surface deformation and extract the deformation rate and cumulative deformation under a preset period; for example: the deformation rate is 40 mm / year and the cumulative deformation is 300 mm; drones equipped with camera probes are used to collect high-resolution surface images; terrain data are collected online through the Internet of Things technology (or LiDAR); GIS software (such as ArcGIS or QGIS) is used to analyze the terrain of the target area and generate factor layers such as slope and aspect, which include the required slope layer; rainfall and temperature in meteorological data are obtained through meteorological stations on the public network; among them, for multi-source data sets, most of the data can be obtained through robot field data collection, which further ensures the real-time and validity of the data.

[0071] S1.2. Hazard marking:

[0072] Mark the known geological disaster risk points in the target area to obtain several marking points;

[0073] Since the geological disaster type targeted by this embodiment is landslide, the landslide location needs to be marked; for example, there are 18 geological disaster potential points in the target area A (usually determined independently based on known information);

[0074] S1.3 Data preprocessing:

[0075] Preprocessing operations are performed on the original multi-source datasets; the preprocessing operations include at least: denoising, registration, and spatiotemporal alignment with meteorological data; for example, using the GAMMA software platform to connect the spatiotemporal baselines of SAR data and synchronize the meteorological data with remote sensing images.

[0076] S2. Identification of high-risk areas and calculation of primary risk index:

[0077] S2.1. Slope analysis and identification:

[0078] Under the target area, based on the slope layer in the geographic information, identify areas with a slope greater than 45° as high-risk areas. The number of high-risk areas can be one or more.

[0079] Example: Generate a slope layer in ArcGIS to identify high-risk areas with slopes greater than 45°, and select them in the image to obtain a complete image of irregular or regular high-risk areas;

[0080] S2.2. Feature Screening Summary:

[0081] Based on the multi-source data set and hidden danger annotation, the feature set is screened for any high-risk area; the feature set includes at least: cumulative deformation (unit: mm), deformation rate (unit: mm / year) and marker density (unit: points / km 2 );

[0082] S2.3. Primary risk index modeling:

[0083] Based on the feature set, the primary risk index R_py corresponding to each high-risk area is calculated in the following way:

[0084] ;

[0085] Where x i : the i-th eigenvalue, such as any value in the feature set;

[0086] f i(): Normalization or standardization function, used to eliminate dimensional differences; specifically, Min-Max normalization or Z-score normalization can be used, w i : The weight of each eigenvalue is determined by the AHP or the coefficient of variation method; i = 1, 2, ..., n, where n is the total number of eigenvalues ​​involved in modeling. In this embodiment, the feature set is: cumulative deformation, deformation rate, and marker density, so n = 3; the cumulative deformation, deformation rate, and marker density are all collected within a preset period;

[0087] It should be noted that, in this embodiment, the weight is determined by the coefficient of variation method, which is a method of assigning weights to each indicator based on the degree of variation between the current value of each evaluation indicator and the target value. If the numerical difference of a certain indicator is large, it can clearly distinguish the evaluated objects, indicating that the indicator has rich discrimination information, and thus the indicator should be given a larger weight. On the contrary, if the numerical difference of each evaluated object on a certain indicator is small, then the ability of this indicator to distinguish the evaluation objects is weak, and thus the indicator should be given a smaller weight. This method directly uses the information contained in each indicator to obtain the weight of the indicator through calculation, and therefore is objective.

[0088] Effect description: By adopting a technical solution of multi-source data set fusion, a comprehensive and effective risk assessment of geological hazard potential points was achieved, achieving the effect of high-precision risk identification. This not only improved the detection capability of potential landslide areas, but also solved the problem of inaccurate risk assessment caused by a single data source in traditional methods.

[0089] S3. Landslide probability prediction:

[0090] S3.1. GBDT model construction:

[0091] Using the GBDT model, the input is the rainfall and temperature change, and the output is the probability of landslide occurrence;

[0092] Among them, the Gradient Boosted Decision Tree (GBDT) algorithm is used in the GBDT model to train a classification model based on historical landslide events and meteorological conditions;

[0093] When the GBDT model is run:

[0094] Input variables: cumulative rainfall P_rain (unit: mm) and temperature change (amplitude) T_change (unit: °C) over a preset historical period; output variable: probability of landslide occurrence P_landslide, where P_landslide∈[0,1];

[0095] Example: Given the rainfall (over 100 mm) and temperature change (average temperature below 10°C) over the past week, the GBDT model predicts a 70% probability of landslide occurrence in target area A.

[0096] S4. Modify the primary risk index to obtain the secondary risk index:

[0097] S4.1. Construct a risk index correction model to generate a secondary risk index R_sy for high-risk areas based on the probability of landslide occurrence. The method is as follows:

[0098] ;

[0099] Where, α is the correction coefficient, which reflects the impact of landslide occurrence probability on risk index;

[0100] Logical explanation: The higher the landslide probability, the more appropriately the risk index should be amplified. The two are positively correlated. The logarithmic form makes the model more robust and avoids extreme value disturbances. Specifically, the logarithmic function is used to prevent the index from exploding when the landslide probability approaches 1. The weight α is regionally adjusted based on the regional risk situation, triggering the parameter adjustment sub-model built into the risk index correction model.

[0101] The recommended value range of the correction coefficient α is greater than 0;

[0102] Effect description: The technical solution of predicting the probability of landslides with the help of GBDT model and correcting the primary risk index has realized the dynamic adjustment from the primary risk index to the secondary risk index, achieving further precise risk quantification effect. This linkage mechanism not only takes into account geological conditions, but also incorporates the impact of real-time meteorological changes, effectively solving the problem that traditional fixed parameter models cannot adapt to complex environmental changes; for example, through the analysis of rainfall and temperature changes, the possibility of landslides can be updated in a timely manner, and then the secondary risk index can be adjusted to ensure the timeliness and accuracy of the early warning system.

[0103] In this embodiment, the operation of the parameter adjustment sub-model is to adjust the correction coefficient α based on the number of high-risk areas in the entire target area and the primary risk index of each high-risk area. This parameter adjustment method can more accurately reflect the overall geological disaster risk situation in the area. The operation process of the parameter adjustment sub-model is as follows:

[0104] S4.1.1. Calculate the number of high-risk areas and the distribution characteristics of the primary risk index:

[0105] Collect the number of all identified high-risk areas N_r and their corresponding primary risk index R_py in the target area jwherein, j = 1, 2,..., N_r, j is a positive integer, representing the number corresponding to the high-risk area; the mean R_py_avg and the standard deviation σ_py of the primary risk index R_py are calculated j ;

[0106] The mean is: ;

[0107] The standard deviation is: ;

[0108] The adjustment formula for the correction coefficient a is determined as follows:

[0109] ;

[0110] wherein, a0: initial weight, the value range is [0.2, 0.4];

[0111] q1 and q2: adjustment coefficients, used to control the influence degree of the number of high-risk areas and the standard deviation of the primary risk index on a, the value range of each is [0, 1], and the recommended values are 0.1 and 0.2 respectively; N_total: the number of grids or units under the target area; therefore, the adjustment logic of the correction coefficient a is:

[0112] Logic 1: if the number of high-risk areas N_r is large and the standard deviation σ_py of the primary risk index is large, it indicates that there is a significant risk difference in the target area, and the correction coefficient a needs to be increased to enhance the influence of the landslide occurrence probability on the secondary risk index, so as to ensure that the high-risk area is fully valued; Logic 2: on the contrary, if the number of high-risk areas N_r or the standard deviation σ_py of the primary risk index is small, it indicates that the risk in the target area is relatively uniform, and the correction coefficient a can be appropriately reduced to avoid overemphasizing the influence of the local high-risk area;

[0113] In summary, this can better capture the potential geological disaster risk in the area;

[0114] Effect description: the primary risk index plays a basic assessment role in geological disaster identification and early warning, which is a preliminary quantitative expression of the potential geological disaster risk in the area, and comprehensively considers key features such as deformation rate, slope, historical marker point density, etc. Through the fusion of these features, an initial index reflecting the stability degree of the high-risk area is formed;

[0115] On the one hand, the primary risk index provides the basic input for risk modeling of subsequent models, enabling them to deeply identify whether there is high geological instability in high-risk areas while considering the influence of meteorological or linkage effects; on the other hand, the primary risk index can also serve as an important basis for dynamic parameter adjustment, used to guide the adaptive adjustment of the correction coefficient α. By counting the number of high-risk areas and their corresponding primary risk index distribution (such as mean and standard deviation), the degree of risk concentration and diversity in the entire target area can be judged, thereby determining the weight of the landslide probability in the calculation of the secondary risk index. This mechanism effectively solves the problem that fixed parameters in traditional methods cannot adapt to complex regional differences, realizes the transition from "static assessment" to "dynamic response", and improves the sensitivity of the secondary risk index to local high-risk events and the robustness of the overall system.

[0116] S5. Comprehensive similarity comparison and risk level benchmarking:

[0117] S5.1. Comprehensive similarity calculation:

[0118] Compare the multi-source dataset corresponding to the secondary risk index R_sy in any high-risk area with similar data from the early stages of different historical geological disasters (referring to landslides), and calculate the comprehensive similarity S_siy:

[0119] ;

[0120] Where, X i1 cu : The i1th factor value in the current multi-source dataset (including at least cumulative deformation, deformation rate, slope, historical marker density, rainfall and temperature), X i1 hi : The corresponding i1-th factor value in the historical multi-source dataset, i1: the corresponding weight coefficient, i1=1, 2,..., n1, n1 represents the number of factor values ​​(i.e., factor dimension). If the factor values ​​include: cumulative deformation, deformation rate, slope, historical marker density, rainfall, and temperature; then n1=6;

[0121] X i1 max and X i1 min : The historical maximum and minimum values ​​of the factor;

[0122] The final output S_sit∈[0,1] is a value where the larger the value, the more similar it is. If the two are completely identical, the value is 1; if they are completely different, the value is close to 0. This conversion method is intuitive and easy to set the threshold for subsequent judgment.

[0123] S5.2. Risk level benchmarking mechanism:

[0124] Set the benchmark threshold S_th, for example: set S_th=0.75;

[0125] Traverse all geological disaster event samples in the historical database, screen out several similar samples that meet S_siy≥S_th, use the KNN algorithm or fuzzy membership function for level mapping, and output the preliminary disaster risk level corresponding to the current high-risk area; Among them, the existing geological disaster risk level is divided into four levels: low, medium, high and very high. The rules for setting this level are existing, so I will not elaborate on it here.

[0126] In addition, the confidence level can be output, supporting interpretations such as "medium-level probability 80%";

[0127] The required warning levels corresponding to different risk levels can be obtained by referring to Table 1: Table 1: Comparison of risk levels and warning levels:

[0128]

[0129] As can be seen from Table 1, different risk levels display different warning levels;

[0130] Example: In this embodiment, the KNN algorithm is used for level mapping. Five samples with a similarity greater than 0.75 with the current high-risk area are found in the historical database. Their disaster levels are medium, medium, medium, high, and medium respectively.

[0131] Therefore, the voting results are: "Medium" accounts for 4 votes, "High" accounts for 1 vote → the current high-risk area is judged to be "Medium" level, so the preliminary disaster risk level is "Medium".

[0132] Effect description: By using the technical solution of comprehensive similarity calculation and KNN algorithm or fuzzy membership function, we can realize the intelligent matching of current high-risk areas with historical disaster cases, achieve the effect of intelligent level mapping, and improve the scientificity and reliability of disaster level determination. By comparing the similarity of current feature values ​​with historical disaster samples, we can quickly locate the closest historical case and determine the risk level of the current area accordingly, further enhancing the adaptive ability and response speed of the early warning system.

[0133] Unlike the subjectivity and limitations of traditional manually set thresholds (with the advantage of faster response speed), and the existing solution's improvement over manually set thresholds, which uses comprehensive calculations based on historical data to derive the required thresholds (with the advantage of higher accuracy), this method can effectively balance the accuracy and response speed of risk assessments to meet actual needs.

[0134] S6. Analysis of linkage effects between adjacent regions:

[0135] S6.1. Regional interval and geological conditions analysis:

[0136] Analyze the interval between two adjacent excessive secondary risk indices and determine whether the interval and the two adjacent excessive secondary risk indices meet the set constraints. The constraints are:

[0137] Condition 1: Are the geological conditions similar or identical?

[0138] Condition 2: Whether the maximum distance between two high-risk areas with excessive secondary risk indices is lower than the calibration value;

[0139] Among them, for several high-risk areas with a historical corresponding risk level of "medium", their respective secondary risk indexes are calculated, and the average is taken as the standard index. If the secondary risk index exceeds the standard index, it is recorded as exceeding the standard;

[0140] When determining condition one, the following methods can be used:

[0141] Acquire key layer data: including geological maps, slope maps (generated from DEM), and InSAR deformation maps; extract target area attributes: for two adjacent high-risk areas and the area between them, extract their stratum lithology (from geological maps), average slope (from DEM), and average annual deformation rate (from InSAR); compare and analyze three indicators: lithology consistency: if the three areas have the same lithology type (for example, all sandstone or shale), the geological foundation is consistent; slope difference: if the slope difference is less than 5°, the terrain conditions are considered similar; deformation rate proximity: if the rate difference is less than 5 mm / yr, the surface stability trends are consistent; comprehensive judgment: if any two of the three indicators are consistent or highly similar, it can be determined that the geological conditions of the interval area and the adjacent area are the same or similar.

[0142] Note: The above method is simple to operate and quick to respond. It is suitable for the rapid identification of spatially continuous areas in automated early warning methods. In actual application, other methods can also be used to determine whether the geological conditions are the same or similar. However, it should be noted that if the interval area is discontinuous, it is directly determined that condition one is not met. The reason is that if the area is discontinuous, a landslide in one of the high-risk areas will have no influence or limited influence on the adjacent high-risk areas.

[0143] In the second judgment condition, the specific value of the calibration value can be set to 1km, which can be set independently according to actual needs.

[0144] S6.2. Assessment of linkage effects:

[0145] When the constraints are met, the linkage risk assessment mechanism is triggered (to consider the impact of a high-risk area with a larger secondary risk index on another high-risk area, and to assess whether there is a linkage effect. If so, it may have the effect of expanding the risk level). A linkage risk index model is constructed, and the two exceeded secondary risk indices and the geographical distance between the center points of the high-risk areas corresponding to the two exceeded secondary risk indices are input, and the joint risk index R_cd is output;

[0146] When running the linkage risk index model, the formula is as follows:

[0147] ;

[0148] Where, R_sy A and Résy B : The secondary risk index of exceeding the standard for two adjacent high-risk areas, β: The linkage enhancement coefficient, the value range is [0, 1], γ: The distance attenuation coefficient, which controls the speed at which the linkage effect weakens with distance, the value range is [0, 0.1]; d: The geographical distance value;

[0149] Logical explanation: The joint risk index formula scientifically quantifies the linkage effect of geological disasters between regions by comprehensively considering the secondary risk index and spatial distance relationship of two adjacent high-risk areas. The larger risk value is used as the basic term in the formula to reflect the influence of the dominant risk area. At the same time, the average risk and the linkage enhancement coefficient β are introduced to construct the enhancement term, and the exponential decay function is combined to reflect the weakening effect of distance on the linkage effect, thereby realizing a dynamic assessment of the potential induced risks between adjacent regions. This method can not only identify the mutual influence between high-risk areas, but also adjust the linkage intensity according to the actual geographical distance. While improving the accuracy of the geological disaster early warning system, it enhances the prediction and prevention and control capabilities of regional chain disasters.

[0150] S6.3. Modified disaster level:

[0151] Compare the joint risk index with the standard index threshold. If the joint risk index exceeds the standard index threshold, it indicates that there is a disaster linkage effect between the two high-risk areas. Execute the disaster level correction action, mark the two adjacent high-risk areas as a whole area, and increase the initial disaster risk level of the lower one by one to obtain the final disaster risk level. Refer to Table 1 and issue the corresponding warning level based on the final disaster risk level.

[0152] For example, if two adjacent high-risk areas with a "medium" level are marked as a whole area, the final disaster level will be "high". If one area is "medium" and the other is "high", the "medium" level will be upgraded to "high", and the final disaster level for the whole area will be "high".

[0153] If the joint risk index does not exceed the standard index threshold, it means that there is no disaster linkage effect between the two high-risk areas or the disaster linkage effect can be ignored. No response action is taken, and the corresponding preliminary disaster risk levels of the two adjacent high-risk areas are used as the final disaster risk levels. Here, it is also necessary to refer to Table 1 and send the corresponding warning level based on the final disaster risk level.

[0154] For example, two adjacent high-risk areas with a "medium" level will both receive a "medium" disaster level.

[0155] Among them, the method of obtaining the standard index threshold is the same as the principle of obtaining the above-mentioned standard indicators, so it is not elaborated here. The standard index threshold is used to determine whether the disaster linkage effect is serious enough to require re-determination of the disaster risk level.

[0156] Effect description: Based on the joint risk index model and the technical solution of correcting disaster levels, a quantitative assessment of potential chain reactions between adjacent high-risk areas is achieved, achieving the effect of a regional risk prevention and control closed loop. This solution solves the problem of traditional isolated analysis of high-risk points ignoring the surrounding impacts, making the early warning system more comprehensive and flexible; specifically, by calculating the joint risk index between two exceeded secondary risk indices and adjusting the linkage strength according to the actual geographical distance, it is possible to effectively predict and respond to possible chain disaster events, thereby changing the original risk level and sending corresponding warnings to attract higher levels of attention, thereby improving the robustness and foresight of the overall solution.

[0157] S7. AI adaptive optimization mechanism as data accumulates (can be AI adaptive optimization without triggering S6 or adjacent region linkage effect analysis):

[0158] S7.1. Data accumulation and range value formation:

[0159] As data continues to accumulate, the secondary risk index will form a range of values ​​rather than a single value. Specifically, any AI model such as GPR / BNN / Random Forest will be used to generate the probability distribution interval of the secondary risk index;

[0160] Example: The secondary risk index R_sy corresponding to a high-risk area is initially 40 → after accumulation it becomes [38, 46];

[0161] S7.2. More effective benchmarking:

[0162] Regularly update the database used in the KNN algorithm and retrain the model with new samples to improve generalization ability;

[0163] Specifically, the application of AI technology is reflected in the following: Gaussian process regression (GPR) or Bayesian neural network (BNN) can be used to model the distribution characteristics of the secondary risk index; the model outputs a distribution with mean and variance based on input features (such as deformation rate, slope, rainfall, etc.); the final output is a confidence interval (for example: 95% confidence interval [35, 45]), rather than a single value; as new data is added, the model continuously updates parameters, so that the prediction interval gradually converges and stabilizes; through more historical data, the matching between the secondary risk index and the known preliminary disaster risk level is more accurate; example: assuming that 100 similar cases have been accumulated and it is found that when the secondary risk index is between 35 and 45, the disaster level is mostly "medium", then the risk level of the current area can be more accurately matched, and at this time S5.2 and the risk level matching mechanism play a verification role; realizing the intelligent evolution from "static rule judgment" to "data-driven + AI-driven".

[0164] Example 2:

[0165] Based on Example 1, this embodiment further provides a geological disaster identification, monitoring and early warning system based on AI technology, which includes:

[0166] Data acquisition module: collects multi-source data sets in the target area, pre-processes the multi-source data sets, and annotates hidden dangers to obtain a number of marked points. The multi-source data sets include at least: SAR data, drone aerial images, geographic information, and meteorological data;

[0167] Risk Identification Module: Within the target area, areas with slopes greater than 45° are identified as high-risk areas based on geographic information. Based on multi-source datasets and hidden danger annotations, a feature set is screened for any high-risk area, and the primary risk index corresponding to each high-risk area is calculated based on the feature set.

[0168] Probability prediction module: adopts GBDT model, with meteorological data as input and landslide occurrence probability as output;

[0169] Risk correction module: Constructs a risk index correction model to correct the primary risk index based on the probability of landslide occurrence, and derives the secondary risk index corresponding to the high-risk area;

[0170] Comprehensive benchmarking module: Compare the multi-source data set corresponding to any secondary risk index with similar data from the early stages of different geological disasters in history, and calculate the comprehensive similarity; set the benchmarking threshold, traverse all geological disaster event samples in the historical database, screen several similar samples, use the KNN algorithm for level mapping, and output the preliminary disaster risk level;

[0171] Linkage Analysis Module: Analyzes the interval between two adjacent excessive secondary risk indices, determines whether the interval and the two adjacent excessive secondary risk indices meet the set constraints, and triggers the linkage risk assessment mechanism to generate a joint risk index if the constraints are met; determines whether the joint risk index exceeds the standard index threshold;

[0172] If not, the preliminary disaster risk level will be used as the final disaster risk level;

[0173] If so, the disaster level correction action is performed and the corresponding warning level is sent based on the final disaster risk level;

[0174] Adaptive optimization module: Under the condition of continuous accumulation of the secondary risk index R_sy, use the AI ​​model to generate the probability distribution interval of the secondary risk index R_sy; regularly update the database used in the KNN algorithm, and retrain the AI ​​model with new samples; among them, the AI ​​model is: any one of GPR, BNN and random forest.

[0175] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those skilled in the art will appreciate that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution.

[0176] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, and may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment as needed.

[0177] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. A method for identifying, monitoring and warning geological hazards based on AI technology, comprising: Collect a multi-source data set in the target area, pre-process the multi-source data set, and mark hidden dangers to obtain a number of marking points; wherein the multi-source data set includes at least: SAR data, drone aerial images, geographic information and meteorological data; the method is characterized in that: Within the target area, areas with slopes greater than 45° are identified as high-risk areas based on geographic information. Based on multi-source datasets and hidden danger annotations, feature sets are screened within any high-risk area, and the primary risk index corresponding to each high-risk area is calculated based on the feature sets. The GBDT model is used, with meteorological data as input and the output as the probability of landslide occurrence; Construct a risk index correction model to correct the primary risk index based on the probability of landslide occurrence, and derive the secondary risk index corresponding to the high-risk area; Compare the multi-source data set corresponding to any secondary risk index with similar data from the early stages of different geological disasters in history to calculate the comprehensive similarity; set a benchmarking threshold, traverse all geological disaster event samples in the historical database, screen several similar samples, use the KNN algorithm for level mapping, and output the preliminary disaster risk level; Analyze the interval between two adjacent excessive secondary risk indices and determine whether the interval and the two adjacent excessive secondary risk indices meet the set constraints. If so, trigger the linkage risk assessment mechanism to generate a joint risk index; determine whether the joint risk index exceeds the standard index threshold; If not, the preliminary disaster risk level will be used as the final disaster risk level; If so, the disaster level correction action is performed and the corresponding warning level is sent based on the final disaster risk level.

2. The method for identifying, monitoring and warning geological disasters based on AI technology according to claim 1, characterized in that: The process of preprocessing multi-source datasets includes at least: denoising, registration, and spatiotemporal alignment in combination with meteorological data; SAR data at least include: cumulative deformation and deformation rate; UAV aerial images include at least: several surface images; Geographic information shall include at least: terrain data and slope layers; Meteorological data include at least: rainfall and temperature.

3. The method for identifying, monitoring and warning geological disasters based on AI technology according to claim 2, characterized in that: The feature set includes at least: cumulative deformation, deformation rate and marker density; The primary risk index R_py corresponding to each high-risk area is calculated based on the following method: ； Where x i : the i-th eigenvalue, f i (): normalization function, w i : The weight of each eigenvalue, i=1, 2, ..., n, where n is the total number of eigenvalues.

4. The method for identifying, monitoring and warning geological disasters based on AI technology according to claim 1, characterized in that: The secondary risk index R_sy corresponding to the high-risk area is generated according to the following method: ； Where, α: correction coefficient, P_landslide: probability of landslide occurrence; Collect the number of all identified high-risk areas N_r and their corresponding primary risk index R_py in the target area j Where j = 1, 2, ..., N_r, calculate the primary risk index R_py j The mean R_py_avg and standard deviation σ_py; The correction factor α is adjusted as follows: ； Where α0 is the initial weight; q1 and q2 are adjustment coefficients, both ranging from [0, 1]; N_total is the number of grids in the target area.

5. The method for identifying, monitoring and warning geological disasters based on AI technology according to claim 1, characterized in that: Calculate the comprehensive similarity S_siy and select the conditions for several similar samples as follows: The comprehensive similarity S_siy ≥ the benchmark threshold S_th is satisfied.

6. The method for identifying, monitoring and warning geological disasters based on AI technology according to claim 1, characterized in that: The constraints are: Condition 1: Are the geological conditions similar or identical? Condition 2: Whether the maximum distance between two high-risk areas with excessive secondary risk indices is lower than the calibration value.

7. The method for identifying, monitoring and warning geological disasters based on AI technology according to claim 1, characterized in that: The process of triggering the linkage risk assessment mechanism is as follows: Construct a linkage risk index model, input two exceeded secondary risk indices and the geographical distance between the center points of the high-risk areas corresponding to the two exceeded secondary risk indices, and output the joint risk index R_cd; When running the linkage risk index model, the method is as follows: ； Where, R_sy A and Résy B : The secondary risk index of exceeding the standard in two adjacent high-risk areas, β: linkage enhancement coefficient, ranging from [0, 1], γ: distance attenuation coefficient, ranging from [0, 0.1]; d: geographical distance value.

8. The method for identifying, monitoring and warning geological disasters based on AI technology according to claim 1, characterized in that: The process of performing the disaster level correction action is to mark two adjacent high-risk areas as one overall area, and to increase the initial disaster risk level of the area with the lower level by one level to obtain the final disaster risk level.

9. The method for identifying, monitoring and warning geological disasters based on AI technology according to claim 1, characterized in that: The method also includes: under the condition that the secondary risk index R_sy continues to accumulate, using the AI ​​model to generate a probability distribution interval of the secondary risk index R_sy; regularly updating the database used in the KNN algorithm, and retraining the AI ​​model with new samples; wherein the AI ​​model is: any one of GPR, BNN and random forest.

10. A geological disaster identification, monitoring and early warning system based on AI technology, the system comprising: Data acquisition module: collects multi-source data sets in the target area, pre-processes the multi-source data sets, and marks hidden dangers to obtain a number of marked points; wherein the multi-source data sets include at least: SAR data, drone aerial images, geographic information and meteorological data; the system is characterized in that: Risk Identification Module: Within the target area, areas with slopes greater than 45° are identified as high-risk areas based on geographic information. Based on multi-source datasets and hidden danger annotations, a feature set is screened for any high-risk area, and the primary risk index corresponding to each high-risk area is calculated based on the feature set. Probability prediction module: adopts GBDT model, with meteorological data as input and landslide occurrence probability as output; Risk correction module: Constructs a risk index correction model to correct the primary risk index based on the probability of landslide occurrence, and derives the secondary risk index corresponding to the high-risk area; Comprehensive benchmarking module: Compare the multi-source data set corresponding to any secondary risk index with similar data from the early stages of different geological disasters in history, and calculate the comprehensive similarity; set the benchmarking threshold, traverse all geological disaster event samples in the historical database, screen several similar samples, use the KNN algorithm for level mapping, and output the preliminary disaster risk level; Linkage Analysis Module: Analyzes the interval between two adjacent excessive secondary risk indices, determines whether the interval and the two adjacent excessive secondary risk indices meet the set constraints, and triggers the linkage risk assessment mechanism to generate a joint risk index if the constraints are met; determines whether the joint risk index exceeds the standard index threshold; If not, the preliminary disaster risk level will be used as the final disaster risk level; If so, the disaster level correction action is performed and the corresponding warning level is sent based on the final disaster risk level; Adaptive optimization module: Under the condition of continuous accumulation of the secondary risk index R_sy, use the AI ​​model to generate the probability distribution interval of the secondary risk index R_sy; regularly update the database used in the KNN algorithm, and retrain the AI ​​model with new samples; among them, the AI ​​model is: any one of GPR, BNN and random forest.

Images