A landslide risk prediction method and system based on data intelligent analysis

Abstract

The present invention belongs to the technical field of risk prediction. The present invention discloses a landslide risk prediction method and system based on data intelligent analysis, including: obtaining corresponding multi-source heterogeneous information and performing feature extraction and data fusion on it to obtain a mountain monitoring basic data set; constructing a mountain twin feature model based on the mountain monitoring basic data set, and simulating landslide scenarios based on it to obtain corresponding risk feature information; constructing a trigger factor association network based on the risk feature information, and performing zoning assessment and grading of target landslide risks in combination with the mountain monitoring basic data set; constructing a corresponding graded risk zoning map; obtaining landslide risk dynamic prediction results based on the graded risk zoning map; and constructing a corresponding graded early warning rule library; performing real-time monitoring of the target mountain area based on the graded early warning rule library and providing feedback on the monitoring results. The present invention provides a comprehensive and efficient solution for the prevention and control of landslide disasters.

Description

Technical Field

[0001] The present invention relates to the technical field of risk prediction, and more specifically, to a method and system for predicting landslide risks based on data intelligent analysis. Background Art

[0002] Landslides are a common geological disaster; as human activities continue to impact the natural environment, their frequency and severity are increasing. Therefore, establishing an efficient and accurate landslide risk prediction system is of great significance for disaster prevention and mitigation.

[0003] Compared with existing technologies, the traditional risk prediction process has the following main problems:

[0004] Traditional landslide monitoring systems typically use a single or limited data source. The information obtained is limited and cannot fully reflect the actual state of the mountain. Existing landslide risk assessment methods are mostly based on empirical formulas or simplified models, which cannot accurately simulate landslide processes under complex geological conditions. Traditional early warning systems usually use a single indicator or a simple threshold method for early warning, such as cumulative rainfall or simple displacement rate judgment. These early warning indicators lack scientific basis, and the threshold setting is mostly based on empirical judgment. In addition, different regions often simply apply the same standards, resulting in low accuracy and reliability of early warnings.

[0005] In view of this, the present invention proposes a landslide risk prediction method and system based on data intelligent analysis to solve the above problems. Summary of the Invention

[0006] In order to overcome the above-mentioned defects of the prior art and achieve the above-mentioned objectives, the present invention provides the following technical solutions:

[0007] A landslide risk prediction system based on data intelligent analysis, including:

[0008] The data acquisition module is used to monitor and collect data on the geological and meteorological parameters of the target mountain area in real time, obtain corresponding multi-source heterogeneous information, and perform feature extraction and data fusion on the collected multi-source heterogeneous information to obtain the corresponding mountain monitoring basic data set;

[0009] The risk simulation module builds a mountain twin feature model based on the basic mountain monitoring data set, and simulates multiple landslide scenarios based on it to obtain the corresponding simulated risk data. Based on the simulated risk data, it obtains the risk feature information corresponding to different landslide scenarios;

[0010] The risk assessment module builds a trigger factor association network based on risk characteristic information, and combines the basic mountain monitoring data set to conduct zoning assessment and classification of target landslide risks; and constructs the corresponding graded risk zoning map;

[0011] The risk prediction module conducts spatiotemporal evolution analysis and prediction analysis based on the graded risk zoning map to obtain the corresponding dynamic prediction results of landslide risks; and establishes an early warning indicator system and sets thresholds based on the dynamic prediction results of landslide risks to obtain the corresponding graded early warning rule library; based on the graded early warning rule library, it conducts real-time monitoring of the target mountain area and provides feedback on the monitoring results.

[0012] Furthermore, the process of obtaining the basic mountain monitoring dataset includes:

[0013] Performing data verification and outlier screening on the obtained multi-source heterogeneous information; and performing data unification on the multi-source heterogeneous data after outlier screening, wherein the data unification includes time alignment and spatial registration;

[0014] Multi-source data fusion is performed on the unified multi-source heterogeneous data, and the multi-source fused data is structured and organized to obtain the corresponding mountain monitoring basic data set.

[0015] Furthermore, the basic mountain monitoring dataset contains three dimensions: time, space, and attributes; the time dimension records the observation time series, the space dimension contains geographic coordinates and elevation information, and the attribute dimension contains various monitoring parameters and derived indicators.

[0016] Furthermore, the process of obtaining risk characteristic information includes:

[0017] Performing three-dimensional modeling based on the mountain monitoring basic data set to obtain a mountain twin feature model corresponding to the target mountain area;

[0018] Collect and organize historical landslide event data; and obtain known landslide scenarios based on them;

[0019] Conduct landslide scenario simulation based on the mountain twin characteristic model to obtain simulated risk data under different types of landslide scenarios;

[0020] Acquire simulated risk data corresponding to several corresponding landslide scenarios; construct a two-dimensional rectangular coordinate system with the monitoring parameters in the simulated risk data as the vertical axis and time as the horizontal axis, and map the acquired simulated risk data into the two-dimensional rectangular coordinate system to obtain the corresponding risk evolution; extract risk characteristic information in the risk evolution diagram, the risk characteristic information including parameter change trend, mutation characteristics, and critical point characteristics;

[0021] Based on the above risk characteristic information extraction process, the risk characteristic information corresponding to other landslide scenarios is obtained.

[0022] Furthermore, the process of constructing the trigger factor association network includes:

[0023] Extract monitoring indicators from the acquired historical landslide event data and mountain monitoring basic data to obtain corresponding mountain monitoring indicators; and obtain indicator characteristic information corresponding to each mountain monitoring indicator;

[0024] Combined with risk characteristic information, the correlation between mountain monitoring indicators and landslide scenarios is evaluated to obtain the support and confidence between each mountain monitoring indicator and the corresponding landslide scenario; and based on this, the correlation between each mountain monitoring indicator and the corresponding landslide scenario is obtained;

[0025] Based on the obtained correlation, the characteristic importance of the corresponding mountain monitoring indicators is analyzed to obtain the contribution of the mountain monitoring indicators to the corresponding landslide scenario; based on the obtained contribution, the corresponding mountain monitoring indicators are divided into direct triggering factors and indirect triggering factors, and a hierarchical structure corresponding to the direct triggering factors and the indirect triggering factors is established;

[0026] Randomly construct several factor combinations based on direct trigger factors and indirect trigger factors;

[0027] Conduct synergy analysis on the constructed factor combinations to obtain the corresponding synergy effects; and obtain the joint probability and conditional probability corresponding to each factor combination;

[0028] Obtain the temporal dependency between triggering factors and landslide scenarios based on historical landslide event data;

[0029] A trigger factor association network is constructed based on the obtained contribution, hierarchical structure, synergistic effect and temporal dependency; the trigger factors include direct trigger factors and indirect trigger factors.

[0030] Furthermore, the process of obtaining the graded risk zoning map includes:

[0031] Based on the constructed trigger factor association network and mountain twin feature model, the target mountain area is divided into regions to obtain several evaluation units;

[0032] Perform sensitivity assessment on the divided assessment units to obtain sensitivity assessment indexes corresponding to the assessment units; construct a coupling model based on the trigger factors and sensitivity indexes, wherein the model input of the coupling model includes the trigger factors and sensitivity indexes involved in each assessment unit, and the model output of the coupling model is the instability probability of each assessment unit;

[0033] Based on the probability of instability, each assessment unit is divided into risk levels to obtain the corresponding graded risk zoning map.

[0034] Furthermore, the process of obtaining the dynamic prediction results of landslide risk includes:

[0035] Perform time series decomposition and trend analysis on the real-time collected mountain monitoring basic data set to obtain corresponding time feature components, and build a time series prediction basic model based on the time feature components. Then, integrate the constructed time prediction basic model to obtain the corresponding time evolution prediction framework.

[0036] Based on the hierarchical risk zoning map, a spatial correlation model and a spatial association network are constructed, and spatial diffusion simulation and boundary analysis are performed through the spatial association network to obtain a spatial evolution prediction framework;

[0037] Integrate the constructed temporal evolution prediction framework and spatial evolution prediction framework to obtain the corresponding mountain spatiotemporal prediction model;

[0038] Based on the obtained mountain spatiotemporal prediction model, scenario prediction is performed to obtain scenario prediction results corresponding to the target mountain area, and confidence calculation is performed on the obtained scenario prediction results to obtain corresponding confidence intervals;

[0039] The confidence interval is integrated with the corresponding scenario prediction results to obtain the corresponding landslide risk dynamic prediction results.

[0040] Furthermore, the construction process of the hierarchical warning rule base includes:

[0041] Based on the obtained landslide risk dynamic prediction results, the obtained trigger factors are updated, and the synergistic effects between the updated trigger factors are obtained, and the corresponding early warning indicator system is constructed based on them;

[0042] Conduct data distribution analysis and critical value research on the obtained early warning indicator system to obtain the corresponding preliminary indicator threshold scheme;

[0043] Based on the historical landslide event data obtained, the effectiveness of the initial indicator threshold scheme obtained is evaluated; and based on the effectiveness evaluation results, whether to adjust the initial indicator threshold scheme is selected. If not, the initial indicator threshold scheme is output as the optimal indicator threshold scheme; if it is adjusted, the corresponding initial indicator threshold scheme is adjusted based on a pre-set adjustment strategy;

[0044] Then, the optimal indicator threshold scheme is divided into levels to obtain a warning classification framework, and warning trigger conditions and combination logic of different levels are formulated according to the warning classification framework to obtain a hierarchical warning rule base.

[0045] Furthermore, the process of real-time monitoring of the target mountain area and feedback of monitoring results based on the hierarchical warning rule library includes:

[0046] Feature extraction is performed on the real-time collected mountain basic monitoring data set to obtain the indicator characteristic information corresponding to each trigger factor, and conditional evaluation is performed on it and compared with the constructed hierarchical warning rule library; based on the conditional evaluation results, the warning level is converted to obtain the corresponding warning state vector; and based on it, the corresponding warning signal and comprehensive risk level are generated to obtain the corresponding mountain warning results and feedback to the management department of the target mountain area.

[0047] Furthermore, a landslide risk prediction method based on data intelligent analysis includes:

[0048] Step 1: Real-time monitoring and data collection of geological and meteorological parameters of the target mountain area are carried out to obtain corresponding multi-source heterogeneous information. Feature extraction and data fusion are then performed on the collected multi-source heterogeneous information to obtain the corresponding mountain monitoring basic data set.

[0049] Step 2: Build a mountain twin feature model based on the mountain monitoring basic data set, and simulate multiple landslide scenarios based on it to obtain the corresponding simulated risk data. Based on the simulated risk data, obtain the risk characteristic information corresponding to different landslide scenarios;

[0050] Step 3: Construct a trigger factor association network based on risk characteristic information, and combine it with the basic mountain monitoring data set to conduct zoning assessment and classification of the target landslide risk; and construct the corresponding graded risk zoning map;

[0051] Step 4: Conduct spatiotemporal evolution analysis and prediction analysis based on the graded risk zoning map to obtain the corresponding dynamic prediction results of landslide risk; establish an early warning indicator system and set thresholds based on the dynamic prediction results of landslide risk to obtain the corresponding graded early warning rule library; conduct real-time monitoring of the target mountain area based on the graded early warning rule library and provide feedback on the monitoring results.

[0052] The technical effects and advantages of the landslide risk prediction method and system based on data intelligent analysis of the present invention are as follows:

[0053] By constructing a twin feature model of a mountain, multiple landslide scenarios are simulated, and a trigger factor association network is established in combination with risk feature information extraction. Then, a refined zoning assessment is carried out on the target mountain area to obtain a graded risk zoning map. Based on spatiotemporal evolution analysis, a prediction model is constructed to obtain dynamic prediction results of landslide risks. Finally, a scientific early warning indicator system and a graded early warning rule library are established to achieve real-time monitoring and early warning, improve prediction accuracy and early warning timeliness, reduce disaster losses, optimize emergency resource allocation, and realize full-cycle intelligent management of landslide risks, providing a systematic solution for geological disaster prevention and control. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1Schematic diagram of a landslide risk prediction system based on data intelligent analysis according to the present invention;

[0055] Figure 2 This is a schematic diagram of a landslide risk prediction method based on data intelligent analysis according to the present invention. DETAILED DESCRIPTION

[0056] The following will clearly and completely describe 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. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0057] Example 1

[0058] See also Figure 1 As shown, the landslide risk prediction system based on data intelligent analysis described in this embodiment includes:

[0059] The data acquisition module is used to monitor and collect data on the geological and meteorological parameters of the target mountain area in real time, obtain corresponding multi-source heterogeneous information, and perform feature extraction and data fusion on the collected multi-source heterogeneous information to obtain the corresponding mountain monitoring basic data set;

[0060] The risk simulation module builds a mountain twin feature model based on the basic mountain monitoring data set, and simulates multiple landslide scenarios based on it to obtain the corresponding simulated risk data. Based on the simulated risk data, it obtains the risk feature information corresponding to different landslide scenarios;

[0061] The risk assessment module builds a trigger factor association network based on risk characteristic information, and combines the basic mountain monitoring data set to conduct zoning assessment and classification of target landslide risks; and constructs the corresponding graded risk zoning map;

[0062] The risk prediction module conducts spatiotemporal evolution analysis and prediction analysis based on the hierarchical risk zoning map to obtain dynamic prediction results of corresponding landslide risks. Based on the dynamic prediction results of landslide risks, it establishes an early warning indicator system and sets thresholds to obtain a corresponding hierarchical early warning rule base. Based on the hierarchical early warning rule base, it conducts real-time monitoring of the target mountain area and provides feedback on the monitoring results.

[0063] The modules are connected via wired and / or wireless means to achieve data transmission between modules.

[0064] It should be further explained that, in the specific implementation process, the process of acquiring multi-source heterogeneous information includes:

[0065] A distributed sensor network is used to collect data on geological parameters within the target mountain area to obtain a corresponding basic geological data set, which includes geological survey data, topographic data, and soil property data. The distributed sensor network is composed of multiple sensor terminals buried at different depths and locations, including soil moisture sensor terminals, pore water pressure sensor terminals, and displacement monitoring sensor terminals. The sensor terminals are wirelessly networked to achieve regular data collection and transmission.

[0066] Obtain meteorological monitoring data corresponding to the target mountain area through local meteorological stations, wherein the meteorological monitoring data includes meteorological parameters such as rainfall, temperature changes, and humidity changes;

[0067] Based on satellite remote sensing images and drone aerial photography data, remote sensing images corresponding to the target mountain area are obtained, and image recognition is performed on them to obtain corresponding terrain data. The terrain data includes static parameters such as slope, slope aspect, vegetation coverage, and dynamic parameters such as surface deformation;

[0068] Aggregate the collected remote sensing images, topographic data, basic geological data sets and meteorological monitoring data to obtain corresponding multi-source heterogeneous information;

[0069] One embodiment of the present invention includes, for example: installing tilt sensors and groundwater level sensors in boreholes inside the mountain to monitor internal deformation and groundwater level fluctuations; deploying soil moisture sensors and rainfall sensors on the slope to monitor rainfall infiltration; installing stress and strain sensors near potential sliding surfaces to monitor the stress state of the rock and soil; deploying automatic weather stations to collect meteorological data such as rainfall, temperature, and humidity; and integrating high-resolution remote sensing image data to obtain large-scale surface coverage and deformation information.

[0070] It should be further explained that, in the specific implementation process, the acquisition process of the basic mountain monitoring data set includes:

[0071] Data validation and outlier screening are performed on the obtained multi-source heterogeneous information. Data validation includes data format verification and data integrity verification. The outlier screening process involves using the 3σ rule and parameter thresholds set based on expert experience to identify outliers in the data parameters within the multi-source heterogeneous information and identify corresponding potential outliers. For example, if soil moisture content increases by more than 20% in a short period of time and there is no rainfall record, it can be marked as a potential outlier.

[0072] Then, the degree of deviation between the corresponding potential abnormal point and the corresponding parameter threshold range is obtained, and different processing measures are taken based on the degree of deviation;

[0073] One embodiment of the present invention includes: if the deviation is less than 5%, a linear interpolation algorithm can be used to quickly repair the potential outlier point; if the deviation is not less than 5%, different outlier processing strategies are adopted based on the data type corresponding to the potential outlier point; for example, for data with time series correlation such as displacement and moisture content, the ARIMA time series prediction algorithm is used for repair; for data with strong spatial correlation, such as surface temperature distribution, the Kriging spatial interpolation method is used for data completion;

[0074] Since there are deviations in data upload frequency and collection location during the data collection process from different data sources, the multi-source heterogeneous data after outlier screening is unified. The data unification refers to unifying the data from different sources into a standard spatiotemporal reference range, which is determined in advance by personnel in this field based on industry standards. The data unification includes time alignment and spatial registration. The time alignment solves the data time synchronization problem caused by the deviation in data upload frequency by adopting the nearest time matching algorithm. Spatial registration ensures that the data at different spatial locations have spatial consistency during analysis by adopting coordinate transformation and unified geographic reference.

[0075] Multi-source data fusion is performed on the unified multi-source heterogeneous data, and the fused data is structured to construct a basic mountain monitoring dataset with three dimensions: time, space, and attributes. The time dimension records the observation time series, the space dimension contains geographic coordinates and elevation information, and the attribute dimension contains various monitoring parameters and derived indicators. The derived indicators represent composite indicators composed of multiple monitoring parameters, such as the product of slope and soil moisture content (indicating slope saturation), the ratio of accumulated rainfall to permeability coefficient (indicating infiltration capacity), and a comprehensive indicator of displacement rate and acceleration (indicating instability trends).

[0076] It should be further explained that the multi-source data fusion integrates the same type of data within multi-source heterogeneous data through a weighted averaging algorithm to improve data reliability. For example, the displacement data measured by GNSS and InSAR technologies can be dynamically weighted and fused according to weather conditions and observation angles to obtain more reliable displacement measurement results.

[0077] It should be further explained that, in the specific implementation process, the acquisition process of simulated risk data includes:

[0078] Three-dimensional modeling is performed based on the mountain monitoring basic data set to obtain a mountain twin feature model corresponding to the target mountain area. Three-dimensional modeling refers to the process of using 3D modeling software and digital twin technology to construct a digital twin model (i.e., a mountain twin feature model) corresponding to the target mountain area. The corresponding three-dimensional modeling process is prior art and will not be elaborated in detail in the present invention.

[0079] Collect and organize historical landslide event data, including the time, location, scale, type, and environmental conditions before the occurrence of the landslide; historical landslide event data sources include local chronicles, historical disaster archives, scientific research literature, and field investigation reports;

[0080] Acquiring known landslide scenarios based on the historical landslide event data, wherein the landslide scenarios include rainfall-induced landslides, earthquake-induced landslides, and progressive landslides;

[0081] Furthermore, landslide scenario simulation is performed based on the mountain twin characteristic model to obtain simulated risk data under different types of landslide scenarios; wherein, the landslide scenario simulation refers to simulating the occurrence process of different types of landslide scenarios by adjusting factors such as soil moisture content, geological structure, rainfall parameters, and earthquake parameters within the mountain twin characteristic model; for example, taking rainfall-type landslide as an example; the rainfall-type landslide refers to the landslide phenomenon caused by the increase of mountain moisture content due to continuous rainfall, the increase of soil weight and the decrease of strength; in the mountain twin characteristic model, by adjusting the rainfall intensity and duration parameters, the changes in the water content inside the mountain under different rainfall conditions can be simulated, and the response changes of various monitoring parameters in the mountain area, including changes in moisture content, pore water pressure, displacement, etc., can be obtained, and used as simulated risk data for rainfall-type landslides;

[0082] It should be further explained that, in the specific implementation process, the process of obtaining risk characteristic information includes:

[0083] Taking a certain landslide scenario as an example, obtain the simulated risk data corresponding to several corresponding landslide scenarios;

[0084] A two-dimensional rectangular coordinate system is constructed with the monitoring parameters in the simulated risk data as the vertical axis and time as the horizontal axis, and the obtained simulated risk data is mapped into the two-dimensional rectangular coordinate system to obtain a corresponding risk evolution diagram, which is a curve;

[0085] Then, the risk characteristic information in the risk evolution diagram is extracted. The risk characteristic information includes parameter change trend, mutation characteristics, and critical point characteristics. Among them, the parameter change trend refers to the overall upward or downward direction of each monitoring parameter in the risk evolution diagram, indicating the overall change trend of the monitoring parameters over time; the mutation characteristic refers to the point where the monitoring parameters in the risk evolution diagram change drastically, indicating that the monitoring parameters have changed significantly at a specific time point; the critical point characteristic refers to the inflection point where the change pattern of the monitoring parameters in the risk evolution diagram changes from one state to another, indicating the critical moment when the mountain state undergoes a qualitative change.

[0086] Furthermore, based on the above risk characteristic information extraction process, the risk characteristic information corresponding to different types of landslide scenarios is obtained.

[0087] It should be further explained that, in the specific implementation process, the construction process of the trigger factor association network includes:

[0088] Based on the principal component analysis algorithm, monitoring indicators are extracted from the acquired historical landslide event data and mountain monitoring basic data to obtain corresponding mountain monitoring indicators. The mountain monitoring indicators include topographic and geological characteristics such as slope, aspect, elevation, lithology, and fault distribution, as well as rainfall intensity, cumulative rainfall, soil moisture change rate, surface displacement rate, etc.; and indicator characteristic information corresponding to each mountain monitoring indicator is obtained. The indicator characteristic information includes parameter change trends, mutation characteristics, and critical point characteristics of each mountain monitoring indicator in the mountain monitoring basic data;

[0089] Combined with risk feature information, an association rule mining algorithm is used to evaluate the correlation between the corresponding mountain monitoring indicators and the corresponding landslide scenarios, and the support and confidence between each mountain monitoring indicator and the corresponding landslide scenario are obtained; and based on the correlation between each mountain monitoring indicator and the corresponding landslide scenario, the correlation between each mountain monitoring indicator and the corresponding landslide scenario is obtained, wherein the higher the support and confidence, the higher the corresponding correlation; the association rule mining algorithm used in the present invention is the FP-growth algorithm; the correlation evaluation refers to setting the minimum support and confidence thresholds based on the FP-growth algorithm, extracting strong correlation rules, and combining temporal correlation analysis and spatial correlation analysis to obtain correlation patterns and spatial correlation patterns at different times, and quantify them into support and confidence; wherein, the temporal correlation analysis adopts a sliding time window method to identify correlation patterns under different time lags; the spatial correlation analysis adopts a geographically weighted statistical method to identify spatial correlation patterns;

[0090] Then, based on the obtained correlation degree, a characteristic importance analysis is performed on the corresponding mountain monitoring indicators to obtain the contribution of the mountain monitoring indicators to the corresponding landslide scenario. The characteristic importance analysis quantifies the contribution of each mountain monitoring indicator to the prediction of the probability of landslide occurrence by combining the correlation degree using the permutation importance method, with the contribution value ranging from 0 to 100.

[0091] Then, based on the obtained contribution, the corresponding mountain monitoring indicators are divided into direct triggering factors and indirect triggering factors, and a hierarchical structure corresponding to direct triggering factors and indirect triggering factors is established; the direct triggering factors generally include factors that can directly trigger landslides, such as rainfall intensity, accumulated rainfall, and earthquake acceleration; indirect triggering factors include intrinsic factors that affect landslide susceptibility, such as slope, lithology, and vegetation coverage; the hierarchical results are represented by a directed acyclic graph, which can be used to reflect the dependency and mutual influence relationship between direct and indirect triggering factors;

[0092] Furthermore, a number of factor combinations are randomly constructed based on the direct trigger factors and the indirect trigger factors. The factor combination may be composed of two direct trigger factors or indirect trigger factors, or may be composed of one direct trigger factor and one indirect trigger factor;

[0093] Based on the second-order interaction test method, a synergistic analysis is performed on the constructed factor combinations to obtain the corresponding synergistic effects, which include positive and negative effects. For example, if high rainfall and steep slopes in a mountain area significantly increase the risk of landslides, the synergistic effect of the corresponding factor combination is a negative effect. The joint probability and conditional probability corresponding to each factor combination are obtained. The joint probability refers to the probability of the corresponding factor combination occurring in a landslide scenario based on historical landslide event data; the conditional probability refers to the change in the probability of the landslide scenario occurring under the corresponding factor combination. For example, the probability of a landslide scenario increasing under the factor combination of high rainfall and steep slopes.

[0094] Based on the historical landslide event data, a time dependency analysis is performed on the triggering factors to obtain a corresponding temporal dependency relationship, wherein the temporal dependency relationship represents the temporal relationship between the sequence of changes in the corresponding triggering factors and the occurrence of the landslide scenario; the time dependency analysis refers to the process of identifying the time series pattern of the triggering factors and the occurrence of the landslide scenario through a Granger causality test algorithm; wherein the triggering factors include direct triggering factors and indirect triggering factors;

[0095] Furthermore, a trigger factor association network is constructed based on the obtained contribution, hierarchical structure, synergistic effect and temporal dependency. The trigger factor association network is represented by a graph data structure. The nodes in the trigger factor association network represent various types of trigger factors, and the edges represent the association relationships between trigger factors.

[0096] It should be further explained that, during the specific implementation process, the target landslide risk is assessed and graded; the process of constructing the corresponding graded risk zoning map includes:

[0097] Based on the constructed trigger factor association network and mountain twin feature model, the target mountain area is divided into several evaluation units. The regional division is achieved by combining the geological unit method and the regular grid method to ensure that the geological characteristics within each evaluation unit are relatively uniform.

[0098] A sensitivity assessment is performed on the divided assessment units to obtain a sensitivity assessment index corresponding to the assessment units. The sensitivity assessment represents a process of quantitatively evaluating the geological sensitivity within the corresponding assessment unit by comprehensively considering the influence of inherent factors such as slope, lithology, and structural surface characteristics, and using a multi-factor weighted scoring method. The geological sensitivity assessment index is used to quantitatively characterize the susceptibility of landslides within the mountain area, reflecting the potential sensitivity of the mountain area itself to landslide hazards without considering triggering factors. The higher the sensitivity index, the higher the corresponding sensitivity.

[0099] Furthermore, a coupling model is constructed based on the triggering factors and sensitivity indices. The model input of the coupling model includes the triggering factors and sensitivity indices involved in each assessment unit. The model output of the coupling model is the instability probability of each assessment unit. The instability probability represents the probability of landslide instability occurring in the corresponding assessment unit under the corresponding triggering factors and sensitivity indices.

[0100] The coupling model is constructed using machine learning methods such as logistic regression or support vector machines. The construction process is prior art and will not be elaborated in detail in this invention. The training data collected during the construction of the coupling model are historical landslide event data and historical mountain monitoring basic data corresponding to the target mountain area being in a safe state.

[0101] Furthermore, each assessment unit is divided into risk levels based on the probability of instability, and a corresponding graded risk zoning map is obtained.

[0102] It should be further explained that, in the specific implementation process, the process of constructing a spatiotemporal evolution model and prediction analysis framework based on the hierarchical risk zoning map and obtaining the corresponding dynamic landslide risk prediction results includes:

[0103] Perform time series decomposition and trend analysis on the real-time collected mountain monitoring basic data set to obtain the corresponding time feature components, and build a time series prediction basic model based on the time feature components to obtain a time evolution prediction framework; the time series decomposition adopts the STL decomposition method to decompose the time series data into trend terms, seasonal terms, and residual terms;

[0104] Specifically, the time series corresponding to the monitoring parameters in the real-time mountain monitoring basic data set are decomposed into trend terms, residual terms and seasonal terms based on the STL decomposition method;

[0105] Then, a trend analysis is performed on the obtained trend term, residual term, and seasonal term; and based on the trend analysis results, corresponding time feature components are obtained, which include trend characteristics (long-term change direction and rate), period characteristics (seasonal fluctuations and cyclical patterns), mutation characteristics (abnormal points and structural changes), and residual characteristics (random fluctuation parts) of the corresponding monitoring parameters. Each feature component includes the mean, variance, skewness, kurtosis, autocorrelation coefficient, etc. of the corresponding monitoring parameter;

[0106] Furthermore, based on each feature component in the time feature component, the corresponding time prediction basic model is constructed. For example, for data with obvious linear trends, the ARIMA model is used, and the model order automatically selects the optimal parameter combination through the AIC criterion; for data with obvious seasonality, the SARIMA model is used to capture seasonality and holiday effects; for complex nonlinear time series, the LSTM deep learning model is used to automatically learn the long-term and short-term dependencies of the time series; wherein, the construction process of the corresponding time prediction basic model is prior art and will not be elaborated in this application;

[0107] Integrating the constructed time prediction basic model to obtain a corresponding time evolution prediction framework; wherein the time evolution prediction framework can be used to predict monitoring parameters at different time scales;

[0108] Based on the hierarchical risk zoning map and spatial interpolation technology, a spatial correlation model and spatial association network are constructed. Spatial diffusion simulation and boundary analysis are performed using the spatial association network to obtain a spatial evolution prediction framework. The spatial correlation model is constructed using the geographically weighted regression (GWR) method to describe the spatial dependency structure between various triggering factors. The spatial association network is presented in the form of a graph structure to represent the spatial correlation relationship between different assessment units within the mountain area. Spatial diffusion simulation simulates the propagation process of landslide scenarios in the target mountain area based on the cellular automaton method. The diffusion process in the simulation process must take into account constraints such as topography, hydrology, and geology. For example, the risk will not spread to higher places, but will spread faster along the direction of water flow. Boundary analysis refers to the identification of spatial boundaries and boundary barrier effects during the landslide risk diffusion process. Spatial boundaries include natural boundaries (such as ridgelines, rivers, geological boundaries) and artificial boundaries (such as roads, dams, and protective projects). Boundary barrier effects are assessed based on boundary type and strength, and are divided into strong barriers (completely blocking risk diffusion), moderate barriers (significantly slowing diffusion speed), and weak barriers (slightly affecting diffusion direction).

[0109] It should be further explained that the spatial evolution prediction framework supports a variety of spatial prediction functions, including: risk diffusion path prediction, dynamic change prediction of impact range, location prediction of new landslides, and expansion direction prediction of existing landslides;

[0110] The constructed temporal evolution prediction framework and spatial evolution prediction framework are integrated to obtain a mountain spatiotemporal prediction model with multi-level prediction functions.

[0111] Then, scenario prediction is performed based on the obtained mountain spatiotemporal prediction model to obtain scenario prediction results corresponding to the target mountain area. The scenario prediction results include the probability of occurrence of various landslide scenarios and the corresponding scenario risk zoning map and potential impact range under the corresponding landslide scenario. The content reflected in the scenario risk zoning map is the probability of landslide risk in each assessment unit under the corresponding landslide scenario.

[0112] Then, the confidence level of the obtained scenario prediction results is calculated based on the uncertainty quantification algorithm to obtain the corresponding confidence interval;

[0113] Integrating the confidence interval with the corresponding scenario prediction results to obtain a corresponding dynamic prediction result of landslide risk; the dynamic prediction result of landslide risk includes a time dimension, a space dimension and a probability dimension;

[0114] The time dimension is used to display the time series of predictions at different time scales, including daily predictions, weekly predictions and monthly predictions; the prediction results at the corresponding time point at each time scale include risk level, probability of occurrence and confidence interval; the spatial dimension includes a hierarchical risk zoning map for displaying the spatial distribution of risks; the probability dimension displays the uncertainty and possibility distribution of the prediction, including the probability value, confidence interval width and coefficient of variation of each risk level.

[0115] It should be further explained that, in the specific implementation process, the construction process of the hierarchical warning rule base includes:

[0116] Based on the obtained landslide risk dynamic prediction results, the obtained triggering factors are updated, and the synergistic effects between the updated triggering factors are obtained;

[0117] Then, a corresponding early warning indicator system is constructed based on the triggering factors and synergistic effects;

[0118] Data distribution analysis and critical value research are conducted on the obtained early warning indicator system to obtain the corresponding preliminary indicator threshold scheme; the data distribution analysis refers to obtaining the statistical distribution characteristics of each triggering factor before the landslide based on the historical monitoring basic data set; the historical monitoring basic data set consists of the historical mountain monitoring basic data set before the landslide scenario and the mountain safety period; then, the statistical distribution characteristics corresponding to the landslide scenario before the landslide scenario and the mountain safety period are compared to obtain the risk threshold range corresponding to the landslide risk; the critical value research refers to finding the optimal early warning threshold point corresponding to each triggering factor using the Youden index maximization principle; the preliminary indicator threshold scheme includes four levels of thresholds corresponding to each triggering factor; for example: the four levels of thresholds for 24-hour cumulative rainfall are 50mm (blue), 80mm (yellow), 120mm (orange) and 180mm (red); the four levels of thresholds for displacement rate are 5mm / day (blue), 10mm / day (yellow), 20mm / day (orange) and 30mm / day (red);

[0119] Furthermore, the effectiveness of the obtained initial indicator threshold scheme is evaluated based on the obtained historical landslide event data. The effectiveness evaluation comprehensively evaluates the practicality of the corresponding initial indicator threshold scheme by calculating the missed alarm rate, false alarm rate and warning lead time.

[0120] If the effectiveness evaluation results meet the expected requirements, the optimal indicator threshold solution will be output accordingly. If the effectiveness evaluation results do not meet the expected requirements, the corresponding initial indicator threshold solution will be adjusted based on the pre-set adjustment strategy; for example, lowering the indicator threshold with a high missed alarm rate, raising the indicator threshold with a high false alarm rate, adjusting the threshold interval, etc.

[0121] Furthermore, based on the multi-level warning principles published by industry standards, the corresponding optimal indicator threshold schemes are hierarchically divided to obtain a warning grading framework, and different levels of warning trigger conditions and combination logic are formulated according to the warning grading framework to obtain a hierarchical warning rule library.

[0122] It should be further explained that, in the specific implementation process, the process of real-time monitoring of the target mountain area and feedback of monitoring results based on the hierarchical early warning rule library includes:

[0123] Feature extraction is performed on the real-time collected mountain basic monitoring data set to obtain the indicator characteristic information corresponding to each trigger factor, and conditional evaluation is performed on it in combination with the constructed hierarchical warning rule library. The conditional evaluation includes single condition evaluation and composite condition evaluation. Single condition evaluation refers to comparing the parameter value of the corresponding trigger factor with its corresponding four-level threshold value. Composite condition evaluation determines the priority of each trigger factor based on the synergistic effect between the trigger factors and the indicator characteristic information.

[0124] Furthermore, the warning level is converted based on the conditional assessment results to obtain the corresponding warning status vector; the warning status vector is used to describe the current warning status of different assessment units in the mountain area; and based on it, the corresponding warning signal and comprehensive risk level are generated to obtain the corresponding mountain warning result, which is fed back to the management department of the target mountain area so that different emergency measures can be taken according to the mountain warning results.

[0125] This system uses distributed sensor networks, meteorological stations, and remote sensing technology to acquire and fuse multi-source heterogeneous data. It then constructs a digital twin model of the mountain to simulate multiple landslide scenarios. It then establishes a trigger factor correlation network for zoning risk assessment. Finally, it predicts spatiotemporal evolution and establishes a hierarchical early warning rule base. The system's beneficial effects include significantly improving data quality and reliability; enabling accurate risk simulation under complex geological conditions; establishing a scientific and rational zoning early warning indicator system; enabling dynamic spatiotemporal prediction of landslide risk; reducing false alarms and missed alarms, and extending early warning lead times, providing a scientific basis for disaster prevention and mitigation decision-making.

[0126] Example 2

[0127] See also Figure 2 As shown, for the parts not described in detail in this embodiment, please refer to the description of Example 1. A method for predicting landslide risk based on data intelligent analysis is provided, including:

[0128] Step 1: Real-time monitoring and data collection of geological and meteorological parameters of the target mountain area are carried out to obtain corresponding multi-source heterogeneous information. Feature extraction and data fusion are then performed on the collected multi-source heterogeneous information to obtain the corresponding mountain monitoring basic data set.

[0129] Step 2: Build a mountain twin feature model based on the mountain monitoring basic data set, and simulate multiple landslide scenarios based on it to obtain the corresponding simulated risk data. Based on the simulated risk data, obtain the risk characteristic information corresponding to different landslide scenarios;

[0130] Step 3: Construct a trigger factor association network based on risk characteristic information, and combine it with the basic mountain monitoring data set to conduct zoning assessment and classification of the target landslide risk; and construct the corresponding graded risk zoning map;

[0131] Step 4: Conduct spatiotemporal evolution analysis and prediction analysis based on the graded risk zoning map to obtain the corresponding dynamic prediction results of landslide risk; establish an early warning indicator system and set thresholds based on the dynamic prediction results of landslide risk to obtain the corresponding graded early warning rule library; conduct real-time monitoring of the target mountain area based on the graded early warning rule library and provide feedback on the monitoring results.

[0132] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art will be able to modify the technical solutions described in the foregoing embodiments or to substitute equivalents for some of the technical features. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

[0133] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or apparatus comprising the element.

[0134] In the description of the present invention, it should be understood that the terms "first", "second", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0135] In the description of the present invention, unless otherwise specified, "plurality" means two or more.

[0136] In the description of the present invention, “several” means one or more, and “a large number” means two or more.

[0137] Throughout this specification, reference to terms such as "one embodiment," "some embodiments," "examples," "specific examples," or "some examples" means that a specific feature, structure, material, or characteristic described in conjunction with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0138] The formulas in this manual are all dimensionless and calculated using numerical values. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters and thresholds in the formulas are set by technicians in this field based on actual conditions.

[0139] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to the embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the claims and their equivalents.

Claims

1. A landslide risk prediction system based on data intelligent analysis, characterized in that: include: The data acquisition module is used to monitor and collect data on the geological and meteorological parameters of the target mountain area in real time, obtain corresponding multi-source heterogeneous information, and perform feature extraction and data fusion on the collected multi-source heterogeneous information to obtain the corresponding mountain monitoring basic data set; The risk simulation module builds a mountain twin feature model based on the basic mountain monitoring data set, and simulates multiple landslide scenarios based on it to obtain the corresponding simulated risk data. Based on the simulated risk data, it obtains the risk feature information corresponding to different landslide scenarios; The risk assessment module builds a trigger factor association network based on risk characteristic information and combines it with the basic mountain monitoring data set to conduct zoning assessment and classification of target landslide risks; Construct corresponding hierarchical risk zoning maps; The process of building the trigger factor association network includes: Extract monitoring indicators from the acquired historical landslide event data and mountain monitoring basic data to obtain corresponding mountain monitoring indicators; and obtain indicator characteristic information corresponding to each mountain monitoring indicator; Combined with risk characteristic information, the correlation between mountain monitoring indicators and landslide scenarios is evaluated to obtain the support and confidence between each mountain monitoring indicator and the corresponding landslide scenario; and based on this, the correlation between each mountain monitoring indicator and the corresponding landslide scenario is obtained; Based on the obtained correlation, the characteristic importance of the corresponding mountain monitoring indicators is analyzed to obtain the contribution of the mountain monitoring indicators to the corresponding landslide scenario; based on the obtained contribution, the corresponding mountain monitoring indicators are divided into direct triggering factors and indirect triggering factors, and a hierarchical structure corresponding to the direct triggering factors and the indirect triggering factors is established; Randomly construct several factor combinations based on direct trigger factors and indirect trigger factors; Conduct synergy analysis on the constructed factor combinations to obtain the corresponding synergy effects; and obtain the joint probability and conditional probability corresponding to each factor combination; Obtain the temporal dependency between triggering factors and landslide scenarios based on historical landslide event data; Constructing a trigger factor association network based on the obtained contribution, hierarchical structure, synergistic effect and temporal dependency; the trigger factors include direct trigger factors and indirect trigger factors; The risk prediction module conducts spatiotemporal evolution analysis and prediction analysis based on the graded risk zoning map to obtain the corresponding dynamic prediction results of landslide risks; and establishes an early warning indicator system and sets thresholds based on the dynamic prediction results of landslide risks to obtain the corresponding graded early warning rule library; based on the graded early warning rule library, it conducts real-time monitoring of the target mountain area and provides feedback on the monitoring results.

2. A landslide risk prediction system based on data intelligent analysis according to claim 1, characterized in that: The process of obtaining the basic mountain monitoring data set includes: Performing data verification and outlier screening on the obtained multi-source heterogeneous information; and performing data unification on the multi-source heterogeneous data after outlier screening, wherein the data unification includes time alignment and spatial registration; Multi-source data fusion is performed on the unified multi-source heterogeneous data, and the multi-source fused data is structured and organized to obtain the corresponding mountain monitoring basic data set.

3. The landslide risk prediction system based on data intelligent analysis according to claim 2 is characterized in that: The basic mountain monitoring dataset contains three dimensions: time, space, and attributes. The time dimension records the observation time series, the space dimension contains geographic coordinates and elevation information, and the attribute dimension contains various monitoring parameters and derived indicators.

4. A landslide risk prediction system based on data intelligent analysis according to claim 3, characterized in that: The process of obtaining risk characteristic information includes: Performing three-dimensional modeling based on the mountain monitoring basic data set to obtain a mountain twin feature model corresponding to the target mountain area; Collect and organize historical landslide event data; and obtain known landslide scenarios based on them; Conduct landslide scenario simulation based on the mountain twin characteristic model to obtain simulated risk data under different types of landslide scenarios; Acquire simulated risk data corresponding to several corresponding landslide scenarios; construct a two-dimensional rectangular coordinate system with the monitoring parameters in the simulated risk data as the vertical axis and time as the horizontal axis, and map the acquired simulated risk data into the two-dimensional rectangular coordinate system to obtain the corresponding risk evolution; extract risk characteristic information in the risk evolution diagram, the risk characteristic information including parameter change trend, mutation characteristics, and critical point characteristics; Based on the above risk characteristic information extraction process, the risk characteristic information corresponding to other landslide scenarios is obtained.

5. The landslide risk prediction system based on data intelligent analysis according to claim 1 is characterized in that: The process of obtaining the graded risk zoning map includes: Based on the constructed trigger factor association network and mountain twin feature model, the target mountain area is divided into regions to obtain several evaluation units; Perform sensitivity assessment on the divided assessment units to obtain sensitivity assessment indexes corresponding to the assessment units; construct a coupling model based on the trigger factors and sensitivity indexes, wherein the model input of the coupling model includes the trigger factors and sensitivity indexes involved in each assessment unit, and the model output of the coupling model is the instability probability of each assessment unit; Based on the probability of instability, each assessment unit is divided into risk levels to obtain the corresponding graded risk zoning map.

6. The landslide risk prediction system based on data intelligent analysis according to claim 5 is characterized in that: The process of obtaining dynamic landslide risk prediction results includes: Perform time series decomposition and trend analysis on the real-time collected mountain monitoring basic data set to obtain corresponding time feature components, and build a time series prediction basic model based on the time feature components. Then, integrate the constructed time prediction basic model to obtain the corresponding time evolution prediction framework. Based on the hierarchical risk zoning map, a spatial correlation model and a spatial association network are constructed, and spatial diffusion simulation and boundary analysis are performed through the spatial association network to obtain a spatial evolution prediction framework; Integrate the constructed temporal evolution prediction framework and spatial evolution prediction framework to obtain the corresponding mountain spatiotemporal prediction model; Based on the obtained mountain spatiotemporal prediction model, scenario prediction is performed to obtain scenario prediction results corresponding to the target mountain area, and confidence calculation is performed on the obtained scenario prediction results to obtain corresponding confidence intervals; The confidence interval is integrated with the corresponding scenario prediction results to obtain the corresponding landslide risk dynamic prediction results.

7. The landslide risk prediction system based on data intelligent analysis according to claim 6 is characterized in that: The construction process of the hierarchical warning rule base includes: Based on the obtained landslide risk dynamic prediction results, the obtained trigger factors are updated, and the synergistic effects between the updated trigger factors are obtained, and the corresponding early warning indicator system is constructed based on them; Conduct data distribution analysis and critical value research on the obtained early warning indicator system to obtain the corresponding preliminary indicator threshold scheme; Based on the historical landslide event data obtained, the effectiveness of the initial indicator threshold scheme obtained is evaluated; and based on the effectiveness evaluation results, whether to adjust the initial indicator threshold scheme is selected. If not, the initial indicator threshold scheme is output as the optimal indicator threshold scheme; if it is adjusted, the corresponding initial indicator threshold scheme is adjusted based on a pre-set adjustment strategy; Then, the optimal indicator threshold scheme is divided into levels to obtain a warning classification framework, and warning trigger conditions and combination logic of different levels are formulated according to the warning classification framework to obtain a hierarchical warning rule base.

8. The landslide risk prediction system based on data intelligent analysis according to claim 7 is characterized in that: The process of real-time monitoring of the target mountain area and feedback of monitoring results based on the hierarchical early warning rule library includes: Feature extraction is performed on the real-time collected mountain basic monitoring data set to obtain the indicator characteristic information corresponding to each trigger factor, and conditional evaluation is performed on it and compared with the constructed hierarchical warning rule library; based on the conditional evaluation results, the warning level is converted to obtain the corresponding warning state vector; and based on it, the corresponding warning signal and comprehensive risk level are generated to obtain the corresponding mountain warning results and feedback to the management department of the target mountain area.

9. A landslide risk prediction method based on data intelligent analysis, which is based on the landslide risk prediction system based on data intelligent analysis according to any one of claims 1 to 8, characterized in that: include: Step 1: Real-time monitoring and data collection of geological and meteorological parameters of the target mountain area are carried out to obtain corresponding multi-source heterogeneous information. Feature extraction and data fusion are then performed on the collected multi-source heterogeneous information to obtain the corresponding mountain monitoring basic data set. Step 2: Build a mountain twin feature model based on the mountain monitoring basic data set, and simulate multiple landslide scenarios based on it to obtain the corresponding simulated risk data. Based on the simulated risk data, obtain the risk characteristic information corresponding to different landslide scenarios; Step 3: Construct a trigger factor association network based on risk characteristic information, and combine it with the basic mountain monitoring data set to conduct zoning assessment and classification of the target landslide risk; and construct the corresponding graded risk zoning map; Step 4: Based on the hierarchical risk zoning map, conduct spatiotemporal evolution analysis and prediction analysis to obtain the corresponding landslide risk dynamic prediction results; Based on the dynamic prediction results of landslide risk, an early warning indicator system is established and thresholds are set to obtain a corresponding graded early warning rule base; Based on the hierarchical early warning rule library, the target mountain area is monitored in real time and the monitoring results are fed back.

Images