Landslide disaster risk assessment method and system based on multiple spatial-temporal scales

Abstract

The invention discloses a landslide disaster risk assessment method and system based on multiple temporal and spatial scales, and relates to the field of landslide disaster risk assessment, and the method comprises a capturing module which is used for collecting geological, terrain, weather, vegetation and human activity associated data of continuous time dimensions and spatial dimensions in a preset landslide potential influence region, and transmitting the data to a database; the data is forwarded to the establishment module in real time; the establishing module is used for receiving the data acquired by the capturing module and performing denoising, format standardization and time-space dimension alignment processing on the acquired data so as to establish a standardized time-space data set under a unified coordinate system; according to the method, multi-dimensional spatio-temporal data associated with geology, terrain, weather, vegetation and human activities are accurately collected, the data quality is optimized through denoising, standardization and time-space dimension alignment processing, influence characteristic parameters of different time scales and space granularities are comprehensively extracted, an evaluation model adaptive to various spatio-temporal scenes is constructed, and the evaluation accuracy is improved. And predictive support is provided for disaster prevention and control planning and decision making.

Description

Technical Field

[0001] This invention relates to the field of landslide disaster risk assessment technology, specifically a landslide disaster risk assessment method and system based on multiple spatiotemporal scales. Background Technology

[0002] Landslides are geological hazards caused by the sliding of rock and soil masses along a defined sliding surface on a slope under the influence of gravity, precipitation, earthquakes, and other factors. They often damage roads, buildings, and other facilities, threatening people's lives and property. They mostly occur in mountainous areas, valleys, or areas with fragile geological structures. Current technologies enable prevention through monitoring, early warning, and engineering measures.

[0003] The invention patent application with application number 202310768265.5 discloses a landslide hazard monitoring and early warning method and system based on real-scene 3D. The application aims to solve the problem that "the existing technology lacks a method for fine and realistic multi-scale 3D reconstruction of complex landslide scenes, supports the analysis of individual, physical, and semantic features of landslide scenes, and realizes accurate early warning of landslide disasters with multiple indicators considering spatiotemporal dynamic correlation".

[0004] However, existing technologies for landslide disasters are mostly based on warnings, alerts, and short-term predictions, and there are no technologies for long-term prediction and early warning.

[0005] To this end, we propose a landslide disaster risk assessment method and system based on multiple spatiotemporal scales. Summary of the Invention

[0006] In view of the above-mentioned shortcomings of the existing technology, the present invention provides a landslide disaster risk assessment method and system based on multiple spatiotemporal scales, which can effectively solve the problems of the existing technology.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions; This invention discloses a landslide disaster risk assessment system based on multiple spatiotemporal scales, comprising: The system comprises the following modules: a capture module for collecting geological, topographical, meteorological, vegetation, and human activity-related data in continuous time and space within a pre-defined potential landslide impact area, and forwarding this data to the establishment module in real time; an establishment module for receiving data from the capture module, performing noise reduction, format standardization, and time-space alignment to establish a standardized spatiotemporal dataset in a unified coordinate system; an extraction module for extracting landslide hazard impact characteristic parameters at different time scales and spatial granularities from the standardized spatiotemporal dataset, and constructing a feature parameter library; a modeling module for applying the feature parameter library to construct landslide hazard risk assessment models adapted to different spatiotemporal scenarios; a prediction module for calling the risk assessment model, inputting long-term spatiotemporal feature prediction data, and performing quantitative prediction of long-term landslide hazard risk; and an output module for receiving the quantitative prediction results from the prediction module and generating a risk notification message. The capture module is interconnected with the establishment module via a wireless network. The establishment module is interconnected with the extraction module and the modeling module via a wireless network. The modeling module is interconnected with the prediction module via a wireless network. The prediction module and the modeling module are interconnected with the output module via a wireless network.

[0008] Furthermore, the geological data collected by the capture module includes soil and rock cohesion, internal friction angle, fault activity rate, and groundwater depth; topographic data includes slope, aspect, and elevation variation coefficient; meteorological data includes cumulative rainfall, rainfall intensity, and temperature change rate; vegetation data includes vegetation coverage and vegetation type normalization index; and human activity-related data includes engineering construction disturbance intensity and cultivated land development ratio. The capture module collects data in real time through distributed sensors deployed in the preset potential landslide impact area. The temporal and spatial resolution of the collected data are dynamically adjusted according to the preset spatiotemporal scale requirements.

[0009] Furthermore, the noise reduction processing of the collected data in the establishment module follows the following principle: ; In the formula: The data is standardized and located at spatial coordinates (i,j) and time t after denoising. The original data collected before noise reduction; , These are time and space weighting coefficients; This is the error correction term.

[0010] Furthermore, the spatiotemporal alignment processing in the establishment module includes spatial coordinate alignment and timestamp synchronization: Spatial coordinate alignment uses the reference control points of a pre-defined unified coordinate system as a reference, and transforms the original coordinates of data collected from different sources to the unified coordinate system using affine transformation formulas: ; In the formula: The target coordinates in the unified coordinate system after transformation; The original coordinates for collecting the data; Spatial resolution adaptation coefficient; , denoted as the error compensation amount of the original coordinates in the X and Y axes; a, b, d, and e are the linear coefficients of the affine transformation; c and f are the translation coefficients of the affine transformation. Timestamp synchronization is based on UTC standard time. By calculating the time deviation value of each acquisition device, the timestamp of the acquired data is corrected to make the time dimension of all data uniform.

[0011] Furthermore, the landslide disaster impact characteristic parameters extracted by the extraction module include spatiotemporal sensitivity parameters, dynamic induction parameters, and vulnerability parameters of the disaster-bearing body, wherein the spatiotemporal sensitivity parameters are: ; In the formula: For spatial coordinates (i,j) and time t, the spatiotemporal sensitivity parameter is used. The total number of factors affecting sensitivity; The weight coefficient of the k-th factor; Let be the standardized value of the k-th factor at spatial coordinates (i,j) and time t; This is a time-scale correction factor; This is the spatial granularity correction factor; in, Including slope factor, vegetation cover factor, soil and rock stability factor, and fault influence factor. The values ​​of all are within the range of (0,1), and , Determined based on the ratio of the current time scale to the reference time scale. Determined based on the ratio of the current spatial granularity to the reference spatial granularity.

[0012] Furthermore, the landslide disaster risk assessment model constructed by the modeling module is a spatiotemporal hybrid fusion model, including a time series prediction sub-model and a spatial heterogeneity adaptation sub-model; Among them, the time series prediction sub-model is based on an improved long short-term memory network, which dynamically adjusts the weights of features at different time scales by introducing spatiotemporal gating units into the hidden layers of the LSTM. The spatial heterogeneity adaptor sub-model is based on the geographic weighted random forest algorithm, which incorporates spatial coordinates as covariates into the model training and adaptively adjusts the weights of different spatial region features. The modeling module trains a time series prediction sub-model using the time series feature dataset in the feature parameter library to obtain the time dimension risk contribution value; it trains a spatial heterogeneity adaptation sub-model using the spatial feature dataset to obtain the spatial dimension risk contribution value; and then superimposes the time dimension risk contribution value and the spatial dimension risk contribution value using preset fusion weights to obtain a comprehensive risk assessment model adapted to different spatiotemporal scenarios. ; In the formula: The risk contribution value of the time dimension corresponding to spatial coordinates (i,j) and time t; The Sigmoid activation function is used to map the spatiotemporal gated output to the [0,1] interval; It is a spacetime gating factor; This represents the total number of features on the time scale. is the dynamic weighting coefficient of the p-th time-scale feature; Let be the standardized value of the p-th time-scale feature at the spatiotemporal location (i,j,t); This is a time trend correction term; The spatial dimension risk contribution value corresponding to spatial coordinates (i,j) and time t; This represents the total number of spatial characteristic factors; Let be the local weight coefficient of the k-th spatial feature factor at position (i,j); Let be the standardized value of the k-th spatial feature factor at its spatiotemporal location (i,j,t); Given the current position (i,j) and the spatial reference point The Euclidean distance; This is the spatial attenuation threshold; This is the spatial error correction term.

[0013] Furthermore, the ultra-long-term spatiotemporal feature prediction data input into the prediction module is obtained through the following methods: ; In the formula: For spatial coordinates (i,j) and future time... Corresponding ultra-long-term spatiotemporal feature prediction data, This is a preset ultra-long-term time interval; The number of historical data periods used for trend projection; The weighting coefficients for the p-th historical period data; Spatial coordinates (i,j), historical time Standardized observation data, where T is the length of the historical data period; The trend correction coefficient for the p-th historical period data; This represents the number of types of dynamic inducing factors; The influence weight of the q-th type of inducing factor; For the qth type of inducing factor in the future time The predicted value; Let be the regional adaptation coefficient of the q-th type of inducing factor in spatial coordinates (i,j); This is the spatiotemporal coordination correction coefficient.

[0014] Furthermore, the risk notification message output by the output module includes the prediction results of the prediction module, the ultra-long-term spatiotemporal feature prediction data input by the prediction module, and the risk level determined by comparing the preset threshold with the quantitative prediction results. The system uses a parallel task processing mechanism to apply to several target landslide disaster risk assessment areas. The prediction results of each processing task are collected in the output module. When the output module outputs the risk notification message, it also includes an electronic map with the prediction results of each target landslide disaster risk assessment area.

[0015] On the other hand, a landslide hazard risk assessment method based on multiple spatiotemporal scales includes: Geological, topographical, and meteorological data were collected within the pre-defined potential landslide impact area. The collected data underwent denoising and format standardization, and spatiotemporal alignment was achieved through coordinate transformation and timestamp synchronization to establish a standardized spatiotemporal dataset under a unified coordinate system. Spatiotemporal sensitivity, dynamic induction, and vulnerability characteristics of the affected body were extracted from the standardized spatiotemporal dataset to construct a feature parameter library covering different spatiotemporal scales. A spatiotemporal hybrid fusion model was constructed based on the feature parameter library, training a time-series prediction sub-model and a spatial heterogeneity adaptation sub-model separately, and then fusing them to obtain a risk assessment model adaptable to multiple spatiotemporal scenarios. Long-term spatiotemporal feature prediction data was acquired and input into the risk assessment model, and the long-term risk of landslide disasters was quantitatively predicted through model computation. The quantitative prediction results were collected, and a risk notification message was generated by combining the long-term input data and risk level determination results.

[0016] Compared with the known prior art, the technical solution provided by this invention has the following beneficial effects: This invention accurately collects multi-dimensional spatiotemporal data related to geology, topography, meteorology, vegetation, and human activities. After denoising, standardization, and spatiotemporal dimension alignment, the data quality is optimized. It comprehensively extracts the influence feature parameters at different time scales and spatial granularities, constructs an assessment model adapted to diverse spatiotemporal scenarios, and achieves ultra-long-term risk quantification prediction by dynamically adjusting the influence weights of spatiotemporal dimensions. It accurately defines the risk level and outputs a risk notification message with an electronic map annotating the prediction results for multiple regions. It supports parallel processing to improve efficiency and significantly improves the accuracy, timeliness, and adaptability of landslide disaster risk assessment, providing predictive support for disaster prevention and control planning and decision-making. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0018] Figure 1 This is a schematic diagram of a landslide disaster risk assessment system based on multiple spatiotemporal scales; Figure 2 This is a flowchart illustrating a landslide hazard risk assessment method based on multiple spatiotemporal scales. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0020] The present invention will be further described below with reference to embodiments.

[0021] Example 1: This embodiment presents a landslide disaster risk assessment system based on multiple spatiotemporal scales, such as... Figure 1 As shown, it includes: The capture module is used to collect geological, topographical, meteorological, vegetation and human activity-related data in the continuous time and spatial dimensions within the preset potential landslide impact area, and forward it to the establishment module in real time. The geological data collected by the capture module includes soil and rock cohesion, internal friction angle, fault activity rate, and groundwater depth; topographic data includes slope, aspect, and elevation variation coefficient; meteorological data includes cumulative rainfall, rainfall intensity, and temperature change rate; vegetation data includes vegetation coverage and vegetation type normalization index; and human activity-related data includes engineering construction disturbance intensity and farmland development ratio. The capture module collects data in real time through distributed sensors deployed in the preset potential landslide impact area. The temporal and spatial resolution of the collected data are dynamically adjusted according to the preset spatiotemporal scale requirements. Among them, the elevation variation coefficient is the ratio of the standard deviation of all elevation sampling point data under the preset spatial granularity in the preset landslide potential impact area to the mean of elevation data in the spatial granularity; the rainfall intensity is the ratio of the cumulative rainfall in the preset time interval collected by the capture module to the duration of the time interval, and is obtained by inversion calculation through real-time data collection by distributed sensors; the vegetation type normalization index is based on the vegetation spectral feature data inverted from high-resolution remote sensing images, and is obtained by normalization processing by mapping the spectral response values ​​of different vegetation types to the [0,1] interval; the engineering construction disturbance intensity is the ratio of the weighted area of ​​various engineering construction disturbance areas in the preset spatial granularity to the total area of ​​the spatial granularity. A module is established to receive data collected by the capture module, and to perform noise reduction, format standardization, and spatiotemporal dimension alignment on the collected data in order to establish a standardized spatiotemporal dataset under a unified coordinate system. The noise reduction processing of the collected data in the module follows the following rules: ; In the formula: The data is standardized and located at spatial coordinates (i,j) and time t after denoising. The original data collected before noise reduction; , These are time and space weighting coefficients; This is the error correction term; The above formula fully considers the decay law of data reliability with distance from the reference time point in the time dimension and the weakening characteristic of data correlation with distance from the reference coordinate in the spatial dimension. It introduces time exponential decay weight and spatial distance decay weight to perform dual calibration of the original data. At the same time, it calculates the deviation between the original data and the overall mean in adjacent time windows as an error correction term to make up for the systematic deviation in the data acquisition process, realizes accurate noise removal of the original data at different time and space locations, and makes the standardized data more in line with the real state of the actual geological environment. in, ; In the formula: As a reference time point, the time node with the best data stability within the collection period was selected; The time decay constant; This is the spatial attenuation coefficient; This is the Euclidean distance between the current spatial coordinates (i,j) and the adjacent reference coordinates; This is the overall mean of the collected data, that is, the statistical mean of the original data collected at all spatiotemporal points; The length of the adjacent time window; This is the original data collected at the k-th time point before the current time t, with spatial coordinates (i,j). in, The preset value range is (0,1]. The larger the spatial scale of the potential landslide impact area, the smaller its value, and vice versa. The core of this denoising process lies in the synergistic effect of dual weighting coefficients, with the time weighting coefficient being... The spatial weighting coefficient is used to characterize the change in data reliability over time using an exponential decay function. The distance decay function reflects the correlation strength of spatially adjacent location data; the product of the two functions acts on the original data, achieving information preservation and noise suppression in both spatiotemporal dimensions. Error correction term. Based on the statistical deviation between the data within adjacent time windows and the overall mean, the systematic deviation in the acquisition process is compensated. This weighted correction method not only preserves the true signal characteristics but also effectively filters out random noise interference. The spatiotemporal alignment process in the module includes spatial coordinate alignment and timestamp synchronization: Spatial coordinate alignment uses the reference control points of a pre-defined unified coordinate system as a reference, and transforms the original coordinates of data collected from different sources to the unified coordinate system using affine transformation formulas: ; In the formula: The target coordinates in the unified coordinate system after transformation; The original coordinates for collecting the data; The spatial resolution adaptation coefficient is the ratio of the spatial resolution of the unified coordinate system to the spatial resolution of the original data. , , where represents the error compensation amount of the original coordinates in the X and Y axes; a, b, d, and e are the linear coefficients of the affine transformation, obtained by solving a system of equations simultaneously with the original coordinates and target coordinates of at least three non-collinear reference control points; c and f are the translation coefficients of the affine transformation, representing the offset of the coordinate origin. The above formula uses the reference control points of the preset unified coordinate system as a reference to construct an affine transformation relationship including linear coefficients and translation coefficients. At the same time, it incorporates the spatial resolution adaptation coefficient and the coordinate axis error compensation amount that integrates the nominal error of the equipment, the influence of environmental interference and the reference deviation. By solving the key parameters by using the coordinates of at least three non-collinear reference control points, it ensures that when the original coordinates are transformed to the unified coordinate system, it not only satisfies the accurate mapping of spatial position, but also adapts to the resolution differences of different data, thereby improving the consistency and usability of multi-source spatial data. in, The X-axis coordinate error compensation amount is calculated by weighting the nominal positioning error of the acquisition device in the X-axis direction, the influence coefficient of environmental interference on the X-axis positioning, and the statistical value of the X-axis deviation between the original coordinates and the reference control point according to the preset weights. The Y-axis coordinate error compensation amount is calculated by weighting the nominal positioning error of the acquisition device in the Y-axis direction, the influence coefficient of environmental interference on the Y-axis positioning, and the statistical value of the Y-axis deviation between the original coordinates and the reference control point according to the preset weights. Timestamp synchronization is based on UTC standard time. By calculating the time deviation value of each acquisition device, the timestamp of the acquired data is corrected to make the time dimension of all data uniform. The time deviation value is periodically calibrated by satellite time signal, and the calibration cycle is set according to the real-time requirements of the data. The extraction module is used to extract landslide disaster impact feature parameters at different time scales and spatial granularities from standardized spatiotemporal datasets and to build a feature parameter library. The landslide hazard impact characteristic parameters extracted by the extraction module include spatiotemporal sensitivity parameters, dynamic induction parameters, and vulnerability parameters of the affected body. Among them, the spatiotemporal sensitivity parameters are: ; In the formula: For spatial coordinates (i,j) and time t, the spatiotemporal sensitivity parameter is used. The total number of factors affecting sensitivity; The weight coefficient of the k-th factor; Let be the standardized value of the k-th factor at spatial coordinates (i,j) and time t; This is a time-scale correction factor; This is the spatial granularity correction factor; The above formula integrates key influencing factors such as slope and vegetation coverage. By reasonably allocating the weights of each factor and combining them with their standardized values, a comprehensive calculation is performed. At the same time, the influence of different time scales and spatial granularities on sensitivity assessment is fully considered, and corresponding correction coefficients are introduced for dynamic adjustment. This comprehensively quantifies the landslide sensitivity basis at different time and space locations, so that the calculation results can accurately reflect the comprehensive effect of various factors on landslide sensitivity in different time and space scenarios. The dynamic induced parameters are obtained by weighting the real-time change rate and cumulative effect of the dynamic induced factors. The dynamic induced factors include the change rate of rainfall intensity, the increment of engineering construction disturbance intensity, and the change rate of groundwater level. The calculation logic is as follows: taking a preset time window as the unit, the instantaneous change contribution value and the cumulative superposition contribution value of each dynamic induced factor within the window are calculated respectively. Then, the two types of contribution values ​​are weighted and summed by the dynamic weight coefficient determined based on the factor correlation analysis to obtain the dynamic induced parameters corresponding to spatial coordinates (i,j) and time t. The dynamic weight coefficient is adaptively adjusted with the change amplitude of the factor within the time window. The larger the change amplitude, the higher the weight coefficient of the corresponding factor. The vulnerability parameters of disaster-bearing bodies are obtained by weighted calculation of vulnerability factors of different types of disaster-bearing bodies. Disaster-bearing bodies include buildings, infrastructure, population and ecosystems. Vulnerability factors include seismic resistance level of buildings, anti-slide capacity of infrastructure, population density and ecosystem resilience. The calculation logic is as follows: First, the vulnerability factors of each disaster-bearing body are standardized. Then, the static weight of each factor is determined according to the importance of the disaster-bearing body. Combined with the distribution density coefficient of the disaster-bearing body at spatial coordinates (i,j), the standardized vulnerability factors are weighted and summed to obtain the vulnerability parameters of the disaster-bearing body at that spatial location. The distribution density coefficient is determined by the ratio of the spatial distribution data of the disaster-bearing body to the area of ​​the region. in, Including slope factor, vegetation cover factor, soil and rock stability factor, and fault influence factor. The values ​​of all values ​​are within the range of (0,1), and , Determined based on the ratio of the current time scale to the reference time scale. Determined based on the ratio of the current spatial granularity to the reference spatial granularity; The modeling module is used to build landslide disaster risk assessment models that are adapted to different spatiotemporal scenarios by applying a feature parameter library; The landslide disaster risk assessment model constructed by the modeling module is a spatiotemporal hybrid fusion model, including a time series prediction sub-model and a spatial heterogeneity adaptation sub-model; Among them, the time series prediction sub-model is based on an improved long short-term memory network, which dynamically adjusts the weights of features at different time scales by introducing spatiotemporal gating units into the hidden layers of the LSTM. The spatial heterogeneity adaptor sub-model is based on the geographic weighted random forest algorithm, which incorporates spatial coordinates as covariates into the model training and adaptively adjusts the weights of different spatial region features. The modeling module trains a time series prediction sub-model using the time series feature dataset in the feature parameter library to obtain the time dimension risk contribution value; it trains a spatial heterogeneity adaptation sub-model using the spatial feature dataset to obtain the spatial dimension risk contribution value; and then, by using preset fusion weights, it superimposes the time dimension risk contribution value and the spatial dimension risk contribution value to obtain a comprehensive risk assessment model adapted to different spatiotemporal scenarios. ; In the formula: The risk contribution value of the time dimension corresponding to spatial coordinates (i,j) and time t; The Sigmoid activation function is used to map the spatiotemporal gated output to the [0,1] interval; It is a spacetime gating factor; This represents the total number of features on the time scale. is the dynamic weighting coefficient of the p-th time-scale feature; Let be the standardized value of the p-th time-scale feature at the spatiotemporal location (i,j,t); This is a time trend correction term; The above formula addresses the risk contribution value in the time dimension by employing an improved long short-term memory network. By introducing a spatiotemporal gating unit and combining it with an activation function, the weight ratio of features at different time scales is dynamically adjusted. A correction term based on fitting historical data time trends is incorporated to accurately capture the dynamic changes in landslide risk in the time dimension. For the risk contribution value in the spatial dimension, a geographically weighted random forest algorithm is used, incorporating spatial coordinates as an influencing factor into the model. Through the distance decay effect between the current location and the spatial benchmark point and local weight allocation, the influence intensity of different spatial region features is adaptively adjusted. The mean of the prediction residuals of surrounding neighboring samples is used for error correction, thereby achieving accurate decomposition and calculation of the risk contribution value in the spatiotemporal dimension. The spatial dimension risk contribution value corresponding to spatial coordinates (i,j) and time t; This represents the total number of spatial characteristic factors; Let be the local weight coefficient of the k-th spatial feature factor at position (i,j); Let be the standardized value of the k-th spatial feature factor at its spatiotemporal location (i,j,t); Given the current position (i,j) and the spatial reference point The Euclidean distance; This is the spatial attenuation threshold; This is a spatial error correction term; in, Take each time and The mean of the sum of the products of; The value range of is (0,1), and ; , This represents the trend correction coefficient, determined based on the time trend fitting results of historical data. Indicates the start time of the time series; The value range is (0,1), and ; Based on the spatial continuity of the regional geological structure; Its value is the mean of the prediction residuals of all sample points within a preset neighborhood around the location; The spatiotemporal gating unit adds a spatial coupling gate to the traditional LSTM gating mechanism. It receives the concatenated input of the spatial feature vector of the current spatiotemporal location and the temporal hidden state vector, and outputs the spatiotemporal coupling coefficient through the Sigmoid activation function, dynamically adjusting the sensitivity of the temporal hidden state to the spatial location; the spatiotemporal gating factor... The model integrates the hidden state output of LSTM with the output of spatial coupling gates, enabling it to dynamically adjust the retention and forgetting of historical information based on the geological characteristics of different spatial locations when processing long-term series, thus achieving collaborative modeling of temporal evolution and spatial heterogeneity. The determination of the fusion weights is based on the spatiotemporal contribution variance analysis of the training samples. During the model training phase, the variance of the time-dimensional risk contribution value output by the time series prediction sub-model is calculated over all samples. And the variance of the spatial dimension risk contribution value output by the spatial heterogeneity adaptor model over all samples. Then, the fusion weights are determined based on the variance proportions, i.e., the fusion weights in the time dimension. Spatial dimension fusion weight The overall risk output is The spatiotemporal coordination correction coefficient is determined based on the ratio of the ultra-long-term time span to the training data time span, as well as the complexity index of the geological structure of the target spatial location, and is obtained by looking up a preset correction coefficient table; The prediction module is used to call the risk assessment model, input long-term spatiotemporal characteristic prediction data, and perform quantitative prediction of long-term landslide disaster risk. The ultra-long-term spatiotemporal feature prediction data input into the prediction module are obtained through the following methods: ; In the formula: For spatial coordinates (i,j) and future time... Corresponding ultra-long-term spatiotemporal feature prediction data, This is a preset ultra-long-term time interval; The number of historical data periods used for trend analysis; The weighting coefficients for the p-th historical period data; Spatial coordinates (i,j), historical time Standardized observation data, where T is the length of the historical data period; The trend correction coefficient for the p-th historical period data; This refers to the number of dynamic inducing factors, including inducing factors such as rainfall, temperature, and human activities; The influence weight of the q-th type of inducing factor; For the qth type of inducing factor in the future time The predicted value; Let be the regional adaptation coefficient of the q-th type of inducing factor in spatial coordinates (i,j); For spatiotemporal coordination correction coefficients; The above formula, by taking into account both the trend continuity of historical data and the future impact of dynamic inducing factors, assigns weights to data from different historical periods according to their correlation and combines them with trend correction coefficients to capture the effect of historical patterns on the long-term. At the same time, it integrates the future predicted values, regional adaptation coefficients, and impact weights of dynamic inducing factors such as rainfall and temperature to quantify the future effects of various inducing factors. Then, it adapts the coupling effect of spatiotemporal heterogeneity and uncertainty in the long-term through spatiotemporal collaborative correction coefficients, comprehensively reflecting the real changing trend of spatiotemporal characteristics at the long-term time scale, and providing data support for long-term risk prediction. For long-term prediction, this invention adopts a dual-scale time memory mechanism in the model structure. Compared to short-term forecasting, which focuses on capturing recent high-frequency fluctuations and uses a shorter LSTM memory window, ultra-long-term forecasting uses a dual-path LSTM structure with both long and short paths. The long path branch has a longer memory window to capture seasonal changes and geological evolution trends, while the short path branch retains the ability to respond to recent anomalous events. The outputs of the two paths are fused through adaptive weight coefficients, which are determined based on the ratio of the ultra-long-term time interval to the maximum forecast time span. This ensures that ultra-long-term forecasting maintains its sensitivity to long-term trends without losing its ability to capture abrupt triggering factors. Future predicted values ​​of dynamic induced factors The data acquisition employs a classification-based prediction method. For precipitation-related meteorological inducing factors, medium- and long-term precipitation forecast data released by the regional meteorological department are used, and probability corrections are made based on the statistical distribution characteristics of historical rainfall during the same period. For anthropogenic inducing factors such as engineering construction disturbances, based on the future engineering project layout information provided by the regional land planning department, the disturbance intensity index of each planned project is quantified, and the cumulative disturbance prediction value at each spatial coordinate is obtained through spatial overlay analysis. For groundwater level change-related geological inducing factors, a groundwater flow numerical simulation method is used. Historical observation data, hydrogeological parameters, and future rainfall prediction data are input, and the groundwater flow motion equation is solved to obtain the predicted value of future groundwater level spatial distribution, which is then standardized and used as input. in, The value range is preset to (0,1), and the sum of all period weight coefficients is 1. The value is larger when the correlation between the corresponding historical period data and the ultra-long-term prediction target is higher, and the value is smaller when the correlation is lower. The preset value range is [-0.5, 0.5]. The value is smaller when the predicted trend of the far-term time is more consistent with the trend of the corresponding historical cycle, and vice versa. The value range is preset to (0,1). The higher the contribution of this type of inducing factor to the occurrence of landslide disaster, the larger the value, and vice versa. The value is larger when the predicted intensity, frequency or range of influence of the inducing factor is greater in the long term, and smaller when the predicted intensity, frequency or range of influence of the inducing factor is less in the long term. The value range is preset to (0,2). The value is larger when the regional topography and geological conditions corresponding to the spatial coordinates (i,j) are more sensitive to the response of the qth type of inducing factor, and vice versa. The value range is preset to (0.8, 1.2). When the coupling effect of spatial heterogeneity and temporal uncertainty under the ultra-long-term time scale is smaller, the value is closer to 1, and vice versa. The output module is used to receive the quantitative prediction results from the prediction module and generate a risk notification message. The risk notification message output by the output module includes the prediction results of the prediction module, the ultra-long-term spatiotemporal feature prediction data input by the prediction module, and the risk level determined by comparing the preset threshold with the quantitative prediction results. The system uses a parallel task processing mechanism to apply to several target landslide disaster risk assessment areas. The prediction results of each processing task are collected in the output module. When the output module outputs the risk notification message, it also includes an electronic map with the prediction results of each target landslide disaster risk assessment area marked. The capture module interacts with the establishment module via a wireless network. The establishment module interacts with the extraction module and the modeling module via a wireless network. The modeling module interacts with the prediction module via a wireless network. The prediction module and the modeling module interact with the output module via a wireless network.

[0022] In this embodiment, the capture module collects geological, topographical, meteorological, vegetation, and human activity-related data in the continuous time and spatial dimensions within a preset potential landslide impact area, and forwards it to the establishment module in real time. The establishment module receives the data collected by the capture module and performs noise reduction, format standardization, and spatiotemporal alignment on the collected data to establish a standardized spatiotemporal dataset under a unified coordinate system. The extraction module further extracts landslide disaster impact feature parameters of different time scales and spatial granularities from the standardized spatiotemporal dataset to construct a feature parameter library. Then, the modeling module uses the feature parameter library to construct a landslide disaster risk assessment model adapted to different spatiotemporal scenarios. The prediction module calls the risk assessment model, inputs ultra-long-term spatiotemporal feature prediction data, and performs quantitative prediction of ultra-long-term landslide disaster risk. Finally, the output module receives the quantitative prediction results from the prediction module and generates a risk notification message.

[0023] In the above embodiments, the system integrates and optimizes multi-dimensional geological, meteorological and other data to accurately adapt to different time and space scenarios for landslide disaster risk assessment. It can effectively quantify long-term risks, clearly present risk levels and regional distribution, and provide reliable support for disaster prevention and mitigation decisions. This can achieve the effect of deploying protective measures in advance, reducing disaster losses, and protecting people's lives, property and ecological environment safety.

[0024] Example 2: At the implementation level, based on Example 1, this example refers to... Figure 2 A further detailed explanation of the landslide hazard risk assessment system based on multiple spatiotemporal scales in Example 1 is provided below: A landslide hazard risk assessment method based on multiple spatiotemporal scales includes: Collect geological, topographical, and meteorological data related to the pre-defined potential landslide impact area; The collected data is denoised and formatted, and spatiotemporal alignment is achieved through coordinate transformation and timestamp synchronization to establish a standardized spatiotemporal dataset in a unified coordinate system. Spatiotemporal sensitivity, dynamic induction, and vulnerability characteristics of disaster-bearing bodies are extracted from standardized spatiotemporal datasets to construct a feature parameter library covering different spatiotemporal scales; A spatiotemporal hybrid fusion model is constructed based on a feature parameter library. The time series prediction sub-model and the spatial heterogeneity adaptation sub-model are trained separately and then fused to obtain a risk assessment model that is adapted to multiple spatiotemporal scenarios. Acquire long-term spatiotemporal characteristic prediction data and input them into the risk assessment model; then, through model calculation, perform quantitative prediction of long-term landslide disaster risk. The system aggregates quantitative forecast results and combines them with long-term input data and risk level assessment results to generate a risk notification message.

[0025] System application example 3: Taking a potential landslide-affected area in a mountainous region in Southwest China as the assessment object, this multi-temporal and spatial scale landslide disaster risk assessment system was used to carry out risk assessment work.

[0026] First, the system's capture module collects relevant data in real time across continuous time and space dimensions through distributed sensors deployed in the area. This includes geological data such as soil cohesion and internal friction angle, topographic data such as slope and aspect, meteorological data such as cumulative rainfall and rainfall intensity, vegetation data such as vegetation coverage and vegetation type normalization index, and human activity-related data such as engineering construction disturbance intensity and cultivated land development ratio. During the collection process, the time and spatial resolution are dynamically adjusted according to the preset time and space scale requirements, and the collected data is forwarded to the establishment module in real time.

[0027] After the module receives the data, it first performs noise reduction on the raw collected data. After time weighting, spatial weighting adjustment and error correction, standardized data for each spatiotemporal point is obtained. Then, the format standardization process is completed, and spatiotemporal dimension alignment is carried out. Spatially, the original coordinates of data from different sources are transformed to the unified coordinate system using the reference control points of the preset unified coordinate system. Temporally, UTC standard time is used as the reference, and the time deviation of each acquisition device is periodically calibrated through satellite time synchronization signal to achieve timestamp synchronization of all data. Finally, a standardized spatiotemporal dataset under the unified coordinate system is established.

[0028] The extraction module extracts landslide hazard impact characteristic parameters from the dataset and constructs a characteristic parameter library. Among them, the spatiotemporal sensitivity parameter, after being weighted by relevant factors and corrected for time scale and spatial granularity, has a value of 0.73 for a certain spatial coordinate at a specific time; the dynamic induction parameter, combining the instantaneous changes and cumulative superposition contributions of factors such as rainfall intensity change rate and engineering construction disturbance intensity increment, has a value of 0.67 for this spatiotemporal location after dynamic weighted summation; the vulnerability parameter of the disaster-bearing body integrates relevant vulnerability factors of buildings, infrastructure, population and ecosystem, and after being weighted according to the static weight of the importance of the disaster-bearing body and the distribution density coefficient, has a value of 0.59 for this area.

[0029] The modeling module constructs a spatiotemporal hybrid fusion model based on a feature parameter library. It trains an improved long short-term memory network time series prediction sub-model using a time series feature dataset, obtaining a time dimension risk contribution value of 0.45 for the region. It trains a spatial heterogeneity adaptation sub-model using a geographically weighted random forest algorithm using a spatial feature dataset, obtaining a spatial dimension risk contribution value of 0.41. The two contribution values ​​are then superimposed according to preset fusion weights to form a comprehensive risk assessment model adapted to the spatiotemporal scenario of the region.

[0030] The prediction module takes into account the ultra-long-term spatiotemporal characteristic prediction data (which is obtained by combining historical cycle data trends, future prediction values ​​of dynamic inducing factors such as rainfall, and regional adaptability, and after collaborative correction), calls the comprehensive risk assessment model to perform ultra-long-term risk quantification prediction, and finally obtains an ultra-long-term landslide disaster risk quantification value of 0.64 for the region.

[0031] After receiving the quantitative prediction results, the output module generates a risk notification message. The message includes the quantitative prediction results, the input long-term spatiotemporal feature prediction data, and the "medium risk" level determined by comparison with preset thresholds. It also includes an electronic map that marks the prediction results of the potential impact area of ​​the landslide in the mountainous area.

[0032] In summary, the system in the above embodiments accurately collects multi-dimensional spatiotemporal data related to geology, topography, meteorology, vegetation, and human activities. After denoising, standardization, and spatiotemporal dimension alignment, the data quality is optimized. The system comprehensively extracts the influence feature parameters at different time scales and spatial granularities, constructs an assessment model adapted to diverse spatiotemporal scenarios, and achieves ultra-long-term risk quantification prediction by dynamically adjusting the influence weights of spatiotemporal dimensions. It accurately defines the risk level, outputs a risk notification message with an electronic map annotating the prediction results for multiple regions, supports parallel processing to improve efficiency, and significantly improves the accuracy, timeliness, and adaptability of landslide disaster risk assessment, providing predictive support for disaster prevention and control planning and decision-making.

[0033] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A landslide hazard risk assessment system based on multiple spatiotemporal scales, characterized in that, include: The capture module is used to collect geological, topographical, meteorological, vegetation and human activity-related data in the continuous time and spatial dimensions within the preset potential landslide impact area, and forward it to the establishment module in real time. A module is established to receive data collected by the capture module, and to perform noise reduction, format standardization, and spatiotemporal dimension alignment on the collected data in order to establish a standardized spatiotemporal dataset under a unified coordinate system. The extraction module is used to extract landslide disaster impact feature parameters at different time scales and spatial granularities from standardized spatiotemporal datasets and to build a feature parameter library. The modeling module is used to build landslide disaster risk assessment models that are adapted to different spatiotemporal scenarios by applying a feature parameter library; The prediction module is used to call the risk assessment model, input long-term spatiotemporal characteristic prediction data, and perform quantitative prediction of long-term landslide disaster risk. The output module is used to receive the quantitative prediction results from the prediction module and generate a risk notification message.

2. The landslide disaster risk assessment system based on multiple spatiotemporal scales according to claim 1, characterized in that, The geological data collected by the capture module includes soil and rock cohesion, internal friction angle, fault activity rate, and groundwater depth; topographic data includes slope, aspect, and elevation variation coefficient; meteorological data includes cumulative rainfall, rainfall intensity, and temperature change rate; vegetation data includes vegetation coverage and vegetation type normalization index; and human activity-related data includes engineering construction disturbance intensity and cultivated land development ratio. The capture module collects data in real time through distributed sensors deployed in the preset potential landslide impact area. The temporal and spatial resolution of the collected data are dynamically adjusted according to the preset spatiotemporal scale requirements.

3. The landslide disaster risk assessment system based on multiple spatiotemporal scales according to claim 1, characterized in that, The noise reduction processing of the collected data in the establishment module follows the following principle: ； In the formula: The data is standardized and located at spatial coordinates (i,j) and time t after denoising. The original data collected before noise reduction; , These are time and space weighting coefficients; This is the error correction term.

4. The landslide disaster risk assessment system based on multiple spatiotemporal scales according to claim 3, characterized in that, The spatiotemporal alignment process in the establishment module includes spatial coordinate alignment and timestamp synchronization: Spatial coordinate alignment uses the reference control points of a pre-defined unified coordinate system as a reference, and transforms the original coordinates of data collected from different sources to the unified coordinate system using affine transformation formulas: ； In the formula: The target coordinates in the unified coordinate system after transformation; The original coordinates for collecting the data; Spatial resolution adaptation coefficient; , This represents the error compensation amount for the original coordinates in the X and Y axis directions; a, b, d, and e are the linear coefficients of the affine transformation; c and f are the translation coefficients of the affine transformation. Timestamp synchronization is based on UTC standard time. By calculating the time deviation value of each acquisition device, the timestamp of the acquired data is corrected to make the time dimension of all data uniform.

5. The landslide disaster risk assessment system based on multiple spatiotemporal scales according to claim 1, characterized in that, The landslide disaster impact characteristic parameters extracted by the extraction module include spatiotemporal sensitivity parameters, dynamic induction parameters, and vulnerability parameters of the disaster-bearing body, wherein the spatiotemporal sensitivity parameters are: ； In the formula: For spatial coordinates (i,j) and time t, there are spatiotemporal sensitivity parameters. The total number of factors affecting sensitivity; The weight coefficient of the k-th factor; Let be the standardized value of the k-th factor at spatial coordinates (i,j) and time t; This is a time-scale correction factor; This is the spatial granularity correction factor; in, Including slope factor, vegetation cover factor, soil and rock stability factor, and fault influence factor. The values ​​of all values ​​are within the range of (0,1), and , Determined based on the ratio of the current time scale to the reference time scale. Determined based on the ratio of the current spatial granularity to the reference spatial granularity.

6. The landslide disaster risk assessment system based on multiple spatiotemporal scales according to claim 1, characterized in that, The landslide disaster risk assessment model constructed by the modeling module is a spatiotemporal hybrid fusion model, including a time series prediction sub-model and a spatial heterogeneity adaptation sub-model; Among them, the time series prediction sub-model is based on an improved long short-term memory network, which dynamically adjusts the weights of features at different time scales by introducing spatiotemporal gating units into the hidden layers of the LSTM. The spatial heterogeneity adaptor sub-model is based on the geographic weighted random forest algorithm, which incorporates spatial coordinates as covariates into the model training and adaptively adjusts the weights of different spatial region features. The modeling module trains a time series prediction sub-model using the time series feature dataset in the feature parameter library to obtain the time dimension risk contribution value; it trains a spatial heterogeneity adaptation sub-model using the spatial feature dataset to obtain the spatial dimension risk contribution value; and then superimposes the time dimension risk contribution value and the spatial dimension risk contribution value using preset fusion weights to obtain a comprehensive risk assessment model adapted to different spatiotemporal scenarios. ； In the formula: The risk contribution value of the time dimension corresponding to spatial coordinates (i,j) and time t; The Sigmoid activation function is used to map the spatiotemporal gated output to the [0,1] interval; It is a spacetime gating factor; This represents the total number of features on the time scale. is the dynamic weighting coefficient of the p-th time-scale feature; Let be the standardized value of the p-th time-scale feature at the spatiotemporal location (i,j,t); This is a time trend correction term; The spatial dimension risk contribution value corresponding to spatial coordinates (i,j) and time t; This represents the total number of spatial characteristic factors; Let be the local weight coefficient of the k-th spatial feature factor at position (i,j); Let be the standardized value of the k-th spatial feature factor at its spatiotemporal location (i,j,t); Given the current position (i,j) and the spatial reference point The Euclidean distance; This is the spatial attenuation threshold; This is the spatial error correction term.

7. The landslide disaster risk assessment system based on multiple spatiotemporal scales according to claim 1, characterized in that, The ultra-long-term spatiotemporal feature prediction data input into the prediction module is obtained through the following methods: ； In the formula: For spatial coordinates (i,j) and future time... Corresponding ultra-long-term spatiotemporal feature prediction data, This is a preset ultra-long-term time interval; The number of historical data periods used for trend projection; The weighting coefficients for the p-th historical period data; Spatial coordinates (i,j), historical time Standardized observation data, where T is the length of the historical data period; The trend correction coefficient for the p-th historical period data; This represents the number of types of dynamic inducing factors; The influence weight of the q-th type of inducing factor; For the qth type of inducing factor in the future time The predicted value; Let be the regional adaptation coefficient of the q-th type of inducing factor in spatial coordinates (i,j); This is the spatiotemporal coordination correction coefficient.

8. The landslide disaster risk assessment system based on multiple spatiotemporal scales according to claim 1, characterized in that, The risk notification message output by the output module includes the prediction results of the prediction module, the ultra-long-term spatiotemporal feature prediction data input by the prediction module, and the risk level determined by comparing the preset threshold with the quantitative prediction results. The system uses a parallel task processing mechanism to apply to several target landslide disaster risk assessment areas. The prediction results of each processing task are collected in the output module. When the output module outputs the risk notification message, it also includes an electronic map with the prediction results of each target landslide disaster risk assessment area.

9. A landslide hazard risk assessment system based on multiple spatiotemporal scales according to claim 1, characterized in that, The capture module is interconnected with the establishment module via a wireless network. The establishment module is interconnected with the extraction module and the modeling module via a wireless network. The modeling module is interconnected with the prediction module via a wireless network. The prediction module and the modeling module are interconnected with the output module via a wireless network.

10. A landslide hazard risk assessment method based on multiple spatiotemporal scales, wherein the method is an implementation method of the landslide hazard risk assessment system based on multiple spatiotemporal scales as described in any one of claims 1-9, characterized in that, include: Collect geological, topographical, and meteorological data related to the pre-defined potential landslide impact area; The collected data is denoised and formatted, and spatiotemporal alignment is achieved through coordinate transformation and timestamp synchronization to establish a standardized spatiotemporal dataset in a unified coordinate system. Spatiotemporal sensitivity, dynamic induction, and vulnerability characteristics of disaster-bearing bodies are extracted from standardized spatiotemporal datasets to construct a feature parameter library covering different spatiotemporal scales; A spatiotemporal hybrid fusion model is constructed based on a feature parameter library. The time series prediction sub-model and the spatial heterogeneity adaptation sub-model are trained separately and then fused to obtain a risk assessment model that is adapted to multiple spatiotemporal scenarios. Acquire long-term spatiotemporal characteristic prediction data and input them into the risk assessment model; then, through model calculation, perform quantitative prediction of long-term landslide disaster risk. The system aggregates quantitative forecast results and combines them with long-term input data and risk level assessment results to generate a risk notification message.

Images