Geological safety risk dynamic assessment method based on multi-source data fusion

Abstract

The invention relates to the technical field of geological engineering, and discloses a geological safety risk dynamic assessment method based on multi-source data fusion, which comprises the following specific steps: step 1, collecting and standardizing multi-source geological data; 2, performing semantic fusion and conflict resolution on the geologic features; 3, constructing a dynamic risk assessment model; 4, risk situation real-time updating and early warning are carried out; the satellite remote sensing system in the first step adopts the synthetic aperture radar interference measurement technology, the spatial resolution is better than 3 meters, the revisit period is shorter than 7 days, and the earth surface deformation monitoring precision reaches the millimeter level. Through standardized processing and semantic fusion of the multi-source geological data, the problem of heterogeneous data integration is effectively solved, the data basic quality of risk assessment is improved, a space-time coupling neural network model is adopted, nonlinear features and space-time correlation characteristics of geological risk evolution are accurately captured, and prediction precision and timeliness are improved.

Description

Technical Field

[0001] This invention relates to the field of geological engineering technology, specifically to a dynamic assessment method for geological safety risks based on multi-source data fusion. Background Technology

[0002] With the increasing demand for geological safety risk prevention and control, risk assessment technology based on multi-source data fusion has become a core means to support the site selection of major projects, regional ecological security and disaster emergency response. Geological safety risks have significant multi-factor coupling, spatiotemporal heterogeneity and dynamic evolution characteristics. Its assessment needs to integrate multi-dimensional information such as geological structure, surface deformation, meteorology and hydrology and human engineering activities to achieve accurate identification and trend prediction of disaster events such as landslides, collapses and ground subsidence. Especially against the background of intensified global climate change and high-intensity development, the state of geological bodies shows a high frequency of disturbance and nonlinear abrupt change trend, which puts forward higher requirements for the real-time performance, spatial coverage and multi-source information coordination capabilities of assessment methods.

[0003] Among them, multi-source data fusion technology is regarded as a key path to improve the accuracy of geological safety risk assessment. This technology aims to integrate heterogeneous information such as satellite remote sensing imagery, UAV aerial survey point cloud, ground sensor time-series data, and historical geological survey data, and construct a unified risk characterization system through spatial alignment and feature complementarity. However, existing fusion methods still face systemic bottlenecks in practical applications: on the one hand, multi-source data have significant differences in format, coordinate system, sampling frequency, and quality level. For example, remote sensing imagery uses GeoTIFF format while lidar data is mostly in LAS structure, and data from different sources often have inconsistent coordinate references (such as the mixing of WGS84 and local independent coordinate systems) and a large number of missing values; on the other hand, current fusion strategies mostly rely on static weighting or simple stitching, lacking a standardized verification mechanism for the physical semantics of geological data, making it difficult to effectively resolve data conflicts and redundancy, resulting in fusion results deviating from the true geological state.

[0004] While existing technologies attempt to introduce dynamic evaluation frameworks, they still fall short in terms of real-time response and evolution prediction, specifically: Newly added monitoring data is difficult to integrate efficiently with historical knowledge bases, especially in data-sparse areas (such as remote mountainous areas or monitoring blind spots), where models are prone to misjudgments due to insufficient input information; at the same time, the computational complexity of high-dimensional multi-source data increases dramatically, which restricts the online update capability of the assessment system.

[0005] In addition, existing models generally use linear or shallow nonlinear assumptions to characterize the risk evolution process, which cannot accurately capture the threshold mutations and cascade effects in the process of geological disaster incubation. This causes the assessment results to lag behind the actual risk situation, making it difficult to meet the timeliness and reliability requirements of geological research institutes in the safety supervision of major projects and early warning of disasters. Summary of the Invention

[0006] (a) Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a dynamic assessment method for geological safety risks based on multi-source data fusion, which solves the problem that the aforementioned technologies are still insufficient in terms of real-time response and evolution prediction.

[0007] (II) Technical Solution To achieve the above objectives, the present invention provides a method for dynamic assessment of geological safety risks based on multi-source data fusion, comprising the following specific steps: Step 1: Multi-source geological data acquisition and standardization processing. Time series data of regional surface deformation are obtained through satellite remote sensing system, high-precision topographic point cloud data are collected by UAV aerial survey platform, and real-time monitoring data of ground sensor network and historical geological survey database are accessed at the same time to obtain multi-source heterogeneous data. Coordinate system transformation, missing value interpolation and repair and quality consistency verification are performed on multi-source heterogeneous data. Step 2: Geological feature semantic fusion and conflict resolution. Based on the multi-source heterogeneous data processed in Step 1, geological tectonic activity indicators, slope stability parameters, hydrogeological characteristic factors and human engineering activity intensity index are extracted. Cross-source data association is established through multi-level feature matching algorithm. A weighted fusion strategy based on geological and physical constraints is adopted to eliminate data redundancy and semantic conflicts to obtain the fusion feature matrix. Step 3: Construct a dynamic risk assessment model. Input the fusion feature matrix obtained in Step 2 into a spatiotemporal coupled neural network. The spatiotemporal coupled neural network includes a long short-term memory module to capture temporal evolution patterns, and an integrated graph convolution module to model spatial correlation characteristics. Through end-to-end training, the nonlinear mapping relationship of geological risks is learned, thereby obtaining the dynamic risk assessment model. Step 4: Real-time updates and early warnings of risk situation. Based on the dynamic risk assessment model built in Step 3, online reasoning is performed on the accessed real-time monitoring data stream to generate a regional geological safety risk level distribution map. When the risk index is detected to exceed the preset threshold or a sudden trend is observed, a multi-level early warning mechanism is automatically triggered. Step 5: Visualization of assessment results and decision support. The regional geological safety risk level distribution map obtained in Step 4 is dynamically rendered using a 3D geographic information system, providing functions such as risk evolution trajectory tracing, sensitivity analysis, and intervention measure simulation to support scientific decision-making in geological safety management.

[0008] Preferably, in step one, the satellite remote sensing system adopts synthetic aperture radar interferometry technology, with a spatial resolution better than 3 meters, a revisit period of less than 7 days, and a surface deformation monitoring accuracy of millimeters. The UAV aerial survey platform is equipped with a multispectral camera and a lidar sensor, with a point cloud density of more than 200 points per square meter and an elevation measurement error of less than 5 centimeters. The ground sensor network includes crack gauges, inclinometers, and pore water pressure gauges, with a sampling frequency of once per minute. The data is transmitted to the central processing unit in real time via an IoT gateway.

[0009] Preferably, in step one, the coordinate system transformation adopts the seven-parameter Bursa model to perform a high-precision transformation between the local independent coordinate system and the geodetic coordinate system, with the transformation residual controlled within 0.1 meters; The missing value imputation and repair adopts the spatiotemporal kriging interpolation algorithm, which comprehensively considers spatial autocorrelation and temporal continuity; The quality consistency verification is achieved through multi-source data cross-validation, and outliers exceeding 3 times the standard deviation are automatically marked and manually reviewed.

[0010] Preferably, the geological tectonic activity indicators in step two include fault density, joint development degree, and seismic activity frequency; The slope stability parameters include slope, aspect, shear strength of soil and rock mass, and groundwater depth; The hydrogeological characteristic factors include rainfall infiltration coefficient, runoff modulus, and soil saturated hydraulic conductivity; The human engineering activity intensity index is quantified by combining nighttime light data, engineering excavation scale, and traffic load density.

[0011] Preferably, the multi-level feature matching algorithm in step two adopts an improved Hungarian algorithm to establish the optimal correspondence under the dual constraints of feature space and geographic space; A weighted fusion strategy based on geophysical constraints introduces a fuzzy comprehensive evaluation method. The weight coefficients are dynamically adjusted according to the reliability, timeliness, and spatial representativeness of the data source. The correlation coefficient between the fusion result and the field verification data is greater than 0.9.

[0012] Preferably, in step three, the spatiotemporal coupled neural network adopts an encoder-decoder architecture. The encoder part contains three layers of long short-term memory units for extracting temporal features, and the decoder part integrates two layers of graph convolutional networks for modeling spatial dependencies. The network training adopts an adaptive moment estimation algorithm, with an initial learning rate of 0.001, which decays to 0.8 times the original rate every 50 training rounds. The training termination condition is that the validation set loss function no longer decreases for 10 consecutive rounds.

[0013] Preferably, in step three, the end-to-end training uses more than 100,000 sets of historical geological disaster case data, and the samples cover the main disaster types including but not limited to landslides, collapses, and ground subsidence. Five-fold cross-validation is used during the training process to prevent overfitting.

[0014] Preferably, the online inference engine in step four adopts a streaming computing framework, which supports processing data inflows from 1,000 monitoring points per second; The risk level distribution map is divided into four levels: extremely high risk, high risk, medium risk, and low risk, with a spatial grid accuracy of 10 meters by 10 meters. The multi-level early warning mechanism is divided into four levels: blue, yellow, orange, and red, based on the degree to which the risk index exceeds the threshold. Early warning information is released through multiple channels, including SMS, platform push notifications, and audible and visual alarms.

[0015] Preferably, the mutation trend detection in step four adopts a change point analysis algorithm, which uses Bayesian online change point detection technology to identify structural change points in the risk sequence in real time with a detection delay of less than 5 minutes. It has the ability to effectively identify both slow-accumulating and sudden geological disasters.

[0016] Preferably, in step five, the three-dimensional geographic information system supports multi-dimensional data overlay display, including elevation models, geological profiles, monitoring points, and risk cloud maps; The risk evolution trajectory backtracking function can reproduce the risk distribution at any point in the past 72 hours. The sensitivity analysis assesses the stability of risk outcomes by perturbing key input parameters; The intervention simulation can predict the improvement effect of different engineering governance schemes on the risk level.

[0017] (III) Beneficial Effects This invention provides a dynamic assessment method for geological safety risks based on multi-source data fusion. It has the following beneficial effects: (1) This invention effectively solves the problem of heterogeneous data integration through the standardized processing and semantic fusion of multi-source geological data, improves the data foundation quality of risk assessment, and adopts a spatiotemporal coupled neural network model to accurately capture the nonlinear characteristics and spatiotemporal correlation characteristics of geological risk evolution, thereby improving the prediction accuracy and timeliness.

[0018] (2) The present invention can quickly respond to changes in geological conditions through the established real-time update and early warning mechanism, providing timely decision support for disaster prevention and control. Furthermore, the integrated visualization and decision support functions enhance the intuitiveness and practicality of the assessment results, and strongly support the improvement of the scientific and intelligent level of geological safety management. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the multi-source geological data acquisition and standardization processing flow of the present invention; Figure 2 This is a schematic diagram of the coordinate system transformation operation process of the present invention; Figure 3 This is a schematic diagram of the geological feature semantic fusion and conflict resolution process of the present invention; Figure 4 This is a schematic diagram of the spatiotemporal coupled neural network architecture design of the present invention; Figure 5 This is a schematic diagram of the real-time risk status update and early warning process of the present invention. Detailed Implementation

[0020] 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 embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Please see Figure 1 - Figure 5 This invention provides a dynamic assessment method for geological safety risks based on multi-source data fusion, specifically including the following steps: Step 1: Multi-source geological data acquisition and standardization processing. Specifically, regional surface deformation time series data are acquired through satellite remote sensing systems, high-precision topographic point cloud data are collected by UAV aerial survey platforms, and real-time monitoring data from ground sensor networks and historical geological survey databases are simultaneously accessed to obtain multi-source heterogeneous data. Coordinate system transformation, missing value interpolation and repair, and quality consistency verification are performed on the multi-source heterogeneous data.

[0022] Furthermore, in step one, the satellite remote sensing system adopts synthetic aperture radar interferometry technology, with a spatial resolution better than 3 meters, a revisit period of less than 7 days, and a surface deformation monitoring accuracy of millimeters. The UAV aerial survey platform is equipped with a multispectral camera and a lidar sensor, with a point cloud density of more than 200 points per square meter and an elevation measurement error of less than 5 centimeters. The ground sensor network includes crack gauges, inclinometers, and pore water pressure gauges, with a sampling frequency of once per minute. The data is transmitted to the central processing unit in real time via an IoT gateway.

[0023] The coordinate system transformation adopts the seven-parameter Bursa model to perform high-precision transformation between the local independent coordinate system and the geodetic coordinate system, with the transformation residual controlled within 0.1 meters; The missing value imputation and repair adopts the spatiotemporal kriging interpolation algorithm, which comprehensively considers spatial autocorrelation and temporal continuity; Quality consistency verification is achieved through cross-validation of multi-source data, and outliers exceeding 3 times the standard deviation are automatically marked and manually reviewed.

[0024] Furthermore, the multi-source geological data acquisition process in step one involves a multi-platform collaborative operation mechanism, specifically: Satellite remote sensing systems periodically scan target areas with a fixed orbital period to acquire temporal data on surface deformation. The data format is GeoTIFF, which includes timestamps, deformation values, and confidence scores. After receiving the mission instructions, the UAV aerial survey platform conducts low-altitude flight scans of key areas according to the preset flight path, generates point cloud data in LAS format, and records flight attitude, GPS coordinates and sensor status information at the same time. Each monitoring device in the ground sensor network uploads real-time data to the edge computing node via LoRa or NB-IoT communication protocols, and then aggregates it to the central processing unit via 4G / 5G networks.

[0025] The historical geological survey database is stored in the form of a structured relational database, containing fields such as regional stratigraphy and lithology, tectonic background, historical disaster records and prevention and control engineering archives. Before entering the standardized processing flow, all raw data first undergoes metadata parsing and format recognition, and a unified data access interface specification is established to ensure that data from different sources can be correctly parsed and scheduled by the system.

[0026] The coordinate system transformation process in step one includes the following steps: 1. Based on control points, perform initial registration between the local independent coordinate system and the WGS84 geodetic coordinate system to obtain 7 transformation parameters (3 translation parameters, 3 rotation parameters and 1 scale parameter). 2. The parameters are optimized using the least squares method to make the root mean square error of the transformation residual less than 0.1 meters. Finally, the transformation model is applied to all multi-source data to generate a spatial dataset unified in the WGS84 coordinate system.

[0027] Furthermore, the missing value imputation and repair adopts the spatiotemporal kriging interpolation algorithm, the core of which lies in constructing a spatiotemporal variogram model. This model considers the influence of spatial distance and time interval on data correlation. During the interpolation process, the neighborhood search radius and time window length are dynamically adjusted to ensure that high-confidence imputation results can still be obtained in sparse data areas. Quality consistency verification is achieved through a multi-source data cross-validation mechanism. For example, the surface deformation data obtained by satellite remote sensing is compared with the crack gauge readings in the ground sensor network. If the difference in the change trend of the two in the same location and time period exceeds a preset threshold (such as 3 times the standard deviation), the data point is marked as an anomaly and a manual review process is triggered.

[0028] Step Two: Geological Feature Semantic Fusion and Conflict Resolution. Based on the multi-source heterogeneous data processed in Step One, geological tectonic activity indicators, slope stability parameters, hydrogeological characteristic factors, and human engineering activity intensity indices are extracted. Cross-source data associations are established using a multi-level feature matching algorithm. A weighted fusion strategy based on geological and physical constraints is employed to eliminate data redundancy and semantic conflicts. Specifically: In step two, the geological tectonic activity indicators include fault density, joint development degree, and earthquake frequency; slope stability parameters cover slope, aspect, shear strength of soil and rock mass, and groundwater depth. Hydrogeological characteristics include rainfall infiltration coefficient, runoff modulus, and soil saturated hydraulic conductivity; The Human Engineering Activity Intensity Index is quantified by combining nighttime light data, engineering excavation scale, and traffic load density.

[0029] The multi-level feature matching algorithm adopts an improved Hungarian algorithm to establish the optimal correspondence under the dual constraints of feature space and geographic space; A weighted fusion strategy based on geophysical constraints introduces a fuzzy comprehensive evaluation method. The weight coefficients are dynamically adjusted according to the reliability, timeliness, and spatial representativeness of the data source. The correlation coefficient between the fusion result and the field verification data is greater than 0.9.

[0030] Step two, the geological feature extraction process, first involves spatially gridding the standardized multi-source data, dividing the entire assessment area into regular 10-meter by 10-meter grid units. For each grid unit, the surface deformation rate and cumulative deformation are extracted from satellite remote sensing data; slope, aspect, and terrain roughness are calculated from UAV point cloud data; crack width variation rate, dip angle variation, and pore water pressure dynamics are obtained from ground sensor data; and the lithology, fault distance, and historical disaster records of the unit are retrieved from historical geological databases. Geological tectonic activity indicators are calculated using spatial kernel density estimation methods to determine the spatial distribution of faults and joints. Earthquake density was calculated, and the frequency of earthquakes per unit time was statistically analyzed in conjunction with the regional earthquake catalog. The shear strength of the soil and rock mass in the slope stability parameters was obtained by spatial interpolation of laboratory test data and field investigation reports. The rainfall infiltration coefficient in the hydrogeological characteristic factors was determined by referring to tables based on soil type and vegetation cover, the runoff modulus was calculated by hydrological model simulation, and the soil saturated hydraulic conductivity was assigned a value based on soil texture classification. The intensity index of human engineering activities was comprehensively quantified by extracting the brightness value after radiometric calibration of nighttime light remote sensing images, and combining it with high-resolution optical image identification of the area of ​​engineering excavation area and traffic network density, using a weighted summation method.

[0031] The multi-level feature matching algorithm operates under dual constraints of feature space and geospatial constraints. Feature space constraints require that the same type of geological features (such as slope) extracted from different data sources should be numerically comparable, while geospatial constraints require that the matched feature points should be spatially adjacent. The improved Hungarian algorithm introduces a distance decay function and feature similarity weights to construct a cost matrix. By solving the minimum cost matching problem, the optimal correspondence between cross-source data is established. The weighted fusion strategy based on geophysical constraints adopts a fuzzy comprehensive evaluation method. First, the evaluation index set (reliability, timeliness, and spatial representativeness) of each data source is defined. Then, the membership function of each index is determined based on expert experience or historical verification data. Finally, the comprehensive weight of each data source is obtained through fuzzy synthesis. For example, for risk assessment tasks with high real-time requirements, the timeliness weight of ground sensor data is assigned a higher value; while in data-sparse areas, the spatial representativeness weight of satellite remote sensing data is correspondingly increased. Each element of the fused feature matrix is ​​a weighted average, which retains the complementary advantages of multi-source information while effectively suppressing noise and bias from a single data source.

[0032] Step 3: Construct a dynamic risk assessment model. Input the fused feature matrix obtained in Step 2 into a spatiotemporally coupled neural network. This network includes a long short-term memory module to capture temporal evolution patterns, and an integrated graph convolution module to model spatial correlation characteristics. Through end-to-end training, it learns the nonlinear mapping relationship of geological risks. Specifically: The spatiotemporal coupled neural network adopts an encoder-decoder architecture. The encoder part contains three layers of long short-term memory units for extracting temporal features, and the decoder part integrates two layers of graph convolutional network for modeling spatial dependencies. The network training adopts the adaptive moment estimation algorithm, with an initial learning rate of 0.001, which decays to 0.8 times the original value every 50 training rounds. The training termination condition is that the validation set loss function no longer decreases for 10 consecutive rounds. End-to-end training uses over 100,000 sets of historical geological disaster case data, covering major disaster types such as landslides, collapses, and ground subsidence. Five-fold cross-validation is used during training to prevent overfitting.

[0033] The spatiotemporal coupled neural network architecture design in step three is as follows: The encoder receives the fused feature vector sequence of each grid cell over the past T time steps (T=30, corresponding to 30 days of historical data), extracts high-level temporal features layer by layer through 3 stacked long short-term memory units, and finally outputs a fixed-dimensional context vector. The decoder uses this context vector as its initial state, combines it with the spatial adjacency graph at the current moment (constructed by the geographical adjacency between grid cells), and performs spatial information propagation and aggregation through a 2-layer graph convolutional network, ultimately outputting the geological safety risk index of each grid cell.

[0034] The graph convolution operation uses a normalized adjacency matrix to ensure smooth information propagation in space. During network training, the loss function uses weighted mean square error, and high-risk samples are given higher weights to improve the model's sensitivity to disaster events. The adaptive moment estimation algorithm is used to optimize network parameters, with momentum parameters β1 set to 0.9, β2 set to 0.999, and weight decay coefficient set to 1e-4.

[0035] The training dataset is divided into training, validation, and test sets in chronological order, with a ratio of 7:2:1. Five-fold cross-validation is performed within the training set to ensure the model's generalization ability. After training, the model is deployed on a high-performance computing server, supporting batch inference and online updates.

[0036] Step Four: Real-time Risk Situation Updates and Early Warnings. Based on the assessment model built in Step Three, online inference is performed on the accessed real-time monitoring data stream to generate a regional geological safety risk level distribution map. When the risk index exceeds a preset threshold or a sudden trend is detected, a multi-level early warning mechanism is automatically triggered. Specifically: The online inference engine uses a streaming computing framework, supporting the processing of data inflows from 1,000 monitoring points per second; The risk level distribution map is divided into four levels: extremely high risk, high risk, medium risk, and low risk, with a spatial grid accuracy of 10 meters by 10 meters. The multi-level early warning mechanism categorizes risks into four levels—blue, yellow, orange, and red—based on the degree to which the risk index exceeds a threshold. Early warning information is disseminated through multiple channels, including SMS, platform push notifications, and audible and visual alarms. The mutation trend detection uses a change point analysis algorithm, which identifies structural change points in the risk sequence in real time through Bayesian online change point detection technology. The detection delay is less than 5 minutes, and it has the ability to effectively identify both slow-accumulating and sudden geological disasters.

[0037] The online inference engine in step four is built on the Apache Flink streaming computing framework. After the data flows in, it first undergoes real-time standardization processing, using the standardized parameters trained in step one. Then, it is fed into the spatiotemporally coupled neural network model deployed in step three for inference. The inference results are output in GeoJSON format, containing the risk index value of each grid cell.

[0038] The risk level classification adopts the quantile method: grid units with the risk index in the top 5% are classified as extremely high risk, 5% to 20% as high risk, 20% to 50% as medium risk, and the rest as low risk.

[0039] The threshold settings for the multi-level early warning mechanism are based on statistical analysis of historical disaster data: a blue warning corresponds to a risk index exceeding the historical average plus one standard deviation, yellow for two standard deviations, orange for two and a half standard deviations, and red for three standard deviations. After the warning information is generated, it is distributed to multiple notification channels through a message queue: the SMS platform sends warning SMS messages to the designated responsible persons, the Web platform highlights the warning area on the map and pushes pop-up notifications, and the on-site deployed audible and visual alarms emit different frequencies of audible and visual signals according to the warning level. The sudden change trend detection module operates independently, monitoring the risk index time series of each grid unit in real time. The Bayesian online change point detection algorithm identifies structural change points in the sequence by calculating the posterior probability. Once a change point is detected and the risk index continues to rise thereafter, an early warning is immediately triggered to ensure a rapid response to sudden disasters.

[0040] Step 5: Visualization of Assessment Results and Decision Support. The risk assessment results obtained in Step 4 are dynamically rendered using a 3D geographic information system, providing functions such as risk evolution trajectory tracing, sensitivity analysis, and intervention measure simulation to support scientific decision-making in geological safety management. Specifically: In step five, the 3D geographic information system supports multi-dimensional data overlay display, including elevation models, geological profiles, monitoring points, and risk cloud maps; the risk evolution trajectory backtracking function can reproduce the risk distribution at any time in the past 72 hours. Sensitivity analysis assesses the stability of risk outcomes by perturbing key input parameters; intervention simulation can predict the improvement effect of different engineering governance schemes on risk levels.

[0041] The three-dimensional geographic information system in step five is developed based on the CesiumJS engine and supports web access. The system integrates a high-precision digital elevation model (DEM) with a resolution better than 2 meters. It overlays realistic texture images generated by UAV aerial surveys. The risk assessment results are displayed in the form of a semi-transparent heat map, with the color gradually changing from green (low risk) to red (extremely high risk).

[0042] Users can select any historical moment through the timeline control, and the system will instantly render the risk distribution map at that moment to realize the backtracking of risk evolution trajectory. The geological profile function allows users to cut the three-dimensional scene along any line segment to display the vertical profile of the underground rock structure and risk distribution. The sensitivity analysis module provides an interactive interface, where users can select a certain input parameter and set the disturbance range. The system will automatically run batch simulations and output the change range and spatial distribution of risk results to help identify key sensitive factors.

[0043] The intervention simulation function has built-in templates for common engineering governance solutions. After the user selects a solution and specifies the implementation area, the system estimates the governance effect based on the physical model and re-runs the risk assessment model to predict the change in risk level after governance, providing a quantitative basis for decision-making.

[0044] Furthermore, this invention also includes constructing an adaptive learning mechanism to continuously optimize the assessment model based on new disaster cases and expert feedback. The optimization process adopts an incremental learning strategy, triggering model fine-tuning every 100 new samples to ensure that the assessment capability continuously improves over time. The adaptive learning mechanism includes three stages: sample selection, model fine-tuning, and performance verification. New samples first undergo quality assessment, eliminating low-confidence data; then, an elastic weight solidification algorithm is used to fine-tune the spatiotemporally coupled neural network to prevent catastrophic forgetting; finally, the model performance is tested on the retained validation set, and if the performance deteriorates, it is rolled back to the previous version. This mechanism ensures that the model can continuously adapt to long-term changes in the geological environment and the emergence of new disaster patterns.

[0045] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A dynamic assessment method for geological safety risks based on multi-source data fusion, characterized in that, Includes the following steps: Step 1: Multi-source geological data acquisition and standardization processing. Time series data of regional surface deformation are obtained through satellite remote sensing system, high-precision topographic point cloud data are collected by UAV aerial survey platform, and real-time monitoring data of ground sensor network and historical geological survey database are accessed at the same time to obtain multi-source heterogeneous data. Coordinate system transformation, missing value interpolation and repair and quality consistency verification are performed on multi-source heterogeneous data. Step 2: Geological feature semantic fusion and conflict resolution. Based on the multi-source heterogeneous data processed in Step 1, geological tectonic activity indicators, slope stability parameters, hydrogeological characteristic factors and human engineering activity intensity index are extracted. Cross-source data association is established through multi-level feature matching algorithm. A weighted fusion strategy based on geological and physical constraints is adopted to eliminate data redundancy and semantic conflicts to obtain the fusion feature matrix. Step 3: Construct a dynamic risk assessment model. Input the fusion feature matrix obtained in Step 2 into a spatiotemporal coupled neural network. The spatiotemporal coupled neural network includes a long short-term memory module to capture temporal evolution patterns, and an integrated graph convolution module to model spatial correlation characteristics. Through end-to-end training, the nonlinear mapping relationship of geological risks is learned, thereby obtaining the dynamic risk assessment model. Step 4: Real-time updates and early warnings of risk situation. Based on the dynamic risk assessment model built in Step 3, online reasoning is performed on the accessed real-time monitoring data stream to generate a regional geological safety risk level distribution map. When the risk index is detected to exceed the preset threshold or a sudden trend is observed, a multi-level early warning mechanism is automatically triggered. Step 5: Visualization of assessment results and decision support. The regional geological safety risk level distribution map obtained in Step 4 is dynamically rendered using a 3D geographic information system, providing functions such as risk evolution trajectory tracing, sensitivity analysis, and intervention measure simulation.

2. The dynamic assessment method for geological safety risks based on multi-source data fusion according to claim 1, characterized in that: In step one, the satellite remote sensing system uses synthetic aperture radar interferometry, which has a spatial resolution better than 3 meters, a revisit period of less than 7 days, and a surface deformation monitoring accuracy of millimeters. The UAV aerial survey platform is equipped with a multispectral camera and a lidar sensor, with a point cloud density of more than 200 points per square meter and an elevation measurement error of less than 5 centimeters. The ground sensor network includes crack gauges, inclinometers, and pore water pressure gauges, with a sampling frequency of once per minute. The data is transmitted to the central processing unit in real time via an IoT gateway.

3. The dynamic assessment method for geological safety risks based on multi-source data fusion according to claim 1, characterized in that: In step one, the coordinate system transformation adopts the seven-parameter Bursa model to perform a high-precision transformation between the local independent coordinate system and the geodetic coordinate system, with the transformation residual controlled within 0.1 meters; The missing value imputation and repair adopts the spatiotemporal kriging interpolation algorithm, which comprehensively considers spatial autocorrelation and temporal continuity; The quality consistency verification is achieved through multi-source data cross-validation, and outliers exceeding 3 times the standard deviation are automatically marked and manually reviewed.

4. The dynamic assessment method for geological safety risks based on multi-source data fusion according to claim 1, characterized in that: The geological tectonic activity indicators in step two include fault density, joint development degree, and seismic activity frequency; The slope stability parameters include slope, aspect, shear strength of soil and rock mass, and groundwater depth; The hydrogeological characteristic factors include rainfall infiltration coefficient, runoff modulus, and soil saturated hydraulic conductivity; The human engineering activity intensity index is quantified by combining nighttime light data, engineering excavation scale, and traffic load density.

5. The dynamic assessment method for geological safety risks based on multi-source data fusion according to claim 1, characterized in that: In step two, the multi-level feature matching algorithm adopts an improved Hungarian algorithm to establish the optimal correspondence under the dual constraints of feature space and geographic space. The weighted fusion strategy based on geological and physical constraints introduces a fuzzy comprehensive evaluation method, which dynamically adjusts the weight coefficients according to the reliability, timeliness and spatial representativeness of the data source. The correlation coefficient between the fusion result and the field verification data is greater than 0.

9.

6. The dynamic assessment method for geological safety risks based on multi-source data fusion according to claim 1, characterized in that: In step three, the spatiotemporal coupled neural network adopts an encoder-decoder architecture. The encoder part contains three layers of long short-term memory units for extracting temporal features, and the decoder part integrates two layers of graph convolutional networks for modeling spatial dependencies. The network training adopts the adaptive moment estimation algorithm, with an initial learning rate of 0.001, which decays to 0.8 times the original value every 50 training rounds. The training termination condition is that the validation set loss function no longer decreases for 10 consecutive rounds.

7. The dynamic assessment method for geological safety risks based on multi-source data fusion according to claim 1, characterized in that: In step three, the end-to-end training uses more than 100,000 sets of historical geological disaster case data. The samples cover major disaster types including but not limited to landslides, collapses, and ground subsidence. Five-fold cross-validation is used during the training process to prevent overfitting.

8. The dynamic assessment method for geological safety risks based on multi-source data fusion according to claim 1, characterized in that: In step four, the online inference engine adopts a streaming computing framework, which supports processing data inflows from 1,000 monitoring points per second. The risk level distribution map is divided into four levels: extremely high risk, high risk, medium risk, and low risk, with a spatial grid accuracy of 10 meters by 10 meters. The multi-level early warning mechanism is divided into four levels—blue, yellow, orange, and red—based on the degree to which the risk index exceeds the threshold. Early warning information is released through multiple channels, including SMS, platform push notifications, and audible and visual alarms. The mutation trend detection adopts a change point analysis algorithm, which uses Bayesian online change point detection technology to identify structural change points in the risk sequence in real time, with a detection delay of less than 5 minutes.

9. The dynamic assessment method for geological safety risks based on multi-source data fusion according to claim 1, characterized in that: In step five, the three-dimensional geographic information system supports multi-dimensional data overlay display, including elevation models, geological profiles, monitoring points, and risk cloud maps. The risk evolution trajectory backtracking function can reproduce the risk distribution at any point in the past 72 hours. The sensitivity analysis assesses the stability of risk outcomes by perturbing key input parameters; the intervention simulation can predict the improvement effect of different engineering governance schemes on the risk level.

10. The dynamic assessment method for geological safety risks based on multi-source data fusion according to claim 1, characterized in that: It also includes building an adaptive learning mechanism to continuously optimize the assessment model based on new disaster cases and expert feedback. The optimization process adopts an incremental learning strategy, triggering model fine-tuning every 100 new samples to ensure that the assessment capability continues to improve over time. The adaptive learning mechanism includes three stages: sample selection, model fine-tuning, and performance verification.

Images