A geological disaster intelligent monitoring and early warning system based on multi-source remote sensing data

Abstract

The application discloses a kind of geological disaster intelligent monitoring and early warning system based on multi-source remote sensing data, it is related to geological disaster monitoring and early warning technical field, the system includes: multi-source data acquisition module integrates multiple data and is stored into repository according to label classification;Multi-modal data preprocessing module removes interference and extracts multiple features, generates multidimensional matrix;Precursor causality and parameter inversion module identifies causal relationship, and parameter distribution is inverted;Disaster chain threshold value joint generation module constructs associated model, dynamically adjusts threshold value and forms comprehensive table;Multi-disaster collaborative early warning output module determines grade, outputs information and feedback;The application is by constructing multi-source data fusion and feature extraction mechanism, integrates multiple data to form standardized library, extracts key precursor feature, improves disaster identification reliability;Establish precursor causality and disaster chain linkage model, analyze cascade relationship, generate associated model and form comprehensive threshold table, realize multi-disaster collaborative early warning, enhance emergency response efficiency.

Description

Technical Field

[0001] This invention relates to the field of geological disaster monitoring and early warning technology, specifically to an intelligent geological disaster monitoring and early warning system based on multi-source remote sensing data. Background Technology

[0002] With the rapid development of remote sensing and ground monitoring technologies, multi-source data is increasingly widely used in geological disaster monitoring, providing a data foundation for comprehensively capturing disaster precursor information. At the same time, disasters are often accompanied by various precursor features such as crack expansion, abnormal seepage, and accelerated deformation. Furthermore, there are close causal relationships between different disaster types, forming a complex disaster chain system. Traditional monitoring models are gradually developing towards an intelligent direction of "data fusion - feature extraction - causal analysis - dynamic early warning". How to achieve standardized integration of multi-source data, accurate extraction of multi-modal features, and correlation analysis of each link in the disaster chain has become a key technical direction for improving the accuracy and timeliness of geological disaster monitoring and early warning, and has also laid the technical foundation for building an integrated intelligent monitoring and early warning system.

[0003] Traditional geological disaster monitoring technologies have significant limitations in dealing with complex disaster scenarios, making it difficult to meet the demands for precise and dynamic early warning. Firstly, traditional monitoring relies heavily on single-type data, lacking systematic integration and standardized processing of multi-source data. This results in limited data coverage or insufficient correlation between data points, failing to comprehensively reflect the multi-dimensional characteristics of disaster precursors. Secondly, in the data preprocessing stage, the filtering effect on interference factors such as cloud cover and sensor noise is poor, and feature extraction often focuses on a single precursor type, failing to achieve fusion analysis of multiple features such as cracks, seepage, deformation, and temperature, thus affecting the accuracy of interpreting precursor information. Furthermore, traditional technologies struggle to effectively identify the cascading causal relationships between precursory features and changes in soil and rock mechanical parameters. Mechanical parameter inversion is largely based on static geological data, failing to dynamically match the disaster development process. Simultaneously, early warnings for landslide-debris flow-barrier lake disaster chains are often conducted in isolation, lacking threshold linkage mechanisms between different disaster types. This results in fixed early warning thresholds, making it difficult to adapt to the dynamic evolution of the disaster chain. In addition, some early warning systems have relatively singular output channels and lack feedback mechanisms between early warning results and actual disaster conditions, hindering continuous optimization of monitoring data and early warning models and easily leading to delayed or misjudged warnings. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and provide an intelligent monitoring and early warning system for geological disasters based on multi-source remote sensing data. The system first integrates remote sensing, ground monitoring, and geological and historical disaster data through a multi-source data acquisition module to form a standardized raw database. Then, a multi-modal data preprocessing module removes interference, extracts multiple types of precursor features, and fuses them to generate a multi-dimensional feature matrix. Subsequently, a precursor causality and parameter inversion module identifies the cascade causal relationship of precursor features and inverts the spatial distribution of soil and rock mechanical parameters. A disaster chain threshold linkage generation module constructs a disaster chain association model, dynamically adjusts the early warning threshold, and forms a comprehensive threshold table. Finally, a multi-disaster collaborative early warning module determines the early warning level, outputs information through multiple channels, and feeds back the results to the database.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a geological disaster intelligent monitoring and early warning system based on multi-source remote sensing data, the system comprising:

[0006] Multi-source data acquisition module: Collects remote sensing, ground monitoring, and geological and historical disaster data, classifies and stores them according to time-space-data type-disaster chain association tags, and forms a standardized raw database;

[0007] The multimodal data preprocessing module removes interference from the data in the standardized raw database and filters out cloud and sensor noise through a comparative learning model. Based on regional characteristics, it extracts crack features, seepage features, deformation features, and temperature features, and fuses them to generate a multi-dimensional feature matrix of time × space × precursor type.

[0008] Precursor causality and parameter inversion module: Based on a multi-dimensional feature matrix, it identifies the cascade causal relationship between crack propagation, seepage anomalies, deformation acceleration, and decay of rock and soil mechanical parameters through a causal transmission intensity algorithm; combined with geological structural data, it inverts the spatial distribution map of rock and soil cohesion and elastic modulus through a dynamic inversion algorithm of mechanical parameters.

[0009] The disaster chain threshold linkage generation module: Based on causal relationships, distribution maps of soil and rock mechanical parameters, and historical disaster chain records, it constructs a disaster chain correlation model of landslide-debris flow-barrier lake; it dynamically adjusts the downstream debris flow early warning threshold according to the upstream soil and rock mechanical parameters and deformation data through a disaster chain threshold linkage algorithm; it calculates the probability of barrier lake blockage by combining the upstream landslide volume and river slope with a barrier lake risk probability algorithm, and integrates them to form a comprehensive threshold table;

[0010] Multi-disaster collaborative early warning output module: compares real-time monitoring data with a comprehensive threshold table to determine the early warning level of a disaster; outputs early warning information through web-based maps, mobile apps, and on-site audible and visual alarms, and feeds back the early warning results and actual disaster conditions to a standardized raw database.

[0011] Furthermore, in the multi-source data acquisition module: remote sensing data is acquired through satellite receiving stations, including InSAR deformation fields, Sentinel-2 optical images, Sentinel-1 SAR data, and thermal infrared data; ground monitoring data is acquired through the deployment of piezometers, GNSS displacement stations, and connection to meteorological interfaces, including seepage rate, GNSS displacement value, rainfall, and temperature; geological and historical disaster data are acquired through the collection of regional geological maps and historical archives, including stratigraphic lithology, fault strike and dip angle, and the occurrence time and impact range of historical landslides, debris flows, and barrier lakes.

[0012] Furthermore, the contrastive learning model adopts the SimCLR framework. During training, the positive samples are multimodal data from the 7 days prior to the historical disaster, and include features such as crack length ≥0.5cm, daily increase in seepage rate ≥10%, and deformation rate ≥2mm / month. The negative samples are multimodal data from the disaster-free period, and include optical images with cloud and fog obscuration area ≥30% and monitoring data with sensor noise value ≥5%. The model training objective is that the mutual information of positive samples is ≥0.8, and the mutual information of negative samples is ≤0.2.

[0013] Furthermore, the multi-modal data preprocessing module generates a multi-dimensional feature matrix: the time dimension is in days, covering nearly three months; the spatial dimension is a 10m×10m grid; the precursor types include crack features, seepage features, deformation features, and temperature features; crack features include length, density, and propagation rate; seepage features include rate, daily variation rate, and peak pore water pressure; deformation features include cumulative value and maximum daily rate; and temperature features include gradient and anomalous area; all feature values ​​are normalized to the 0-1 range.

[0014] Furthermore, in the precursor causality and parameter inversion module, the expression for the causal transmission strength algorithm is: ,in, These are early warning signs. arrive The strength of causal transmission, Intervention When it happens, The conditional probability of occurrence Intervention When it does not occur, The conditional probability of occurrence yes and The maximum probability of occurrence. As a time lag factor, For dynamic weights.

[0015] Furthermore, in the precursor causality and parameter inversion module, the expression for the dynamic inversion algorithm of mechanical parameters is: ,in, yes time Geomechanical parameters of the location. These are the initial mechanical parameters of the soil and rock mass. It is the number of precursor features participating in the inversion. It is the characteristic influence coefficient. yes Time of the first The total strength of the causal link of each precursor feature, yes time Position No. The current value of each precursor feature, It is the position number The initial values ​​of the precursor features, It is the first The historical maximum value of a precursor feature.

[0016] Furthermore, in the disaster chain threshold linkage generation module, the landslide-debris flow-barrier lake disaster chain correlation model includes: Landslide and debris flow correlation rules: When the cohesion of the upstream soil and rock mass is ≤15kPa and the deformation rate is ≥5mm / month, a downstream debris flow warning threshold reduction mechanism is triggered. The reduction magnitude is positively correlated with the upstream landslide volume; for every 1000m³ increase in volume, the downstream debris flow 24-hour rainfall warning threshold is reduced by 5mm; Landslide and barrier lake correlation rules: Based on the upstream landslide volume and river... The basic probability of blockage is calculated based on the slope. When the volume is ≥5000m³ and the slope is ≥10°, the initial probability of blockage is 30%. For every additional 5000m³ of volume or 5° of slope, the probability increases by 15%. The association rule between debris flow and landslide dammed lake is as follows: when the downstream debris flow scale is ≥1000m³, it is superimposed on the blockage probability of the landslide dammed lake. The superimposed value is 10% of the ratio of debris flow scale to river volume. By integrating the above rules, the model outputs the debris flow threshold adjustment amount and the landslide dammed lake probability correction value corresponding to different landslide states.

[0017] Furthermore, in the disaster chain threshold linkage generation module, the expression for the disaster chain threshold linkage algorithm is: ,in, yes Downstream debris flow warning threshold at any given time This is the initial warning threshold for debris flow. It is a threshold adjustment coefficient, set according to the regional geological characteristics. yes Real-time mechanical parameters of the soil and rock mass in the upstream area. These are the initial mechanical parameters of the rock and soil mass in the upstream area. yes The sum of the causal link strengths of all precursor features in the upstream region at a given time. This indicates the number of precursor features involved in the calculation of the disaster chain threshold. yes Real-time rainfall impact coefficient Indicates the first A few early warning features Total strength of the causal link at any given moment.

[0018] Furthermore, in the disaster chain threshold linkage generation module, the expression for the landslide dam risk probability algorithm is: ,in, yes The probability of a landslide dam becoming blocked at any given time. The volume of the upstream landslide. For the river slope, The total causal strength of the disaster chain. For real-time river water levels, and These are model coefficients, calibrated based on historical landslide dam cases. When the value is less than 0.3, it is considered low risk; when 0.3 ≤ When the value is ≤0.6, it is judged as medium risk; when A value greater than 0.6 is considered high risk.

[0019] Furthermore, the specific contents of the comprehensive threshold table in the disaster chain threshold linkage generation module are as follows:

[0020] When the disaster type is landslide, the monitoring indicators are soil cohesion and deformation rate. The dynamic threshold range is as follows: soil cohesion greater than 20 kPa and deformation rate less than 2 mm / month corresponds to low risk; soil cohesion less than or equal to 20 kPa and deformation rate less than or equal to 5 mm / month corresponds to medium risk; soil cohesion less than or equal to 15 kPa and deformation rate greater than or equal to 5 mm / month corresponds to high risk. The description of the associated disaster impact is that it triggers the adjustment of the downstream debris flow warning threshold, and the adjustment range is positively correlated with the landslide volume.

[0021] When the disaster type is debris flow, the monitoring indicator is 24-hour rainfall; the dynamic threshold range is based on the initial threshold, with the threshold lowered by 5mm for every additional 1000m³ of landslide volume upstream. After the reduction, a threshold greater than 80mm corresponds to low risk, 60mm less than or equal to the reduced threshold less than or equal to 80mm corresponds to medium risk, and a threshold less than 60mm corresponds to high risk; the explanation of the associated disaster impact is that the threshold is affected by the mechanical parameters and deformation data of the upstream soil and rock mass, and when the debris flow scale is greater than or equal to 1000m³, it has an additional impact on the probability of blockage of the landslide dam.

[0022] When the disaster type is a landslide dam, the monitoring indicator is the blockage probability; the dynamic threshold range is as follows: a blockage probability less than 0.3 corresponds to low risk, 0.3 less than or equal to a blockage probability less than or equal to 0.6 corresponds to medium risk, and a blockage probability greater than 0.6 corresponds to high risk; the description of the associated disaster impact is as follows: the initial value of the blockage probability is calculated from the upstream landslide volume greater than or equal to 5000 m³ and the river slope greater than or equal to 10°. For every 5000 m³ increase in volume or every 5° increase in slope, the probability increases by 15%. When the downstream debris flow scale is greater than or equal to 1000 m³, the probability value is superimposed by the ratio of debris flow scale to river volume multiplied by 10%.

[0023] The comprehensive threshold table is updated every 15 days in sync with the distribution map of soil and rock mechanical parameters. The updates are based on newly collected multi-dimensional feature matrices, geological structure data, and historical disaster chain records.

[0024] Compared with existing technologies, this intelligent monitoring and early warning system for geological disasters based on multi-source remote sensing data has the following advantages:

[0025] I. This invention achieves three-dimensional and precise geological disaster monitoring by constructing a multi-source data fusion and multi-modal feature extraction mechanism. The system integrates remote sensing images, ground sensors, and historical geological data to form a standardized database covering multiple dimensions of time and space. It also uses a contrastive learning model to eliminate interference from clouds, fog, noise, etc., and extracts key precursor features such as cracks, seepage, and deformation. By constructing a feature matrix of time-space-precursor types, the system breaks through the limitations of traditional single data sources and significantly improves the sensitivity and reliability of disaster precursor identification. This data processing mode provides richer feature evidence for disaster early warning and enhances the system's adaptability to complex geological environments.

[0026] Second, this invention achieves intelligent and dynamic multi-hazard collaborative early warning by establishing a model linking precursor causal transmission with dynamic thresholds in the disaster chain. Based on the causal transmission intensity algorithm, the system analyzes the cascade relationship between precursors such as crack propagation and abnormal seepage and the attenuation of mechanical parameters of soil and rock. Combined with dynamic inversion technology of mechanical parameters, it generates a correlation model of landslides, debris flows, and barrier lakes. Through the disaster chain threshold linkage algorithm, the system can adjust the downstream early warning threshold in real time according to the state of upstream soil and rock and integrate the risk probability of barrier lakes to form a comprehensive threshold table. This disaster chain-like early warning mechanism breaks through the limitation of independent early warning of a single disaster and significantly improves the accuracy of early warning and the efficiency of emergency response in scenarios with multiple disasters occurring simultaneously.

[0027] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from an examination of the following, or may be learned from the practice of the invention. Attached Figure Description

[0028] 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.

[0029] Figure 1 This is a flowchart illustrating the workflow of an intelligent monitoring and early warning system for geological disasters based on multi-source remote sensing data.

[0030] Figure 2 This is a general framework diagram of a geological disaster intelligent monitoring and early warning system based on multi-source remote sensing data. Detailed Implementation

[0031] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0032] Example 1: Collaborative monitoring and early warning of landslides, debris flows, and barrier lakes in mountainous areas during the rainy season.

[0033] In response to the high incidence of geological disasters in mountainous areas during the rainy season, the system's multi-source data acquisition module was activated. First, various types of remote sensing data were acquired via satellite receiving stations, including InSAR deformation field data, Sentinel-2 optical imagery, Sentinel-1 SAR data, and thermal infrared data. Then, piezometers and GNSS displacement stations were deployed in the monitoring area and connected to a regional meteorological interface to collect real-time ground monitoring data such as seepage rate, GNSS displacement values, rainfall, and temperature. Simultaneously, regional geological maps and historical disaster archives for the mountainous area were collected, extracting stratigraphic lithology, fault strike and dip, and the timing and impact range of historical landslides, debris flows, and barrier lakes. Finally, all collected data was stored according to a classification rule of "time-space-data type-disaster chain association tag," forming a standardized raw database. This ensures that subsequent modules can quickly locate the required data type when accessing data, avoiding analysis delays caused by data confusion. Figure 1 As shown.

[0034] The multimodal data preprocessing module is invoked to process the data in the standardized raw database. First, interference removal is performed using a contrastive learning model based on the SimCLR framework to filter cloud and fog interference and sensor noise from the data. Positive samples selected during training are multimodal data from the 7 days prior to historical rainy season disasters in the mountainous area, and must include precursory features such as crack length ≥0.5cm, daily seepage rate increase ≥10%, and deformation rate ≥2mm / month. Negative samples are multimodal data from disaster-free periods in the mountainous area, and must include optical images with cloud and fog obscuration area ≥30% and monitoring data with sensor noise values ​​≥5%. The model training objective is to ensure a mutual information of ≥0.8 for positive samples and ≤0.2 for negative samples, thereby effectively removing interference and ensuring the purity of the data for subsequent feature extraction. Subsequently, based on the regional characteristics of the mountainous area, four core precursor features were extracted: crack features, seepage features, deformation features, and temperature features. These features were then fused to generate a multi-dimensional feature matrix. The time dimension was measured in days, covering nearly three months of continuous data. The spatial dimension was divided into 10m×10m grids, precisely corresponding to each point in the monitoring area. The precursor types included the above four types of features, and all feature values ​​were normalized to the 0-1 range, providing standardized and highly available data support for subsequent modules.

[0035] The precursor causality and parameter inversion module is activated, and analysis is conducted based on the multi-dimensional feature matrix generated above. First, the cascading causal relationships of disaster precursors within the monitoring area are identified using the causal transmission strength algorithm. The expression for the causal transmission strength algorithm is: ,in, These are early warning signs. arrive The strength of causal transmission, Intervention When it happens, The conditional probability of occurrence Intervention When it does not occur, The conditional probability of occurrence yes and The maximum probability of occurrence. As a time lag factor, Using dynamic weighting, the algorithm focuses on capturing the transmission chain of "crack propagation → abnormal seepage → accelerated deformation → attenuation of soil and rock mechanical parameters." This algorithm can clearly determine the correlation strength between various precursors, avoiding the risk of misjudgment caused by isolated analysis of a single precursor. Subsequently, combined with previously collected geological structure data of mountainous areas, a dynamic inversion algorithm for mechanical parameters is used to invert the key mechanical parameters of the soil and rock in this area. The expression of the dynamic inversion algorithm for mechanical parameters is: ,in, yes time Geomechanical parameters of the location. These are the initial mechanical parameters of the soil and rock mass. It is the number of precursor features participating in the inversion. It is the characteristic influence coefficient. yes Time of the first The total strength of the causal link of each precursor feature, yes time Position No. The current value of each precursor feature, It is the position number The initial values ​​of the precursor features, It is the first The historical maximum values ​​of several precursor features, specifically the spatial distribution map of cohesion and elastic modulus, can intuitively present the differences in mechanical stability of soil and rock masses in different regions, providing clear mechanical parameter support for subsequent disaster chain analysis and ensuring that disaster risk assessment has a quantitative basis.

[0036] The disaster chain threshold linkage generation module was invoked to construct a disaster chain correlation model for landslide-debris flow-barrier lake in the mountainous area. Firstly, based on previously obtained causal relationships, distribution maps of soil and rock mechanical parameters, and historical disaster chain records, the disaster chain correlation rules were determined: Regarding the correlation between landslides and debris flows, when the cohesion of the upstream soil and rock mass is ≤15kPa and the deformation rate is ≥5mm / month, a mechanism for lowering the downstream debris flow warning threshold is triggered. The reduction magnitude is positively correlated with the volume of the upstream landslide body; that is, for every 1000m³ increase in the landslide volume, the 24-hour rainfall warning threshold for downstream debris flows is lowered by 5mm. The correlation between landslides and barrier lakes... The basic probability of blockage is calculated based on the upstream landslide volume and river slope. When the landslide volume is ≥5000 m³ and the river slope is ≥10°, the initial blockage probability is 30%. Thereafter, for every 5000 m³ increase in landslide volume or 5° increase in river slope, the blockage probability increases by 15%. Regarding the correlation between debris flow and landslide-dammed lakes, when the downstream debris flow scale is ≥1000 m³, 10% of the ratio of debris flow scale to river volume is used as the superposition value and included in the landslide-dammed lake blockage probability. Subsequently, a disaster chain threshold linkage algorithm is used to dynamically adjust the downstream debris flow early warning threshold based on real-time upstream geotechnical mechanical parameters and deformation data. The expression for the disaster chain threshold linkage algorithm is: ,in, yes Downstream debris flow warning threshold at any given time This is the initial warning threshold for debris flow. It is a threshold adjustment coefficient, set according to the regional geological characteristics. yes Real-time mechanical parameters of the soil and rock mass in the upstream area. These are the initial mechanical parameters of the rock and soil mass in the upstream area. yes The sum of the causal link strengths of all precursor features in the upstream region at a given time. This indicates the number of precursor features involved in the calculation of the disaster chain threshold. yes Real-time rainfall impact coefficient Indicates the first A few early warning features The total causal link strength at any given time; the probability of landslide damming is calculated by combining the upstream landslide volume, river slope, and downstream debris flow scale using a landslide dam risk probability algorithm. The expression for the landslide dam risk probability algorithm is: ,in, yes The probability of a landslide dam becoming blocked at any given time. The volume of the upstream landslide. For the river slope, The total causal strength of the disaster chain. For real-time river water levels, and These are model coefficients, calibrated based on historical landslide dam cases. When the value is less than 0.3, it is considered low risk; when 0.3 ≤ When the value is ≤0.6, it is judged as medium risk; when When the threshold is greater than 0.6, it is considered high risk. Finally, all threshold adjustment results and probability values ​​are integrated to form a comprehensive threshold table covering three types of disasters: landslides, debris flows, and barrier lakes. The table is updated every 15 days, combining newly collected multi-dimensional feature matrices, geological structure data, and historical disaster chain records, to provide a comprehensive and timely standard basis for subsequent early warning judgments.

[0037] The multi-hazard collaborative early warning output module is activated, comparing real-time monitoring data with a comprehensive threshold table to determine the early warning level of various disasters: for example, when the cohesion of the soil and rock mass in a certain area is ≤15kPa and the deformation rate is ≥5mm / month, it is determined to be a high-risk landslide; when the 24-hour rainfall warning threshold for debris flow downstream is lowered, if the real-time rainfall reaches the corresponding range of the lowered threshold, it is determined to be the corresponding risk level of debris flow; when the probability of blockage by a landslide dam is >0.6, it is determined to be a high-risk landslide dam. Subsequently, early warning information is output simultaneously through three channels: the location and level of each risk area are intuitively marked on the web map; text and pop-up warnings are pushed to management personnel and affected people through the mobile APP; and on-site audible and visual alarms issue audible and visual alerts at monitoring stations and around villages; at the same time, the results of this early warning and the subsequent actual disaster occurrence are fed back to the standardized raw database.

[0038] In summary, in the scenario of coordinated monitoring and early warning of landslides, debris flows, and barrier lakes during the rainy season in mountainous areas, the system integrates remote sensing, ground monitoring, and historical geological data through a multi-source data acquisition module, providing a comprehensive data source for early warning. The multi-modal data preprocessing module uses the contrastive learning model of the SimCLR framework to remove interference and generate a standardized multi-dimensional feature matrix, laying the foundation for analysis. The precursor causality and parameter inversion module clarifies the precursor correlation through a causal transmission intensity algorithm and obtains the mechanical distribution of soil and rock mass by combining a dynamic inversion algorithm for mechanical parameters. The disaster chain threshold linkage generation module constructs a correlation model and forms a dynamic comprehensive threshold table through a disaster chain threshold linkage algorithm and a barrier lake risk probability algorithm. The multi-disaster coordinated early warning output module pushes early warnings and provides data feedback through multiple channels, adapting to the characteristics of disasters during the rainy season throughout the entire process, and realizing coordinated early warning for three types of disasters.

[0039] Example 2: Intelligent monitoring and early warning of geological disasters in the health service industrial park construction project.

[0040] This embodiment focuses on intelligent monitoring and early warning of geological hazards for a health service industrial park construction project. The project involves excavation of foundation pits, site leveling, and other engineering operations, which are prone to landslides caused by artificial slope excavation and disturbance of soil and rock masses. As a critical construction project, it demands high precision and real-time performance in geological hazard prevention and control. Therefore, this intelligent monitoring and early warning system for geological hazards based on multi-source remote sensing data is implemented to monitor and warn of geological hazards throughout the entire construction cycle. Each module of the system is adapted to the geological hazard prevention and control needs of the construction scenario, achieving intelligent and dynamic monitoring and early warning of landslide risks, ensuring the safety of construction and subsequent operation. Figure 2 As shown.

[0041] Multi-source data acquisition module: Conducts targeted multi-type data acquisition and standardized storage. Remote sensing data is acquired through satellite receiving stations, including InSAR deformation fields, Sentinel-2 optical images, Sentinel-1 SAR data, and thermal infrared data, focusing on capturing macroscopic information such as surface deformation, surface cracks in soil and rock, and temperature anomalies in the industrial park construction area; Ground monitoring data is combined with the characteristics of engineering construction, with piezometers and GNSS displacement stations densely deployed in the foundation pit excavation area, artificial slope treatment area, and key area of ​​site leveling to collect data such as seepage rate and GNSS displacement values ​​in real time. At the same time, it connects to the regional meteorological interface to obtain meteorological data such as rainfall and temperature, accurately grasping the hydrological and meteorological factors affecting the risk of collapse; Geological and historical disaster data are obtained by collecting geological maps, engineering survey data, and historical geological disaster archives of the project area, including regional stratigraphic lithology, fault strike and dip angle, topographic elevation, design excavation depth of foundation pit, design slope of artificial slope, and information on the occurrence time, impact range, and inducing factors of historical collapse disasters in the region. All collected remote sensing, ground monitoring, geological and historical disaster data are classified and stored according to the classification rules of "time-space-data type-disaster chain association tag". Special association tags are added to data sources in key risk areas such as foundation pits and artificial slopes to ensure the targeted and efficient retrieval of data, and finally form a standardized original database adapted to the construction project of the health service industrial park.

[0042] Multimodal data preprocessing module: This module performs interference removal and feature extraction fusion on data from the standardized raw database. First, interference removal is performed using a contrastive learning model based on the SimCLR framework to filter out interference factors such as cloud cover and sensor noise. Positive samples selected for model training are multimodal data from the 7 days prior to a collapse disaster in the engineering construction field, including pre-collapse features such as crack length ≥0.5cm, daily seepage rate increase ≥10%, and deformation rate ≥2mm / month. Negative samples are multimodal data from disaster-free periods, including optical images with cloud cover ≥30% and monitoring data with sensor noise ≥5%. The model training objective is a mutual information of ≥0.8 for positive samples and ≤0.2 for negative samples. After training, the monitoring data for the project area is effectively de-interferenced to ensure data purity. Subsequently, based on the geological environment characteristics of the industrial park construction area, four types of precursor features highly correlated with collapse risk were extracted: crack features, seepage features, deformation features, and temperature features. Among them, crack features focused on the crack length, density, and propagation rate of the foundation pit wall and artificial slope, while deformation features focused on the cumulative deformation of the soil and rock mass in the excavation area and the maximum daily deformation rate. The four types of features were fused to generate a multi-dimensional feature matrix of time × space × precursor type. The time dimension of this feature matrix is ​​based on days, covering nearly three months of continuous monitoring data, and the spatial dimension is divided into 10m × 10m grids, with grid densification in key areas such as foundation pits and artificial slopes. All precursor feature values ​​were normalized to the 0-1 range, providing standardized and highly available data support for subsequent module analysis.

[0043] Precursor Causality and Parameter Inversion Module: Based on the generated multi-dimensional feature matrix, this module identifies the causal relationships of collapse precursors and inverts the mechanical parameters of the soil and rock mass. First, it uses the causal transmission strength algorithm, the expression of which is: ,in, These are early warning signs. arrive The strength of causal transmission, Intervention When it happens, The conditional probability of occurrence Intervention When it does not occur, The conditional probability of occurrence yes and The maximum probability of occurrence. As a time lag factor, Using dynamic weighting, the algorithm identifies the cascading causal relationships of precursor features related to landslide disasters within the project construction area, focusing on capturing the core transmission chain of "crack propagation → abnormal seepage → accelerated deformation → attenuation of soil and rock mechanical parameters." This algorithm clarifies the correlation strength between various precursor features, avoiding misjudgments of landslide risk due to isolated analysis of single precursors, and providing a logical basis for locating the core inducing factors of landslide disasters. Subsequently, combining geological structural data and engineering survey data of the industrial park construction area, a dynamic inversion algorithm for mechanical parameters is employed. The expression for the dynamic inversion algorithm for mechanical parameters is: ,in, yes time Geomechanical parameters of the location. These are the initial mechanical parameters of the soil and rock mass. It is the number of precursor features participating in the inversion. It is the characteristic influence coefficient. yes Time of the first The total strength of the causal link of each precursor feature, yes time Position No. The current value of each precursor feature, It is the position number The initial values ​​of the precursor features, It is the first The historical maximum values ​​of the precursor features are used to invert the spatial distribution map of the cohesion and elastic modulus of the soil and rock in key areas such as the foundation pit excavation area and the artificial slope treatment area within the project area. This visually presents the differences in the mechanical stability of the soil and rock in different areas and accurately marks the high-risk points of collapse with low cohesion and high deformation rate, providing quantitative mechanical parameter support for subsequent collapse risk assessment and early warning threshold setting.

[0044] Disaster Chain Threshold Linkage Generation Module: Constructs a geological disaster association model adapted to the health service industrial park construction project and generates a dynamic comprehensive threshold table. Based on the aforementioned identified precursory causal relationships, the inverted distribution map of soil and rock mechanical parameters, and historical records of landslide disasters caused by engineering construction, a geological disaster association model with landslide risk as the core is constructed, focusing on setting threshold linkage rules related to engineering excavation and soil and rock disturbance; through the disaster chain threshold linkage algorithm, the expression of the disaster chain threshold linkage algorithm is: ,in, yes Downstream debris flow warning threshold at any given time This is the initial warning threshold for debris flow. It is a threshold adjustment coefficient, set according to the regional geological characteristics. yes Real-time mechanical parameters of the soil and rock mass in the upstream area. These are the initial mechanical parameters of the rock and soil mass in the upstream area. yes The sum of the causal link strengths of all precursor features in the upstream region at a given time. This indicates the number of precursor features involved in the calculation of the disaster chain threshold. yes Real-time rainfall impact coefficient Indicates the first A few early warning features The system dynamically adjusts the collapse warning threshold of surrounding areas based on real-time geotechnical parameters and deformation data of the excavation area and artificial slope treatment area. For areas with decreased geotechnical cohesion and increased deformation rate, the collapse warning threshold is simultaneously lowered to improve the accuracy of the warnings. It integrates various monitoring indicators and threshold adjustment rules for collapse hazards to form a comprehensive threshold table tailored to the industrial park construction scenario. This table clarifies the threshold ranges for low, medium, and high risks of collapse hazards. Specifically, a geotechnical cohesion greater than 20 kPa and a deformation rate less than 2 mm / month corresponds to low risk; a cohesion less than or equal to 20 kPa and a deformation rate less than or equal to 5 mm / month corresponds to medium risk; and a cohesion less than or equal to 15 kPa and a deformation rate greater than or equal to 5 mm / month corresponds to high risk. The system also indicates the corresponding engineering construction control requirements for each risk level. The comprehensive threshold table is updated every 15 days in sync with the distribution map of soil and rock mechanical parameters. The updates are based on newly collected multi-dimensional feature matrices, geological structure data of project construction, and collapse risk monitoring records to ensure the timeliness and adaptability of the threshold table.

[0045] Multi-hazard collaborative early warning output module: Conducts landslide risk level assessment, outputs early warning information through multiple channels, and provides data feedback. It compares real-time monitoring data of the industrial park construction area with a comprehensive threshold table to accurately determine the landslide disaster early warning level for each area. It focuses on real-time level assessment of excavated foundation pit areas and artificial slope treatment areas. If the soil cohesion in a certain area is ≤15kPa and the deformation rate is ≥5mm / month, it is directly determined to be at high risk of landslide. If the monitoring data is in the medium-risk range, a risk warning is issued and the monitoring frequency is increased. Early warning information is simultaneously output through three channels. The web-based map intuitively marks the collapse risk level, specific high-risk locations, and risk impact range of various areas such as the foundation pit, artificial slope, and site leveling area, supporting construction management personnel to remotely monitor risk distribution. The mobile APP pushes text, pop-up, and voice warning information to engineering construction management personnel, on-site workers, and project safety management departments to ensure that relevant personnel receive the warning content as soon as possible. On-site audible and visual alarms are deployed at key locations such as around the foundation pit, on both sides of the artificial slope, and at the construction site entrances and exits, issuing different frequencies of audible and visual alerts according to the warning level, providing immediate risk reminders to on-site workers. At the same time, the warning results, warning response measures, and actual collapse disaster occurrences during the construction process (including monitoring records where no disasters occurred) are fed back to the system's standardized raw database, forming a data closed loop. This provides practical engineering data support for the continuous optimization of the system model and the dynamic adjustment of warning thresholds.

[0046] In summary, in the intelligent monitoring and early warning scenario for geological disasters in the health service industrial park construction project, each module of the system is closely adapted to the operational characteristics of the project construction and the needs for geological disaster prevention and control. Addressing the risk of collapses easily triggered by foundation pit excavation and site leveling, the system achieves intelligent and dynamic monitoring and early warning of geological disasters throughout the entire project construction cycle through standardized integration of multi-source data, accurate extraction of multi-modal precursor features, scientific identification of collapse-related causal relationships, dynamic inversion of soil and rock mechanical parameters, and coordinated adjustment of collapse early warning thresholds. Simultaneously, the system ensures efficient transmission of early warning information through multi-channel output of early warning information via web, mobile devices, and on-site audible and visual alarms. Combined with data feedback from early warning results and actual disaster conditions, the system achieves self-optimization, effectively reducing the probability of geological disasters occurring during construction and providing a solid technical guarantee for the construction safety of the health service industrial park project.

[0047] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A geological disaster intelligent monitoring and early warning system based on multi-source remote sensing data, characterized in that, The system includes: Multi-source data acquisition module: Collects remote sensing, ground monitoring, and geological and historical disaster data, classifies and stores them according to time-space-data type-disaster chain association tags, and forms a standardized raw database; The multimodal data preprocessing module removes interference from the data in the standardized raw database and filters out cloud and sensor noise through a comparative learning model. Based on regional characteristics, it extracts crack features, seepage features, deformation features, and temperature features, and fuses them to generate a multi-dimensional feature matrix of time × space × precursor type. Precursor Causality and Parameter Inversion Module: Based on a multi-dimensional feature matrix, this module uses a causal transmission intensity algorithm to identify the cascading causal relationships from crack propagation, seepage anomalies, and accelerated deformation to the decay of soil and rock mechanical parameters. The expression for the causal transmission intensity algorithm is as follows: ,in, These are early warning signs. arrive The strength of causal transmission, Intervention When it happens, The conditional probability of occurrence Intervention When it does not occur, The conditional probability of occurrence yes and The maximum probability of occurrence. As a time lag factor, For dynamic weights; combining geological structural data, a dynamic inversion algorithm for mechanical parameters is used to invert the spatial distribution map of soil and rock cohesion and elastic modulus. The expression for the dynamic inversion algorithm for mechanical parameters is: ,in, yes time Geomechanical parameters of the location. These are the initial mechanical parameters of the soil and rock mass. It is the number of precursor features participating in the inversion. It is the characteristic influence coefficient. yes Time of the first The total strength of the causal link of each precursor feature, yes time Position No. The current value of each precursor feature, It is the position number The initial values ​​of the precursor features, It is the first The historical maximum value of each precursor feature; Disaster chain threshold linkage generation module: Based on causal relationships, distribution maps of soil and rock mechanical parameters, and historical disaster chain records, a landslide-debris flow-barrier lake disaster chain correlation model is constructed; the disaster chain threshold linkage algorithm dynamically adjusts the downstream debris flow early warning threshold according to the upstream soil and rock mechanical parameters and deformation data. The expression of the disaster chain threshold linkage algorithm is as follows: ,in, yes Downstream debris flow warning threshold at any given time. This is the initial warning threshold for debris flow. It is a threshold adjustment coefficient, set according to the regional geological characteristics. yes Real-time mechanical parameters of the soil and rock mass in the upstream area. These are the initial mechanical parameters of the rock and soil mass in the upstream area. yes The sum of the causal link strengths of all precursor features in the upstream region at a given time. This indicates the number of precursor features involved in the calculation of the disaster chain threshold. yes Real-time rainfall impact coefficient Indicates the first A few early warning features The total causal link strength at any given time; the probability of landslide damming is calculated using a landslide dam risk probability algorithm combined with the upstream landslide volume and river slope. The expression for the landslide dam risk probability algorithm is: ,in, yes The probability of a landslide dam becoming blocked at any given time. The volume of the upstream landslide. For the river slope, The total causal strength of the disaster chain. For real-time river water levels, and These are model coefficients, calibrated based on historical landslide dam cases. When the value is less than 0.3, it is considered low risk; when 0.3 ≤ When the value is ≤0.6, it is judged as medium risk; when When the value is greater than 0.6, it is considered high-risk and integrated into a comprehensive threshold table; Multi-disaster collaborative early warning output module: compares real-time monitoring data with a comprehensive threshold table to determine the early warning level of a disaster; outputs early warning information through web-based maps, mobile apps, and on-site audible and visual alarms, and feeds back the early warning results and actual disaster conditions to a standardized raw database.

2. The intelligent monitoring and early warning system for geological disasters based on multi-source remote sensing data according to claim 1, characterized in that, In the multi-source data acquisition module: remote sensing data is acquired through satellite receiving stations, including InSAR deformation fields, Sentinel-2 optical images, Sentinel-1 SAR data, and thermal infrared data; ground monitoring data is acquired through the deployment of piezometers, GNSS displacement stations, and connection to meteorological interfaces, including seepage rate, GNSS displacement value, rainfall, and temperature; geological and historical disaster data are acquired through the collection of regional geological maps and historical archives, including stratigraphic lithology, fault strike and dip angle, and the occurrence time and impact range of historical landslides, debris flows, and barrier lakes.

3. The intelligent monitoring and early warning system for geological disasters based on multi-source remote sensing data according to claim 1, characterized in that, The comparative learning model adopts the SimCLR framework. During training, the positive samples are multimodal data from the 7 days prior to the historical disaster, and include features such as crack length ≥0.5cm, daily increase in seepage rate ≥10%, and deformation rate ≥2mm / month. The negative samples are multimodal data from the disaster-free period, and include optical images with cloud and fog obscuration area ≥30% and monitoring data with sensor noise value ≥5%. The model training objective is that the mutual information of positive samples is ≥0.8 and the mutual information of negative samples is ≤0.

2.

4. The intelligent monitoring and early warning system for geological disasters based on multi-source remote sensing data according to claim 1, characterized in that, The multimodal data preprocessing module generates a multidimensional feature matrix: the time dimension is in days, covering nearly three months; the spatial dimension is a 10m×10m grid; the precursor types include crack features, seepage features, deformation features, and temperature features. Crack characteristics include length, density, and propagation rate; seepage characteristics include rate, daily variation rate, and peak pore water pressure; deformation characteristics include cumulative value and maximum daily rate; and temperature characteristics include gradient and area of ​​anomalous regions. All characteristic values ​​are normalized to the 0-1 range.

5. The intelligent monitoring and early warning system for geological disasters based on multi-source remote sensing data according to claim 1, characterized in that, In the disaster chain threshold linkage generation module, the landslide-debris flow-barrier lake disaster chain correlation model includes: Landslide and debris flow correlation rules: When the cohesion of the upstream soil and rock mass is ≤15kPa and the deformation rate is ≥5mm / month, a downstream debris flow warning threshold reduction mechanism is triggered. The reduction magnitude is positively correlated with the upstream landslide volume; for every 1000m³ increase in volume, the downstream debris flow 24-hour rainfall warning threshold is reduced by 5mm. Landslide and barrier lake correlation rules: Based on the upstream landslide volume and river slope... The basic probability of blockage is calculated as follows: when the volume is ≥5000m³ and the slope is ≥10°, the initial probability of blockage is 30%. For every additional 5000m³ of volume or 5° of slope, the probability increases by 15%. The association rule between debris flow and landslide dammed lake is as follows: when the downstream debris flow scale is ≥1000m³, it is superimposed on the blockage probability of landslide dammed lake, and the superimposed value is 10% of the ratio of debris flow scale to river channel volume. By integrating the above rules, the model outputs the debris flow threshold adjustment amount and landslide dammed lake probability correction value corresponding to different landslide states.

Images