Geological disaster early warning method and system based on machine learning

Abstract

The application discloses a geological disaster early warning method and system based on machine learning, and relates to the technical field of geological disaster early warning. The ground surface deformation of a monitoring area is identified by using synthetic aperture radar interferometry technology, and a geological disaster hidden danger area is circled. Multi-modal sensing devices are deployed in the geological disaster hidden danger area, real-time monitoring data related to geological disasters are collected, preprocessing operations are performed, and a unified data matrix is constructed. The unified data matrix is simultaneously input into a process early warning timing neural network model and a causal inference graph neural network model. The results of the two models are fused by an ensemble learning model to output a joint early warning level and a confidence degree. Dynamic threshold setting is performed based on reinforcement learning, the determination threshold parameters of each early warning level are automatically adjusted according to the feedback reward of historical early warning, and the related weights of the ensemble learning model are updated in real time. The application can adapt to the dynamic change of the geological environment, update the early warning determination in real time, and finally improve the early warning accuracy.

Description

Technical Field

[0001] This invention relates to the field of geological disaster early warning technology, and in particular to a geological disaster early warning method and system based on machine learning. Background Technology

[0002] Geological disasters are characterized by their suddenness, destructiveness, and difficulty in early warning, posing a serious threat to people's lives and property and critical infrastructure. With the rapid development of the Internet of Things, big data, and artificial intelligence technologies, machine learning-based geological disaster early warning methods and systems have become a research hotspot and application direction in the field of disaster prevention and mitigation. Existing machine learning-based geological disaster early warning schemes often use statically set thresholds based on existing monitoring data and expert experience. However, the formation and occurrence of geological disasters are dynamic evolutionary processes involving multiple coupled factors. Rock and soil structures, hydrological environments, and external disturbances are all in a state of continuous change. Early warning mechanisms with fixed static thresholds have fundamental technical defects; existing solutions cannot adjust in real time as monitoring data dynamically increases, and early warning models still have limitations in accuracy and coverage. Therefore, for those skilled in the art, designing a novel geological disaster early warning method with dynamic adaptive early warning and intelligent decision support is an urgent problem to be solved. Summary of the Invention

[0003] The purpose of this invention is to provide a geological disaster early warning method and system based on machine learning to solve the problems mentioned in the background art. It can adapt to dynamic changes in the geological environment, update the early warning judgment in real time, and ultimately improve the accuracy of the early warning.

[0004] To achieve the above objectives, the present invention provides the following solution: On one hand, it provides a geological disaster early warning method based on machine learning, the specific steps of which include the following:

[0005] Synthetic aperture radar interferometry is used to identify surface deformation in the monitored area and delineate potential geological hazard zones.

[0006] Multimodal sensing devices are deployed in the geological hazard-prone areas to collect real-time monitoring data related to geological hazards;

[0007] The real-time monitoring data is preprocessed to construct a unified data matrix;

[0008] The unified data matrix is ​​simultaneously input into the process early warning time-series neural network model and the causal inference graph neural network model;

[0009] By integrating the output results of the process early warning time-series neural network model and the output results of the causal inference graph neural network model through an ensemble learning model, a joint early warning level and confidence level are output.

[0010] Dynamic threshold setting is based on reinforcement learning. The judgment threshold parameters of each warning level are automatically adjusted according to the feedback rewards of historical warnings, and the relevant weights of the ensemble learning model are updated in real time.

[0011] Preferably, the specific steps for identifying surface deformation in the monitoring area using synthetic aperture radar interferometry are as follows:

[0012] Acquire multi-temporal synthetic aperture radar images of the monitoring area, perform registration, terrain phase and atmospheric phase correction processing on the multi-temporal synthetic aperture radar images, and extract the surface deformation time series;

[0013] The deformation rate and deformation acceleration are calculated based on the aforementioned surface deformation time series.

[0014] A deformation anomaly threshold is set, and areas where the deformation rate or deformation acceleration exceeds the deformation anomaly threshold are marked as potential deformation zones. Spatial overlay analysis is then performed by combining optical remote sensing images with digital elevation models to finally delineate the geological hazard hazard zones.

[0015] Preferably, the real-time monitoring data includes geological structural parameter data, topographic and geomorphological parameter data, meteorological parameter data, soil parameter data, and hydrological parameter data; wherein, the geological structural parameter data includes stratigraphic structure, lithological distribution, and fault zones; the topographic and geomorphological parameter data includes surface elevation, slope, aspect, and topographic curvature; the meteorological parameter data includes rainfall, temperature, humidity, and wind speed; the soil parameter data includes soil moisture content, soil type, and permeability; and the hydrological parameter data includes groundwater level and river water level.

[0016] Preferably, the preprocessing operation for the real-time monitoring data includes spatiotemporal alignment, adaptive noise filtering, and feature extraction to construct the unified data matrix.

[0017] Preferably, the process early warning temporal neural network model is constructed by combining a temporal convolutional network with an attention mechanism. It takes the temporal feature sequence in the unified data matrix as input, extracts the dynamic pattern of surface deformation and environmental factors evolving over time, and outputs a first early warning probability vector.

[0018] Preferably, the causal inference graph neural network model is constructed using a graph attention network. Each monitoring point in the geological hazard area is used as a graph node, and the spatial topological relationship or geological structure relationship between the nodes is used as an edge to construct a spatial topological graph. The causal dependency relationship between the features of each node is extracted through graph convolution operation, and a second early warning probability vector is output.

[0019] Preferably, the method also includes simultaneously pushing the early warning information to a digital twin visualization platform to automatically trigger three-dimensional simulation and predict the scope and evolution trend of the disaster's impact.

[0020] Preferably, the step of fusing the output of the process early warning time-series neural network model and the output of the causal inference graph neural network model through an ensemble learning model is as follows: using a stacking ensemble learning strategy, the output of the process early warning time-series neural network model and the output of the causal inference graph neural network model are used as meta-features; the meta-features are input into a preset meta-learner for weighted fusion, and the meta-learner dynamically allocates weights according to the performance of each base model on the validation set, and finally outputs the joint early warning level and the corresponding confidence level.

[0021] On the other hand, a machine learning-based geological disaster early warning system is provided, including a surface deformation identification module, a monitoring data acquisition module, a preprocessing module, a dual-model parallel analysis module, a joint early warning module, and a dynamic threshold adaptive module; among which,

[0022] The surface deformation identification module is used to identify surface deformation in the monitoring area using synthetic aperture radar interferometry technology and delineate areas with potential geological hazards.

[0023] The monitoring data acquisition module is used to deploy multimodal sensing devices in the geological hazard-prone area to collect real-time monitoring data related to geological hazards;

[0024] The preprocessing module is used to preprocess the real-time monitoring data and construct a unified data matrix;

[0025] The dual-model parallel analysis module is used to simultaneously input the unified data matrix into the process early warning time-series neural network model and the causal inference graph neural network model;

[0026] The joint early warning module is used to fuse the output results of the process early warning time-series neural network model and the output results of the causal inference graph neural network model through an ensemble learning model, and output the joint early warning level and confidence level.

[0027] The dynamic threshold adaptive module is used to set dynamic thresholds based on reinforcement learning, automatically adjust the judgment threshold parameters of each warning level according to the feedback rewards of historical warnings, and update the relevant weights of the ensemble learning model in real time.

[0028] Preferably, it also includes a digital twin simulation module, used to receive the joint early warning level and confidence level, automatically trigger three-dimensional simulation, and predict the impact range and evolution trend of geological disasters.

[0029] According to the present invention, a geological disaster early warning method and system based on machine learning is provided, which discloses the following technical effects: A three-dimensional sensing system is constructed by macroscopically delineating potential hazard areas using InSAR technology and combining it with real-time microscopic monitoring using multimodal sensing equipment; an innovative dual-model parallel architecture of a process-based early warning temporal neural network and a causal inference graph neural network is adopted, utilizing ensemble learning to deeply fuse temporal evolution features and spatial causal correlations, significantly improving the accuracy and robustness of early warnings in complex environments; further, a reinforcement learning mechanism is introduced to dynamically adjust the early warning threshold and model weights based on historical feedback, effectively solving the problem of false alarms and missed alarms caused by traditional fixed thresholds, realizing adaptive evolution and intelligent decision-making of the early warning system, and greatly improving the timeliness and reliability of geological disaster early warnings. Attached Figure Description

[0030] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0031] Figure 1 This is a flowchart of the method of the present invention;

[0032] Figure 2 This is a system structure diagram of the present invention. Detailed Implementation

[0033] 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.

[0034] like Figure 1 As shown, the purpose of this invention is to provide a geological disaster early warning method based on machine learning, the specific steps of which include the following:

[0035] S1. Use synthetic aperture radar interferometry to identify surface deformation in the monitoring area and delineate potential geological hazard zones;

[0036] S2. Deploy multimodal sensing equipment in areas prone to geological disasters to collect real-time monitoring data related to geological disasters;

[0037] S3. Perform preprocessing operations on real-time monitoring data to construct a unified data matrix;

[0038] S4. Simultaneously input the unified data matrix into the process early warning time-series neural network model and the causal inference graph neural network model;

[0039] S5. By fusing the output results of the process warning time-series neural network model and the output results of the causal inference graph neural network model through the ensemble learning model, the joint warning level and confidence level are output.

[0040] S6. Dynamic threshold setting based on reinforcement learning: The judgment threshold parameters of each warning level are automatically adjusted according to the feedback rewards of historical warnings, and the relevant weights of the integrated learning model are updated in real time.

[0041] Furthermore, the specific steps for identifying surface deformation in the monitored area using synthetic aperture radar interferometry in S1 are as follows:

[0042] S11. Acquire multi-temporal synthetic aperture radar images of the monitoring area, perform registration, terrain phase and atmospheric phase correction on the multi-temporal synthetic aperture radar images, and extract the surface deformation time series.

[0043] S12. Calculate deformation rate and deformation acceleration based on surface deformation time series;

[0044] S13. Set a deformation anomaly threshold and mark areas where the deformation rate or deformation acceleration exceeds the deformation anomaly threshold as potential deformation areas. To further improve the accuracy of hazard area delineation, combine high-resolution optical remote sensing images with digital elevation models for spatial overlay analysis, eliminate false deformations caused by vegetation cover or buildings, and finally accurately delineate the geological hazard areas.

[0045] Furthermore, within the aforementioned designated geological hazard zones, multimodal sensing equipment, including geological sensors, meteorological monitoring stations, soil monitoring instruments, hydrological monitoring terminals, and surface displacement monitoring equipment, is deployed in a grid pattern based on the zone's extent, topographic complexity, and geological structure distribution. This enables multi-dimensional, all-weather data collection. Real-time monitoring data includes geological structural parameters, topographic parameters, meteorological parameters, soil parameters, and hydrological parameters. The data includes: geological structural parameters: real-time collection of stratigraphic structure data, surface and underground lithological distribution characteristics data, and fault zone location and activity status data in the hazard area, reflecting the stability of the regional geological foundation; topographic parameters: collection of surface elevation, slope, aspect, and curvature data at monitoring points, characterizing the inducing effect of regional topography on geological disasters; meteorological parameters: real-time high-frequency collection of regional rainfall, ambient temperature, air humidity, and wind speed data, with rainfall being a core inducing factor for disasters such as debris flows and landslides; soil parameters: collection of soil moisture content, soil type, and soil permeability data at monitoring points, reflecting soil stability and permeability characteristics; and hydrological parameters: real-time monitoring of groundwater depth and regional river water level changes, as water infiltration and water level fluctuations are important factors inducing slope instability. All sensing devices are set to a sampling frequency of 5-30 minutes / time, with automatic frequency increase under extreme weather conditions to ensure the real-time nature and completeness of the monitoring data.

[0046] Furthermore, since the data collected by multimodal sensing devices is characterized by multi-source heterogeneity and varying sampling frequencies, preprocessing operations are performed on the real-time monitoring data in S3, including spatiotemporal alignment, adaptive noise filtering, and feature extraction, to construct a unified data matrix. Spatiotemporal alignment eliminates timestamp differences and spatial coordinate deviations between different sensors; adaptive noise filtering removes outliers caused by environmental interference; finally, feature extraction and standardization are performed to map the multidimensional heterogeneous data to a unified feature space, thus constructing a unified data matrix.

[0047] Furthermore, in S4, the preprocessed unified data matrix is ​​synchronously input into the process early warning time-series neural network model and the causal inference graph neural network model to achieve bidirectional mining of time-series dynamic features and spatial causal features. The two models can operate in parallel without interfering with each other.

[0048] The process early warning temporal neural network model is constructed by combining temporal convolutional networks with an attention mechanism. It uses the temporal feature sequence from a unified data matrix as input to extract the dynamic patterns of surface deformation and environmental factors evolving over time, and outputs a first early warning probability vector. Specifically, a multi-layer temporal convolutional network is used to extract the dynamic changes in surface deformation, meteorological, hydrological, and soil factors over time, capturing the gradual changes in the disaster incubation process. Then, an attention mechanism is used to assign adaptive weights to key temporal features, enhancing the ability to extract key information about disaster precursors such as sudden increases in rainfall, accelerated surface deformation, and sudden changes in groundwater levels. After feature encoding and temporal dependency learning, the model outputs the first early warning probability vector.

[0049] The causal inference graph neural network model is constructed using a graph attention network. Each monitoring point in the geological hazard area is used as a graph node, and the spatial topological relationship or geological structure relationship between the nodes is used as an edge to construct a spatial topological graph. The causal dependency relationship between the features of each node is extracted through graph convolution operation, and the second early warning probability vector is output.

[0050] Furthermore, in S5, a Stacking ensemble learning strategy is adopted to fuse the output results of the two models, achieving complementary advantages and improving the accuracy and reliability of the early warning results. The specific steps are as follows:

[0051] S51. The first early warning probability vector output by the process early warning time-series neural network model and the second early warning probability vector output by the causal inference graph neural network model are used together as meta-features of Stacking ensemble learning.

[0052] S52. Input the meta-features into the preset meta-learner for weighted fusion. The meta-learner dynamically allocates weights according to the performance of each base model on the validation set, and finally outputs the joint warning level and the corresponding confidence level, thus completing the joint warning result output.

[0053] Furthermore, S6 uses reinforcement learning to dynamically set thresholds, automatically adjusting the judgment threshold parameters for each warning level based on historical warning feedback rewards. Specifically, this includes:

[0054] S61. Construct a reinforcement learning agent, using the joint warning level and confidence level as the state space, and the judgment threshold parameters of each warning level as the action space.

[0055] S62. Based on the comparison between the actual occurrence of geological disasters and the early warning results, construct a composite reward function that includes penalties for missed reporting, penalties for false reporting, and rewards for early warning.

[0056] S63. The reinforcement learning agent maximizes the composite reward function and uses the policy gradient algorithm or the deep deterministic policy gradient algorithm to optimize online and automatically adjust the decision threshold parameter.

[0057] Furthermore, it also includes simultaneously pushing early warning information to a digital twin visualization platform to automatically trigger three-dimensional simulation and predict the scope and evolution trend of disaster impact.

[0058] On the other hand, a geological disaster early warning system based on machine learning is provided, such as... Figure 2 As shown, it includes a surface deformation identification module, a monitoring data acquisition module, a preprocessing module, a dual-model parallel analysis module, a joint early warning module, and a dynamic threshold adaptive module; among which,

[0059] The surface deformation identification module is used to identify surface deformation in the monitoring area using synthetic aperture radar interferometry technology, and to delineate areas with potential geological hazards.

[0060] The monitoring data acquisition module is used to deploy multimodal sensing devices in areas prone to geological disasters to collect real-time monitoring data related to geological disasters;

[0061] The preprocessing module is used to preprocess real-time monitoring data and construct a unified data matrix;

[0062] The dual-model parallel analysis module is used to simultaneously input a unified data matrix into the process early warning time-series neural network model and the causal inference graph neural network model;

[0063] The joint early warning module is used to fuse the output results of the process early warning time-series neural network model and the output results of the causal inference graph neural network model through an ensemble learning model, and output the joint early warning level and confidence level.

[0064] The dynamic threshold adaptive module is used to set dynamic thresholds based on reinforcement learning. It automatically adjusts the judgment threshold parameters of each warning level according to the feedback rewards of historical warnings and updates the relevant weights of the ensemble learning model in real time.

[0065] It also includes a digital twin simulation module, which is used to receive joint early warning levels and confidence levels, automatically trigger three-dimensional simulation, and predict the impact range and evolution trend of geological disasters.

[0066] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A geological disaster early warning method based on machine learning, characterized in that, The specific steps include the following: Synthetic aperture radar interferometry technology is used to identify surface deformation in the monitoring area and delineate areas prone to geological disasters. Multimodal sensing devices are deployed in the geological hazard-prone areas to collect real-time monitoring data related to geological hazards; The real-time monitoring data is preprocessed to construct a unified data matrix; The unified data matrix is ​​simultaneously input into the process early warning time-series neural network model and the causal inference graph neural network model; By integrating the output results of the process early warning time-series neural network model and the output results of the causal inference graph neural network model through an ensemble learning model, a joint early warning level and confidence level are output. Dynamic threshold setting is based on reinforcement learning. The judgment threshold parameters of each warning level are automatically adjusted according to the feedback rewards of historical warnings, and the relevant weights of the ensemble learning model are updated in real time.

2. The geological disaster early warning method based on machine learning according to claim 1, characterized in that, The specific steps for identifying surface deformation in the monitored area using synthetic aperture radar interferometry are as follows: Acquire multi-temporal synthetic aperture radar images of the monitoring area, perform registration, terrain phase and atmospheric phase correction processing on the multi-temporal synthetic aperture radar images, and extract the surface deformation time series; The deformation rate and deformation acceleration are calculated based on the aforementioned surface deformation time series. A deformation anomaly threshold is set, and areas where the deformation rate or deformation acceleration exceeds the deformation anomaly threshold are marked as potential deformation zones. Spatial overlay analysis is then performed by combining optical remote sensing images with digital elevation models to finally delineate the geological hazard hazard zones.

3. The geological disaster early warning method based on machine learning according to claim 1, characterized in that, The real-time monitoring data includes geological structural parameters, topographic parameters, meteorological parameters, soil parameters, and hydrological parameters. Specifically, the geological structural parameters include stratigraphic structure, lithological distribution, and fault zones; the topographic parameters include surface elevation, slope, aspect, and topographic curvature; the meteorological parameters include rainfall, temperature, humidity, and wind speed; the soil parameters include soil moisture content, soil type, and permeability; and the hydrological parameters include groundwater level and river water level.

4. The geological disaster early warning method based on machine learning according to claim 1, characterized in that, The preprocessing operations for the real-time monitoring data include spatiotemporal alignment, adaptive noise filtering, and feature extraction to construct the unified data matrix.

5. A geological disaster early warning method based on machine learning according to claim 1, characterized in that, The process early warning temporal neural network model is constructed by combining a temporal convolutional network with an attention mechanism. It takes the temporal feature sequence in the unified data matrix as input, extracts the dynamic pattern of surface deformation and environmental factors evolution over time, and outputs the first early warning probability vector.

6. The geological disaster early warning method based on machine learning according to claim 1, characterized in that, The causal inference graph neural network model is constructed using a graph attention network. Each monitoring point in the geological hazard area is used as a graph node, and the spatial topological relationship or geological structure relationship between the nodes is used as an edge to construct a spatial topological graph. The causal dependency relationship between the features of each node is extracted through graph convolution operation, and a second early warning probability vector is output.

7. The geological disaster early warning method based on machine learning according to claim 1, characterized in that, It also includes simultaneously pushing early warning information to a digital twin visualization platform to automatically trigger three-dimensional simulation and predict the scope and evolution trend of disaster impact.

8. A geological disaster early warning method based on machine learning according to claim 1, characterized in that, The step of fusing the output of the process early warning time-series neural network model and the output of the causal inference graph neural network model through an ensemble learning model is as follows: using a stacking ensemble learning strategy, the output of the process early warning time-series neural network model and the output of the causal inference graph neural network model are used as meta-features; The meta-features are input into a preset meta-learner for weighted fusion. The meta-learner dynamically allocates weights based on the performance of each base model on the validation set, and finally outputs the joint warning level and the corresponding confidence level.

9. A geological disaster early warning system based on machine learning, characterized in that, It includes a surface deformation identification module, a monitoring data acquisition module, a preprocessing module, a dual-model parallel analysis module, a joint early warning module, and a dynamic threshold adaptive module; among which, The surface deformation identification module is used to identify surface deformation in the monitoring area using synthetic aperture radar interferometry technology and delineate areas with potential geological hazards. The monitoring data acquisition module is used to deploy multimodal sensing devices in the geological hazard-prone area to collect real-time monitoring data related to geological hazards; The preprocessing module is used to preprocess the real-time monitoring data and construct a unified data matrix; The dual-model parallel analysis module is used to simultaneously input the unified data matrix into the process early warning time-series neural network model and the causal inference graph neural network model; The joint early warning module is used to fuse the output results of the process early warning time-series neural network model and the output results of the causal inference graph neural network model through an ensemble learning model, and output the joint early warning level and confidence level. The dynamic threshold adaptive module is used to set dynamic thresholds based on reinforcement learning, automatically adjust the judgment threshold parameters of each warning level according to the feedback rewards of historical warnings, and update the relevant weights of the ensemble learning model in real time.

10. A geological disaster early warning system based on machine learning, characterized in that, It also includes a digital twin simulation module, which is used to receive the joint early warning level and confidence level, automatically trigger three-dimensional simulation, and predict the impact range and evolution trend of geological disasters.

Images