An extreme environment emergency hidden danger multi-source data fusion processing system

Abstract

The application discloses an extreme environment emergency hidden danger multi-source data fusion processing system and relates to the technical field of extreme environment data processing.The system comprises the following steps: collecting multi-dimensional hidden danger original data, real-time environment parameters and rescue resource data through air-ground-water three-dimensional intelligent detection terminals, adopting a dual-mode redundant transmission architecture to guarantee real-time data return and network interruption continuation; relying on a three-level progressive fusion architecture of data level-feature level-semantic level, combining an emergency hidden danger field knowledge graph and regional geological environment characteristics, deeply fusing multi-source heterogeneous data, extracting hidden danger core features, correlation and semantic information; constructing a hidden danger-environment-intervention ternary dynamic coupling evolution model, combining historical disaster cases and an adaptive iteration mode, deducing hidden danger evolution tracks and risk transmission paths, and generating differentiated intervention schemes; finally, screening and optimizing the minimum intervention decision and visualizing the same. The system realizes precise hidden danger perception, intelligent fusion and scientific decision-making under extreme environment.

Description

Technical Field

[0001] This invention relates to the field of extreme environment data processing technology, and more specifically, to a multi-source data fusion processing system for emergency response risks in extreme environments. Background Technology

[0002] Following a disaster, affected areas often face extreme conditions such as road closures, communication disruptions, and harsh environments. Before emergency rescue forces can reach the site, numerous potential secondary disaster hazards may be hidden within the area, such as landslide debris, unstable rock masses, dam breach risk points, underwater reefs, and sources of toxic gas leaks. If these hidden hazards cannot be quickly, accurately identified, and visualized, they will seriously threaten the lives of rescue personnel and hinder emergency response efficiency. Therefore, a multi-source data fusion and processing system for extreme environment emergency hazards is needed.

[0003] Existing multi-source data fusion technologies for emergency response hazards in extreme environments mostly employ simple splicing or weighted summation methods, failing to fully adapt to the heterogeneous characteristics of air, ground, and water-based detection data. For multimodal data such as images, point clouds, and physical quantities, shallow fusion is only performed at the data or feature layers, lacking semantic-level alignment with knowledge graphs in the emergency response hazard domain, making it difficult to uncover the coupling and correlation patterns between hazards. Furthermore, in extreme environments, issues such as sensor noise, data gaps, and temporal misalignments are prominent. Existing methods are insufficient in completing missing data and suppressing noise, resulting in low accuracy in hazard feature extraction and large deviations in risk identification results, failing to provide reliable data support for subsequent evolutionary projections.

[0004] Existing technologies lack the ability to model the coupled evolution of hazards in extreme environments. They often focus on the independent analysis of single hazards, neglecting the inducing, symbiotic, and coupled effects among multiple hazards such as building collapse, gas leaks, and landslide instability. This makes it difficult to accurately predict risk transmission paths and the evolution trend of secondary disasters. In the intervention decision-making stage, there is still a reliance on experience-based and standardized fixed plans, without dynamic optimization based on real-time rescue resource constraints, hazard coupling intensity, and changes in extreme working conditions. This results in problems such as redundant intervention intensity, unreasonable resource allocation, and poor adaptability to working conditions, making it impossible to achieve precise emergency intervention with "controllable risk and minimal resources," thus restricting the safety and response efficiency of emergency rescue.

[0005] Existing technologies for hazard detection visualization mostly remain at the level of static data display, only presenting basic information such as the location and type of hazards. They lack dynamic visualization of hazard evolution trends, risk transmission paths, and the effects of intervention plans, making it difficult to provide decision-makers with a comprehensive risk situation awareness. At the same time, the system's ability to adapt to real-time scene changes in extreme environments is insufficient, and it has failed to establish a dynamic feedback mechanism between hazards, environment, and intervention. It cannot quickly adjust intervention plans based on real-time changes in the core characteristics of hazards and environmental parameters, resulting in a disconnect between decision-making results and actual on-site needs, making it difficult to cope with the complex emergency response requirements in extreme disaster scenarios. Summary of the Invention

[0006] The main objective of this invention is to disclose a multi-source data fusion processing system for emergency response hazards in extreme environments, comprising: Multi-source heterogeneous data acquisition module: Based on the deployment of an air-ground-water amphibious intelligent detection terminal, multi-source data is collected; the multi-source data includes: original data of multi-dimensional hidden dangers in extreme environments within the target detection area, real-time environmental parameters, and on-site rescue resource data; Furthermore, before acquiring multi-source heterogeneous data, the multi-source heterogeneous data acquisition module divides the area into high-altitude detection sub-regions, ground detection sub-regions, and underwater detection sub-regions, and deploys drone clusters, ground mobile robot formations, and underwater robots. Among them, the multi-dimensional original data of hidden dangers refers to: the form, location and physical characteristics of hidden dangers within the target detection area; the real-time environmental parameters refer to: rainfall, wind speed, soil moisture and water temperature; and the on-site rescue resource data refers to: the number of rescue personnel, equipment type and material reserves.

[0007] Multi-source data fusion module: Adopting a three-level progressive fusion architecture, it combines knowledge graphs in the field of emergency hazards with geological and environmental characteristics of the target detection area to deeply fuse the collected data and extract the core features of hazards, the correlation between hazards, and the semantic association between hazards and the environment; Specifically, the multi-source data fusion module adopts a three-level progressive fusion architecture of data level, feature level, and semantic level. The data level refers to the registration, splicing, and redundancy removal of raw data from similar sensors. The feature level refers to the extraction of core features of various types of data and the generation of a unified hazard feature vector through attention-weighted fusion. The semantic level refers to the mapping of the feature vector to the emergency hazard semantic space to complete the semantic alignment of different modal data. Furthermore, the multi-source data fusion module performs preprocessing on the knowledge graph of the emergency hazard domain, extracts the knowledge subgraph corresponding to the target detection area and removes irrelevant nodes, and filters and optimizes the initial feature vector in combination with geological environment characteristics, retains the core features associated with high-risk hazards, and generates a unified hazard feature vector that is adapted to the characteristics of the target detection area. Among them, the relationships between hidden dangers extracted by the multi-source data fusion module include: induced relationships, coupled relationships, and symbiotic relationships; Furthermore, the extraction of core features of hidden dangers, the correlation between hidden dangers, and the semantic association between hidden dangers and the environment refers to: extracting the iconic core features of each hidden danger, combining the coupling correlation of hidden dangers, analyzing and identifying the coupling, inducing and symbiotic correlation between different hidden dangers, and conducting multi-dimensional correlation analysis between the core feature set of hidden dangers and the correlation network between hidden dangers and regional geological and environmental characteristics to form a semantic association set between hidden dangers and the environment in the target detection area; Furthermore, the correlation between the hidden dangers is identified by extracting key features of structural damage, flammable and explosive, and geological disaster hazards when extracting core features of the hidden dangers. Combined with the hidden danger coupling correlation of the special emergency hidden danger knowledge subgraph of the target detection area, the spatiotemporal correlation analysis identifies the induced correlation, coupling correlation and symbiotic correlation between different hidden dangers.

[0008] Hazard Coupling Evolution Simulation Module: Based on the ternary dynamic coupling evolution model, combined with historical disaster coupling cases, it simulates the coupling evolution trajectory of hazards and risk transmission paths, and outputs full-dimensional simulation data and differentiated intervention plans; Specifically, the ternary dynamic coupling evolution model consists of an input layer, a coupling evolution layer, and an output deduction layer. After initial parameter assignment, the model is initially calibrated based on on-site measured data to ensure that the simulation deviation does not exceed 10%. The model parameters are dynamically adjusted through a real-time fusion data feedback calibration mechanism. Furthermore, the hidden danger coupling evolution simulation module adapts differentiated intervention schemes from a preset set of basic intervention schemes based on the risk level and evolution trend of the hidden dangers. The basic intervention scheme set includes intervention schemes for structural damage hazards, flammable and explosive hazards, and geological hazards under the conditions of rapid escalation of danger level and slow development of warning level, respectively. Furthermore, the aforementioned simulation of the hazard coupling evolution trajectory and risk transmission path refers to: inputting high-quality fused feature data into a ternary dynamic coupling evolution model, setting different environmental operating condition parameters, and using a dynamic coupling simulation algorithm to simulate and generate the hazard coupling evolution trajectory and risk transmission path under different extreme environmental operating conditions.

[0009] Emergency Decision-Making and Visualization Output Module: Used to filter and optimize intervention plans by combining fused feature data, and generate minimum intervention decision results, fused results, extrapolation results and decision results; Specifically, the emergency decision-making and visualization output module takes risk control compliance as a hard constraint and minimizing resource input as an optimization goal. It uses a three-element dynamic coupling evolution model to simulate the intervention effects under different working conditions, screens and optimizes various solutions, and generates the minimum intervention decision result.

[0010] According to this invention, multi-dimensional hazard data is collected comprehensively by deploying an amphibious intelligent detection terminal. A three-level progressive fusion architecture is adopted in conjunction with an emergency hazard knowledge graph to achieve deep fusion of multi-source heterogeneous data and extract the core features and correlations of hazards. A ternary dynamic coupling evolution model of hazard-environment-intervention is constructed to deduce the evolution trajectory of hazards and risk transmission paths and generate differentiated intervention plans. With risk control compliance as a hard constraint and minimizing resource input as the goal, the minimum intervention decision results are selected and generated. This achieves accurate perception, intelligent fusion and scientific decision-making of hazards in extreme environments, optimizes the allocation of rescue resources, and improves the safety and response efficiency of emergency rescue in extreme environments. Attached Figure Description

[0011] Figure 1 This is a system architecture diagram of a multi-source data fusion processing system for emergency response hazards in extreme environments, provided according to an embodiment of the present invention. Detailed Implementation

[0012] The specific implementation of the present invention will now be described in detail with reference to the accompanying drawings.

[0013] like Figure 1 As shown, the main objective of this invention is to disclose a multi-source data fusion processing system for emergency response hazards in extreme environments, comprising: P100 Multi-Source Heterogeneous Data Acquisition Module: Based on the deployment of an air-ground-water amphibious intelligent detection terminal, it acquires multi-source data; the multi-source data includes: original data of multi-dimensional hidden dangers in extreme environments within the target detection area, real-time environmental parameters, and on-site rescue resource data; Furthermore, before collecting multi-dimensional hidden danger raw data, real-time environmental parameters and on-site rescue resource data in extreme environments within the target detection area, the geometric center of the target detection area is used as a reference to divide it into several detection sub-areas according to the grid principle. The high-altitude detection sub-area is divided into 200m×200m, the ground detection sub-area is divided into 100m×100m, and the underwater detection sub-area is divided into 150m×150m. Furthermore, amphibious intelligent detection terminals are deployed in each detection sub-area. In the high-altitude detection sub-area, a cluster of drones equipped with photoelectric, infrared, synthetic aperture radar and meteorological sensors is deployed. In the ground detection sub-area, a formation of ground mobile robots equipped with lidar, gas sensors, vibration sensors and temperature sensors is deployed. In the underwater detection sub-area, underwater robots equipped with side-scan sonar, multibeam echo sounders and water quality sensors are deployed. Specifically, data is collected at preset frequencies: drones collect data every 30 minutes, ground mobile robots collect data every 15 minutes, and underwater robots collect data every 20 minutes. Simultaneously, multi-dimensional raw data on potential hazards, real-time environmental parameters, and on-site rescue resource data are collected within the target detection area. The multi-dimensional raw data on potential hazards includes: hazard morphology, location, and physical characteristics; real-time environmental parameters include: rainfall, wind speed, soil moisture, and water temperature; and on-site rescue resource data includes: the number of rescue personnel, equipment types, and material reserves.

[0014] P110 Multi-Source Data Fusion Module: Adopts a three-level progressive fusion architecture, combining knowledge graphs in the field of emergency hazards, geological and environmental characteristics of the target detection area, to deeply fuse the collected data and extract core features of hazards, relationships between hazards, and semantic associations between hazards and the environment; 1) The three-level progressive fusion architecture is a data-level-feature-semantic-level progressive fusion architecture. The specific fusion process is as follows: Data-level fusion: Register, stitch, and de-redundancy process multiple frames of raw data collected by similar sensors within the target detection area to achieve normalized integration of similar data and improve data integrity and accuracy.

[0015] Feature-level fusion: Extracting core features from various data types, including texture and contour features of image data, spatial coordinates and topological features of point cloud data, and gradient and rate of change features of physical quantity data. Through cross-modal feature extraction and attention-weighted fusion, a unified hazard feature vector is generated. The specific steps are as follows: Step 1: Extract visual features such as texture, contour, and edge from image data; extract geometric features such as spatial coordinates, topology, and density distribution from point cloud data; and extract dynamic features such as gradient, rate of change, and peak value from physical quantity data such as gas, temperature, and vibration. Step 2: Introduce a cross-attention mechanism to adaptively assign weights to multimodal features, giving higher weights to modal features that are more relevant to hazard identification, and normalizing conflict features to eliminate fusion bias caused by scale differences; Step 3: Transform the raw data of different modalities into a unified hazard feature vector that includes information on appearance, space, physical and dynamic changes, so as to achieve organic integration of multi-source data features and improve the accuracy of subsequent hazard identification and correlation analysis.

[0016] Semantic-level fusion: Hazard feature vectors are mapped to a predefined emergency hazard semantic space, completing semantic alignment of data from different modalities and achieving deep fusion of multi-source heterogeneous data within the target detection area. Specific steps are as follows: Step 1: Construct a unified semantic space based on the knowledge graph of the emergency hazard domain, including: core hazard semantic nodes such as structural damage, flammable and explosive materials, and geological disasters, as well as the evolution characteristics, coupling relationships and semantic associations of various hazards; Step 2: The hidden danger feature vector obtained by feature-level fusion is transformed into structured semantic information through a semantic mapping model to achieve consistent semantic expression of multimodal data such as images, point clouds, and physical quantities. Step 3: Combine the target detection area with the specific emergency hazard knowledge subgraph to perform regional calibration of semantic information, ensuring that the semantic description is consistent with the on-site geological environment and the evolution law of hazards, and eliminating semantic conflicts caused by data heterogeneity between different modalities; Specifically, through semantic-level fusion, a deep transformation from multi-source heterogeneous data to structured semantic information is achieved, laying a semantic foundation for subsequent extraction of core features of hidden dangers, analysis of the coupling and correlation of hidden dangers, and full-dimensional evolutionary deduction.

[0017] 2) Combining the knowledge graph of emergency hazard domain with the geological and environmental characteristics of the target detection area, achieve deep fusion of multi-source heterogeneous data within the target detection area. The specific steps are as follows: Step 1: Knowledge Graph Preprocessing: The knowledge graph for the emergency hazard domain is a ternary knowledge network with extreme environment type - hazard type - evolution law - geological environment influencing factors as its core. It includes hazard types such as extreme environment and structural damage, flammable and explosive, and geological disaster. Knowledge subgraphs are extracted for the target detection area, irrelevant knowledge nodes are removed, and an emergency hazard knowledge subgraph exclusive to the target detection area is formed. Step 2: Collect standards, industry norms, disaster prevention manuals and academic research results related to typical extreme environments and emergency hazards such as earthquakes, landslides, rainstorms, gas leaks, and building damage. Summarize the classification system of extreme environment types, structural damage types, flammable and explosive types and geological disaster types, and clarify the typical manifestations, evolution stages and development patterns of various types of hazards. Step 3: Based on a large number of historical disaster cases, measured monitoring data and simulation test results, statistical analysis is conducted on the triggering conditions, evolution rates, risk thresholds and coupling correlations of various hidden dangers under different geological and meteorological conditions, forming quantitative knowledge entries; Step 4: Generate knowledge entries as a triplet structure with extreme environment type - hidden danger type - evolution law - geological environment influencing factors as the core, establish nodes and related edges, clarify node attributes, association strength, influence weight and data feature threshold, and form a structured knowledge network; Step 5: Experts in emergency management, geological engineering, and safety monitoring review and revise the rules, logic, and thresholds of the knowledge graph to ensure that the knowledge graph conforms to the real disaster evolution mechanism and engineering practice, and to determine the domain knowledge graph suitable for the analysis of emergency hazards in extreme environments.

[0018] 3) Extract the geological characteristics of soil and rock types and stratigraphic structure, as well as the environmental characteristics of real-time climate and hydrological conditions in the target detection area. Normalize all characteristic parameters according to a unified standard to form a quantitative dataset of geological environmental characteristics. Specifically, by retrieving geological survey data, geotechnical engineering investigation reports, and historical geographic information data of the target detection area, basic geological information such as soil and rock types, stratigraphic distribution, soil layer thickness, rock mass integrity, and geological structure of the area is obtained; combined with lidar, synthetic aperture radar, and UAV aerial survey data, the physical properties of surface soil and rock are inverted to complete the extraction and quantification of geological characteristics; and considering environmental characteristics such as real-time climate and hydrological conditions, dynamic environmental parameters such as rainfall intensity, ambient temperature, wind speed and direction, soil moisture, surface runoff, and groundwater level in the area are obtained in real time through meteorological sensors carried by UAVs, temperature, humidity and wind speed sensors deployed by ground mobile robots, and water temperature and water flow velocity data collected by underwater robots.

[0019] 4) Extract the core features of each modality of data to construct the initial feature vector. The specific steps are as follows: Step 1: Extract the core features of each modal data of image texture contour, point cloud spatial topology and physical quantity gradient change rate and construct an initial feature vector. Based on the knowledge subgraph of the target detection area, combined with the geological environment characteristic quantification dataset, the initial feature vector is screened and optimized, retaining the core features associated with high-risk hazards and eliminating redundant features, and fusion to generate a unified hazard feature vector that is adapted to the characteristics of the target detection area. Step 2: For image modal data, Canny edge detection and LBP texture operator are used to extract the contour features and texture distribution features of the hidden danger area, respectively; for cloud data, spatial topology, point cloud curvature and spatial density distribution features are extracted by normal estimation and Euclidean clustering algorithm; for physical quantity data such as gas concentration, vibration and displacement, the gradient, rate of change and fluctuation amplitude features of time series data are calculated by sliding window differentiation method.

[0020] 5) Map the hazard feature vectors to the emergency hazard semantic space, and complete the accurate matching of feature vectors and semantic nodes through a knowledge subgraph specific to the target detection area, transforming the quantified feature vectors into structured semantic information. The specific steps are as follows: Step 1: Input the hidden danger feature vector obtained by feature-level fusion into the preset emergency hidden danger semantic space; The semantic space is constructed based on a knowledge graph in the field of emergency hazards, including three core hazard semantic nodes: structural damage, flammable and explosive, and geological disasters. It is further subdivided into standardized semantic entries such as hazard evolution stage, risk level, and coupling association type. Step 2: Through a pre-trained semantic mapping model, numerical feature vectors are accurately matched with standardized semantic nodes in the semantic space, transforming continuously quantified feature data into structured semantic information of hazard type, risk level, and evolution state. For example, the feature vector of "crack width 5.2mm, deformation rate 0.18mm / h" can be mapped to a semantic description of "structural damage hazard - moderate damage - slow evolution"; Step 3: Combine the quantitative dataset of geological environment characteristics to perform regional correction on the semantic information of the preliminary matching. Associate the preliminary semantic results with regional characteristics such as soil and rock type, stratigraphic structure, climate and hydrological conditions, and correct the semantic deviation caused by the mismatch between the general knowledge graph and the local geological conditions on site, so that a clear causal and spatiotemporal relationship is formed between the semantics of hidden dangers and regional geological parameters. For example, in areas with high porosity sandy soil layers, the system will automatically correct the semantic evaluation related to soil stability, so that there is a clear causal and spatiotemporal relationship between the semantics of potential hazards and regional geological parameters, making the semantic expression more in line with engineering practice. Step 4: Based on the identified hazard coupling relationships in the dedicated emergency hazard knowledge subgraph of the target detection area, perform cross-semantic alignment on the semantic information generated by different hazards and different modal data. For the differences and potential conflicts in semantic expression of multi-source data such as images, point clouds, and physical quantities, achieve alignment and fusion of different hazard semantics under a unified semantic standard through semantic consistency verification, conflict resolution rules, and coupling edge constraints of the knowledge subgraph, and eliminate semantic contradictions caused by data heterogeneity, different collection perspectives, or noise interference.

[0021] Step 5: After matching, correction and alignment, multi-source heterogeneous data such as images, point clouds, physical sensors and environmental parameters are unified into a structured semantic form, and this semantic information is consistent with the regional geological environment characteristics and domain prior knowledge.

[0022] 6) Extract the core features of potential hazards, the relationships between hazards, and the semantic associations between hazards and the environment in the target detection area. The specific steps are as follows: Step 1: From the deeply fused feature dataset, extract the key features of each hazard type for areas classified as structural damage, flammable and explosive, and geological disaster. Among them, structural damage type extracts the characteristics of crack size, deformation rate and component damage degree of structural body; flammable and explosive type extracts the characteristics of gas concentration, diffusion range and leakage point location; geological disaster type extracts the characteristics of soil displacement rate, slope gradient and soil compaction. Step 2: Combining the hazard coupling correlation in the specific emergency hazard knowledge subgraph of the target detection area, identify the coupling, inducing and symbiotic correlation relationships between different hazards in the target detection area through spatiotemporal correlation analysis; Step 3: Conduct multi-dimensional correlation analysis between the core feature set of hidden dangers and the correlation network between hidden dangers and regional geological and environmental characteristics, transform the feature data and correlation relationships into structured semantic information, and finally form a semantic set of correlation between hidden dangers and the environment in the target detection area.

[0023] 7) Analyze the correlation of the changing trends of the core characteristics of each hidden danger and determine the strength of the correlation between hidden dangers. The specific steps are as follows: Step 1: Construct dynamic change curves for all quantitative features in the core feature set of hidden dangers according to time series, unify the time dimension and sampling frequency, and form a time series change dataset for each hidden danger feature; Step 2: Using the Pearson correlation coefficient method, calculate the correlation coefficient of the change curves of the core features of any two types of hidden dangers. The time series change datasets of the two types of hidden danger features after the same time dimension and sampling frequency are denoted as time series X and time series Y, respectively. The feature values ​​at the same time are used as a group of sample points to construct N sets of time series corresponding sample pairs. Calculate the mean, covariance and standard deviation of the two types of time series. The correlation coefficient r is obtained by the ratio of the covariance to the product of the standard deviations of the two types of series. Where r takes values ​​in the range of [-1, 1], a positive value of r indicates that the two types of hidden danger characteristics change in the same direction, a negative value of r indicates that they change in opposite directions, the closer |r| is to 1, the stronger the correlation, and the closer |r| is to 0, the weaker the correlation. Specifically, the correlation coefficient is calculated segment by segment using time-series data within the sliding window to obtain the time-series correlation change curve, thus comprehensively quantifying the dynamic correlation degree of the core characteristics of the two types of hidden dangers throughout the entire monitoring period; Among them, a correlation coefficient r ≥ 0.8 indicates a strong association, 0.5 ≤ r < 0.8 indicates a moderate association, 0.3 ≤ r < 0.5 indicates a weak association, and r < 0.3 indicates no association.

[0024] 8) The correlations between potential hazards extracted by the multi-source data fusion module include: If the core characteristics of two types of hazards change with a time lag, and meet the causal triggering of hazard A to hazard B in the regional exclusive emergency hazard knowledge subgraph, that is, the core characteristics of hazard A change abnormally first, and the core characteristics of hazard B change significantly within the lag time window, and if there is no abnormality in the characteristics of hazard A, then hazard B has no obvious trend of change, then it is determined to be an induced association. If the core characteristics of the two types of hazards change without significant time lag, show synchronous and unidirectional or reverse abnormal trends, and conform to the interaction between hazards in the region-specific emergency hazard knowledge subgraph, and the characteristic changes of the two types of hazards influence and reinforce each other, without a clear causal triggering relationship, then it is judged as a coupled association. If the core characteristics of two types of hidden dangers show a synchronous abnormal trend, and they are consistent with the same geological environment triggering multiple types of hidden dangers in the regional exclusive emergency hidden danger knowledge sub-map, and the two types of hidden dangers have no direct interaction relationship, and are both caused by the same geological and environmental parameter anomalies in the target detection area, then they are judged to be symbiotic associations. Specifically, the identified relationships, their strength, and their types are recorded in a structured manner.

[0025] P120 Hazard Coupling Evolution Simulation Module: Based on the high-quality fusion feature data of the output target detection area, it constructs a three-element dynamic coupling evolution model of hazard-environment-intervention within the target detection area. It adopts a dynamic coupling simulation + intervention strategy adaptive iteration mode to capture the coupling correlation effect of multiple types of hazards in the target detection area, simulates the hazard coupling evolution trajectory and risk transmission path under different environmental conditions, outputs full-dimensional simulation data of the target detection area, and adapts to the real-time scenario of the target detection area and dynamically adjustable differentiated intervention schemes. The specific steps for constructing the ternary dynamic coupling evolution model are as follows: Step 1: Based on high-quality fused feature data, combined with regional-specific emergency hazard knowledge subgraphs, historical disaster coupling cases, and geological environment characteristic quantitative datasets, determine the model input elements and constraints; Step 2: A unified parameterized representation of the three core elements—hazards, environment, and intervention—is implemented. Hazard elements include: type identification of structural damage hazards, flammable and explosive hazards, and geological disaster hazards, along with quantified values ​​of core characteristics such as crack size, displacement rate, and gas concentration. Environmental elements include: geological parameters such as soil and rock type and stratigraphic structure, as well as real-time environmental parameters such as rainfall, wind speed, and soil moisture. Intervention elements include: intervention methods such as support, sealing, and monitoring, and their intensity levels. All parameters are assigned three-dimensional spatial coordinates and timestamps to achieve a unified expression in the spatiotemporal dimension; Step 3: Based on a large amount of historical disaster measurement data, multiple linear regression is used to obtain the initial interaction coefficients among hidden dangers, environment, and intervention. The coefficients are corrected through indoor physical simulation experiments to make them more consistent with the evolution of real disasters. The coupling association weights in the region-specific emergency hidden danger knowledge subgraph are calibrated to obtain the final interaction coefficients with values ​​between 0 and 1. A 3×3 coupling relationship quantification matrix is ​​then constructed to intuitively represent the mutual influence intensity among the three elements. Step 4: Construct a three-layer simulation architecture consisting of an input layer, a coupled evolution layer, and an output deduction layer. Assign initial parameter values ​​and perform initial calibration using measured data to ensure that the simulation results deviate from the real data by no more than 10%. Introduce a real-time fusion data feedback calibration mechanism to dynamically adjust model parameters based on the actual site conditions. Combine regional geological environment characteristics to set autonomous evolution rules for hidden dangers and intervention response rules, thus completing the construction of the ternary dynamic coupled evolution model.

[0026] Furthermore, the specific steps for outputting the full-dimensional inference data of the target detection area are as follows: Step 1: Input high-quality fused feature data of the target detection area into the ternary dynamic coupling evolution model, set different environmental conditions such as rainfall, wind speed and soil moisture, and use dynamic coupling simulation algorithm to simulate and generate the hazard coupling evolution trajectory and risk transmission path under different extreme environmental conditions; Step 2: During the simulation, capture the coupled evolution trajectory of multiple types of hidden dangers, the direction, speed and range of risk transmission in real time, and output the risk peak, outbreak time, secondary disaster trigger probability and risk impact range of the hidden dangers under each working condition; Step 3: Integrate the hazard evolution data, environmental parameter change data, and intervention measure simulation data from the model simulation process, and organize them in a structured manner according to the spatiotemporal dimension, risk dimension, and evolution dimension to generate full-dimensional simulation data of the target detection area, including data tables, evolution curves, and risk heat maps.

[0027] Step 4: The hazard coupling evolution simulation module adapts differentiated intervention schemes from the preset basic intervention scheme set based on the risk level and evolution trend of the hazards.

[0028] Specifically, the steps for adapting to the real-time scenario of the target detection area and dynamically adjustable differentiated intervention schemes are as follows: extract the risk levels and evolution trends of three core hidden dangers in the target detection area: structural damage, flammable and explosive, and geological disasters; determine the risk levels and evolution trends; and adapt differentiated intervention schemes accordingly. Among them, the risk levels include warning level and danger level, and the evolution trends include rapid escalation and slow development. The real-time scene of the target detection area and the dynamically adjustable differentiated intervention schemes include: 1) If the risk level of structural damage hazards is dangerous and the evolution trend is rapid escalation, then generate a set of basic intervention schemes for steel plate support and deep grouting; The foundation intervention scheme set, which includes steel plate support + deep grouting, includes: 16mm thick steel plates were used to vertically support the walls in the collapse hazard area, with fixing bolts spaced 50cm apart. High-pressure deep grouting (grouting depth ≥5m, grouting pressure 2.5MPa) was used simultaneously to fill the cracks and voids in the wall and prevent the collapse from spreading. For dilapidated frame structures, steel clamps and steel plates are used for beams and columns. Reinforcing steel plates are added at beam-column joints, and deep grouting (the grouting material is cement grout with a water-cement ratio of 1:1) is used to enhance the structural bearing capacity. Steel sheet piles (8m long, 1.2m spacing) were used to support the foundations of buildings near the slope, combined with deep grouting to reinforce the foundation and prevent the structure from collapsing due to slope displacement.

[0029] 2) If the risk level of structural damage hazards is at the warning level and the evolution trend is slow development, then generate a set of basic intervention schemes for component reinforcement and shallow grouting; The foundation intervention scheme set, which includes component reinforcement and shallow grouting, includes: For cracks in building walls, carbon fiber cloth is used for reinforcement (cloth width 30cm, number of layers 2), and shallow grouting is carried out simultaneously (grouting depth ≤2m, grouting pressure 1.0MPa) to seal surface cracks and slow down the development of damage; Honeycomb and pitted areas on the surface of concrete components are removed and repaired, reinforced with epoxy mortar, and filled with shallow grouting to prevent rainwater penetration and further damage. Rust removal treatment is carried out on the rusted areas of the steel structure components, anti-corrosion coating is applied, weak parts are locally reinforced with steel plates, and shallow grouting is used to seal the joints of the components.

[0030] 3) If the risk level of flammable and explosive hazards is dangerous and the evolution trend is rapid escalation, then a basic set of intervention plans for emergency sealing, gas dilution and personnel evacuation will be generated; The basic intervention plan set, which includes emergency containment, gas dilution, and personnel evacuation, includes: Emergency sealing of gas pipeline leaks was carried out using explosion-proof sealing agents, and explosion-proof fans were activated to force ventilation and dilute the leak area (ventilation volume ≥ 1000m³). 3 / h), demarcate a 50m warning zone, and organize on-site personnel to evacuate in an orderly manner along the safety passage; For leaks in hazardous chemical storage tanks, airbags are used to seal the leaks, neutralizing agents are sprayed to dilute the leaked medium, emergency sprinkler systems are activated to reduce the temperature in the area, personnel are assigned to guide evacuation to safe assembly points, and surrounding ignition sources are simultaneously cut off. For the damaged oil and gas pipeline, use quick clamps to seal the leak, turn on inert gas (nitrogen) to dilute the leaked oil and gas, set up warning signs, organize rescue and evacuate unrelated personnel to prevent explosion and poisoning accidents.

[0031] 4) If the risk level of flammable and explosive hazards is at the warning level and the evolution trend is slow development, then generate a set of basic intervention schemes for precise sealing and real-time monitoring of the surrounding environment; The basic intervention scheme set, which combines precise containment with real-time monitoring of the surrounding environment, includes: Special sealant is used to precisely seal tiny leaks at gas pipeline interfaces. Three gas concentration sensors (monitoring range 0-100% LEL) are installed around the leak point to monitor gas concentration changes in real time and record data every 10 minutes. To address the potential aging of seals on hazardous chemical containers, replace them with specialized sealing gaskets for precise sealing. Install dual sensors for temperature, humidity, and gas concentration around the containers to monitor environmental parameters in real time and issue an immediate warning if any exceedances are detected. For minor leaks in oil and gas pipeline valves, valve-specific sealant is used to seal them, and pipeline pressure sensors and gas monitoring terminals are installed to upload monitoring data in real time, enabling dynamic management and control of potential hazards.

[0032] 5) If the risk level of geological disaster hazards is dangerous and the evolution trend is rapid escalation, then generate a set of basic intervention schemes for deep grouting reinforcement and slope anti-slide pile installation; The foundation intervention scheme set, which includes deep grouting reinforcement and slope anti-slide pile installation, includes: The landslide body was reinforced by deep grouting (grouting depth ≥ 10m, grouting material is cement mortar, grouting pressure 3.0MPa), and anti-slide piles (pile diameter 1.2m, pile length 15m, spacing 4m) were installed simultaneously to curb the sliding of the landslide body; For areas with potential ground subsidence, deep grouting is used to fill underground cavities (grouting depth 8-12m), and anti-slide piles are installed to stabilize the surrounding soil and prevent the subsidence area from expanding. To address potential slope instability, deep grouting is used to reinforce the internal soil and rock mass of the slope, combined with anti-slide piles and anchor cables for support, thereby enhancing the slope's anti-slide capacity and preventing the transmission of landslide risks.

[0033] 6) If the risk level of geological disaster hazards is at the early warning level and the evolution trend is slow development, then generate a set of basic intervention schemes for shallow grouting, slope protection and real-time monitoring of soil displacement; The basic intervention scheme set, which includes shallow grouting, slope protection, and real-time soil displacement monitoring, includes: Shallow grouting was carried out on the slope surface (grouting depth ≤ 3m, grouting pressure 1.5MPa), geogrid was laid for slope protection, and three soil displacement sensors were installed at the top of the slope to monitor displacement changes in real time. For areas with potential small landslide hazards, shallow grouting is used to reinforce the surface soil, slope protection mortar is sprayed for slope protection, and tilt sensors are installed to monitor changes in slope tilt in real time and capture the evolution trend of potential hazards in a timely manner. For areas with potential ground subsidence risks, shallow grouting is used to fill surface voids, anti-slip netting is laid for slope protection, and subsidence monitoring points are set up to collect subsidence data every 15 minutes to achieve dynamic monitoring of potential risks.

[0034] P130 Emergency Decision-Making and Visualization Output Module: Based on the full-dimensional simulation data and differentiated intervention plans output by the P120 Hazard Coupling Evolution Simulation Module, combined with the fusion feature data, the intervention plans are screened and optimized to generate the minimum intervention decision results. At the same time, the fusion results, simulation results and decision results are presented in a visual form. The specific steps for screening and optimizing intervention plans to generate the minimum intervention decision result are as follows: Step 1: Based on the generated differentiated intervention schemes, input the schemes into the ternary dynamic coupling evolution model to simulate the intervention effects under different extreme environmental conditions, and obtain the risk reduction rate, resource utilization rate, probability of secondary risk induction, and coupling hazard blocking rate of each scheme. Step 2: With risk control compliance as a hard constraint and minimizing resource input as the optimization goal, the solution that simultaneously meets the requirements of risk control effectiveness, minimizes rescue resource consumption, and optimizes adaptability to extreme environmental conditions, and is at the Pareto optimal frontier in the three dimensions of risk control, resource input, and condition adaptability, is taken as the minimum intervention decision alternative. Among them, the risk control effectiveness meets the standards: after intervention, the risk level of the corresponding structural damage, flammable and explosive, and geological disaster hazards can be reduced to the safe level, and the coupling correlation between hazards can be reduced to weak correlation or below; the minimum resource consumption means that the required personnel, equipment, and materials do not exceed the on-site reserves and the input is the lowest, and the resource utilization rate is the best; the optimal adaptability to working conditions means that, through simulation verification by the ternary dynamic coupling evolution model, under extreme environmental working conditions, the risk control effectiveness decay rate is no more than 10%, the resource consumption increment is no more than 5%, and there is no risk induced by secondary hazards; Specifically, within a unified 3D geographic information scene, a high-precision electronic map and terrain model of the target detection area are overlaid and displayed. The location, type, and risk level of potential hazards obtained from the fusion of multi-source data are intuitively marked with markers of different colors and sizes. The results of the coupled evolution of potential hazards are displayed in the form of dynamic time-series curves, risk heat maps, and risk transmission path animations, presenting the development trend of potential hazards, the scope of risk impact, and the probability of secondary disasters in real time. The final generated minimum intervention decision plan is output in the form of a text list, a handling flowchart, a zonal handling strategy map, and a resource allocation distribution map. It also supports interactive operations such as querying fused data, replaying the simulation process, and exporting decision plans.

[0035] This invention achieves comprehensive collection of multi-dimensional hidden danger data in extreme environments by deploying an amphibious intelligent detection terminal. It adopts a three-level progressive fusion architecture of data level, feature level, and semantic level, combined with an emergency hidden danger domain knowledge graph, to achieve deep fusion of multi-source heterogeneous data and extract the core features and correlations of hidden dangers. It constructs a three-element dynamic coupling evolution model of hidden danger, environment, and intervention, and combines historical disaster cases to deduce the evolution trajectory of hidden dangers and risk transmission paths, and generates differentiated intervention plans. With risk control compliance as a hard constraint and resource input as the goal, it generates minimum intervention decision results through simulation screening and visualization, realizing accurate perception, intelligent fusion, and scientific decision-making of hidden dangers in extreme environments, optimizing the allocation of rescue resources, and improving the safety and response efficiency of emergency rescue in extreme environments.

[0036] The above-disclosed embodiments are merely a few specific examples of the present invention. However, the present invention is not limited thereto, and any variations that can be conceived by those skilled in the art should fall within the protection scope of the present invention.

Claims

1. A multi-source data fusion processing system for emergency response hazards in extreme environments, characterized in that, include: Multi-source heterogeneous data acquisition module: Based on the deployment of an air-ground-water amphibious intelligent detection terminal, multi-source data is acquired; The multi-source data includes: original data on multi-dimensional potential hazards in extreme environments within the target detection area, real-time environmental parameters, and on-site rescue resource data; Multi-source data fusion module: Adopting a three-level progressive fusion architecture, it combines knowledge graphs in the field of emergency hazards with geological and environmental characteristics of the target detection area to deeply fuse the collected data and extract the core features of hazards, the correlation between hazards, and the semantic association between hazards and the environment; Hazard Coupling Evolution Simulation Module: Based on the ternary dynamic coupling evolution model, combined with historical disaster coupling cases, it simulates the coupling evolution trajectory of hazards and risk transmission paths, and outputs full-dimensional simulation data and differentiated intervention plans; Emergency Decision-Making and Visualization Output Module: Used to combine fused feature data to screen and optimize intervention plans, and generate minimum intervention decision results, fused results, extrapolation results and decision results.

2. The multi-source data fusion processing system for extreme environment emergency hazards according to claim 1, characterized in that, Before acquiring multi-source heterogeneous data, the multi-source heterogeneous data acquisition module divides the area into high-altitude detection sub-regions, ground detection sub-regions, and underwater detection sub-regions, and deploys drone clusters, ground mobile robot formations, and underwater robots.

3. The multi-source data fusion processing system for extreme environment emergency hazards according to claim 1, characterized in that, The multi-dimensional raw data on potential hazards refers to: the morphology, location, and physical characteristics of potential hazards within the target detection area; the real-time environmental parameters refer to: rainfall, wind speed, soil moisture, and water temperature; and the on-site rescue resource data refers to: the number of rescue personnel, the type of equipment, and the reserves of supplies.

4. The multi-source data fusion processing system for extreme environment emergency hazards according to claim 1, characterized in that, The multi-source data fusion module adopts a three-level progressive fusion architecture of data level, feature level, and semantic level. The data level refers to the registration, splicing, and redundancy removal of raw data from similar sensors. The feature level refers to the extraction of core features of various types of data and the generation of a unified hazard feature vector through attention-weighted fusion. The semantic level refers to the mapping of the feature vector to the emergency hazard semantic space to complete the semantic alignment of different modal data.

5. The multi-source data fusion processing system for extreme environment emergency hazards according to claim 1, characterized in that, The multi-source data fusion module performs preprocessing on the knowledge graph of the emergency hazard domain, extracts the knowledge subgraph corresponding to the target detection area and removes irrelevant nodes, and filters and optimizes the initial feature vector in combination with geological environment characteristics, retains the core features associated with high-risk hazards, and generates a unified hazard feature vector adapted to the characteristics of the target detection area.

6. The multi-source data fusion processing system for extreme environment emergency hazards according to claim 1, characterized in that, The multi-source data fusion module extracts the following relationships among potential hazards: induced relationships, coupled relationships, and symbiotic relationships. The extraction of core features of potential hazards, relationships among potential hazards, and semantic relationships between potential hazards and the environment refers to: extracting the distinctive core features of each potential hazard, combining the coupling relationships of potential hazards, analyzing and identifying the coupling, induced, and symbiotic relationships among different potential hazards, and conducting multi-dimensional correlation analysis between the core feature set of potential hazards and the relationship network among potential hazards and the regional geological and environmental characteristics to form a semantic set of relationships between potential hazards and the environment in the target detection area.

7. The multi-source data fusion processing system for extreme environment emergency hazards according to claim 6, characterized in that, The correlation between hidden dangers is identified by extracting key features of hidden dangers such as structural damage, flammable and explosive materials, and geological disasters. Combined with the hidden danger coupling correlation of the knowledge subgraph of the specific emergency hidden danger in the target detection area, the spatiotemporal correlation analysis identifies the induced correlation, coupling correlation and symbiotic correlation between different hidden dangers.

8. The multi-source data fusion processing system for extreme environment emergency hazards according to claim 1, characterized in that, The ternary dynamic coupling evolution model consists of an input layer, a coupling evolution layer, and an output deduction layer. After initial parameter assignment, it is initially calibrated based on field measured data to control simulation deviations. The model parameters are dynamically adjusted through a real-time fusion data feedback calibration mechanism.

9. The multi-source data fusion processing system for extreme environment emergency hazards according to claim 1, characterized in that, The hidden danger coupling evolution simulation module adapts differentiated intervention schemes from a preset set of basic intervention schemes based on the risk level and evolution trend of the hidden dangers. The basic intervention scheme set includes intervention schemes for structural damage hazards, flammable and explosive hazards, and geological hazards under the conditions of rapid escalation of danger level and slow development of warning level, respectively. The aforementioned simulation of the hazard coupling evolution trajectory and risk transmission path refers to: inputting high-quality fused feature data into a ternary dynamic coupling evolution model, setting different environmental operating parameters, and using a dynamic coupling simulation algorithm to simulate and generate the hazard coupling evolution trajectory and risk transmission path under different extreme environmental operating conditions.

10. The multi-source data fusion processing system for extreme environment emergency hazards according to claim 1, characterized in that, The emergency decision-making and visualization output module takes risk control compliance as a hard constraint and minimizing resource input as an optimization goal. It uses a ternary dynamic coupling evolution model to simulate the intervention effects under different working conditions, screens and optimizes various solutions, and generates the minimum intervention decision result.

Images