Yellow River upstream main stream reservoir dam group section multi-disaster big data early warning system and method

Abstract

The invention relates to a Yellow River upstream main stream reservoir dam group section multi-disaster big data early warning system and method. A geological disaster monitoring module obtains surface displacement and deformation data, low-altitude remote sensing data and high-resolution remote sensing image data in real time; the data processing and storage module performs cleaning, space-time alignment and semantic mapping on the multi-dimensional monitoring data based on an ETL technology to form a standardized data set, and stores the standardized data set in a distributed database; the multi-disaster early warning calculation module triggers a specific disaster model according to spatial-temporal characteristics of input data, and performs single-disaster risk probability calculation and cross-disaster risk superposition probability analysis; the early warning issuing module can convert the calculation result into early warning information and send the early warning information to an early warning department in real time; and the real scene display module can convert a disaster risk early warning result into a visual three-dimensional scene. According to the invention, the early warning timeliness of the sudden disaster chain can be effectively improved, and the whole-process intelligent management from data acquisition to emergency response and the accurate prediction of the disaster chain are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of multi-disaster monitoring for the Yellow River water conservancy, and particularly relates to a multi-disaster big data early warning system and method for the main stream reservoir-dam group section in the upper reaches of the Yellow River. Background Art

[0002] As a key barrier for flood control safety in the Yellow River Basin, the stability of the main stream reservoir-dam group in the upper reaches of the Yellow River is directly related to regional ecological security and people's livelihood. However, this area is located on the eastern edge of the Qinghai-Tibet Plateau, with strong geological tectonic activities, complex geomorphic conditions, frequent occurrence of chain disasters such as earthquakes, collapses, landslides, debris flows, and disasters such as flash floods induced by short-term heavy rainfall. The multi-disaster coupling effect and the evolution characteristics of the disaster chain are significant.

[0003] Although monitoring and early warning systems have been established in many places for flash floods, geological disasters, etc. at present, the existing early warning system research still has the following technical bottlenecks. For example, at the data level, in recent years, provincial flash flood disaster data synchronization and sharing systems, geological disaster monitoring and early warning systems based on WebGIS, CORS, Internet of Things, etc., and loess landslide monitoring and early warning systems based on universal monitoring equipment, etc., although they can realize the unified management and dynamic update of geological disaster point information from different sources and batches, they adopt heterogeneous data standards, resulting in difficult fusion of spatio-temporal resolution difference data such as hydrological stations, InSAR deformation, GNSS monitoring, etc., and the utilization rate of multi-source data is less than 37%. The spatio-temporal resolution and format differences limit the comprehensive analysis ability; in terms of model construction, for example, the disaster early warning cloud computing platform designed by Li Lan, which mines the collected monitoring data through the disaster event trigger inspection process and uses clustering algorithms and MapReduce parallel computing frameworks to improve the timeliness of disaster early warning, but the types of monitored disasters of this platform are not clear and lack pertinence; Wu Runze et al. took the Three Gorges Reservoir Area as an example and developed a geological disaster trend prediction and early warning system, which integrated the commonly used geological disaster susceptibility evaluation models at present, and selected a suitable model by comparing the model training accuracy, but this platform focuses on the analysis of the occurrence trend of geological disasters and the early warning function is relatively single.

[0004] Generally speaking, the existing research still faces the following bottlenecks: data fragmentation, lack of standardized integration of multi-source heterogeneous data (remote sensing, meteorology, hydrology, geology), and spatio-temporal resolution and format differences limit the comprehensive analysis ability; model isolation, there are mostly single-disaster early warning models, few cross-disaster coupling risk prediction models, and the construction of the model library lags behind, making it difficult to support dynamic collaborative computing; response lag, the traditional threshold-driven early warning mechanism relies on historical data and has weak real-time deduction and active defense capabilities for sudden disaster chains.

[0005] Based on this, it is necessary to study a multi-disaster big data early warning system and method for the main stream reservoir-dam group section in the upper reaches of the Yellow River. Summary of the Invention

[0006] In view of this, the purpose of the present invention is to provide a multi-disaster big data early warning system and method for the main stream reservoir-dam group section in the upper reaches of the Yellow River, which effectively solves the problems of existing data fragmentation, model isolation, and response lag.

[0007] To achieve the above purpose, the technical solution adopted by the present invention is: a multi-disaster big data early warning system for the main stream reservoir-dam group section in the upper reaches of the Yellow River, including a geological disaster monitoring module, a data processing and storage module, a multi-disaster early warning calculation module, an early warning release module, a real-scene display module, and a communication module for data transmission between modules; the geological disaster monitoring module includes GNSS stations, low-altitude remote sensing monitoring by unmanned aerial vehicles, and high-resolution satellite series monitoring. The GNSS stations are used to obtain surface displacement and deformation data in real time. The low-altitude remote sensing monitoring by unmanned aerial vehicles is used to obtain low-altitude monitoring data. The high-resolution satellite series monitoring is used to obtain high-resolution remote sensing image data, and the above multi-dimensional monitoring data is transmitted to the data processing and storage module in real time through the communication module; the data processing and storage module performs multi-source heterogeneous cleaning on the above multi-dimensional monitoring data based on ETL technology, and then performs spatio-temporal alignment and semantic mapping to form a standardized data set, and stores the standardized data set in a distributed database to provide dynamic data input for the multi-disaster early warning calculation module; the multi-disaster early warning calculation module triggers a specific disaster model according to the spatio-temporal characteristics of the input data, calculates the single-disaster risk probability, and realizes the cross-disaster risk superposition probability analysis based on the complex network deduction algorithm in the model, generating a multi-scenario early warning plan and the prediction result of the disaster chain evolution; the early warning release module is used to receive the calculation result of the multi-disaster early warning calculation module, convert the calculation result into an early warning message and send it to the early warning department in real time to realize cross-departmental data sharing and emergency order linkage; the real-scene display module receives the model output result of the multi-disaster early warning calculation module in real time, and converts the disaster risk early warning result into an intuitive three-dimensional scene to realize the visual deduction and quantitative analysis of the disaster process.

[0008] Furthermore, the geological disaster monitoring module is based on the sky-air-ground multi-dimensional perception technology to collect the monitoring data of landslides, collapses, debris flows, and mountain floods. The monitoring data covers the disaster basic data, remote sensing data, regional disaster data, and disaster point data of the main stream reservoir-dam group section in the upper reaches of the Yellow River.

[0009] Furthermore, the basic data includes basic topographic data and basic geographic data. The remote sensing data includes remote sensing image data of high-resolution satellite series; the regional disaster data includes meteorological, hydrological, and seismic data; the disaster point data includes the location, size, and development status of geological disaster points such as landslides, collapses, and debris flows.

[0010] Further, based on the data transmitted by the geological disaster monitoring module, the data processing and storage module constructs a multi-hazard risk early warning multi-source database by sorting out the monitoring data sources, storage methods, structures, spatio-temporal sequence relationships, and attribute logical relationships of earthquakes, landslides, debris flows, rainstorms, and mountain floods in the reservoir-dam group section of the upper reaches of the Yellow River, with the themes of disaster risk assessment, regional disaster early warning, single-disaster early warning, and disaster situation assessment.

[0011] Further, the multi-hazard risk early warning multi-source database adopts a distributed database environment, integrating spatial data management, image data management, disaster data management, and comprehensive analysis functions; the multi-hazard risk early warning multi-source database is configured with unified data and interface standards, and uses data warehouse technology to perform ETL operations on the data for realizing automatic real-time extraction and conversion of the data.

[0012] Further, the multi-hazard early warning calculation module integrates the risk early warning probability models for mountain floods, earthquakes, and landslides and debris flows, and realizes the early warning of mountain floods, earthquakes, and landslides and debris flow disasters and the cross-disaster risk coupling analysis by integrating the adaptive particle filter algorithm, adversarial neural network, and discrete element-finite element coupled numerical simulation technology.

[0013] Further, based on the risk early warning probability model for mountain floods, it uses the adaptive particle filter algorithm to dynamically assimilate the multi-source data of rainstorm mountain floods, inversely calculates the roughness coefficient parameters, water levels, and flows in the flood evolution in real time, and combines the multi-scenario ensemble simulation to evaluate the risks of single and overall reservoir-dam groups; based on the risk early warning probability model for earthquakes, it uses the adversarial neural network to generate ground motion time history signals that conform to the characteristics of the upper reaches of the Yellow River, constructs a non-linear mapping relationship between magnitude, intensity, and dam body damage by mining historical earthquake damage data, and uses the reliability theory to analyze the failure thresholds of different dam types to realize the efficient calculation of the reliability of the seismic response of single dams; based on the risk early warning probability model for landslides and debris flows, it establishes a fluid-solid coupling constitutive model for soil and debris, combines the discrete element method to simulate particle motion and the finite element method to analyze the stress and strain of rock and soil masses, and forms an erosion parameter and collapse mechanical model under the interaction of water flow and rock and soil for dynamically predicting the evolution path of the landslide and debris flow disaster chain.

[0014] Further, the early warning release module is used to convert the calculation results of the multi-hazard risk early warning model into standardized and operable early warning instructions. Based on the risk probability, disaster chain evolution path, and three-dimensional deduction results output by the corresponding model, the early warning release module automatically generates multi-level early warning information using a rule engine, and realizes the multi-channel real-time push of early warning information through an information gateway and an information sharing interface, and generates a structured disaster report to realize cross-departmental instruction linkage and resource optimization allocation.

[0015] Furthermore, the real-scene display module combines the Cesium visualization technology framework with the UE5 rendering engine, and uses the virtual simulation physics engine to simulate the disaster process in combination with the model calculation results of the multi-disaster warning calculation module, completing the high-fidelity three-dimensional dynamic restoration of the disaster process.

[0016] The present invention also provides a multi-disaster big data warning method for the main stream reservoir-dam group section in the upper reaches of the Yellow River, including the following steps: Step 1: Data collection Based on the space-air-ground integrated monitoring network, multi-dimensional and multi-source heterogeneous data are synchronously collected by using GNSS stations, low-altitude remote sensing emergency monitoring by unmanned aerial vehicles, and high-resolution satellite series monitoring, and the multi-source heterogeneous data are transmitted to the data processing and storage module in real time through the communication module; Step 2: Data processing After the data processing and storage module uses ETL technology to clean, transform, and perform spatio-temporal alignment operations on the multi-source heterogeneous data, a standardized data set is formed and stored in a distributed database to ensure that the data can be called in real time; Step 3: Multi-disaster risk calculation The multi-disaster warning calculation module triggers the corresponding model library according to the standardized data. Among them, the mountain flood model uses the adaptive particle filter algorithm to assimilate rainstorm data, inversely calculates the roughness coefficient parameter and the flood evolution process in real time, and evaluates the risks of individual and overall reservoir-dam groups; the earthquake model generates regional ground motion signals based on the adversarial neural network, establishes the magnitude-intensity-dam body damage mapping relationship in combination with historical earthquake damage data, and analyzes the dam type failure threshold through the reliability theory; the landslide and debris flow model integrates the discrete element method and the finite element method to simulate the erosion parameters and collapse mechanism under the coupling action of soil-water-rock; in addition, the multi-disaster warning calculation module can also integrate the single-disaster results based on the complex network deduction algorithm, calculate the cross-disaster risk superposition probability, and thus generate multi-scenario warning plans and disaster chain evolution paths; Step 4: Warning information generation and release The warning release module converts the risk probability and disaster chain path output by the above multi-disaster warning calculation module into multi-level warning instructions through the rule engine, and pushes them to the warning department in seconds using the short message gateway and the government cloud API, and automatically generates a structured disaster report; Step 5: Three-dimensional real-scene deduction The real-scene display module, based on the Cesium framework and the UE5 engine, receives the model output results of the above multi-disaster warning calculation module in real time, synchronizes the deduction results, and realizes the high-fidelity three-dimensional dynamic simulation of the disaster process.

[0017] The beneficial effects of the above technical solution are as follows: The multi-disaster big data early warning system and method for the main stream reservoir dam group section in the upper reaches of the Yellow River provided by the present invention, through the sky-air-ground multi-dimensional perception network of the geological disaster monitoring module, realizes the real-time collection and dynamic tracking of multi-source heterogeneous data such as surface displacement, deformation, meteorology, and hydrology, effectively improves the data utilization rate, and realizes the real-time monitoring and data visualization of disasters such as landslides, collapses, and debris flows; the early warning model calculation breaks through the limitation of single disasters, and constructs a risk model library for earthquakes, landslides, debris flows, mountain floods, and multi-disaster coupling; the three-dimensional visualization integrates the Cesium and UE5 engines, and highly accurately simulates the flood evolution and the process of landslide and debris flow disasters, and at the same time provides spatial interaction tools such as inundation analysis and profile measurement.

[0018] The present invention constructs a complete closed loop of "monitoring - analysis - early warning - decision-making", realizing the full-process intelligent management from data collection to emergency response; through the sky-air-ground integrated monitoring network and ETL technology, effectively improving the utilization rate of multi-source heterogeneous data and reducing the spatio-temporal alignment error; integrating a variety of intelligent algorithm models, realizing the coupling analysis of multi-disasters such as earthquakes, mountain floods, and landslides and debris flows, and greatly improving the early warning accuracy; adopting a five-layer cloud service architecture and Kubernetes intelligent scheduling, effectively shortening the early warning response time of complex disaster chains; integrating the Cesium and UE5 engines, dynamically simulating the disaster process, and greatly enhancing the intuitiveness of decision-making; through the intelligent rule engine and standardized interfaces, realizing the release of early warning information within seconds, effectively solving the problem of response lag; and the system of the present invention adopts a modular design and standardized interfaces, which is also convenient for function expansion and system upgrade, so as to adapt to the disaster prevention needs of different basins.

[0019] When the early warning system is actually applied, it can be linked with the emergency command and decision-making one-map of the China Earthquake Emergency Search and Rescue Center, effectively improving the timeliness of early warning for sudden disaster chains, enhancing the intuitiveness of emergency decision-making through three-dimensional real-scene deduction, optimizing resource scheduling and response efficiency, and providing strong support for intelligent disaster prevention and mitigation in the main stream reservoir dam group section of the upper reaches of the Yellow River. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a schematic structural diagram of the early warning system of the present invention; Figure 2 It is a distribution map of the basic geographical information of the disaster points collected by the geological disaster monitoring module; Figure 3 It is a schematic structural diagram of the composition of the model library in the multi-disaster early warning calculation module; Figure 4 It is a schematic diagram of the model library architecture in the multi-disaster early warning calculation module; Figure 5 It is a schematic diagram of the three-dimensional data fusion effect in the real-scene display module; Figure 6 It is a deduction diagram of the simulation effect of landslides and debris flows by the real-scene display module; Figure 7 For the partial simulation and deduction diagram in the real-scene display module; Figure 8 It is a schematic flowchart of the early warning method of the present invention. Specific implementation manners

[0021] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific implementation manners: Embodiment 1. This embodiment aims to provide a multi-disaster big data early warning system for the main stream reservoir-dam group section in the upper reaches of the Yellow River, mainly used to realize the full-chain dynamic early warning of multi-disaster risks and emergency decision-making support. Aiming at the problems of data fragmentation, model isolation, and response lag in the existing monitoring and early warning systems, this embodiment provides a multi-disaster big data early warning system for the main stream reservoir-dam group section in the upper reaches of the Yellow River, as Figure 1 shown. This system covers functions in multiple aspects such as dynamic monitoring of geological disasters, calculation of multi-disaster early warning models, early warning release, and real-scene three-dimensional model deduction. It integrates and combines multi-source heterogeneous data such as remote sensing, ground monitoring, meteorology, and hydrology, constructs a sky-air-ground multi-dimensional data baseboard, and realizes dynamic monitoring of disasters, risk early warning, and emergency decision-making support through the collaborative work of each module, so as to provide an intelligent solution for multi-disaster risk identification, early warning release, and emergency command in the main stream reservoir-dam group section in the upper reaches of the Yellow River.

[0022] As Figure 1-7 shown, the multi-disaster big data early warning system for the main stream reservoir-dam group section in the upper reaches of the Yellow River provided in this embodiment includes a geological disaster monitoring module, a data processing and storage module, a multi-disaster early warning calculation module, an early warning release module, a real-scene display module, and a communication module for data transmission between each module; Specifically, as Figure 2 shown, the geological disaster monitoring module is used to collect multi-dimensional monitoring data of disasters such as landslides, collapses, debris flows, and mountain floods in real time, so as to provide high-precision and high-timeliness data support for the probability calculation of each early warning model. In terms of ground monitoring, it includes GNSS station monitors deployed at the landslide bodies of the Maruo Hydropower Station, Laxiwa Hydropower Station, and Longyangxia Hydropower Station to obtain dynamic parameters such as landslide displacement and deformation; in terms of aerial monitoring, emergency monitoring and image acquisition of disaster points are realized through low-altitude aerial photography by drones; in addition, this module also includes high-altitude data monitoring, integrating high-resolution satellite series (Gaofen 1, 2, 4, 6) and Sentinel-2 remote sensing images (update cycle ≤ 12 days) to provide large-scale surface deformation and disaster feature data.

[0023] The above geological disaster monitoring module is based on the multi-dimensional perspective of space-air-ground, forming a space-air-ground multi-dimensional perception network monitoring architecture. The data collected covers the disaster basic data, remote sensing data, regional disaster data, and disaster point data of the main stream reservoir dam group section in the upper reaches of the Yellow River. Among them, the basic data includes basic topographic data and basic geographical data, and the remote sensing data includes remote sensing image data collected by Gaofen-1, Gaofen-2, Gaofen-4, Gaofen-6, and Sentinel-2. The disaster point data includes the location, size, development status, etc. of geological disaster points such as landslides, collapses, and debris flows. Thus, it provides multi-dimensional disaster image monitoring data with full-domain coverage, and transmits the collected multi-dimensional monitoring data to the data processing and storage module in real time through the communication module.

[0024] To ensure the efficient utilization of data, in the present invention, the data processing and storage module first performs ETL (extraction, transformation, loading) operations on the received monitoring data using data warehouse technology, realizes the automatic extraction and transformation of data, thus forming a standardized data set, and stores the standardized data set in a distributed database to ensure that the data can be called in real time to provide a real-time data source for subsequent calculations.

[0025] Specifically, the process of realizing the full-process standardization of multi-source heterogeneous data by this module is as follows: First, extract the original data from space-air-ground monitoring devices such as GNSS stations, UAV aerial photography, and high-resolution satellites. Here, the original data includes displacement sequences, remote sensing images, meteorological hydrology, etc. For the differences in spatio-temporal resolution, such as the second-level of GNSS data and the day-level of satellite images, the sliding window interpolation and spatio-temporal gridding methods are used to align the data timestamps and spatial coordinates. Then, through the semantic mapping rule library, such as the geological disaster ontology model, fields from different sources, such as displacement and deformation rate, are uniformly mapped to the standardized data model to eliminate semantic ambiguity. Then, the distributed computing framework (Spark) is used to perform parallel cleaning on the massive data, removing outliers and noise (such as GNSS signal drift, remote sensing cloud interference), and at the same time dynamically repairing data missing based on the rule engine (such as rainfall data interpolation). The processed standardized data set is stored in the database by theme stratification. In addition, this module can also be linked with the multi-hazard early warning model calculation module in real time through the Kafka message queue, dynamically pushing standardized data streams, such as displacement mutation alarms, rainfall intensity threshold breakthrough events, etc., to support the model library to call data slices on demand, thus forming an integrated process of "data preprocessing-intelligent quality inspection-efficient supply", effectively improving the utilization rate of multi-source data and providing high-quality input for cross-hazard coupling analysis.

[0026] Furthermore, the module systematically sorts out the input multi-source heterogeneous data, and then constructs a multi-hazard risk warning multi-source database with themes such as disaster risk assessment, regional disaster warning, single-disaster warning, and disaster situation assessment according to the data sources, storage methods, structures, spatio-temporal sequence relationships, and attribute logical relationships of earthquake, landslide, debris flow, rainstorm, and mountain flood monitoring data in the target area. The multi-hazard risk warning multi-source database adopts a distributed database environment, integrating functions such as spatial data management, image data management, disaster data management, and comprehensive analysis. The construction of this database can become a data aggregator for insights into disasters in the main stream reservoir-dam group section of the upper Yellow River from multiple dimensions of multiple hazards, various historical periods, and various spatial positions, forming an open, active, continuously calibrated and updated data element asset, thereby providing data support for the entire warning system.

[0027] The multi-hazard warning calculation module triggers a specific disaster model according to the spatio-temporal characteristics of the data transmitted from the above data processing and storage module. The multi-hazard warning calculation module is equipped with a multi-hazard warning model, and a model library as shown in Figure 3 is constructed. Based on the received environmental parameter information about different regions, this model can calculate the earthquake risk warning probability, landslide and debris flow risk warning probability, mountain flood disaster risk warning probability, and multi-hazard coupling risk probability.

[0028] In terms of mountain flood disaster risk warning, the module integrates a mountain flood disaster risk warning probability model. The adaptive particle filter algorithm is used to assimilate the rainstorm and mountain flood data to achieve real-time inversion and correction of roughness parameters, water levels, and flows during the flood evolution process. The antecedent precipitation, rainfall amount per event, rainfall pattern, and underlying surface conditions are used as multi-source input conditions to evaluate the risks of single reservoir dams and reservoir-dam groups under different mountain flood disaster conditions, and realize mountain flood disaster risk warning based on multi-scenario ensemble simulation.

[0029] In terms of earthquake risk warning, an earthquake risk warning probability model is integrated. By collecting historical earthquake monitoring data in the upper Yellow River and its surrounding areas, the characteristics of earthquake acceleration time history in the upper Yellow River and its surrounding areas are extracted, and an earthquake ground motion signal that conforms to the regional characteristics is generated using an adversarial neural network. Data mining and analysis are carried out on the domestic and foreign dam damage datasets, a non-linear mapping relationship between magnitude, intensity, and dam damage is established, and a dam damage risk assessment model is constructed. The reliability theory is used to analyze the key control indicators of the failure mechanisms of different dam types, and the reliability of the seismic response of a single dam is efficiently calculated to achieve earthquake risk warning.

[0030] In terms of landslide risk warning, a landslide risk warning probability model is integrated. By collecting field data of landslide tests, a soil debris constitutive model considering fluid-solid coupling is established. The influence of soil debris particle size characteristics and physical and mechanical properties on erosion parameters such as erosion velocity, erosion index and erosion shear stress is analyzed to form a mechanical mechanism model of rock and soil collapse and destruction under the action of water flow. Using numerical simulation techniques such as discrete element method and finite element method, the simulation of landslide disasters considering the rock-soil-water interaction mechanism is realized, which is used to dynamically predict the evolution path of the landslide disaster chain.

[0031] In addition, this module can also integrate the above-mentioned single-disaster models and introduce complex network deduction theory through the distributed model library dynamic scheduling algorithm to realize the calculation of the probability of cross-disaster risk superposition, thereby supporting the comparison of multiple plans and minute-level deduction of disaster chains, generating multi-scenario warning plans and disaster chain evolution prediction results, which can greatly improve the warning accuracy compared with the traditional single model.

[0032] Furthermore, in order to meet the computing requirements of the model library in the multi-hazard early warning calculation module, this module has also developed a model interface and computing engine adapted to the cloud service architecture, and realized the interaction and task scheduling of the multi-hazard risk warning probability model through Web Service technology, supporting model selection, combination generation, and simulation calculation configuration, and ultimately realizing cloud computing and efficient data interaction of multiple types of models.

[0033] like Figure 4 As shown in the figure, the multi-disaster early warning computing module realizes efficient model collaboration based on the five-layer cloud computing platform architecture. At the request access layer, the gRPC protocol is used to open the standardized interfaces of deep learning frameworks such as TensorFlow and PyTorch, supporting users to call model instances online through IP / port. At the same time, a model service proxy layer is built to realize front-end and back-end interactive protocol conversion and load balancing to ensure efficient and compatible interfaces. At the model instance layer, it encapsulates disaster early warning models trained by different frameworks (such as mountain torrent particle filter algorithm and earthquake adversarial neural network) as Docker containerized services to realize independent model deployment, resource isolation and rapid migration, effectively improving resource utilization. The Kubernetes management layer optimizes the predictive scaling algorithm (based on LSTM network prediction of load trends), replaces the traditional threshold trigger mechanism, dynamically schedules CPU / GPU resources and monitors the container status in real time, that is, upgrades the traditional threshold-based responsive strategy to a prediction-based active strategy, shortening the resource allocation response time from minutes to seconds. The data layer relies on the distributed storage system to provide real-time data stream access and cross-model sharing, and caches high-frequency access data (such as real-time displacement series and rainfall intensity thresholds) through the memory database (Redis), reducing I / O latency to milliseconds.

[0034] Through hierarchical decoupling and intelligent scheduling, this architecture realizes parallel computing and dynamic coordination of multi-hazard models (mountain torrents, earthquakes, landslides and flows), supports real-time calculation of the probability of disaster chain risk superposition, and successfully responds to sudden landslide and landslide-mountain torrent coupling disaster events in the application of the reservoir and dam group section in the upper reaches of the Yellow River. The hour-level warning cycle of traditional single-model serial calculations is compressed to minutes. This architecture has high robustness and computing efficiency advantages in complex disaster scenarios, breaking through the bottlenecks of traditional single-machine computing power and response efficiency, thereby effectively solving the lag problem of resource scheduling and improving the dynamic response capability of the system.

[0035] The warning release module is used to receive the calculation results of the multi-hazard warning calculation module in real time through the communication module, convert the calculation results into warning information, and send it to the warning department in real time to realize cross-departmental data sharing and emergency command linkage. Based on the risk probability, disaster evolution path and three-dimensional deduction results output by the multi-hazard warning calculation module, this module uses the rule engine to automatically generate multi-level warning information, such as warning level, impact range and disaster avoidance suggestions, and realizes multi-channel real-time push of warning information through SMS gateway and government cloud platform API interface. At the same time, it integrates the responsible person management system, dynamically associates the emergency plan library, and generates a structured disaster report including disaster overview, risk map, disposal suggestions, etc. with one click, supporting online editing and multi-level review. Furthermore, it can also seamlessly connect with the existing "one map for emergency command decision-making" system through standardized data interfaces, such as RESTful API, to synchronize warning information, resource distribution and deduction scenarios in real time, realize cross-departmental command linkage and resource optimization configuration, ensure the closed-loop management of the whole process from warning trigger to emergency action, greatly reduce the manual review of corresponding events, and significantly improve the collaborative disposal capability of the sudden disaster chain.

[0036] The real-scene display module is based on the Cesium framework and the UE5 engine. It can receive the calculation results of the multi-disaster early warning calculation module in real time through the communication module. The real-scene 3D model display function based on this module can realize the visual virtual display of underwater terrain, 3D real scene, important annotations, and disaster process simulation. In the present invention, the real-scene display module uses the UAV oblique photography model to fuse DEM data to construct a 3D data scene in the target area. The fusion effect is as follows: Figure 5 As shown, 3ds Max can be used for preprocessing and optimization when building the real-life model, including position matching, multi-dimensional sub-material settings, and splicing of multiple models to ensure the structural integrity of the model and compatibility with subsequent visualization simulation loading.

[0037] Furthermore, through the virtual simulation physics engine, this module can combine the calculation results of the above multi-hazard early warning calculation module to simulate the disaster process, making the disaster outbreak process more real and intuitive, thus effectively improving the cognition and interaction ability of the collapse disaster scenario and meeting the requirements of disaster prevention and mitigation departments for efficiently obtaining information for decision-making.

[0038] In terms of flood simulation, this module uses the Fluid Flux plug-in of the UE5 rendering engine. Based on the shallow water equations and combined with the water source location, calculations are carried out in a 2D grid projected on the height field (capturing the ground). All terrain undulations are rendered in a top-down projection manner. Finally, a flow field map is generated with reference to the projected terrain and based on physical characteristics, and the 3D water flow effect is restored and simulated on the ground grid.

[0039] In terms of landslide and debris flow disaster simulation, the Chaos Physics engine of UE5 is used to accurately simulate the whole process of landslide and debris flow disasters based on rigid body dynamics, fracture systems, and physical field functions. As Figure 6 and 7 shown, first, the high-risk areas of disaster occurrence are identified through the landslide and debris flow disaster risk warning probability model in the model library. Then, the regional mountain body is cut into potential fragments through the Procedural Mesh Generation tool, and the physical properties (friction coefficient, density, etc.) of different rock layers are defined. Finally, the bond strength of rock masses is simulated through the constraint system, and the collapse is dynamically activated in combination with external triggering conditions (such as seismic waves, rainfall). Chaos Physics supports real-time calculations of large-scale fragment collisions and accumulations, and optimizes performance through LOD grading, asynchronous physics Tick, and GPU acceleration to ensure efficient simulation of high-density fragments. High-fidelity rendering of simulation results is achieved through Nanite and Lumen, while data such as fragment trajectories and impact forces are recorded and synchronized to the real-scene display module, thus providing dynamic data guarantee for the disaster deduction of the entire early warning system.

[0040] In addition, the communication module in the present invention is used for real-time data exchange and transmission, responsible for connecting all functional modules at high speed and reliably, ensuring seamless transfer of data and instructions in all links of monitoring, analysis, early warning, and decision-making, thereby effectively solving the efficiency and stability problems of cross-module transmission of multi-source data. In actual application, a hierarchical protocol architecture can be adopted to achieve efficient data transmission between modules. This module uses the Kafka message queue to build a high-throughput data pipeline, and real-time pushes GNSS displacement data, UAV images, etc. collected by the geological disaster monitoring module to the data processing module; encapsulates the model call interface based on the gRPC protocol to support the multi-hazard early warning calculation module to transmit risk instructions to the early warning release module at the second level; synchronizes the three-dimensional deduction stream for the real-scene display module through the WebSocket duplex channel, and integrates a protocol converter to automatically adapt heterogeneous formats such as satellite images and matrix data, enabling seamless connection of the entire process from monitoring data driving model calculation to early warning instructions triggering emergency response, and ensuring the reliability of data transmission.

[0041] The multi-hazard big data early warning system for the main stream reservoir-dam group section in the upper reaches of the Yellow River provided by the present invention breaks through the limitations of traditional single-hazard early warning in multi-hazard coupled modeling and dynamic collaborative calculation, integrates risk prediction algorithms for multiple hazards such as earthquakes, landslides, debris flows, and mountain floods, and supports the calculation and dynamic deduction of the cross-hazard risk superposition probability model. In terms of high-precision three-dimensional real-scene simulation and interactive deduction, it innovatively integrates the Cesium visualization framework and the UE5 physics engine to complete the high-fidelity three-dimensional dynamic restoration of the disaster process, supports spatial analysis and decision-making interaction, and intuitively empowers emergency command. It integrates and integrates multi-source heterogeneous data such as remote sensing, ground monitoring, meteorology, and hydrology, constructs a multi-dimensional data floor of sky-air-ground, realizes an open and calibrated data resource pool, and supports the dynamic assessment of multi-hazard risks. This system can be linked with the existing emergency command and decision-making single map, and rely on standardized interfaces to achieve seamless docking of cross-departmental early warning intelligent release and emergency command, effectively shortening the decision-making response time, thereby effectively promoting the intelligence and precision of the disaster prevention and mitigation system, and providing a technological innovation for the ecological protection and high-quality development of the Yellow River Basin.

[0042] Embodiment 2, based on Embodiment 1, this embodiment provides a multi-hazard big data early warning method for the main stream reservoir-dam group section in the upper reaches of the Yellow River, as Figure 8 shown, specifically including the following steps: Step 1: Data collection Based on the sky-air-ground integrated monitoring network, use GNSS stations, UAV low-altitude remote sensing emergency monitoring, and high-resolution satellite series monitoring to synchronously collect multi-dimensional multi-source heterogeneous data, and transmit the multi-source heterogeneous data to the data processing and storage module in real time through the communication module; Step 2: Data processing After the data processing and storage module uses ETL technology to clean, transform, and perform spatio-temporal alignment operations on multi-source heterogeneous data, a standardized data set is formed and stored in a distributed database to ensure that the data can be called in real time; Step 3: Multi-hazard risk calculation The multi-hazard early warning calculation module triggers the corresponding model library according to the standardized data. Among them, the mountain flood model uses the adaptive particle filter algorithm to assimilate rainstorm data, inversely calculate the roughness parameters and the flood evolution process in real time, and evaluate the risks of individual and overall reservoir-dam groups; the earthquake model generates regional ground motion signals based on the adversarial neural network, and combines historical earthquake damage data to establish a magnitude-intensity-dam body damage mapping relationship, and analyzes the failure threshold of the dam type through the reliability theory; the landslide and debris flow model integrates the discrete element method and the finite element method to simulate the erosion parameters and collapse mechanism under the coupled action of soil, water, and rock; in addition, the multi-hazard early warning calculation module can also integrate the single-hazard results based on the complex network deduction algorithm, calculate the cross-hazard risk superposition probability, and thus generate multi-scenario early warning plans and disaster chain evolution paths; Step 4: Early warning information generation and release The early warning release module uses a rule engine to convert the risk probability and disaster chain path output by the above multi-hazard early warning calculation module into multi-level early warning instructions, and pushes them to the early warning department in seconds using the SMS gateway and the government cloud API, and automatically generates a structured disaster report; Step 5: 3D real-scene deduction The real-scene display module is based on the Cesium framework and the UE5 engine, and receives the model output results of the above multi-hazard early warning calculation module in real time, synchronizes the deduction results, and realizes the high-fidelity 3D dynamic simulation of the disaster process.

[0043] The above-described embodiments of the present invention do not constitute a limitation on the protection scope of the present invention. The basic concept of the present invention is that the present invention constructs a complete closed loop of "monitoring - analysis - early warning - decision-making" to realize the full-process intelligent management from data collection to emergency response; integrates a variety of intelligent algorithm models to realize the coupled analysis of multiple hazards such as earthquakes, mountain floods, and landslides and debris flows, and significantly improves the early warning accuracy; adopts a five-layer cloud service architecture and Kubernetes intelligent scheduling to effectively shorten the early warning response time of complex disaster chains; integrates the Cesium and UE5 engines to dynamically simulate the disaster process and significantly improve the intuitiveness of decision-making. Any modifications, equivalent replacements, and improvements made within the spirit and principles of the present invention shall be included within the protection scope of the claims of the present invention.

Claims

1. A multi-disaster big data early warning system for the main stream reservoir-dam group section in the upper reaches of the Yellow River, characterized in that: It includes a geological disaster monitoring module, a data processing and storage module, a multi-disaster early warning calculation module, an early warning release module, a real-scene display module, and a communication module for data transmission between modules; The geological disaster monitoring module includes GNSS stations, low-altitude remote sensing monitoring by drones, and high-resolution satellite series monitoring. The GNSS stations are used to obtain surface displacement and deformation data in real time. The low-altitude remote sensing monitoring by drones is used to obtain low-altitude monitoring data. The high-resolution satellite series monitoring is used to obtain high-resolution remote sensing image data, and the above multi-dimensional monitoring data is transmitted to the data processing and storage module in real time through the communication module; The data processing and storage module performs multi-source heterogeneous cleaning on the above multi-dimensional monitoring data based on ETL technology, and then performs spatio-temporal alignment and semantic mapping to form a standardized data set, and stores the standardized data set in a distributed database to provide dynamic data input for the multi-disaster early warning calculation module; The multi-disaster early warning calculation module triggers a specific disaster model according to the spatio-temporal characteristics of the input data, performs single-disaster risk probability calculation, and realizes cross-disaster risk superposition probability analysis based on the complex network deduction algorithm in the model, generating multi-scenario early warning plans and disaster chain evolution prediction results; The early warning release module is used to receive the calculation results of the multi-disaster early warning calculation module, convert the calculation results into early warning information and send them to the early warning department in real time to realize cross-departmental data sharing and emergency order linkage; The real-scene display module receives the model output results of the above multi-disaster early warning calculation module in real time, and converts the disaster risk early warning results into an intuitive three-dimensional scene to realize the visual deduction and quantitative analysis of the disaster process.

2. The multi-disaster big data early warning system for the main stream reservoir-dam group section in the upper reaches of the Yellow River according to claim 1, wherein: The geological disaster monitoring module is based on the sky-air-ground multi-dimensional perception technology to collect monitoring data of landslides, collapses, debris flows, and mountain floods. The monitoring data covers the disaster basic data, remote sensing data, regional disaster data, and disaster point data of the main stream reservoir-dam group section in the upper reaches of the Yellow River.

3. The multi-disaster big data early warning system for the main stream reservoir-dam group section in the upper reaches of the Yellow River according to claim 2, characterized in that: The basic data includes basic topographic data and basic geographic data. The remote sensing data includes remote sensing image data of the high-resolution satellite series. The regional disaster data includes meteorological, hydrological, and seismic data. The disaster point data includes the location, size, and development status of geological disaster points of landslides, collapses, and debris flows.

4. The multi-hazard big data early warning system for the main stream reservoir-dam group section in the upper reaches of the Yellow River according to claim 1, characterized in that: Based on the data transmitted by the geological disaster monitoring module, the data processing and storage module constructs a multi-disaster risk early warning multi-source database with the themes of disaster hazard assessment, regional disaster early warning, single-disaster early warning, and disaster situation assessment by sorting out the data sources, storage methods, structures, spatio-temporal sequence relationships, and attribute logical relationships of multi-disaster monitoring data of earthquakes, landslides, debris flows, rainstorms, and mountain floods in the main stream reservoir-dam group section in the upper reaches of the Yellow River.

5. The multi-disaster big data early warning system for the main stream reservoir-dam group section in the upper reaches of the Yellow River according to claim 4, characterized in that: The multi-hazard risk early warning multi-source database adopts a distributed database environment, integrating spatial data management, image data management, disaster data management, and comprehensive analysis functions. The multi-hazard risk early warning multi-source database is configured with unified data and interface standards, and uses data warehouse technology to perform ETL operations on the data to achieve automated real-time extraction and conversion of the data.

6. The multi-disaster big data early warning system for the main stream reservoir and dam group section in the upper reaches of the Yellow River according to claim 1, characterized in that: The multi-hazard early warning calculation module integrates the risk early warning probability models for mountain flood disasters, earthquake risks, and landslide, collapse, and debris flow risks. By integrating the adaptive particle filter algorithm, adversarial neural network, and discrete element-finite element coupled numerical simulation technology, it realizes the early warning of mountain floods, earthquakes, and landslide, collapse, and debris flow disasters and the coupled analysis of cross-hazard risks.

7. The multi-disaster big data early warning system for the main stream reservoir-dam group section in the upper reaches of the Yellow River according to claim 6, characterized in that: Based on the risk early warning probability model for mountain flood disasters, it uses the adaptive particle filter algorithm to dynamically assimilate multi-source data of rainstorm mountain floods, and inversely calculates the roughness parameters, water levels, and flow rates in real-time during flood evolution. Combining with multi-scenario ensemble simulation, it evaluates the risks of individual and overall reservoir-dam groups. Based on the earthquake risk early warning probability model, it uses the adversarial neural network to generate ground motion time history signals that conform to the characteristics of the upper Yellow River region, constructs a nonlinear mapping relationship between magnitude, intensity, and dam body damage by mining historical earthquake damage data, and uses reliability theory to analyze the failure thresholds of different dam types to achieve efficient calculation of the reliability of single-dam earthquake responses. Based on the risk early warning probability model for landslide, collapse, and debris flow, it establishes a fluid-solid coupling constitutive model for soil and debris, combines the discrete element method to simulate particle motion and the finite element method to analyze the stress and strain of rock and soil, and forms an erosion parameter and collapse mechanics model under the interaction of water flow and rock and soil for dynamically predicting the evolution path of the landslide, collapse, and debris flow disaster chain.

8. The multi-hazard big data early warning system for the main stream reservoir-dam group section in the upper reaches of the Yellow River according to claim 1, characterized in that: The early warning release module is used to convert the calculation results of the multi-hazard risk early warning model into standardized and operable early warning instructions. Based on the risk probability, disaster chain evolution path, and three-dimensional deduction results output by the corresponding model, the early warning release module automatically generates multi-level early warning information using a rule engine, and realizes real-time multi-channel push of early warning information through an information gateway and information sharing interface, and generates a structured disaster report to achieve cross-departmental instruction linkage and optimized allocation of resources.

9. The multi-hazard big data early warning system for the main stream reservoir-dam group section in the upper reaches of the Yellow River according to claim 1, characterized in that: The real-scene display module combines the Cesium visualization technology framework with the UE5 rendering engine, and uses a virtual simulation physics engine to simulate the disaster process in combination with the model calculation results of the multi-hazard early warning calculation module to complete the high-fidelity three-dimensional dynamic restoration of the disaster process.

10. A multi-disaster big data early warning method for the main stream reservoir-dam group section in the upper reaches of the Yellow River, which applies the multi-disaster big data early warning system for the main stream reservoir-dam group section in the upper reaches of the Yellow River described in any one of the above claims 1-9, is characterized in that: Including the following steps: Step 1: Data collection Based on the deployed sky-air-ground integrated monitoring network, using GNSS stations, low-altitude remote sensing emergency monitoring by drones, and high-resolution satellite series monitoring, multi-dimensional and multi-source heterogeneous data are synchronously collected, and the multi-source heterogeneous data is transmitted to the data processing and storage module in real-time through a communication module. Step 2: Data processing After the data processing and storage module uses ETL technology to clean, transform, and perform spatio-temporal alignment operations on the multi-source heterogeneous data, a standardized data set is formed and stored in a distributed database to ensure that the data can be called in real-time. Step 3: Multi-hazard risk calculation The multi-hazard early warning calculation module triggers the corresponding model library according to the standardized data. Among them, the flash flood model uses the adaptive particle filter algorithm to assimilate rainstorm data, inversely calculates the roughness parameters and the flood evolution process in real time, and evaluates the risks of individual and overall reservoir-dam groups; the earthquake model generates regional ground motion signals based on the adversarial neural network, establishes the magnitude-intensity-dam body damage mapping relationship by combining historical earthquake damage data, and analyzes the failure threshold of dam types through the reliability theory; the landslide and debris flow model integrates the discrete element method and the finite element method to simulate the erosion parameters and collapse mechanism under the coupling action of soil, water and rock. In addition, the multi-hazard early warning calculation module can also integrate the single-hazard results based on the complex network deduction algorithm, calculate the cross-hazard risk superposition probability, and thus generate multi-scenario early warning plans and disaster chain evolution paths; Step 4: Generation and release of early warning information The early warning release module converts the risk probability and disaster chain path output by the above multi-hazard early warning calculation module into multi-level early warning instructions through the rule engine, pushes them to the early warning department in seconds using the SMS gateway and the government cloud API, and automatically generates a structured disaster report; Step 5: 3D real-scene deduction The real-scene display module is based on the Cesium framework and the UE5 engine, receives the model output results of the above multi-hazard early warning calculation module in real time, synchronizes the deduction results, and realizes the high-fidelity 3D dynamic simulation of the disaster process.

Images