Geological disaster hidden danger accident point management system and method

Abstract

The invention relates to the technical field of geological disasters, in particular to a geological disaster hidden danger accident point management system and method.The method comprises the steps that a hidden danger accident point management knowledge base is constructed based on hidden danger related data and data collected in a geological disaster monitoring area, and standardized integration of multi-source data is achieved; a management knowledge base is identified and analyzed, a plurality of management dimensions are divided, and hidden danger features are comprehensively described; a plurality of management dimensions are fused, a hidden danger accident point comprehensive management engine is constructed, and association analysis and risk assessment of multi-dimensional data are achieved; and according to a preset management process and a comprehensive management engine, generating an automatic management strategy, determining an execution sequence and a data interaction mode of the sub-engines corresponding to each management dimension, and realizing full-process automatic management from data analysis to decision generation. According to the method, through multi-source data integration, multi-dimensional risk analysis and automatic strategy generation, the accuracy, real-time performance and scientificity of geological disaster hidden danger management are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of geological disaster technology, and in particular to a management system and method for geological disaster hazard sites. Background Technology

[0002] Geological hazards (such as landslides, debris flows, and ground subsidence) pose significant threats to people's lives and property and the stability of infrastructure. Their scientific and meticulous management has always been a core issue in the field of geological hazard prevention and control. With the intensification of global climate change and human engineering activities, the frequency and severity of geological hazards are on the rise. How to efficiently integrate multi-source monitoring data, accurately identify potential risks, and implement dynamic management has become a key challenge facing the industry.

[0003] Traditional methods for managing geological hazard risks primarily rely on manual inspections, discrete analysis of single-point monitoring data, and empirical risk assessments. These methods suffer from low data integration efficiency, limited risk identification dimensions, and delayed management decisions. Specifically, multi-source data, including historical disaster records, geological environmental data, and information from monitoring equipment, are often stored in different systems, lacking unified standardized processing and integration. This makes it difficult to comprehensively characterize the risk characteristics of potential hazard sites. Furthermore, existing management systems are mostly based on single indicators or localized scenarios for assessment, failing to construct a systematic management framework from multiple perspectives, such as geological structure evolution, temporal correlation of monitoring data, and disaster chain transmission paths. This makes it difficult to meet the real-time and automated management needs under complex geological conditions. In addition, industry disaster prevention standards and specifications lack efficient digital mapping methods in practical applications, leading to a disconnect between management strategies and technical standards, further affecting the accuracy and timeliness of hazard prevention and control.

[0004] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] The main objective of this invention is to provide a geological hazard hazard incident site management system and method, which aims to solve the technical problems of traditional geological hazard hazard management methods, such as low efficiency of multi-source data integration, single risk identification dimension, and lagging management decision-making, which cannot meet the needs of multi-source data fusion processing, multi-dimensional risk systematic assessment and automated and precise management under complex geological conditions.

[0006] To achieve the above objectives, the present invention provides a method for managing potential geological disaster sites, the method comprising: Based on the hazard-related data and information collected from the geological disaster monitoring area, a knowledge base for hazard accident point management is constructed. The hazard-related data and information include at least one of the following: historical disaster records, geological environment data, information sensed by monitoring equipment, and industry disaster prevention standards and specifications. By identifying and analyzing the knowledge base for the management of potential geological hazards and accidents, multiple management dimensions are obtained. Different management dimensions are used to comprehensively manage different aspects of potential geological hazards and accidents. By integrating multiple management dimensions, a comprehensive management engine for potential hazards and accident points is constructed. Based on the preset management process and the comprehensive management engine for potential geological hazards and accidents, an automated management strategy for potential geological hazards and accidents is generated. The management process is used to determine the execution order and data interaction method between the sub-engines corresponding to each management dimension in the management engine.

[0007] Optionally, the construction of a knowledge base for the management of potential hazard incident sites based on hazard-related data and information collected from geological disaster monitoring areas includes: The acquired hazard-related data and information are analyzed, and the analysis results are standardized based on a multi-source data fusion tool to generate a structured management data source. The hazard-related data and information include at least one of the following: monitoring station sensor data, UAV remote sensing imagery, geological survey reports, and historical disaster databases. The hazard-related data and information include at least one of the following: time-series monitoring data, unstructured survey text, and professional drawing files. The time-series monitoring data is analyzed based on field extraction rules, the unstructured survey text is analyzed based on natural language processing, and the professional drawing files are analyzed based on image recognition technology. Key hazard entities in the structured management data source are located using predefined geological entity identification rules. Based on the frequency of occurrence, hazard level, and spatial correlation of entities in the data source, the metadata of the management knowledge base is determined from the key hazard entities. A pre-trained geological feature extraction model is used to obtain the disaster feature units and hidden danger feature units included in the structured management data source. Based on the geological disaster evolution model, the risk connotation of the disaster feature units and hidden danger feature units is further mined according to the spatial context information of the disaster feature units and hidden danger feature units, so as to map the hidden danger information in the data source into the corresponding hidden danger feature vector. By using geological data association technology, the original records of the data source, the determined metadata, and the generated hazard feature vectors are associated and integrated through the spatial coordinate identifiers of each data item in the structured management data source to generate the hazard accident point management knowledge base.

[0008] Optionally, by identifying and analyzing the knowledge base for managing potential hazards and accidents, multiple management dimensions corresponding to the knowledge base for managing potential hazards and accidents are obtained, including: Based on the hazard feature vector and the metadata in the hazard accident point management knowledge base, calculate the hazard correlation degree between each data item in the management knowledge base, and divide the data items in the management knowledge base into different management dimension candidate sets based on the hazard correlation degree; The disaster evolution path characteristics and impact range propagation characteristics of the data items included in the candidate sets of each management dimension are analyzed to determine the impact area and risk transmission link of different disaster stages in each data item. Key management indicators are extracted by identifying the disaster type, affected objects and triggering factors in each data item. The key management indicators are used to clarify the core control description information of different stages in each data item. Based on the affected area, the risk transmission link, and the key management indicators, the data items in each of the candidate sets of management dimensions are summarized and a dimension description of each candidate set of management dimensions is generated. The dimension description is used to reflect the role and function of the dimensions in the candidate sets of management dimensions in the management of hidden danger and accident points. Based on the dimension descriptions in each of the management dimension candidate sets, the data item content in each of the management dimension candidate sets, and the preset management specifications, target data fragments in each of the management dimension candidate sets are determined as dimension examples of each of the management dimension candidate sets; Based on the dimension descriptions and dimension examples in each of the candidate sets of management dimensions, dimension management standards for each of the candidate sets of management dimensions are determined from the preset disaster prevention standards. The dimension management standards are used to clarify the management accuracy and timeliness of each dimension in the candidate sets of management dimensions. Based on the consistency of dimension descriptions, dimension examples, and dimension management standards corresponding to each dimension in each candidate set of management dimensions, multiple management dimensions corresponding to the hidden danger and accident point management knowledge base are determined in each candidate set of management dimensions.

[0009] Optionally, the comprehensive management engine for potential hazards and accidents includes a management task scheduling sub-engine, a special management sub-engine corresponding to each of the management dimensions, and a management result verification sub-engine; the process of integrating multiple management dimensions to construct the comprehensive management engine for potential hazards and accidents includes: Based on the dimension description of each management dimension, the management target features, core indicator features, and expected output features of each management dimension are obtained, and the management target features, core indicator features, and expected output features are input into the management model to guide the management model to generate management algorithm parameter templates corresponding to each management dimension. Based on the management algorithm parameter template and the dimension examples of the management dimensions, the initial management model is trained and optimized to generate specialized management sub-engines corresponding to each management dimension. The dimension management standards corresponding to each of the management dimensions are transformed into executable verification rules to construct verification algorithms corresponding to each of the dimension management standards, and management result verification sub-engines for the corresponding management dimensions are generated based on each of the verification algorithms. By deeply exploring the overall management framework of geological disaster hazard points and the dependencies between each management dimension, and based on the overall management framework and the dependencies, the collaborative workflow of each specialized management sub-engine and each management result verification sub-engine is determined, and a management task scheduling sub-engine related to the collaborative workflow is generated. Based on the management task scheduling sub-engine, the special management sub-engine corresponding to each management dimension, and the management result verification sub-engine corresponding to each management dimension, the comprehensive management engine for hidden danger and accident points is constructed.

[0010] Optionally, the step of generating an automated management strategy for geological disaster hazard points based on a preset management process and the comprehensive management engine for hazard points includes: By formally describing the preset management process, the management stages, management nodes, and triggering conditions between the management nodes in the management process are obtained. The management stage represents the execution steps of the management task, the management node represents the execution unit of each sub-engine, and the triggering condition represents the data transfer rules between sub-engines. Based on the management stage, the management node, and the triggering condition, determine the execution order and data interface of each of the special management sub-engines, each of the management result verification sub-engines, and the management task scheduling sub-engine; Based on the execution order and data interface, configure the operating parameters of the comprehensive management engine for potential geological hazards and accidents, and generate an automated management strategy for potential geological hazards and accidents.

[0011] Optionally, the professional drawing file is parsed based on image recognition technology, including: The image data of the professional drawing file is acquired, and the drawing type is determined by the drawing type recognition tool to generate a drawing type identifier. The drawing type includes at least one of geological profile map, contour map, and disaster zoning map. Based on the drawing type identifier, the corresponding graphic recognition parser is invoked to perform vectorization processing and feature recognition on the image data, and to extract the topographic features, geological structure features, and hazard point annotation features from the drawing. The accuracy of the parsed topographic features, geological structures, and hazard point marking features is verified using preset feature verification rules. The positional deviations and attribute errors in the features are corrected by comparing them with field survey data. The verified and corrected elements are converted into a standardized spatial vector format and linked to the corresponding geographic coordinates to generate structured drawing data. The structured drawing data is subjected to cross-data source consistency verification with the time-series monitoring data and the parsing results of the unstructured survey text. Based on the verification results, the fuzzy elements in the structured drawing data are manually annotated and corrected to generate the target parsing result of the professional drawing file.

[0012] Optionally, calculating the correlation degree between data items in the management knowledge base based on the hazard feature vector and the metadata in the hazard accident point management knowledge base includes: Extract the vector corresponding to the disaster stage of each data item from the hidden danger feature vector, and aggregate the vectors of each data item into a data item-level hidden danger vector by weighted average algorithm; The first correlation value between the data item-level hidden danger vectors is calculated based on the cosine similarity algorithm, and the index overlap rate between the data items is calculated as the second correlation value based on the hidden danger index set in the metadata. The first correlation value and the second correlation value are weighted and fused according to a preset weight allocation strategy to generate the potential correlation degree between each data item.

[0013] Optionally, the step of inputting the management target features, the core indicator features, and the expected output features into the management model to guide the management model in generating management algorithm parameter templates corresponding to each management dimension includes: The management objective features are converted into management scenario description statements, the core indicator features are converted into indicator weight constraint lists, and the expected output features are converted into management result format specifications. Based on the preset parameter template generation rules, the management scenario description statement, the indicator weight constraint list, and the management result format specification are sequentially concatenated into an initial parameter configuration sequence. The initial parameter configuration sequence is optimized and logic enhanced using the management model to generate a set of candidate management algorithm parameter templates. Using the dimensional example as a validation sample, the management model generates simulated management results based on the candidate management algorithm parameter template, and calculates the indicator consistency and management trend consistency indicators between the simulated management results and the dimensional example. Candidate management algorithm parameter templates with an indicator matching degree higher than the third preset threshold and a management trend consistency indicator higher than the fourth preset threshold are selected as the management algorithm parameter templates corresponding to the management dimension.

[0014] Optionally, when locating key hazard entities in the structured management data source using predefined geological entity identification rules, the method further includes: Identify monitoring sequences with catastrophic evolution characteristics in the structured management data source, and extract the deformation stages in the monitoring sequences as entity evolution labels; Analyze the time series and causal relationship of each deformation stage within the monitoring sequence, and associate the time series and causal relationship with the entity evolution tag to form a time-series catastrophe path; Detect similar abnormal indicators across monitoring points in the structured management data source, and track the spatial impact range of the key hidden danger entity based on the similar abnormal indicators; The time-series disaster path and the spatial impact range are mapped to a preset disaster evolution map to generate a multi-dimensional disaster behavior model of the key hidden danger entity; Based on the multi-dimensional disaster behavior model, entity attribution is performed on the interrelated hidden danger events in the structured management data source to eliminate duplicate disaster event nodes.

[0015] Furthermore, to achieve the above objectives, the present invention also provides a geological disaster hazard incident site management system, the system comprising: processor; Memory used to store the processor's executable instructions; The processor is configured to execute the instructions to implement the steps of the geological disaster hazard accident site management method as described in any of the above.

[0016] This invention provides a method for managing geological disaster hazard sites. The method constructs a management knowledge base encompassing historical disaster records, geological environmental data, monitoring equipment information, and industry standards. This enables the unified integration and standardized processing of multi-source heterogeneous data, solving the problems of scattered data and inconsistent formats in traditional methods. It provides a comprehensive and structured data source for hazard analysis. By dividing the data into multiple management dimensions, such as disaster evolution paths, risk transmission links, and key management indicators, it overcomes the limitations of single-indicator assessment. It comprehensively characterizes hazard features from multiple perspectives, including geological structure, monitoring timelines, and risk propagation, improving the systematicness and accuracy of risk identification. Finally, it integrates these multi-dimensional management dimensions to construct a comprehensive management engine. Based on preset processes, it generates automated management strategies, achieving full-process automation from data parsing and dimensional analysis to strategy generation. This solves the problems of lagging and inefficient traditional management decision-making and meets the real-time and intelligent management needs under complex geological conditions. It integrates industry disaster prevention standards and specifications into the management knowledge base and assessment dimensions, and uses digital means to achieve automatic mapping and verification of standards, ensuring that management strategies comply with industry standards and improving the scientific and standardized nature of hazard prevention and control. Through multi-dimensional data association and risk feature mining, it forms a full-chain management capability for hazard and accident points from monitoring and assessment to decision-making, providing systematic support for the prevention, emergency response, and long-term governance of geological disasters, and effectively reducing disaster risks and losses. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating an embodiment of the geological disaster hazard incident site management method of the present invention.

[0019] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0021] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0022] Reference Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the method for managing potential geological disaster sites according to the present invention.

[0023] In one embodiment, the method for managing potential geological hazard sites includes the following steps: Step S100: Based on the hazard-related data and information collected from the geological disaster monitoring area, construct a hazard accident point management knowledge base. The hazard-related data and information include historical disaster records, geological environment data, monitoring equipment sensing information, and at least one of the industry disaster prevention standards and specifications.

[0024] Historical disaster records can be structured or unstructured documents documenting the time, location, type, scale, impact range, and handling of past geological disasters, providing historical evolutionary evidence for risk pattern identification of potential hazard sites. Geological environmental data can be a set of static and dynamic parameters describing the natural geological conditions of the area where a geological disaster occurs, including soil and rock properties, topography, hydrology, and tectonic activity, providing geological background support for the formation of potential hazard sites and used to assess regional stability and potential damage patterns. Monitoring equipment sensing information can be real-time or periodic data on changes in physical quantities collected by sensors deployed at or around potential hazard sites, reflecting the dynamic response of surface or subsurface conditions, providing real-time observational evidence of the hazard evolution process, and supporting time-series risk assessment. Industry disaster prevention standards and specifications can be a collection of technical regulations, assessment criteria, and management requirements issued by national or industry authorities to guide geological disaster prevention and control work, providing compliance constraints for management strategy generation and ensuring that assessment logic and prevention and control measures meet legal requirements. The hazard and accident site management knowledge base can be a unified structured data set integrating historical disaster records, geological environmental data, monitoring equipment sensing information, and industry disaster prevention standards and specifications. It possesses semantic association and standardized indexing capabilities, providing a consistent, traceable, and computable data foundation for multi-dimensional analysis and eliminating information silos. Based on this data, an ETL toolchain can be used to parse CSV, JSON, GeoJSON, and PDF documents into unified graph database nodes and edge relationships. Entity recognition technology can then be used to link multi-source records of the same hazard point, achieving semantic integration of multi-source data and providing a consistent data foundation for subsequent analysis.

[0025] Step S200: By identifying and analyzing the knowledge base for the management of potential hazards and accidents, multiple management dimensions are obtained corresponding to the knowledge base. Different management dimensions are used to comprehensively manage different aspects of potential geological hazards and accidents.

[0026] Among these, the management dimension can be an analytical perspective that systematically characterizes geological hazard risks. Each dimension focuses on a specific type of risk influencing factor or evolutionary logic, achieving a multi-faceted deconstruction of hazard characteristics and avoiding the one-sidedness of single-indicator assessment. The disaster evolution path can describe the phased state change sequence and triggering mechanism of a geological disaster from initial deformation to final instability, revealing the dynamic development trajectory of hazard points from the incubation period to the pre-disaster period, assisting in prediction. The risk transmission link can describe the causal path of a geological disaster spreading from a hazard point to adjacent areas or related facilities at the spatial or systemic level, identifying the potential for chain reactions and assessing the scope of secondary disaster risks. Key management indicators can be a set of measurable parameters used to quantify the risk status of hazards and support management decisions. They have clear calculation logic and threshold definitions, providing calculable input variables for automated assessment and supporting the quantitative basis for strategy generation. Through natural language processing and clustering algorithms, semantic analysis of various data types in the knowledge base is performed to identify high-frequency association patterns, summarizing a stable and reusable analytical framework to achieve a systematic deconstruction of hazard characteristics and avoid subjective biases from pre-set dimensions.

[0027] Step S300: Integrate multiple management dimensions to build a comprehensive management engine for potential hazards and accidents.

[0028] The comprehensive management engine for potential hazards and accidents can be a logical computing system composed of multiple management dimension sub-engines, capable of collaborative execution. It integrates multi-source analysis results and outputs a comprehensive judgment. Management dimensions such as disaster evolution paths, risk transmission links, and key management indicators can be connected to the comprehensive management engine as independent sub-engines, with their outputs serving as inputs or constraints for other sub-engines. A workflow engine framework (such as a DAG) is used to orchestrate the computational modules of each dimension sub-engine, setting input / output interfaces and data dependencies to achieve parallel and collaborative processing of multi-dimensional analysis, improving the comprehensiveness and logical consistency of risk identification.

[0029] Step S400: Based on the preset management process and the comprehensive management engine for potential geological hazards and accidents, generate an automated management strategy for potential geological hazards and accidents. The management process is used to determine the execution order and data interaction method between the sub-engines corresponding to each management dimension in the management engine.

[0030] The pre-defined management process can be a template that specifies the execution order, triggering conditions, and data interaction logic of each sub-engine during the management of potential hazards and incidents, ensuring the analysis process is repeatable, logically consistent, and automatically executable. Through process modeling tools (such as BPMN), management steps from expert experience and industry standards are transformed into directed flowcharts, defining the data flow and control flow between nodes. The pre-defined management process acts as a scheduler, controlling the startup order, parallel conditions, and result fusion methods of each sub-engine within the integrated management engine, ultimately outputting standardized management decision instructions.

[0031] Taking the dynamic management of hidden dangers on mountain highway slopes as an example, GNSS displacement meters, rain gauges, and crack sensors are deployed along a mountain highway, and the data is connected to the knowledge base for the management of hidden danger accident points in real time. The system automatically identifies that a small landslide has occurred at this point in the past, and the geological environment shows that the strongly weathered mudstone layer and bedding joints are well developed. The management dimensions are automatically divided into disaster evolution paths (based on changes in displacement rate), risk transmission links (extending downhill along joint surfaces), and key management indicators (cumulative displacement, rainfall intensity index). The comprehensive management engine first assesses the evolution stage according to the preset process, then determines whether it threatens the roadbed below by combining the transmission links, and finally verifies the red warning triggered by deformation rate > 3mm / d and rainfall > 50mm in the industry standard. Finally, it automatically generates an automated management strategy that includes increasing the frequency of patrols, setting up temporary drainage ditches, and activating emergency evacuation plans, and simultaneously outputs a compliance report that complies with the technical specifications for landslide prevention.

[0032] This embodiment provides a method for managing potential geological disaster sites. It constructs a knowledge base for managing potential geological disaster sites based on data and information collected from geological disaster monitoring areas. This knowledge base is then analyzed and divided to obtain multiple management dimensions. These dimensions are integrated to build a comprehensive management engine for potential geological disaster sites. Based on a pre-defined management process and the engine, automated management strategies for potential geological disaster sites are generated. This method achieves comprehensive management of potential geological disaster sites, improves the comprehensiveness and logical consistency of risk identification, and realizes a fully automated closed loop from data access and dimensional analysis to strategy generation. It eliminates the delays and subjectivity of manual judgment and ensures that each output strategy meets legal technical requirements, achieving simultaneous assurance of scientific validity and compliance. Ultimately, it forms an intelligent management capability covering the entire chain of monitoring, analysis, assessment, and decision-making.

[0033] In one embodiment, a knowledge base for hazard incident management is constructed based on hazard-related data and information collected from geological disaster monitoring areas, including: The acquired hazard-related data and information are analyzed, and the analysis results are standardized based on multi-source data fusion tools to generate structured management data sources. Hazard-related data and information include at least one of the following: monitoring station sensor data, UAV remote sensing imagery, geological survey reports, and historical disaster databases. The hazard-related data and information include at least one of the following: time-series monitoring data, unstructured survey text, and professional drawing documents. Time-series monitoring data is analyzed based on field extraction rules, unstructured survey text is analyzed based on natural language processing, and professional drawing documents are analyzed based on image recognition technology.

[0034] The structured data source can be a collection of geological disaster data with a unified field structure and semantic labels, formed after parsing and standardization. It includes time-series monitoring data, entity-based text information, and geometric drawing information. This data can serve as the input basis for feature extraction and evolutionary modeling, ensuring data consistency and computational operability in subsequent processing. In this embodiment, the structured data source converts data from monitoring station sensor data, UAV remote sensing images, geological survey reports, and historical disaster databases into a structured format through field alignment, entity extraction, and image segmentation.

[0035] By using predefined geological entity identification rules, key hazard entities in the structured management data source are located, and based on the frequency of occurrence, hazard level, and spatial relationships of entities in the data source, metadata for the management knowledge base is determined from the key hazard entities.

[0036] The predefined geological entity identification rules can be a set of rules based on geological knowledge bases and expert experience, used to identify and locate key hazard entities in structured management data sources. These rules can be used to extract metadata from the management knowledge base of key hazard entities, providing a basis for subsequent risk assessment and management.

[0037] A pre-trained geological feature extraction model is used to obtain disaster feature units and hidden danger feature units included in the structured management data source. Based on the geological disaster evolution model, the risk connotation of disaster feature units and hidden danger feature units is further explored according to the spatial context information of the disaster feature units and hidden danger feature units, so as to map the hidden danger information in the data source into the corresponding hidden danger feature vector.

[0038] The pre-trained geological feature extraction model can be a deep learning model trained on large-scale geological data, used to automatically identify and extract feature units with hazard indication significance from unstructured or semi-structured geological data. It can be used to achieve automated perception and structured representation of hidden hazard elements in heterogeneous data such as UAV remote sensing imagery, geological survey texts, and professional drawings. In this embodiment, the pre-trained geological feature extraction model extracts hazard feature units and potential hazard feature units from structured data sources using technologies such as CNN, NER, and GCN.

[0039] By leveraging geological data association technology, the original records, defined metadata, and generated hazard feature vectors of the data source are linked and integrated through the spatial coordinate identifiers of each data item in the structured management data source, thereby generating a knowledge base for hazard and accident point management.

[0040] Among these, geological data association technology can be based on spatial coordinates and attribute information to link and integrate the original records, metadata, and hazard feature vectors of the data source. This can be used to generate a knowledge base for hazard and accident site management, providing data support for subsequent risk assessment and management.

[0041] Taking intelligent monitoring of steep rock slopes as an example, multiple drones were deployed on a highway slope in a mountainous area to periodically collect RGB and LiDAR images, simultaneously acquiring displacement and rainfall time-series data from 12 monitoring stations, as well as PDF survey reports and CAD drawings provided by the geological team. The structured data source was analyzed using NLP to identify entities such as strongly weathered mudstone layers, bedding joints, and water-filled fissures. Image recognition was used to extract 37 potential sliding surfaces and 8 surface crack networks. A pre-trained geological feature extraction model transformed the above results into a set of catastrophic feature units. Based on the spatial distribution and rock strata dip of these units, the geological hazard evolution model simulated the stress transmission paths and instability probability evolution trends of 5 high-risk nodes, generating corresponding hazard feature vectors. With the help of geological data association technology, the original monitoring curves, metadata (such as two small collapses that have occurred in the past five years), and 128-dimensional risk vectors of each hazard point were uniformly bound to their WGS84 coordinates, constructing a semantically deep hazard accident point management knowledge base to support subsequent dimension division and strategy generation.

[0042] This embodiment provides a method for managing geological disaster hazard points. By analyzing hazard-related data and information, locating key hazard entities, extracting disaster characteristic units and hazard characteristic units, and linking and integrating data, it can achieve the technical effect of constructing a hazard point management knowledge base with spatiotemporal consistency, semantic computability, and evolutionary reasoning.

[0043] In one embodiment, by identifying and analyzing the knowledge base for managing potential hazards and accidents, multiple management dimensions corresponding to the knowledge base are obtained, including: Based on the hazard feature vector and metadata, the hazard correlation degree between each data item in the management knowledge base is calculated, and the data items in the management knowledge base are divided into different management dimension candidate sets based on the hazard correlation degree. The hazard feature vector can be a high-dimensional numerical vector representing the potential risk state of a specific hazard point, containing semantic dimensions such as evolution trend, transmission potential, and instability probability. It can be used as a quantitative input unit for dividing management dimensions, supporting automatic clustering and semantic grouping based on data correlation. The hazard feature vector is calculated using spatial similarity and semantic distance, serving as the core basis for dividing the candidate set of management dimensions.

[0044] The disaster evolution path characteristics and impact range propagation characteristics of the data items included in the candidate set of each management dimension are analyzed to determine the impact area and risk transmission link of different disaster stages in each data item. Key management indicators are extracted by identifying the disaster type, impact object and triggering factors in each data item. The key management indicators are used to clarify the core control description information of different stages in each data item. In this embodiment, the candidate set of management dimensions can be a group of data items that can potentially serve as management dimensions, automatically clustered from the knowledge base based on the results of hazard correlation calculations. This can provide a data-driven candidate set for the generation of management dimensions, avoiding the one-sidedness and subjectivity of manually pre-defined dimensions. The spatial distribution and evolution path of data items within each candidate set are used to identify their common risk transmission patterns, serving as the basis for the formation of dimension descriptions.

[0045] Based on the area of ​​impact, risk transmission links, and key management indicators, the data items in the candidate sets of each management dimension are summarized and a dimension description of each candidate set of management dimensions is generated. The dimension description is used to reflect the role and function of the dimensions in the candidate sets of management dimensions in the management of hidden danger and accident points. In this embodiment, key management indicators can be a set of measurable parameters used to quantify the status of potential risks and support management decisions. They have clear calculation logic and threshold definitions and can be used as core semantic elements for dimension description, defining the control objectives focused on by the management dimension. Indicators that appear repeatedly in the candidate set (such as displacement rate and cumulative rainfall) are extracted as the key control parameters for that dimension.

[0046] Based on the dimension descriptions in each management dimension candidate set, the data item content in each management dimension candidate set, and the preset management specifications, the target data fragments in each management dimension candidate set are determined as dimension examples for each management dimension candidate set. In this embodiment, the dimension management standard can be a quantifiable and executable specification set for each management dimension, clearly defining its management accuracy and timeliness requirements. It can be used to transform industry disaster prevention standards into verifiable dimension-level operational constraints, ensuring that management dimensions are compliant and executable. A semantic matching algorithm is used to semantically align the clauses in the preset disaster prevention standards with the dimension descriptions and examples, extracting quantifiable requirements such as response time limits, monitoring frequency, and threshold ranges to form a dimension-specific standard set.

[0047] Based on the dimension descriptions and examples in each management dimension candidate set, the dimension management standards for each management dimension candidate set are determined from the preset disaster prevention standards. The dimension management standards are used to clarify the management accuracy and timeliness of each dimension in the management dimension candidate set. In this embodiment, the risk transmission link can be a causal path describing the spread of a geological disaster from a potential hazard point to adjacent areas or related facilities at the spatial or system level. It can be used as an input feature for dimensional description generation to identify common propagation patterns of data items in the candidate set of management dimensions. The transmission paths of all data items within the candidate set are aggregated into typical topological patterns to summarize the propagation characteristics of that dimension.

[0048] Based on the consistency of dimension descriptions, dimension examples, and dimension management standards in each candidate set of management dimensions, multiple management dimensions corresponding to the knowledge base for hazard and accident point management are determined in each candidate set of management dimensions.

[0049] Taking the automatic generation of intelligent management dimensions for regional landslide clusters as an example, the system calculates the semantic similarity of the feature vectors and metadata of 127 hazard points in the hazard accident point management knowledge base, and clusters them to generate 6 candidate sets of management dimensions. One candidate set contains 8 hazard points, whose risk transmission links are all manifested as bedding joints, groundwater seepage, slope toe softening, and overall sliding. The key management indicators are displacement rate, pore water pressure, and rainfall intensity. Based on this, the dimension description is generated as: bedding sliding-hydraulic coupling type. Typical hazard points in this group are selected as dimension examples, and dimension management standards are generated. After consistency verification, the description, examples, and standards of this candidate set are completely matched. Of the remaining 5 candidates, 2 were eliminated due to missing standards, and 3 were merged due to ambiguous descriptions. Finally, 4 stable, compliant, and executable management dimensions are determined to support the construction of the dimension sub-engine of the subsequent comprehensive management engine.

[0050] This embodiment provides a method for managing geological disaster hazard incident sites. It calculates the correlation degree of hazards based on hazard feature vectors and metadata, divides the candidate set of management dimensions, analyzes the disaster evolution path characteristics and impact propagation characteristics of data items, extracts key management indicators, summarizes data items, generates dimension descriptions, determines dimension management standards, and determines multiple final management dimensions by referencing the consistency between dimension descriptions, dimension examples, and dimension management standards. This achieves the technical effect of generating a candidate set of management dimensions through data-driven clustering, integrating risk transmission links and key management indicators to generate dimension descriptions, and semantically aligning dimension descriptions, dimension examples, and industry disaster prevention standards to generate dimension management standards. This ensures that the final management dimensions are completely consistent in semantic expression, typical scenarios, and compliance requirements.

[0051] In one embodiment, the comprehensive management engine for potential hazards and accidents includes a management task scheduling sub-engine, specialized management sub-engines corresponding to each management dimension, and a management result verification sub-engine; it integrates multiple management dimensions to construct the comprehensive management engine for potential hazards and accidents, including: Based on the dimensional descriptions of each management dimension, the management objective features, core indicator features, and expected output features of each management dimension are obtained. These features are then input into the management model to guide the model in generating management algorithm parameter templates for each management dimension. Among these, management objective features can be semantic descriptions of the core management intent of that dimension, such as identifying early signs of slippage, inhibiting chain reactions, and controlling the radius of influence. These features can provide objective guidance for the specialized management sub-engine, directing algorithm design to focus on specific risk control objectives. Understandably, management objective features utilize a natural language understanding model to semantically parse the dimension description text, extracting verbal target phrases and constraints, and structuring them into a computable target vector. Management objective features can be early warning targets, chain reaction suppression targets, spatial isolation targets, etc.

[0052] Core indicator features can be a set of key parameters used to quantify the degree of goal achievement in the management dimension. They have clear calculation logic and threshold definitions, serving as input variables and optimization objectives for the special management sub-engine, supporting the algorithm's sensitivity adjustment in response to risk states. Core indicator features can include deformation rate indicators, hydraulic coupling index, structural instability probability, etc.

[0053] The expected output characteristics can be standardized management results that the management dimensions should produce under ideal operating conditions, such as warning levels, handling suggestions, and inspection frequency. This provides output format constraints for the specialized management sub-engine, ensuring that the generated strategies are compatible with downstream systems. Expected output characteristics can include four-level warning instructions, engineering reinforcement suggestion packages, and emergency evacuation area delineation.

[0054] Based on management target features, core indicator features, and expected output features, inputs can be used to generate management algorithm parameter templates for the management model. This can be achieved by using a multimodal encoder to encode the target text, indicator list, and output samples into vectors, which are then fed into a Transformer architecture for joint modeling. The output parameter template can be generated using an LSTM with an Attention structure, a learning rate of 0.001, and a FocalLoss loss function. Alternatively, a meta-learning framework (MAML) can be used to retrieve similar configurations from a historical parameter template library, fine-tune the gradient based on the current input, and generate an initial parameter set adapted to the new dimension. This transforms management strategies from manually written rules to data-driven automatic parameter generation, improving the efficiency and adaptability of strategy generation.

[0055] Based on the management algorithm parameter template and management dimension examples, the initial management model is trained and optimized to generate specialized management sub-engines corresponding to each management dimension; The management algorithm parameter template is a reusable set of algorithm configurations automatically generated by the management model based on objectives, metrics, and output features. It includes model structure, loss function, regularization terms, and hyperparameter combinations, serving as the initialization skeleton for specialized management sub-engines to automate the generation and parameter adaptation of management strategies. Essentially, the management algorithm parameter template employs a meta-learning framework, taking the objective-metric-output triplet as input, automatically generating the optimal parameter configuration through a neural network search space, and outputting a serializable configuration file.

[0056] Based on management algorithm parameter templates and dimensional examples of management dimensions, an initial management model can be trained and optimized to generate specialized management sub-engines for each management dimension. This can be achieved by using supervised fine-tuning of the model initialized with parameter templates using typical scenario data from the dimensional examples to adapt it to specific risk patterns. Alternatively, a dedicated training set can be constructed for each candidate dimension, inputting historical monitoring data and expert annotation results, and using cross-entropy loss to optimize the consistency between the model output and the examples. Another approach is to employ a reinforcement learning framework, using the compliance and effectiveness of management results as reward signals to iteratively optimize the sub-engine's decision-making strategy in a simulated environment. This enables each sub-engine to possess high-precision response capabilities for specific geological risk scenarios, addressing the problem of insufficient generalization in general models.

[0057] The dimension management standards corresponding to each management dimension are transformed into executable verification rules to construct verification algorithms corresponding to each dimension management standard, and management result verification sub-engines for the corresponding management dimensions are generated based on each verification algorithm. In this context, dimensional management standards can be quantifiable, enforceable specifications set for each management dimension, clearly defining its accuracy and timeliness requirements. These standards serve as the basis for generating verification rules, ensuring that management results comply with legal technical requirements. It is understandable that timeliness clauses, threshold conditions, and response levels in dimensional management standards are parsed into Boolean judgment logic and numerical comparison rules.

[0058] Transforming dimensional management standards into executable verification rules can be achieved by encoding standard clauses into IF-THEN statements or Python functions using a rule engine. For example, if the warning level is red, an evacuation route suggestion must be included; if the displacement rate is greater than 5 mm / d, a 24-hour response should be triggered. A callable verification sub-engine module should be built. This ensures that industry standards are mandatoryly embedded during system runtime, guaranteeing that every output strategy complies with legal requirements.

[0059] By deeply exploring the dependencies between the overall management framework and various management dimensions, and based on the overall management framework and dependencies, the collaborative workflow of each specialized management sub-engine and each management result verification sub-engine is determined, and a management task scheduling sub-engine is generated. This involves analyzing the input-output dependencies, execution timing constraints, and anomaly propagation paths among various management dimensions to construct a Directed Acyclic Graph (DAG) as the scheduling logic. This can be understood as using graph neural networks to analyze shared hazard feature vectors and data flows between dimensions, identifying sequential dependencies such as disaster evolution paths, risk transmission links, and key management indicators, and generating a DAG scheduling graph. Alternatively, it can be based on expert-annotated typical management processes, training a sequence prediction model to predict the sub-engine call order, and automatically generating executable workflow definition files based on constraints.

[0060] By deeply exploring the dependencies between the overall management framework and various management dimensions, the collaborative workflows of each specialized management sub-engine and the management result verification sub-engine can be determined, generating a management task scheduling sub-engine. This enables automated collaborative orchestration of multiple sub-engines, ensuring that the analysis logic aligns with the geological disaster evolution timeline and management logic.

[0061] Based on the management task scheduling sub-engine, the special management sub-engine, and the management result verification sub-engine, a comprehensive management engine for hidden danger and accident points is constructed. This approach integrates the three types of sub-engines into a unified execution framework, coordinating their invocation, execution, and feedback processes through a message bus and state machine. This can be understood as follows: using Apache Airflow to build a DAG workflow, with the scheduling sub-engine acting as a task orchestrator, the specialized sub-engine as a processing node, and the verification sub-engine as a post-verification hook; or building a microservice architecture where each sub-engine is an independent service, communicating via gRPC, with the scheduling sub-engine acting as an API gateway to control the call chain and circuit breaker mechanism.

[0062] By managing task scheduling sub-engines, special project management sub-engines, and management result verification sub-engines, a comprehensive management engine for potential hazards and incidents can be built. This forms a closed-loop intelligent system with four capabilities: perception, decision-making, verification, and scheduling, achieving end-to-end automation and intrinsic compliance. Taking the construction of a comprehensive management engine for regional landslide clusters as an example, the system extracts the target feature for the bedding-hydraulic coupled dimension as suppressing the activation of the sliding surface, with the core indicators being the pore water pressure rise rate + displacement acceleration, and the expected output being a level-three early warning + drainage recommendations. The management algorithm parameter template is generated as XGBoost + time window sliding features. Five historical examples of this dimension are used for fine-tuning to generate a special management sub-engine A. If the water pressure rise rate is >0.1MPa / d, the verification rules for drainage recommendations must be output, and a verification sub-engine B is constructed. Analysis shows that this dimension can only run after the disaster evolution path sub-engine outputs, and must be executed before the risk transmission link sub-engine. Based on this, a scheduling sub-engine C is generated. Finally, A, B, and C are integrated into a unified workflow. When new hazard point data is input, the scheduling sub-engine triggers the A generation strategy in sequence. After A is output, B automatically verifies whether it contains drainage recommendations. If it is missing, it triggers recalculation and alarms. No manual intervention is required throughout the process.

[0063] This embodiment provides a method for managing geological disaster hazard incident sites. It obtains management target characteristics, core indicator characteristics, and expected output characteristics based on the dimensional descriptions of each management dimension, and inputs these into a management model to generate management algorithm parameter templates. By fine-tuning the parameter templates through dimensional examples, each specialized management sub-engine accurately adapts to specific geological risk scenarios, significantly improving the scenario adaptability and accuracy of the strategy. By transforming dimensional management standards into executable verification rules and constructing independent verification sub-engines, it achieves, for the first time in geological disaster management, the mandatory and dynamic embedding of disaster prevention standards during system operation, ensuring 100% compliance of strategy outputs. Through in-depth mining of inter-dimensional dependencies, it automatically generates management task scheduling sub-engines, achieving temporal coordination, data flow control, and anomaly circuit breaking between sub-engines, constructing an end-to-end automated decision-making closed loop. Finally, the three types of sub-engines operate collaboratively under a unified framework, forming a comprehensive management engine with four levels of intelligence: perception, decision-making, verification, and scheduling. This enables the system to have dynamic evolution, self-correction, and standard-inherent capabilities, completely solving the three fundamental problems of decision lag, standard disconnect, and lack of coordination in traditional methods, becoming the core technical architecture for achieving fully automated, intelligent, and standardized management.

[0064] In one embodiment, an automated management strategy for geological disaster hazard points is generated based on a preset management process and a comprehensive management engine for hazard and accident points, including: The pre-defined management process is formally described to obtain the management stages, management nodes, and triggering conditions between management nodes. The management stage represents the execution steps of the management task, the management node represents the execution unit of each sub-engine, and the triggering condition represents the data transfer rules between sub-engines.

[0065] The management phase can be a high-level step unit within a pre-defined management process, representing the logical order of management task execution. Each phase corresponds to a set of collaboratively completed management behaviors and can be used to provide a structured timeline framework for the management process, defining the macro-temporal boundaries of sub-engine execution. In this embodiment, the management phase can be described using a process modeling language (BPMN) or a state machine, decomposing the expert experience process into discrete phases such as data preprocessing, risk assessment, strategy generation, compliance verification, and output release.

[0066] A management node can be the smallest functional node corresponding to a specific sub-engine execution unit in the management process. Each node is bound to a dedicated management sub-engine, management result verification sub-engine, or management task scheduling sub-engine, which can be used to achieve a precise mapping between the management process and the system engine, ensuring that each process action is executed by a specific module. In this embodiment, the management node can be a monitoring and response phase, a multi-dimensional analysis phase, a strategy generation phase, etc.

[0067] Triggering conditions can be logical criteria defining whether data transfer and execution initiation are allowed between management nodes. These criteria are dynamically determined based on the output results of preceding nodes or external states, enabling condition-driven management processes and avoiding ineffective execution or resource waste caused by fixed timing. In this embodiment, triggering conditions could be Boolean logic conditions such as initiating a verification node if the preceding node outputs a warning level ≥ yellow, or skipping risk assessment if the geological environment data missing rate > 10%.

[0068] Based on the management stage, management node, and triggering conditions, determine the execution order and data interface of each special management sub-engine, each management result verification sub-engine, and each management task scheduling sub-engine.

[0069] Among them, the data interface can be a standardized structure definition for data transmission between management nodes, including input parameter format, output field type and data dependency relationship, which can be used to ensure semantic consistency and system compatibility of data exchange between sub-engines and avoid process interruption caused by format mismatch. In this embodiment, the data interface can be based on the API documents and input / output specifications of each sub-engine, defining JSONSchema or Protobuf structure, and specifying contracts such as "Input: Hidden danger feature vector

[128] ; Output: Warning level {0, 1, 2, 3}".

[0070] Based on the execution order and data interface, configure the running parameters of the comprehensive management engine for potential geological hazards and accidents, and generate an automated management strategy for potential geological hazards and accidents.

[0071] Among them, the running parameters can be a set of executable instructions configured for the comprehensive management engine of hidden danger and accident points based on the formal description of the management process. These instructions include the node calling order, interface binding relationship and trigger condition logic, which can be used to directly drive the automated operation of the comprehensive management engine and realize the seamless transformation from process design to system execution.

[0072] In this embodiment, the running parameters can be uniformly encoded into a workflow definition file (such as YAML / JSON) for management stages, nodes, triggering conditions and data interfaces, and loaded into the engine scheduler as runtime configuration. Taking the generation of automated management strategy for landslide groups as an example, the system reads the preset management process document and formalizes it into three stages: monitoring response, multi-dimensional assessment and strategy output; each stage contains three management nodes: data verification node, evolution analysis node and verification output node; the triggering conditions include skipping evolution analysis if displacement data is missing, and retrying strategy generation if the verification node returns non-compliant; the data interface is defined as the output of the evolution analysis node: hidden danger feature vector

[128] , and the input of the verification node: warning level and suggested content; based on this, the system automatically generates running parameters: the node order is [data verification, evolution analysis, standard verification], the triggering conditions are bound to the jump logic between nodes, and the data interface is mapped to each sub-engine API; finally, an executable automated management strategy is generated. When new hidden danger data arrives, the engine automatically calls the special sub-engine and verification sub-engine according to this configuration and outputs the compliance strategy without human intervention.

[0073] This embodiment provides a method for managing geological disaster hazard incident sites. By formalizing the preset management process into management stages, management nodes, and triggering conditions, it achieves machine-readable management logic for the first time in the field of geological disaster management, transforming the process documents that originally relied on manual execution into a computable structured model. By accurately mapping the execution sequence and data interface, the special management sub-engine, management result verification sub-engine, and management task scheduling sub-engine achieve semantic-level linkage, constructing an intelligent workflow with dynamic response and condition-driven capabilities. Finally, through the automatic generation of operating parameters, a seamless closed loop from process design to system execution is completed, making the automated management strategy no longer a static script, but a living decision-making system with context awareness, condition judgment, and anomaly feedback capabilities. This mechanism completely solves the fundamental contradictions in traditional management, such as the disconnect between processes and systems, the reliance on manual judgment for execution, and the inability to embed standards into processes. It enables industry disaster prevention standards to be internalized into the engine's operating rules through triggering conditions and verification nodes, achieving closed-loop control where processes are standards and execution is compliance. This marks a paradigm revolution in geological disaster hazard management, moving from manual process-driven to intelligent process-as-a-service.

[0074] In one embodiment, the professional drawing documents are parsed based on image recognition technology, including: Acquire image data of professional drawing files, and use a drawing type recognition tool to determine the drawing type of the image data and generate a drawing type identifier. The drawing type includes at least one of the following: geological profile map, contour map, and disaster zoning map.

[0075] Among them, professional drawing files can be geological engineering drawings stored in image or vector format to carry high-density, high-precision information on geological structures and hazard distribution. They can serve as an important original data source for geological hazard analysis. Acquisition methods include obtaining drawing image data in formats such as PDF, DWG, and JPG through scanners or digitization platforms, which can then be used as input for image recognition. The drawing type recognition tool can be a machine learning-based classification model used to automatically determine the geological map type to which the professional drawing image belongs. It can achieve adaptive startup of the parsing process, ensuring that subsequent image recognition parsers match the optimal processing strategy according to the map type.

[0076] Based on the drawing type identifier, the corresponding graphic recognition parser is invoked to perform vectorization processing and feature recognition on the image data, extracting topographic features, geological structure features, and hazard point annotation features from the drawing.

[0077] In this embodiment, based on the type label output by the drawing type identification tool, a pre-set image recognition parser is dynamically invoked to perform targeted image processing and feature extraction on the image. For example, when identified as a geological profile, a rock stratum interface tracking module based on Canny edge detection and Hough transform is invoked to extract rock stratum boundaries and label lithology codes; when identified as a contour map, an elevation line vectorization module based on gradient analysis and contour line connectivity algorithms is invoked to generate a contour line topology network; when identified as a disaster zoning map, a region segmentation module based on K-means clustering and semantic masking is invoked to identify red / yellow / blue risk zones and associate them with preset disaster level labels. This step enables automated feature extraction from different types of professional drawings, eliminating the inefficiency and subjectivity of manual transfer and annotation.

[0078] The accuracy of the parsed topographic features, geological structures, and hazard point marking features is verified by using preset feature verification rules. The positional deviations and attribute errors in the features are corrected by comparing them with field survey data.

[0079] In this embodiment, the elements output by graphic recognition are spatially matched and their attributes compared with independently collected field survey points and measured data to identify and correct abnormal elements that do not conform to the verification rules. For example, the identified fault lines are matched with the GNSS measured fault trajectories using a buffer zone. If the fault line offset exceeds 5 meters, it is corrected by translation based on the measured points. Hazard markers are overlaid with slope layers. If the markers are located in areas with a slope <15°, they are marked as suspected mismarks and sent for manual review. Logical verification is performed on the dip angle attributes of rock strata. If a rock stratum is marked with a dip angle of 85° in the profile, but its adjacent rock strata have a dip angle of 15° and are not separated by a fault, a dip angle abrupt change alarm is triggered, and it is recommended to correct it by interpolation based on regional trends. This step can significantly improve the spatial accuracy and geological rationality of the map analysis results and reduce subsequent analysis deviations caused by recognition errors.

[0080] The verified and corrected elements are converted into a standardized spatial vector format and linked to the corresponding geographic coordinates to generate structured drawing data.

[0081] In this embodiment, the identified topographic, structural, and hazard elements are converted into GeoJSON or Shapefile format, bound to the WGS84 geographic coordinate system, and attribute fields such as lithology, dip angle, and label type are added. This step converts the identified topographic, structural, and hazard elements into GeoJSON or Shapefile format, binds them to the WGS84 geographic coordinate system, and adds attribute fields such as lithology, dip angle, and label type, providing a computable and associative spatial element foundation for the geological hazard knowledge base, supporting subsequent feature extraction and evolutionary modeling.

[0082] The structured drawing data is cross-data source consistency verification is performed with the analysis results of time-series monitoring data and unstructured survey text. Based on the verification results, fuzzy elements are manually labeled and corrected to generate the target analysis results of professional drawing files.

[0083] In this embodiment, spatial coordinate alignment is used to compare the semantic consistency between drawing elements and monitoring data trends and text descriptions. Conflicting or ambiguous areas trigger a manual review process, resulting in a reliable final analysis output. For example, if a drawing labels "Landslide Hazard Point A" at the X coordinate, but the GNSS displacement rate at this point has been below 0.1 mm / d for the past three months, while the displacement rate within 20 meters is >5 mm / d, the system marks this label as a location offset and prompts manual verification to determine if it should be corrected to a high displacement area. If the survey text mentions the existence of a south-sloping weak interlayer, but this depth is not marked on the drawing profile, a text-drawing semantic inconsistency alarm is triggered, prompting manual supplementation of the label and association of the evidence chain. This step enables logical closed-loop verification of drawing data and other multi-source information, ensuring that the target analysis results possess cross-data source semantic consistency and engineering credibility.

[0084] Taking the intelligent processing of geological profile maps in a landslide area as an example, the system receives a geological profile map in PDF format. The map type recognition tool determines that it is a geological profile map and automatically calls the profile map element extraction module to identify 4 rock layer interfaces and 2 fault lines. The element verification rules found that one of the fault lines was offset by 8.3 meters from the center of the crack distribution monitored by the field GNSS, and its dip angle was marked as 75°, but the dip angle of the adjacent rock layer was 22°, which violated the geological continuity rules. The system automatically pushed it to the manual review interface. After manual verification, the fault position was corrected to the crack-dense area and the dip angle was adjusted to 68°. Subsequently, the corrected fault line was compared with the rainfall monitoring data of the same period and it was found that the displacement of the section where the fault was located increased sharply after heavy rainfall, which was consistent with the description of water-rich softening of the fault zone in the text report. The system confirmed the semantic consistency, generated the target parsing result, and bound its coordinates to the hidden danger accident point management knowledge base, which became the key input for risk transmission link analysis.

[0085] This embodiment provides a method for managing geological hazard accident sites. Through steps such as acquiring image data from professional drawing documents, drawing type identification, graphic recognition and parsing, element verification, structured data generation, and cross-data source consistency verification, the method automatically classifies geological profiles, contour maps, and hazard zoning maps using a drawing type identification tool. It drives a graphic recognition parser to perform optimal vectorization processing according to map type, achieving, for the first time, large-scale automated extraction of geological elements from professional drawings. By combining element verification rules with field survey data, the method performs dual correction of spatial location and geological attributes on the identification results, significantly improving the accuracy and geological rationality of graphic parsing. Through cross-data source consistency verification... Source consistency verification semantically aligns and resolves conflicts between the drawing analysis results and the time-series monitoring data and exploration texts, forcing ambiguous or contradictory elements into the manual review process, forming target analysis results with spatial consistency, logical self-consistency, and engineering credibility. This process transforms the originally isolated and unstructured drawing data into calculable, verifiable, and associative structured geological assets, upgrading professional drawings from static maps to core components of a dynamic knowledge base. It provides high-precision and high-reliability spatial semantic support for the generation of hazard feature vectors, the construction of risk transmission links, and the compliance verification of management strategies, and is the underlying technical guarantee for achieving deep integration of multi-source heterogeneous data and intelligent decision-making.

[0086] In one embodiment, based on the hazard feature vectors and metadata in the hazard incident point management knowledge base, the hazard correlation degree between each data item in the management knowledge base is calculated, including: The vectors corresponding to the disaster stage of each data item are extracted from the hazard feature vector, and the vectors of each data item are aggregated into a data item-level hazard vector by a weighted average algorithm.

[0087] The hazard feature vector can be a high-dimensional numerical vector characterizing the potential risk state of a specific hazard point, containing semantic dimensions such as evolution trend, transmission potential, and instability probability. It can serve as the semantic basis for disaster stage identification and correlation calculation, supporting the quantitative comparison of risk similarity between data items. In this embodiment, for each data item, the sub-vector components corresponding to the disaster stage are extracted from its hazard feature vector, and a weighted average is performed based on the stage confidence level to form a single comprehensive vector. This may include, but is not limited to, deformation evolution vectors, hydraulic coupling vectors, and structural instability vectors.

[0088] The first correlation value between the hidden danger vectors at the data item level is calculated based on the cosine similarity algorithm, and the second correlation value is calculated based on the set of hidden danger indicators in the metadata.

[0089] The first correlation value can be calculated based on the cosine similarity of the hazard feature vectors, reflecting the geometric similarity between two data items in the morphology of disaster evolution. It can be used to capture the morphological consistency of geological disasters during their physical evolution, such as the similarity of displacement patterns and stress distribution trends. In this embodiment, for any two data item-level hazard vectors, the cosine value of their included angle is calculated as a quantitative indicator of morphological evolution similarity. A sliding window normalization is introduced, and the similarity is calculated after local normalization of the vectors within the spatial neighborhood to eliminate the influence of regional scale differences.

[0090] The first correlation value and the second correlation value are weighted and merged according to the preset weight allocation strategy to generate the potential correlation between each data item.

[0091] The weighted fusion strategy can be a calculation rule or weight allocation mechanism used to determine the proportion of the first and second correlation values ​​in generating the final hazard correlation score. It can be used to balance the contribution weights of morphological similarity and semantic overlap, ensuring that the correlation score calculation takes into account both physical evolution and management logic. In this embodiment, the first and second correlation values ​​are linearly combined according to preset weights to output a single hazard correlation score. A fixed weight can be used: Hazard Correlation Score = 0.6 × First Correlation Value + 0.4 × Second Correlation Value, suitable for areas with stable geological environments. A dynamic weight model can be used: inputting the regional geological type (e.g., rock slope / loess landslide) and the completeness of monitoring data, and outputting the optimal weight combination through a lightweight neural network. Taking the clustering and dimensional classification of regional landslide hazards as an example, the system decomposes the hazard feature vectors of 150 hazard points in a mountainous area into catastrophic stages, extracts the acceleration phase sub-vectors, and generates 150 data item-level hazard vectors by weighted averaging based on confidence levels. The cosine similarity between any two points is calculated to obtain the first correlation value matrix. At the same time, key indicators (such as displacement rate, rainfall intensity, and joint density) are extracted from the metadata of each point, and the Jaccard overlap rate is calculated as the second correlation value. A regional adaptive weighting strategy is adopted (the weights for rock slope areas are 0.7:0.3), and the two are weighted and fused to generate a hazard correlation matrix. Based on this matrix, the DBSCAN clustering algorithm identifies 5 highly correlated clusters, one of which contains 12 points. The first correlation value of these clusters is concentrated above 0.85, and the second correlation value is greater than 0.7, indicating that their evolutionary form is highly consistent with management needs. Finally, they are identified as the candidate set of bedding-sliding-rainfall-triggered management dimensions.

[0092] This embodiment provides a method for managing geological disaster hazard incident sites. By extracting disaster stage sub-vectors from hazard feature vectors and weighting and aggregating them into data item-level hazard vectors, it achieves a unified representation of multi-stage risk states. It calculates a first correlation value using cosine similarity to accurately capture the similarity of geological disasters in their evolutionary forms; it calculates a second correlation value using indicator overlap rate to effectively identify semantic commonalities at the management strategy level; and it dynamically integrates morphological similarity and management overlap using a preset weighted fusion strategy to generate a hazard correlation degree with dual-dimensional explanatory power, solving the one-sided problem of traditional methods relying solely on a single feature for correlation analysis. This mechanism ensures that the division of management dimensions no longer depends on manual experience but is driven by the inherent evolutionary consistency of data and the overlap of control needs. It provides a scientific, reproducible, and verifiable quantitative basis for subsequent automatic generation of management dimensions such as disaster evolution paths and risk transmission links, and is the core computing engine for achieving the two major goals of systematically characterizing hazard features and automatically mapping standards.

[0093] In one embodiment, management target features, core indicator features, and expected output features are input into the management model to guide the model in generating management algorithm parameter templates corresponding to each management dimension, including: The management objective features are converted into management scenario description statements, the core indicator features are converted into indicator weight constraint lists, and the expected output features are converted into management result format specifications. Based on the preset parameter template generation rules, the management scenario description statement, the indicator weight constraint list, and the management result format specification are sequentially concatenated into an initial parameter configuration sequence. The initial parameter configuration sequence is optimized and logic enhanced by the management model to generate a set of candidate management algorithm parameter templates. Using dimensional examples as validation samples, simulated management results are generated by the management model based on candidate management algorithm parameter templates, and the index consistency between the simulated management results and the dimensional examples and the management trend consistency index are calculated. Candidate management algorithm parameter templates with an indicator matching degree higher than the third preset threshold and a management trend consistency indicator higher than the fourth preset threshold are selected as management algorithm parameter templates corresponding to the management dimensions.

[0094] Among them, the management target features can be semantic target descriptions that characterize the intention of geological disaster management, such as identifying landslide precursors, inhibiting chain propagation, and controlling the radius of influence. These can be used as the original input for generating management scenario description statements, supporting the management model's semantic understanding of management intentions. After semantic restatement and structured rewriting, the management target features are transformed into natural language description statements that can be parsed by the model. For example, when the cumulative rainfall in the bedding joint development area exceeds 80 mm for three consecutive days and the displacement acceleration is >2 mm / d, a level-three warning is activated.

[0095] Core indicator characteristics can be a set of key parameters used to quantify the achievement of goals in the management dimension. They have clear calculation logic and threshold definitions, and can be used as the source of the indicator weight constraint list, determining the priority and sensitivity of the algorithm's response to different monitored variables. Each core indicator is normalized, and weights are assigned based on expert experience or correlation analysis of historical data, forming a key-value pair list such as {displacement rate: 0.6, pore water pressure: 0.3, rainfall intensity: 0.1}.

[0096] Expected output features can be standardized management results that management dimensions should produce under ideal operating conditions, such as warning levels, handling suggestions, and inspection frequency. These can be used to provide structural templates for management result format specifications, ensuring compatibility with downstream systems. The structure of expected output features is parsed into a JSON Schema or field mapping table, forming a standardized output format specification.

[0097] In this embodiment, management target features are converted into management scenario description statements, core indicator features are converted into indicator weight constraint lists, and expected output features are converted into management result format specifications. Semantic structuring processing is performed on the three types of features respectively, transforming them into three standardized expressions that can be parsed by machines. This achieves accurate transformation from abstract management intentions to computable semantic expressions, eliminating semantic ambiguity.

[0098] Based on the preset parameter template generation rules, the management scenario description statement, the indicator weight constraint list and the management result format specification are concatenated in sequence to form the initial parameter configuration sequence. The three types of structured content are merged into a single input sequence according to a fixed syntax order for unified processing by the management model. A unified input interface is established, enabling the management model to reuse the same processing flow across dimensions and improve its generalization ability.

[0099] By optimizing the initial parameter configuration sequence and enhancing the logic through the management model, a set of candidate management algorithm parameter templates is generated. The optimal configuration combination is searched in the parameter space using a deep learning model to generate multiple feasible algorithm templates, thereby realizing the automated generation and diversity exploration of management strategies and replacing manual intervention and rule writing.

[0100] Using dimensional examples as validation samples, the management model generates simulated management results based on candidate management algorithm parameter templates, and calculates the index consistency and management trend consistency index between the simulated management results and the dimensional examples. For each candidate template, historical dimensional example data is input to generate simulated output, which is compared with real examples to calculate two validation indicators, thereby achieving dual scientific validation of the candidate templates and ensuring that the strategy conforms to both historical experience and geological laws.

[0101] Candidate management algorithm parameter templates with an indicator matching degree higher than the third preset threshold and a management trend consistency index higher than the fourth preset threshold are selected as management algorithm parameter templates corresponding to the management dimensions. The candidate template set is filtered by double thresholds, and only templates that simultaneously meet the requirements of numerical matching and physical rationality are retained. This ensures that the final template has both data fitting ability and physical interpretability, and eliminates false associations caused by pure statistical fitting.

[0102] Taking the automatic generation of debris flow risk dimension parameter templates as an example, the system receives the management target features of debris flow triggering-hydraulic coupling type dimension to suppress the activation of gully sources. The core indicator features are the proportion of loose solids volume, rainfall intensity, and gully water content. The expected output features are red warning and blocking suggestions. The system converts this into a scenario statement: "When the proportion of loose material sources in the gully is >40% and the 24-hour rainfall is >60mm and the water content continues to rise, initiate the material source blocking project." The system generates an indicator weight list: {Material source proportion: 0.5, Rainfall: 0.3, Water content: 0.5}. 2}; The output format is JSON {Level: Red, Measures: [Dredging, Deployment of Barrier Dams]}; Concatenate them to form the initial sequence input management model; The model generates 5 candidate templates: including decision trees, random forests, GNNs, etc.; Validate using 12 historical debris flow examples, among which 3 templates have a consistency index > 0.87 and a trend consistency > 0.82, which conforms to the typical path of geological disasters such as sudden increase in rainfall, water saturation, and unstable source material; Finally, the one with the highest trend consistency is selected as the management algorithm parameter template for this dimension, which is used to train the special management sub-engine.

[0103] This embodiment provides a method for managing geological disaster hazard sites. It transforms management target characteristics, core indicator characteristics, and expected output characteristics into management scenario description statements, indicator weight constraint lists, and management result format specifications, respectively. These are then concatenated into an initial parameter configuration sequence using preset rules, constructing a unified input interface that enables the management model to be reused across dimensions. The management model optimizes and enhances the sequence's logic, generating diverse candidate templates and overcoming the rigid limitations of traditional manual design strategies. Most importantly, it introduces indicator fit and management trend consistency indicators as dual verification standards. The former ensures accurate consistency between the output and historical experience, while the latter verifies the consistency between the output and historical experience. The comparison of evolutionary paths prevents statistical noise from misleading the model and ensures that the model deduction conforms to the inherent mechanism of geological disasters. Finally, through a dual-threshold screening mechanism, the optimal parameter template that conforms to both data patterns and physical laws is automatically locked, so that each management algorithm parameter template has both data-driven and scientific interpretability. This mechanism internalizes industry standards into quantifiable and verifiable algorithm constraints, enabling the generation of management strategies to leap from trial and error based on expert experience to an intelligent closed loop driven by mechanisms and verified by data. It is the core engine for achieving the two major technical effects of accurately characterizing the features of hidden dangers and automatically mapping standards, and completely solves the problems of traditional methods in strategy generation relying on subjective experience, lacking objective verification, and being difficult to generalize.

[0104] In one embodiment, when locating key hazard entities in a structured management data source using predefined geological entity identification rules, the method further includes: Identify monitoring sequences with catastrophic evolution characteristics in structured management data sources, and extract the deformation stages in the monitoring sequences as entity evolution labels.

[0105] The monitoring sequence can be a time-series dataset used to record the deformation state of geological entities, which can be used to characterize the state evolution process of a single potential hazard entity in the time dimension. In this embodiment, by applying segmented clustering and mutation detection algorithms, stage nodes such as stable period, acceleration period, and critical period are extracted, and causal edges are established based on the dependency relationship between the preceding and following states to form a directed path with time-series labels, thus realizing the time-series catastrophe path.

[0106] Analyze the time series and causal relationships of each deformation stage within the monitoring sequence, and link the time series and causal relationships to the entity evolution tags to form a time-seriesd disaster path.

[0107] The input to the time-series catastrophe path is the monitoring sequence from a structured data source, and the output is a structured stage sequence and causal relationship graph. In this embodiment, by analyzing the time series and causal relationships of each deformation stage, the time series and causal relationships are associated with entity evolution tags to form a time-series catastrophe path.

[0108] Detect similar abnormal indicators across monitoring points in structured management data sources, and track the spatial impact range of key hidden danger entities based on similar abnormal indicators.

[0109] The spatial impact range can be the geographic boundary of the impact of a certain hidden danger entity, revealed by similar anomaly indicators across monitoring points. It can be used to characterize the spatial spread of a disaster. In this embodiment, similar anomaly indicators (such as sudden increases in displacement or crack opening) that occur simultaneously in multiple monitoring points are identified through spatial interpolation and anomaly pattern clustering. The boundary of the impact area is delineated by combining geographic proximity and geological connectivity, thus forming the spatial impact range.

[0110] By mapping the temporal disaster path and spatial impact range to a pre-defined disaster evolution map, a multi-dimensional disaster behavior model of key hidden entities is generated.

[0111] A multi-dimensional disaster behavior model can be a composite model that integrates temporal disaster paths and spatial impact ranges, representing the comprehensive disaster behavior patterns of key hazard entities across both temporal evolution and spatial propagation dimensions. In this embodiment, the temporal disaster path is used as the time axis dimension, and the spatial impact range is used as the spatial dimension. Both are input into a preset disaster evolution map for pattern matching, outputting standardized behavior template codes to generate a multi-dimensional disaster behavior model.

[0112] Based on a multi-dimensional disaster behavior model, entity attribution is performed on interrelated hidden danger events in structured management data sources to eliminate duplicate disaster event nodes.

[0113] This embodiment provides a method for managing geological disaster hazard incident sites. By identifying deformation stages in the monitoring sequence to construct a time-series disaster path, it achieves semantic stage labeling of the hazard evolution process for the first time. By tracking similar anomalies across points, it constructs the spatial impact range, breaking through the local limitations of single-point monitoring and depicting the spatial diffusion boundary of the disaster. Both are mapped to a preset disaster evolution map to generate a multi-dimensional disaster behavior model with evolutionary logic and spatial constraints, transforming the originally isolated monitoring events into dynamic behavior units that can be standardized and matched. Based on this model, it performs semantic-level attribution and merging of repeated events in the structured management data source, completely eliminating the risk of redundant records and misjudgments caused by data fragmentation. This mechanism upgrades the hazard incident site management knowledge base from a static data set to a dynamic evolutionary reasoning engine with the ability to identify behavior, match patterns, and deduplicate entities. It provides an entity foundation with causal explanatory power and uniqueness guarantee for the subsequent accurate division of multi-dimensional management dimensions and the generation of automated strategies, which is a key leap from event recording to behavioral cognition.

[0114] Taking multi-point collaborative early warning of regional slope clusters as an example, for instance, 32 monitoring points were continuously deployed in mountainous areas, and multiple points showed accelerated displacement and crack opening within two weeks. The system extracted the deformation stage sequence of each point and identified stable, accelerated, and critical time-series disaster paths. At the same time, it was found that the time difference of the sudden increase in displacement of 5 adjacent points was less than 4 hours, and the deformation area was connected in a strip shape, delineating the spatial influence range as a strip-shaped area with a length of 2.1km and a width of 300m. The path and range were input into the disaster evolution map and matched with the bedding landslide chain initiation behavior template to generate a unified multi-dimensional disaster behavior model. Based on this, the system attributed the 7 previously scattered independent landslide early warning events to the multi-point response of the same entity, merged them into 1 hidden danger entity, and updated its evolution status to high-risk chain, avoiding duplicate early warning pushes and resource waste.

[0115] Furthermore, this embodiment of the invention also proposes a storage medium storing a geological hazard accident site management program, which, when executed by a processor, implements the steps of the geological hazard accident site management method described above.

[0116] Furthermore, this invention also proposes a geological disaster hazard accident site management system, the system comprising: processor; Memory used to store the processor's executable instructions; The processor is configured to execute the instructions to implement the steps of the geological disaster hazard accident site management method as described in any of the above.

[0117] Other embodiments or specific implementations of the geological disaster hazard accident site management system described in this invention can refer to the above-mentioned method embodiments, and will not be repeated here.

[0118] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A method for managing potential geological disaster sites, characterized in that, The method includes: Based on the hazard-related data and information collected from the geological disaster monitoring area, a knowledge base for hazard accident point management is constructed. The hazard-related data and information include at least one of the following: historical disaster records, geological environment data, information sensed by monitoring equipment, and industry disaster prevention standards and specifications. By identifying and analyzing the knowledge base for the management of potential geological hazards and accidents, multiple management dimensions are obtained. Different management dimensions are used to comprehensively manage different aspects of potential geological hazards and accidents. By integrating multiple management dimensions, a comprehensive management engine for potential hazards and accident points is constructed. Based on the preset management process and the comprehensive management engine for potential geological hazards and accidents, an automated management strategy for potential geological hazards and accidents is generated. The management process is used to determine the execution order and data interaction method between the sub-engines corresponding to each management dimension in the management engine.

2. The method for managing geological disaster hazard sites as described in claim 1, characterized in that, The aforementioned knowledge base for the management of potential hazard incident sites is constructed based on hazard-related data and information collected from geological disaster monitoring areas, including: The acquired hazard-related data and information are analyzed, and the analysis results are standardized based on a multi-source data fusion tool to generate a structured management data source. The hazard-related data and information include at least one of the following: monitoring station sensor data, UAV remote sensing imagery, geological survey reports, and historical disaster databases. The hazard-related data and information include at least one of the following: time-series monitoring data, unstructured survey text, and professional drawing files. The time-series monitoring data is analyzed based on field extraction rules, the unstructured survey text is analyzed based on natural language processing, and the professional drawing files are analyzed based on image recognition technology. Key hazard entities in the structured management data source are located using predefined geological entity identification rules. Based on the frequency of occurrence, hazard level, and spatial correlation of entities in the data source, the metadata of the management knowledge base is determined from the key hazard entities. A pre-trained geological feature extraction model is used to obtain the disaster feature units and hidden danger feature units included in the structured management data source. Based on the geological disaster evolution model, the risk connotation of the disaster feature units and hidden danger feature units is further mined according to the spatial context information of the disaster feature units and hidden danger feature units, so as to map the hidden danger information in the data source into the corresponding hidden danger feature vector. By using geological data association technology, the original records of the data source, the determined metadata, and the generated hazard feature vectors are associated and integrated through the spatial coordinate identifiers of each data item in the structured management data source to generate the hazard accident point management knowledge base.

3. The method for managing geological disaster hazard sites as described in claim 2, characterized in that, The process involves identifying and analyzing the knowledge base for managing potential hazards and accidents, resulting in multiple management dimensions corresponding to the knowledge base, including: Based on the hazard feature vector and the metadata in the hazard accident point management knowledge base, calculate the hazard correlation degree between each data item in the management knowledge base, and divide the data items in the management knowledge base into different management dimension candidate sets based on the hazard correlation degree; The disaster evolution path characteristics and impact range propagation characteristics of the data items included in the candidate sets of each management dimension are analyzed to determine the impact area and risk transmission link of different disaster stages in each data item. Key management indicators are extracted by identifying the disaster type, affected objects and triggering factors in each data item. The key management indicators are used to clarify the core control description information of different stages in each data item. Based on the affected area, the risk transmission link, and the key management indicators, the data items in each of the candidate sets of management dimensions are summarized and a dimension description of each candidate set of management dimensions is generated. The dimension description is used to reflect the role and function of the dimensions in the candidate sets of management dimensions in the management of hidden danger and accident points. Based on the dimension descriptions in each of the management dimension candidate sets, the data item content in each of the management dimension candidate sets, and the preset management specifications, target data fragments in each of the management dimension candidate sets are determined as dimension examples of each of the management dimension candidate sets; Based on the dimension descriptions and dimension examples in each of the candidate sets of management dimensions, dimension management standards for each of the candidate sets of management dimensions are determined from the preset disaster prevention standards. The dimension management standards are used to clarify the management accuracy and timeliness of each dimension in the candidate sets of management dimensions. Based on the consistency of dimension descriptions, dimension examples, and dimension management standards corresponding to each dimension in each candidate set of management dimensions, multiple management dimensions corresponding to the hidden danger and accident point management knowledge base are determined in each candidate set of management dimensions.

4. The method for managing geological disaster hazard sites as described in claim 3, characterized in that, The comprehensive management engine for hidden danger and accident points includes a management task scheduling sub-engine, a special management sub-engine corresponding to each management dimension, and a management result verification sub-engine. The integration of multiple management dimensions to construct a comprehensive management engine for potential hazard and accident points includes: Based on the dimension description of each management dimension, the management target features, core indicator features, and expected output features of each management dimension are obtained, and the management target features, core indicator features, and expected output features are input into the management model to guide the management model to generate management algorithm parameter templates corresponding to each management dimension. Based on the management algorithm parameter template and the dimension examples of the management dimensions, the initial management model is trained and optimized to generate specialized management sub-engines corresponding to each management dimension. The dimension management standards corresponding to each of the management dimensions are transformed into executable verification rules to construct verification algorithms corresponding to each of the dimension management standards, and management result verification sub-engines for the corresponding management dimensions are generated based on each of the verification algorithms. By deeply exploring the overall management framework of geological disaster hazard points and the dependencies between each management dimension, and based on the overall management framework and the dependencies, the collaborative workflow of each specialized management sub-engine and each management result verification sub-engine is determined, and a management task scheduling sub-engine related to the collaborative workflow is generated. Based on the management task scheduling sub-engine, the special management sub-engine corresponding to each management dimension, and the management result verification sub-engine corresponding to each management dimension, the comprehensive management engine for hidden danger and accident points is constructed.

5. The method for managing geological disaster hazard sites as described in claim 4, characterized in that, The process of generating an automated management strategy for geological disaster hazard points based on a preset management process and the comprehensive management engine for hazard points includes: By formally describing the preset management process, the management stages, management nodes, and triggering conditions between the management nodes in the management process are obtained. The management stage represents the execution steps of the management task, the management node represents the execution unit of each sub-engine, and the triggering condition represents the data transfer rules between sub-engines. Based on the management stage, the management node, and the triggering condition, determine the execution order and data interface of each of the special management sub-engines, each of the management result verification sub-engines, and the management task scheduling sub-engine; Based on the execution order and data interface, configure the operating parameters of the comprehensive management engine for potential geological hazards and accidents, and generate an automated management strategy for potential geological hazards and accidents.

6. The method for managing geological disaster hazard sites as described in claim 2, characterized in that, The professional drawing files are parsed based on image recognition technology, including: The image data of the professional drawing file is acquired, and the drawing type is determined by the drawing type recognition tool to generate a drawing type identifier. The drawing type includes at least one of geological profile map, contour map, and disaster zoning map. Based on the drawing type identifier, the corresponding graphic recognition parser is invoked to perform vectorization processing and feature recognition on the image data, and to extract the topographic features, geological structure features, and hazard point annotation features from the drawing. The accuracy of the parsed topographic features, geological structures, and hazard point marking features is verified using preset feature verification rules. The positional deviations and attribute errors in the features are corrected by comparing them with field survey data. The verified and corrected elements are converted into a standardized spatial vector format and linked to the corresponding geographic coordinates to generate structured drawing data. The structured drawing data is subjected to cross-data source consistency verification with the time-series monitoring data and the parsing results of the unstructured survey text. Based on the verification results, the fuzzy elements in the structured drawing data are manually annotated and corrected to generate the target parsing result of the professional drawing file.

7. The method for managing geological disaster hazard sites as described in claim 3, characterized in that, The step of calculating the correlation degree between data items in the management knowledge base based on the hazard feature vector and the metadata in the hazard accident point management knowledge base includes: Extract the vector corresponding to the disaster stage of each data item from the hidden danger feature vector, and aggregate the vectors of each data item into a data item-level hidden danger vector by weighted average algorithm; The first correlation value between the data item-level hidden danger vectors is calculated based on the cosine similarity algorithm, and the index overlap rate between the data items is calculated as the second correlation value based on the hidden danger index set in the metadata. The first correlation value and the second correlation value are weighted and fused according to a preset weight allocation strategy to generate the potential correlation degree between each data item.

8. The method for managing geological disaster hazard sites as described in claim 4, characterized in that, The step of inputting the management target features, the core indicator features, and the expected output features into the management model to guide the management model in generating management algorithm parameter templates corresponding to each management dimension includes: The management objective features are converted into management scenario description statements, the core indicator features are converted into indicator weight constraint lists, and the expected output features are converted into management result format specifications. Based on the preset parameter template generation rules, the management scenario description statement, the indicator weight constraint list, and the management result format specification are sequentially concatenated into an initial parameter configuration sequence. The initial parameter configuration sequence is optimized and logic enhanced using the management model to generate a set of candidate management algorithm parameter templates. Using the dimensional example as a validation sample, the management model generates simulated management results based on the candidate management algorithm parameter template, and calculates the indicator consistency and management trend consistency indicators between the simulated management results and the dimensional example. Candidate management algorithm parameter templates with an indicator matching degree higher than the third preset threshold and a management trend consistency indicator higher than the fourth preset threshold are selected as the management algorithm parameter templates corresponding to the management dimension.

9. The method for managing geological disaster hazard sites as described in claim 2, characterized in that, When locating key hazard entities in the structured management data source using predefined geological entity identification rules, the method further includes: Identify monitoring sequences with catastrophic evolution characteristics in the structured management data source, and extract the deformation stages in the monitoring sequences as entity evolution labels; Analyze the time series and causal relationship of each deformation stage within the monitoring sequence, and associate the time series and causal relationship with the entity evolution tag to form a time-series catastrophe path; Detect similar abnormal indicators across monitoring points in the structured management data source, and track the spatial impact range of the key hidden danger entity based on the similar abnormal indicators; The time-series disaster path and the spatial impact range are mapped to a preset disaster evolution map to generate a multi-dimensional disaster behavior model of the key hidden danger entity; Based on the multi-dimensional disaster behavior model, entity attribution is performed on the interrelated hidden danger events in the structured management data source to eliminate duplicate disaster event nodes.

10. A geological disaster hazard incident site management system, characterized in that, The system includes: processor; Memory used to store the processor's executable instructions; The processor is configured to execute the instructions to implement the steps of the geological disaster hazard accident site management method as described in any one of claims 1-9.

Images