Emergency decision support method based on cross-domain knowledge fusion

Abstract

The emergency decision support method based on cross-domain knowledge fusion comprises the steps of building an emergency semantic model, building an emergency response graph, building an emergency decision auxiliary model based on the graph and the like, a full-cycle emergency disposal knowledge graph oriented to multiple disaster types is provided, and various emergency disposal knowledge can be effectively integrated and stored. Meanwhile, an emergency decision-making auxiliary model is constructed on the downstream based on the atlas, disaster grade division, multi-hop path query analysis and decision-making recommendation can be carried out according to real-time multi-dimensional disaster data, and the cross-domain text retrieval efficiency is greatly improved; meanwhile, a plan chain recommendation model based on an event chain is provided, disaster chain prediction can be carried out based on rules, refined plan pushing can be given, and a vulnerable community is helped to make emergency risk avoiding and planning in advance in a complex scene.

Description

Technical Field

[0001] The present invention relates to the field of knowledge graph technology, and in particular to an emergency decision support method based on cross-domain knowledge fusion. Background Art

[0002] Accidents and disasters are characterized by multiple types, rapid onset, high frequency, wide distribution, and strong destructiveness. With the massive accumulation of data, it is inevitable that problems such as inconsistent data representation, insufficient data relevance, and inefficient cross-domain retrieval will arise.

[0003] Knowledge graph technology has gradually become an important branch of artificial intelligence. Knowledge graph is a semantic network that expresses various semantic relationships between entities / concepts, provides necessary background knowledge for understanding natural language, and is a key common basic technology for crossing the semantic gap and realizing cognitive intelligence. In recent years, many scholars have conducted research on knowledge graphs in the field of emergency management. Among them, in the construction of graphs, EPSOnto established the emergency plan system architecture from a macro perspective based on SUMO ontology. Zhou Yiqi et al. constructed a knowledge graph of power grid emergency plans from the perspective of emergency departments and their responsibilities to improve the efficiency of emergency response to power grid accidents; Qiu Qinjun et al. analyzed the mechanism of geological disasters from the perspective of disaster chain and constructed a knowledge graph of geological disaster chain. He Haixia et al. directly analyzed the elements and relationship construction model layers of earthquake disasters, extracted data from different sources to construct an earthquake disaster knowledge graph, and Liu Yongli and Wang Haitao constructed a coal mine disaster knowledge graph based on departmental emergency response plans to achieve cross-departmental collaborative emergency response methods. In terms of graph application, Kontopoulos E et al. proposed a lightweight climate crisis management ontology to provide support for decision-making in the field of climate crisis management. JedsadaPhengsuwan et al. verified and predicted landslide disasters on a large amount of time series data based on the landslide ontology. The ORFFM model provided intelligent solutions for farmland irrigation and flood warning. Y Jiaod et al. constructed an earthquake disaster knowledge graph and proposed a rescue decision-making algorithm based on knowledge graph reasoning.

[0004] However, the current research on emergency decision-making assistance technology based on knowledge graphs still has the following bottlenecks:

[0005] 1) Single corpus type: Many works only focus on a certain type of text data such as emergency plans, ignoring other disaster classification standards, rescue regulations, and emergency knowledge in historical emergency cases.

[0006] 2) Single disaster type: Most existing knowledge systems only focus on information on a single disaster type and have not yet achieved the construction of an event chain-oriented emergency decision-making knowledge system under the risk of secondary and derivative accidents.

[0007] 3) Rough reasoning granularity: Most existing emergency decision support models are based on keyword matching of documents or paragraphs, such as plan recommendation or paragraph retrieval, which cannot meet the needs of refined reasoning;

[0008] 4) Weak context-response linkage: The decision support mechanisms in previous studies only responded to a single context, ignoring the triggering conditions, evolutionary mechanisms, and the interrelationships between consecutive events, lacking the correlation between events, situations, and actions.

[0009] This application establishes a knowledge representation model by introducing three types of concepts, namely, concrete, abstract and temporal, at the pattern layer, mapping static normative texts and event-driven case data to construct a knowledge graph for multi-disaster, full-cycle emergency response. Based on the ontology framework, an emergency knowledge representation model and semantic specifications are constructed to reorganize and standardize the more fine-grained emergency response elements (such as emergency resources, disaster levels, and key points of response) in multiple types of texts (plans, specifications, and battle examples), so as to achieve more refined decision recommendations throughout the entire emergency response process. By constructing the interaction and alternating evolution of emergency events and emergency situations under constraints in the pattern layer, the correlation and integrity between emergency events are reflected. Finally, based on the multi-disaster emergency response knowledge graph, two types of emergency decision support service models are constructed: 1) efficient cross-file knowledge retrieval 2) disaster chain response plan recommendation, which are applied to actual emergency command platform products to verify the feasibility and effectiveness of the method in this paper. Summary of the invention

[0010] The present invention proposes an emergency decision support method based on cross-domain knowledge fusion, which solves the existing lack of knowledge graphs that can be used for multi-disaster and full-cycle emergency response. The technical solution of the present invention is implemented as follows: an emergency decision support method based on cross-domain knowledge fusion, comprising the following steps:

[0011] Step S1: Constructing emergency semantic model

[0012] 1) Define the emergency semantic model

[0013] Based on the ABC Ontology framework, the four-tuple representation of emergency knowledge is defined as:

[0014] <Concepts,Relations,Rules,Instances> ;

[0015] Among them, Concepts: = {Actualities, Abstractions, Temporalities} represents a concept set; Actualities represents a concrete concept set, describing the reality of the objective world; Abstractions represents an abstract concept set, describing abstract concepts without objective reality; Temporalities represents a temporal concept set, describing concepts with spatiotemporal existence; and each concept contains multiple properties.

[0016] Relations: = {relation, inheritance} represents a set of relations, associated relation

[0017] Mainly including reference, compliance, utilization, execution, initiation, and accompanying non-inheritance relationships. Inheritance refers to the parent-child relationship.

[0018] Rlues:={R:N1*N2*N3*...*N n}→N n+1 |N1,N2,...,N n ∈Concepts∪Relations}

[0019] is a set of emergency decision rules, where the multiplication sign is defined as an association constraint between entities, indicating that the n+1th entity can be uniquely determined based on n entities and relationships in the emergency response system, and Instances represents a set of instances, i.e., the instantiation of concepts and relationships in emergency cases and emergency documents;

[0020] 2) Knowledge representation model construction

[0021] By extracting and dividing the knowledge of emergency plans, rescue specifications, classification standards and emergency cases, a concept set of the knowledge representation model is constructed, which is derived from concrete, abstract and timely classes;

[0022] Step S2: Construction of emergency response map

[0023] Referring to the knowledge representation model framework of the reference mode layer, we first collect and integrate data, extract knowledge from the four types of data, and then map and fuse the extracted knowledge into the knowledge representation model to generate triples. After manual quality assessment, they are imported into the graph database for storage to build a complete knowledge graph.

[0024] Step S3: Construction of emergency decision-making support model based on graph

[0025] Construct an emergency decision support model for efficient cross-file knowledge retrieval and a disaster chain disposal plan recommendation emergency decision support model.

[0026] As an optimal technical solution, the emergency decision-making rule set consists of two categories: grading rules Grades and rescue norms Norms. The grading rules include multiple grading standards such as disaster grade, emergency response grade, and rescue action grade. The rescue norms mainly include action norms and safety warning rules.

[0027] As a preferred technical solution, step S2 specifically includes the following steps:

[0028] Step a: Data acquisition: data are divided into static text data and event data containing spatiotemporal attributes according to data characteristics. The static text data is directly acquired through web crawlers from relevant Wikipedia, official websites of provincial and municipal people's governments, and documents published on public accounts of various emergency organizations, and saved in unstructured text format. The event data containing spatiotemporal attributes is acquired through internal real data and relevant Wikipedia open records;

[0029] Step b: Knowledge extraction and mapping. The knowledge extraction method based on the ChatGLM large language model refers to the pattern layer to determine the corresponding concept entities, attributes and relationships that need to be extracted. A portion of the data is manually annotated as prompt examples. The performance and expression of the large model in the knowledge extraction scenario are improved through context learning. Information extraction is completed from unstructured text. After knowledge extraction, knowledge mapping needs to be combined with the pattern layer.

[0030] Step c: Knowledge storage, using neo4j graph database to support emergency knowledge management and high-performance query, relying on the py2neo toolkit to write data import script 1, and batch import the extracted nodes and attributes CSV into neo4j.

[0031] As a preferred technical solution, the construction of a cross-file knowledge retrieval and reasoning model includes the following steps: by encapsulating the rule-based knowledge stored in the knowledge graph into a function, after extracting information from the user input, the situation information, monitoring indicators and decision-making goals are passed as parameters to the Cypher query statement, and the results are visualized and returned to the user through intelligent analysis and query.

[0032] As a preferred technical solution, the construction of a recommended model for handling plans for multi-hazard event chains includes the following steps: combining the temporal phenomenon class definitions in the model layer, the real-time environmental data and historical monitoring data in the data layer, mathematically modeling the prediction generation process of the event chain, and then the system will match the event level and event type according to the generated event chain, generate the corresponding emergency plan chain based on the graph, and accurately push it to relevant authorities and vulnerable communities; based on the current situation and the historical feedback on relevant action measures in the knowledge graph, the system will also recommend refined proactive risk reduction measures such as chain breaking, personnel evacuation, and rescue material dispatch.

[0033] As a preferred technical solution, the emergency event Ei at the event chain layer includes various attributes of the event and its on-site measurement values, expressed as E i =(V i ), V i That is, the event feature vector, such as disaster characteristics such as rainstorm intensity and range, and the disaster-affected environment situation S i Contains the environmental spatiotemporal context of the current event detected on site, denoted as S i =(Obj i , Env i ), where Obj i Indicates the characteristic vector of the disaster-prone body in the current time and space, such as the soil looseness coefficient of the mountain affected by heavy rain, Env i Represents the characteristic vector of the disaster-prone environment in the current space and time, such as topography and vegetation density. The steps to generate the event chain are as follows:

[0034] (1) The event chain begins, and the original emergency event E1 occurs;

[0035] (2) The field-related attribute values ​​of measurement E1 constitute vector V1;

[0036] (3) Determine the disaster-affected environmental situation S1 based on the changes in the objective environment caused by E1;

[0037] (4) measure the spatiotemporal context data of S1 (Obj1, Env1);

[0038] (5) Predict the next emergency event E1×S1→E2 based on the rule matching of each measurement value of E1 and S1;

[0039] (6) Loop through steps (2)-(5) to predict the subsequent event chain E2→…→E n ;

[0040] (7) The event chain construction process ends when the result of predicting the next possible emergency event during the rule matching process is empty.

[0041] As a preferred technical solution, the concrete class Actuality is used to express objectively existing people and things, involving rescue personnel, organizations, and materials. It is the context of the emergency situation and is connected to the situation through isContext; it includes Agent rescue personnel, Organization emergency organization, Resource emergency support resources, and Document emergency file system.

[0042] As a preferred technical solution, the abstract class Abstraction describes concepts that have no objective reality in the emergency response process, including Warning special warnings and Duty task responsibilities.

[0043] As a preferred technical solution, the temporal phenomenon class is divided into three subcategories: emergency situation, emergency event, and emergency action. It is connected to time and place through atTime and inPlace respectively to describe its temporal and spatial attributes.

[0044] As a preferred technical solution, the emergency semantic model assists the construction of the knowledge representation model through the Web ontology language OWL specification.

[0045] Compared with the existing technology, this solution has the following beneficial effects: constructing knowledge ontology based on concrete, temporal, and abstract categories, proposing a knowledge graph for emergency response for multiple types of disasters and the entire cycle, which can effectively integrate and store various types of emergency response knowledge. At the same time, building an emergency decision-making auxiliary model based on the graph in the downstream, it can classify disaster levels, perform multi-hop path query analysis, and make decision recommendations based on real-time multi-dimensional disaster data, greatly improving the efficiency of cross-domain text retrieval; at the same time, proposing a plan chain recommendation model based on event chains, which can predict disaster chains based on rules and give refined plan push, helping vulnerable communities to prepare for emergency risk avoidance and planning in advance in complex scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0047] Figure 1 A schematic diagram of a knowledge representation model of an emergency decision support method based on cross-domain knowledge fusion according to the present invention;

[0048] Figure 2 It is an emergency action expansion division of an emergency decision support method based on cross-domain knowledge fusion in the present invention;

[0049] Figure 3 This is a graph construction process based on cross-domain knowledge fusion in the present invention;

[0050] Figure 4 A schematic diagram of knowledge extraction and mapping based on cross-domain knowledge fusion in the present invention;

[0051] Figure 5 This is a schematic diagram of emergency knowledge structuring based on cross-domain knowledge fusion according to the present invention;

[0052] Figure 6 The present invention is an emergency decision support flow chart of an emergency decision support method based on cross-domain knowledge fusion;

[0053] Figure 7 The invention provides a model for generating a plan chain for an emergency decision support method based on cross-domain knowledge fusion. DETAILED DESCRIPTION

[0054] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] Reference Figure 1 , Figure 4 and Figure 7 , an emergency decision support method based on cross-domain knowledge fusion, comprising the following steps:

[0056] Step S1: Constructing emergency semantic model

[0057] 1) Define the emergency semantic model

[0058] Based on the ABC Ontology framework, the four-tuple representation of emergency knowledge is defined as:

[0059] <Concepts,Relations,Rules,Instances> ;

[0060] Among them, Concepts: = {Actualities, Abstractions, Temporalities} represents a concept set; Actualities represents a concrete concept set, describing the reality of the objective world; Abstractions represents an abstract concept set, describing abstract concepts without objective reality; Temporalities represents a temporal concept set, describing concepts with spatiotemporal existence; and each concept contains multiple properties.

[0061] Relations: = {Relation, inheritance} represents a set of relations. Relations mainly include reference, compliance, utilization, execution, initiation, and accompanying non-inheritance relations. Inheritance represents the parent-child relationship.

[0062] Rlues:={R:N1*N2*N3*...*N n}→N n+1 |N1,N2,...,N n ∈Concepts∪Relations}

[0063] is a set of emergency decision-making rules, where the multiplication sign is defined as an association constraint between entities, indicating that the n+1th entity can be uniquely determined based on n entities and relationships in the emergency response system. Instances represents a set of instances, i.e., the instantiation of concepts and relationships in emergency cases and emergency documents. The rule set consists of two categories: grading rules Grades and rescue norms Norms. The grading rules include disaster grading, emergency response grading, rescue action grading and other grading standards. The rescue norms mainly include action norms and safety warning rules. Here, a grading rule for forest fire is taken as an example:

[0064]

[0065] 2) Knowledge representation model construction

[0066] By extracting and dividing knowledge of emergency plans, rescue specifications, classification standards and emergency cases, a concept set of the knowledge representation model is constructed, which is derived from concrete classes, abstract classes and image classes; the semantic relationship structure between concepts in the above figure is as follows Figure 1 shown.

[0067] Specifically, the actuality class is used to express objectively existing people and things, including rescue personnel, organizations, and materials. It is the context of the emergency situation and is connected to the situation through isContext. It includes:

[0068] 1) Agent rescuer

[0069] Agent is both a participant in emergency events and an executor of emergency actions.

[0070] 2) Organization

[0071] The subclass extended from Agent represents the group organization participating in the emergency action, including two subclasses: emergency organization and emergency department. The emergency department is the government department involved in the emergency action, such as the agricultural department, water conservancy department, etc. Among them, the emergency organization extends to three subclasses: emergency command organization, emergency linkage organization and expert group. The command organization is responsible for emergency command and overall arrangement, such as the joint defense command center, etc.; the linkage organization is responsible for following the arrangement to carry out on-site rescue and disaster relief operations, such as the fire brigade and the armed police brigade, etc.; the expert group is responsible for using professional knowledge to provide assistance and suggestions based on real-time feedback from the scene. The emergency organization can use various emergency guarantees provided on site to complete the rescue work, which is connected to Resource through uses. At the same time, it must comply with the special warning Warning and clarify its own task responsibilities Duty, which are connected to the two through abides and hasDuty respectively.

[0072] 3) Resource emergency support resources

[0073] It indicates the various resource requirements required by emergency organizations when carrying out emergency actions, such as material support, medical support, life support and communication platform support, etc. It is used to record the reference support quantity currently required and the rescue support quantity prepared in advance, so as to complete the logistics support work in time.

[0074] 4) Document emergency file system

[0075] Contains emergency plans, classification standards, rescue specifications and other emergency document data. The emergency document system will be associated with other concepts. For example, the action specifications for different emergency stages in the rescue specification will be connected to the emergency action with a referTo relationship, so as to provide accurate auxiliary specifications in stages.

[0076] The abstract class Abstraction describes some concepts without objective reality in the emergency response process. It includes:

[0077] 1) Warning

[0078] Warning refers to the specific warning information given for specific disaster events in the rescue specification file, and gives precautions before the emergency organization takes action. It mainly includes attributes such as applicable emergency event type, disaster level, warning content, and source file name.

[0079] 2) Duty

[0080] Duty refers to the emergency document system and provides reference task indicators and responsibilities for emergency organizations, including attributes such as responsibility content and reference document name.

[0081] The temporal class combines static text data such as emergency plans with spatiotemporal data such as emergency cases by introducing a set of temporal class concepts. The temporal class is divided into three subclasses: emergency situation, emergency event, and emergency action. They are connected to time and place through atTime and inPlace respectively to describe their spatiotemporal attributes. They include:

[0082] 1) Situation

[0083] Each situation corresponds to an environmental context, which includes two subcategories: loss situation and disaster-affected environment situation. Loss situation refers to damage situations such as casualties, property losses, and house collapses, which are reported to relevant agencies for disaster assessment and subsequent information release. The disaster-affected environment situation describes the state changes of the objective environment after the disaster, and is used to infer the probability of subsequent sub-events.

[0084] 2) Event Emergency

[0085] Event is the center of the entire emergency response, including the transition between events, and is associated with the emergency situation through precedences and follows. The interaction between emergency events and emergency situations is the basis for building event chains. The context transition of emergency events is triggered by the participating subject Agent.

[0086] 3) Action

[0087] Connected to the participant Agent through hasPaticipant. With reference to the relevant stage classification standards and naming of the domestic overall emergency plan and the foreign PPRR model, this paper divides it into four subcategories: prevention, reconnaissance and evaluation, response, and recovery. Each subcategory will be further divided into one subcategory to better describe the various stages of emergency actions. The division results are as follows: Figure 2 shown.

[0088] The Action class contains four attributes: name, explanation, action target (taskAim), and feedback information (feedback). Name refers to the actual name of the current action, explanation describes the action content, taskAim describes the purpose of the action, and feedback records the changes in the emergency situation after the action (divided into positive feedback and negative feedback).

[0089] Step S2: Construction of emergency response map

[0090] The overall construction process of the knowledge graph is as follows: Figure 3 As shown in the figure, the knowledge representation model framework of the reference mode layer first collects and integrates data, extracts knowledge from four types of data, and then maps and integrates the extracted knowledge into the knowledge representation model to generate triples. After manual quality evaluation, it is imported into the graph database for storage to build a complete knowledge graph. Specifically, it includes the following sub-steps:

[0091] 2.1 Data Acquisition

[0092] At the data level, data can be divided into static text data and event data with spatiotemporal attributes according to data characteristics. Static data such as emergency plans, rescue specifications, and classification standards mainly come from relevant Wikipedia on the Internet, official websites of provincial and municipal people's governments, and documents published on the public accounts of various emergency organizations. This part of the data is directly obtained through web crawlers and saved in unstructured text format. Rescue process disposal records and emergency case data mainly come from internal real data, relevant Wikipedia open records, etc.

[0093] 2.2 Knowledge Extraction and Mapping

[0094] This article uses a knowledge extraction method based on the ChatGLM large language model. Referring to the pattern layer, determine the corresponding concept entities, attributes, and relationships that need to be extracted, manually annotate a portion of the data as prompt examples, and use in-context learning (ICL) to improve the performance and performance of the large model in the knowledge extraction scenario, and complete information extraction from unstructured text. After knowledge extraction, it is necessary to combine the pattern layer for knowledge mapping. In order to demonstrate the process of knowledge extraction and mapping and the overall structure of the data layer, some example fragments are intercepted from emergency plans, rescue specifications, classification standards, and emergency cases, such as Figure 4 As shown in the figure. The selected emergency plans are the organizational command system and guarantee measures of the "Jiangsu Province Typhoon Emergency Plan", which correspond to the emergency organization and emergency guarantee of the presentation layer after structuring; the rescue specifications are the three parts of the fire fighting guide of a cigarette factory, namely, disaster investigation, fire fighting measures and special warnings, which correspond to the disaster characteristics, action specifications and special warnings of the presentation layer after structuring; the classification standard is the "Hydrological Information Forecast Specification", which structures the disaster type, grade classification and classification basis characteristics; the emergency case takes the typhoon "Dusurui" event that has caused widespread influence and attention as an example, and the data comes from the case records of the event in Fujian Province on July 28 and July 29 crawled from Wikipedia, showing the mapping from emergency case data to events, situations and actions under the time phenomenon category.

[0095] In order to facilitate subsequent knowledge import, the structured knowledge is first saved as a node table and a relationship table. The table data structure is as follows: Figure 5 As shown in the figure. The node table mainly contains concept names, labels, unique identifier entity ids and multiple parallel attribute values; the relationship table is in the standard triple format, i.e. entity id-relationship-entity id. Given that different data sources have different terminologies for entities and relationships, and that some data have abbreviations, inconsistent data formats, missing values ​​and other issues, it is necessary to perform unified data preprocessing and quality assessment on entity and relationship tables before subsequent knowledge storage.

[0096] 2.3 Knowledge Storage

[0097] This paper adopts neo4j graph database to support emergency knowledge management and high-performance query. Relying on py2neo toolkit, the data import script 1 is written to batch the extracted nodes and attributes into CSV 1. The relevant code can be found at https: / / github.com / loyalty-fox / emergency_decision / blob / main / module_1.

[0098] Import the data into neo4j.

[0099] Step S3: Construction of emergency decision-making support model based on graph

[0100] A cross-file efficient knowledge retrieval emergency decision support model and a disaster chain disposal plan recommendation emergency decision support model were constructed. Relying on the semantic understanding ability and data association characteristics of the emergency response knowledge graph, two types of emergency decision support models were constructed: 1) cross-file efficient knowledge retrieval, 2) disaster chain disposal plan recommendation.

[0101] Model 1: Based on cross-document knowledge retrieval and reasoning model

[0102] The overall process is as follows Figure 6 As shown in the figure, by encapsulating the rule-based knowledge stored in the knowledge graph into functions, after extracting information from the user input, the situation information, monitoring indicators, and decision-making goals are passed as parameters to the Cypher query statement, and the results are visualized and returned to the user through intelligent analysis and query. Taking a geological disaster event in a certain place as an example, the knowledge graph is used to store historical entity data and relationships, and the real-time data is intelligently reasoned and retrieved based on the auxiliary decision-making model:

[0103]

[0104] The overall steps of the model responding to the commander's decision-making needs are as follows:

[0105] (1) Extract information from the decision-making requirements input by the user and extract keywords for disaster description and decision-making objectives, namely, geological disaster geologic_disaster and rescue personnel dispatch rescue_dispatch;

[0106] (2) Extract the query indicators required to construct the cypher statement, which requires two indicators: the current disaster type and the disaster level;

[0107] (3) Determine whether the conditions for generating a cypher query statement are met based on the extracted message. The user input contains the disaster type of geological disaster geologic_disaster, but the user has not yet entered the disaster level. According to the knowledge graph, the geological disaster type needs to be classified according to the multi-dimensional disaster monitoring indicators;

[0108] (4) Prompting the user to input additional information on designated disaster monitoring indicators;

[0109] (5) Generate a multi-hop query statement to query the personnel scheduling indicators and return them, convert them into readable answers and print them out to the user.

[0110] Implementing the above emergency assistance steps involves a multi-step reasoning process across documents. The first step is to classify the disaster level based on multi-dimensional feature data query, which involves the classification indicators in the classification standard file. The second step is to jump to the personnel dispatch part in the rescue specification file according to the disaster level and demand category, and query the number of rescue personnel that need to be dispatched under the current level.

[0111] In addition, the model will synchronously return the names of relevant files referenced in each step to improve the credibility of the recommendation results.

[0112] Model 2: Recommendation model for multi-hazard event chains

[0113] Major disasters often trigger a series of secondary and derivative disasters, forming a complex and changeable chain structure, which is called sudden disasters.

[0114] When dealing with different emergencies, it is necessary to activate the corresponding emergency plans, thus forming a corresponding plan chain.

[0115] Therefore, the generation of the plan chain depends on the construction of the event chain. To this end, this paper combines the temporal class definition in the model layer, the real-time environmental data and historical monitoring data in the data layer, and mathematically models the prediction generation process of the event chain, such as Figure 7 As shown. Then, the system will match the event level and event type according to the generated event chain, generate the corresponding emergency plan chain based on the map, and accurately push it to the relevant authorities and vulnerable communities. In addition, based on the current situation and the historical feedback of relevant action measures in the knowledge map, the system will also recommend refined proactive risk reduction measures such as chain breaking, personnel evacuation, and relief material dispatch.

[0116] The emergency event Ei at the event chain layer contains various attributes of the event and its on-site measurement values, expressed as, that is, the event feature vector, such as disaster characteristics such as rainstorm intensity and range. The disaster-affected environment situation contains the environmental spatiotemporal context of the current event detected on-site, expressed as =, where represents the feature vector of the disaster-bearing body in the current time and space, such as the soil looseness coefficient of the mountain body affected by the rainstorm, represents the feature vector of the disaster-prone environment in the current time and space, such as topography and vegetation density. The steps for generating an event chain are as follows:

[0117] (1) The event chain begins, and the original emergency event E1 occurs;

[0118] (2) The field-related attribute values ​​of measurement E1 constitute vector V1;

[0119] (3) Determine the disaster-affected environmental situation S1 based on the changes in the objective environment caused by E1;

[0120] (4) measure the spatiotemporal contextual data of S1;

[0121] (5) Predict the next emergency event based on the rule matching of each measurement value of E1 and S1;

[0122] (6) Loop through steps (2)-(5) to predict the subsequent chain of events;

[0123] (7) The event chain construction process ends when the result of predicting the next possible emergency event during the rule matching process is empty.

[0124] Taking a typhoon event in a certain urban area as an example, we try to construct a possible event chain by combining environmental trends, events and knowledge graphs.

[0125] Event 1: A typhoon occurs, and the characteristic values ​​of the typhoon's impact range, path, speed, etc. are measured. Within the impact range, the disaster-affected environment situation such as the atmospheric environment, topography, and water vapor supply in the area is measured, and the rules are matched through the knowledge graph to predict that the next event is likely to be heavy rain:

[0126]

[0127] Event 2: Heavy rain occurs. Query the relevant attributes of the knowledge graph and combine the historical data of the area in the graph to estimate the possible precipitation intensity and impact range. In the affected area, collect relevant drainage system, terrain, surface hardening degree and other situation data. After matching the rules, predict that the next event is likely to be urban waterlogging:

[0128]

[0129] Event 3: Waterlogging occurs, water levels rise, the flooded area increases, and the emergency plan level increases. Residents in the basin area need to be reminded to prepare for disasters, and disaster reduction measures such as drainage system dredging and maintenance are recommended to the corresponding vulnerable communities. After the event chain is generated, the relevant classification standards are matched in the knowledge graph according to the event type, the disaster level is inferred based on the event feature value, and the emergency plan chain is matched. The relevant government departments and threatened areas are located for plan push, and disaster level warnings, personnel evacuation work, and refined disaster reduction chain-breaking measures are recommended. Through this model, continuous response in dynamic scenarios of situation, decision-making, and action can be achieved, and the support function of emergency decision-making auxiliary applications in complex scenarios can be realized.

[0130] At present, there are still a large amount of diverse data such as plans, specifications, standards and cases distributed in the field of emergency response in my country. This paper constructs a knowledge ontology based on concrete, temporal and abstract categories, and proposes a multi-disaster, full-cycle emergency response knowledge graph that can effectively integrate and store various emergency response knowledge.

[0131] At the same time, an emergency decision-making assistance model is built based on the graph in the downstream, which can classify disaster levels, perform multi-hop path query analysis and make decision recommendations based on real-time multi-dimensional disaster data, greatly improving the efficiency of cross-domain text retrieval; at the same time, an event chain-based plan chain recommendation model is proposed, which can predict disaster chains based on rules and give refined plan push, helping vulnerable communities to prepare for emergency avoidance and planning in advance in complex scenarios.

[0132] The above two application models illustrate that our method for emergency decision support based on cross-domain knowledge fusion can greatly improve the efficiency of cross-domain text retrieval. At the same time, we propose a plan chain recommendation model based on event chain, which can predict disaster chains based on rules and give refined plan push, helping vulnerable communities to prepare for emergency avoidance and planning in advance in complex scenarios, and has broad application prospects.

[0133] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. An emergency decision support method based on cross-domain knowledge fusion, characterized in that: The following steps are involved: Step S1: Constructing emergency semantic model 1) Define the emergency semantic model Based on the ABC Ontology framework, the emergency knowledge representation quadruple is defined as: <Concepts,Relations,Rules,Instances> ; Among them, Concepts:={Actualities,Abstractions,Temporalities} represents a concept set; Actualities represents a concrete concept set, describing the reality of the objective world; Abstractions represents an abstract concept set, describing abstract concepts without objective reality; Temporalities represents a temporal concept set, describing concepts with spatiotemporal existence; and each concept contains multiple properties. Relations: = {relation, inheritance} represents a set of relationships. Relationships mainly include reference, compliance, utilization, execution, initiation, and accompanying non-inheritance relationships. Inheritance represents parent-child relationships. Rlues:={R:N1*N2*N3*...*N n }→ N n+1 |N1, N2,..., N n ∈Concepts∪Relations} is a set of emergency decision rules, where the multiplication sign is defined as an inter-entity association constraint, indicating that the n+1th entity can be uniquely determined based on the n entities and relationships in the emergency response system; Instances represent a collection of instances, i.e., instantiations of concepts and relations in emergency cases and emergency documents; 2) Knowledge representation model construction By extracting and dividing the knowledge of emergency plans, rescue specifications, classification standards and emergency cases, a concept set of the knowledge representation model is constructed, which is derived from concrete, abstract and timely classes; Step S2: Construction of emergency response map Referring to the knowledge representation model framework of the reference mode layer, we first collect and integrate data, extract knowledge from the four types of data, and then map and fuse the extracted knowledge into the knowledge representation model to generate triples. After manual quality assessment, they are imported into the graph database for storage to build a complete knowledge graph. Step S3: Construction of emergency decision-making support model based on graph Construct an emergency decision support model for efficient cross-file knowledge retrieval and a disaster chain disposal plan recommendation emergency decision support model.

2. The emergency decision support method based on cross-domain knowledge fusion as claimed in claim 1, characterized in that: The emergency decision rule set consists of two categories: grading rules Grades and rescue norms Norms. The grading rules include multiple grading standards such as disaster grading, emergency response grading, and rescue action grading. The rescue norms mainly include action norms and safety warning rules.

3. The emergency decision support method based on cross-domain knowledge fusion as claimed in claim 1 is characterized in that: The step S2 specifically includes the following steps: Step a: Data acquisition: data are divided into static text data and event data containing spatiotemporal attributes according to data characteristics. The static text data is directly acquired through web crawlers from relevant Wikipedia, official websites of provincial and municipal people's governments, and documents published on public accounts of various emergency organizations, and saved in unstructured text format. The event data containing spatiotemporal attributes is acquired through internal real data and relevant Wikipedia open records; Step b: Knowledge extraction and mapping. The knowledge extraction method based on the ChatGLM large language model refers to the pattern layer to determine the corresponding concept entities, attributes and relationships that need to be extracted. A portion of the data is manually annotated as prompt examples. The performance and expression of the large model in the knowledge extraction scenario are improved through context learning. Information extraction is completed from unstructured text. After knowledge extraction, knowledge mapping needs to be combined with the pattern layer. Step c: Knowledge storage, using neo4j graph database to support emergency knowledge management and high-performance query, relying on the py2neo toolkit to write data import script 1, and batch import the extracted nodes and attributes CSV into neo4j.

4. The emergency decision support method based on cross-domain knowledge fusion as claimed in claim 1, characterized in that: The construction of cross-file knowledge retrieval and reasoning model includes the following steps: by encapsulating the rule-based knowledge stored in the knowledge graph into functions, after extracting information from user input, the situation information, monitoring indicators and decision goals are passed as parameters to the Cypher query statement, and the results are visualized and returned to the user through intelligent analysis and query.

5. The emergency decision support method based on cross-domain knowledge fusion as claimed in claim 1, characterized in that: The construction of a recommended model for handling plans for multi-hazard event chains includes the following steps: combining the temporal phenomenon class definitions in the pattern layer, the real-time environmental data and historical monitoring data in the data layer, mathematically modeling the prediction generation process of the event chain, and then the system will match the event level and event type according to the generated event chain, generate the corresponding emergency plan chain based on the graph, and accurately push it to relevant authorities and vulnerable communities; based on the current situation and the historical feedback on relevant action measures in the knowledge graph, the system will also recommend refined proactive risk reduction measures such as chain breaking, personnel evacuation, and rescue material dispatch.

6. The emergency decision support method based on cross-domain knowledge fusion as claimed in claim 5, characterized in that: Emergency event E at the event chain level i Contains various attributes of the event and its on-site measurement values, denoted as E i =(V i ), V i That is, the event feature vector, the disaster environment situation S i Contains the environmental spatiotemporal context of the current event detected on site, denoted as S i =(Obj i , Env i ), where Obj i Indicates the characteristic vector of the disaster-prone body in the current space and time, Env i Represents the disaster-prone environmental feature vector in the current space-time. The steps to generate the event chain are as follows: (1) The event chain begins, and the original emergency event E1 occurs; (2) The field-related attribute values ​​of measurement E1 constitute vector V1; (3) Determine the disaster-affected environmental situation S1 based on the changes in the objective environment caused by E1; (4) measure the spatiotemporal context data of S1 (Obj1, Env1); (5) Predict the next emergency event E1×S1→E2 based on the rule matching of each measurement value of E1 and S1; (6) Loop through steps (2)-(5) to predict the subsequent event chain E2→…→E n ; (7) The event chain construction process ends when the result of predicting the next possible emergency event during the rule matching process is empty.

7. The emergency decision support method based on cross-domain knowledge fusion as claimed in claim 1, characterized in that: The concrete class Actuality is used to express objectively existing people and things, involving rescue personnel, organizations, and materials. It is the context of the emergency situation and is connected to the situation through isContext; it includes Agent rescue personnel, Organization emergency organization, Resource emergency support resources, and Document emergency file system.

8. The emergency decision support method based on cross-domain knowledge fusion as claimed in claim 1, characterized in that: The abstract class Abstraction describes concepts that have no objective reality in the emergency response process, including Warning special warnings and Duty task responsibilities.

9. The emergency decision support method based on cross-domain knowledge fusion as claimed in claim 1, characterized in that: The temporal phenomenon class is divided into three subclasses: emergency situation, emergency event, and emergency action, which are respectively connected to time and place through atTime and inPlace to describe their temporal and spatial attributes.

10. The emergency decision support method based on cross-domain knowledge fusion according to claim 1, characterized in that: The emergency semantic model assists the construction of the knowledge representation model through the Web ontology language OWL specification.

Images