Accident scene construction method, processing method and system based on knowledge graph

Abstract

The present disclosure relates to the technical field of big data application, and particularly discloses a method and system for constructing and processing accident scenarios based on a knowledge graph. The method comprises the following steps: extracting a plurality of historical scenario elements from historical scenario data selected from historical accident cases; constructing a historical accident graph based on the historical scenario elements; constructing and optimizing a scenario sample library and an emergency rescue scheme sample library based on the historical accident graph, thereby automatically updating the construction process of the scenario sample library. When a scenario judgment and disposal of a target accident is needed, a similarity comparison can be made between target scenario elements contained in the target accident scenario and historical scenario elements of historical accident scenarios in the scenario sample library, and an emergency rescue scheme corresponding to a historical accident scenario with a similarity satisfying a preset requirement is determined as a matching scheme for the target accident, thereby timely formulating an effective emergency rescue scheme.

Description

Technical Field

[0001] This disclosure relates to the field of big data application technology, and in particular to a method, processing method and system for constructing accident scenarios based on knowledge graphs. Background Technology

[0002] Production accidents are characterized by their suddenness, uncertainty, variability, and low probability. When an accident occurs, accurately predicting its evolution and promptly implementing targeted countermeasures are crucial for effective emergency response and minimizing losses. Current accident evolution simulations suffer from simplistic, superficial, and inaccurate approaches. Actual emergency response relies heavily on expert experience, impacting efficiency and effectiveness. Therefore, there is an urgent need to establish scientific methods for constructing accident scenario evolution, rapidly and accurately grasping the patterns of accident evolution, and developing specific emergency response plans. This is of great significance for improving emergency response capabilities. Summary of the Invention

[0003] In view of the above-mentioned problems of the prior art, the present disclosure provides a method, processing method and system for constructing accident scenarios based on knowledge graphs, aiming to automatically build historical accident scenarios based on knowledge graphs, and to grasp the evolution law of target accidents based on the built historical accident scenarios, and to formulate timely and effective emergency rescue plans.

[0004] In a first aspect, embodiments of this disclosure provide a method for constructing accident scenarios based on knowledge graphs, including: Historical accident cases are obtained, and the historical scenario information contained in the obtained historical accident cases is filtered to remove duplicate historical accident cases. Field integrity verification is performed on unstructured historical scenario information to filter out invalid historical scenario information and obtain the filtered historical scenario information. Extract multiple historical scenario elements from the filtered historical scenario information corresponding to historical accidents; The historical scenario information is divided into two stages: internal handling and external rescue, and a modular system for emergency response is formed. According to the modular system, the extracted historical scenario elements are used as key entities to construct a historical accident map. A knowledge graph-based multimodal clustering analysis algorithm is used to group the historical scenario elements in the historical accident map; feature vectors are used to represent the features of the grouped historical scenario elements to construct the ontological features, disaster chain features and emergency response features of each historical accident; and dynamic weighted spectrum clustering and scenario clustering are used to identify the commonalities and differences of different categories of historical accident scenarios. Based on the identification results, a scenario sample library and an emergency rescue plan sample library are constructed and optimized. The scenario sample library contains historical scenario elements of historical accident scenarios, and the emergency rescue plan sample library contains emergency rescue elements of emergency rescue plans corresponding to the historical accident scenarios.

[0005] Secondly, embodiments of this disclosure provide a method for handling accident scenarios, including: Extract multiple target scenario elements from the target scenario information corresponding to the target accident; The scenario state S, initial disaster-causing factor C, disaster consequence state I, and emergency rescue activities E of the target accident are extracted from the multiple target scenario elements to form scenario nodes and their variables in the evolution of the target accident, thereby obtaining the evolution process of the target accident. The emergency rescue process is then integrated into the evolution process of the target accident to obtain the target accident scenario. The similarity between the target scenario elements in the target accident scenario and the historical scenario elements in the scenario sample library of historical accident scenarios constructed by the knowledge graph-based accident scenario construction method described in the first aspect is calculated, wherein, based on the attributes of each scenario element in the accident scenario, the weight of each scenario element in the similarity calculation process is determined by a weight adjustment method. The calculated similarity is compared with a preset similarity threshold. If the similarity is greater than or equal to the preset similarity threshold, the emergency rescue plan corresponding to the historical accident scenario corresponding to the similarity is output from the emergency rescue plan sample library as the matching plan for the target accident.

[0006] Thirdly, embodiments of this disclosure provide a knowledge graph-based accident scenario construction system, including: The data filtering module is used to obtain historical accident cases, filter the historical scenario information contained in the obtained historical accident cases, remove duplicate historical accident cases, and perform field integrity verification on unstructured historical scenario information to filter invalid historical scenario information and obtain the filtered historical scenario information. The first element extraction module is used to extract multiple historical scenario elements from the filtered historical scenario information corresponding to historical accidents. The map construction module is used to divide the historical scenario information into two stages: internal handling and external rescue, and form a modular system for emergency response. According to the modular system, the extracted historical scenario elements are used as key entities to construct a historical accident map. The sample optimization module is used to group the historical scenario elements in the historical accident map using a knowledge graph-based multimodal clustering analysis algorithm; it uses feature vectors to express the features of the grouped historical scenario elements, constructs the ontology features, disaster chain features, and emergency response features of each historical accident, and uses dynamic weighted spectrum clustering and scenario clustering to identify the commonalities and differences of different categories of historical accident scenarios; An optimization module is constructed to build and optimize a scenario sample library and an emergency rescue plan sample library based on the identification results. The scenario sample library contains historical scenario elements of historical accident scenarios, and the emergency rescue plan sample library contains emergency rescue elements of emergency rescue plans corresponding to the historical accident scenarios.

[0007] Fourthly, embodiments of this disclosure provide a system for handling accident scenarios, including: The second element extraction module is used to extract multiple target scenario elements from the target scenario information corresponding to the target accident. The scenario evolution module is used to extract the scenario state S, initial disaster-causing factor C, disaster consequence state I, and emergency rescue activities E of the target accident from the multiple target scenario elements to form scenario nodes and their variables in the evolution of the target accident, obtain the evolution process of the target accident, and thus integrate the emergency rescue process in the evolution process of the target accident to obtain the target accident scenario. The calculation module is used to calculate the similarity between the target scenario element in the target accident scenario and the historical scenario element in the scenario sample library of the accident scenario construction method based on knowledge graph described in the first aspect, wherein the weight of each scenario element in the similarity calculation process is determined by a weight adjustment method based on the attributes of each scenario element in the accident scenario. The processing module is used to compare the calculated similarity with a preset similarity threshold. If the similarity is greater than or equal to the preset similarity threshold, the emergency rescue plan corresponding to the historical accident scenario corresponding to the similarity is output from the emergency rescue plan sample library as the matching plan for the target accident.

[0008] The knowledge graph-based accident scenario construction and processing method provided in this application extracts multiple historical scenario elements from historical scenario information selected from historical accident cases; divides the historical scenario information into two stages: internal handling and external rescue, forming a modular system for emergency response; constructs a historical accident graph using the extracted historical scenario elements as key entities according to the modular system; groups the historical scenario elements in the historical accident graph using a knowledge graph-based multimodal clustering analysis algorithm; expresses the features of the grouped historical scenario elements in the form of feature vectors, constructing the ontology features, disaster chain features, and emergency response features of each historical accident; and identifies the commonalities and differences of different categories of historical accident scenarios using dynamic weighted spectral clustering and scenario clustering; and constructs and optimizes a scenario sample library and an emergency rescue plan sample library based on the identification results. The scenario sample library contains historical scenario elements of historical accident scenarios, and the emergency rescue plan sample library contains emergency rescue elements of emergency rescue plans corresponding to historical accident scenarios, thereby automating the construction and updating of the scenario sample library and the emergency rescue plan sample library. When it is necessary to predict the scenario and carry out emergency rescue for the current target accident, the target accident scenario can be obtained first. Then, the similarity of the target scenario elements contained in the target accident scenario with the historical scenario elements of each historical accident scenario in the scenario sample library is compared to determine the historical accident scenario whose similarity meets the preset requirements. The emergency rescue plan corresponding to the historical accident scenario in the emergency rescue plan sample library is used as the matching plan for the target accident, thereby formulating a timely and effective emergency rescue plan.

[0009] The technical solutions of this disclosure will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0010] The accompanying drawings, which form part of this specification, illustrate embodiments of this disclosure and, together with the description, serve to explain the principles of this disclosure.

[0011] This disclosure will become clearer with reference to the accompanying drawings and the following detailed description, wherein: Figure 1a This is a flowchart illustrating a knowledge graph-based accident scenario construction method according to an embodiment of the present disclosure. Figure 1b This is a schematic diagram of a historical accident map implemented according to this disclosure; Figure 2 This is a flowchart illustrating a method for handling accident scenarios according to an embodiment of the present disclosure; Figure 3 This is a schematic flowchart of a specific step in step S220 of the accident scenario handling method according to an embodiment of the present disclosure; Figure 4This is a schematic diagram illustrating the interaction relationships of scenario elements according to an embodiment of the present disclosure; Figure 5 This is a schematic diagram illustrating the scenario construction evolution process according to embodiments of the present disclosure; Figure 6 This is a flowchart illustrating a method for handling an accident scenario according to an embodiment of the present disclosure; Figure 7 This is a flowchart illustrating a method for handling an accident scenario according to an embodiment of the present disclosure; Figure 8 This is a flowchart illustrating a method for handling an accident scenario according to an embodiment of the present disclosure; Figure 9 This is a schematic diagram of the structure of a knowledge graph-based accident scenario construction system according to an embodiment of the present disclosure; Figure 10 This is a schematic diagram of the structure of an accident scenario handling system according to an embodiment of the present disclosure; Figure 11 This is a block diagram of an electronic device according to an embodiment of the present disclosure. Detailed Implementation

[0012] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the various embodiments of this application will be described in detail below with reference to the accompanying drawings. However, those skilled in the art will understand that many technical details have been provided in the various embodiments of this application to help readers better understand this application. However, the technical solutions claimed in this application can be implemented even without these technical details and various changes and modifications based on the following embodiments. The division of the various embodiments below is for the convenience of description and should not constitute any limitation on the specific implementation of this application. The various embodiments can be combined with and referenced by each other without contradiction.

[0013] Example 1 Figure 1a This is a flowchart illustrating a knowledge graph-based accident scenario construction method according to an embodiment of the present disclosure. Figure 1a As shown, the method may specifically include the following steps S110-S150.

[0014] Step S110: Obtain historical accident cases, filter the historical scenario information contained in the obtained historical accident cases, remove duplicate historical accident cases, and perform field integrity verification on unstructured historical scenario information to complete the filtering of invalid historical scenario information and obtain the filtered historical scenario information.

[0015] In this article, "accident" generally refers to safety accidents, such as unexpected events that occur suddenly during the production and operation activities (including activities related to production and operation) of production and operation units, causing personal injury and health, damage to equipment and facilities, or economic losses, leading to the temporary suspension or permanent termination of the original production and operation activities (including activities related to production and operation), such as gas explosions in the coal mining industry. Of course, it can also refer to other accidents.

[0016] Those skilled in the art will understand that the "scenario information" in this document refers to accident-related data, primarily including the time and location of the event, the event severity, the event's course, emergency response plans, and event losses. Examples include various accident records, accident investigation reports, and accident case analysis documents. It differs from the "scenario elements" mentioned below in that "scenario elements" can be considered specific pieces of information describing the accident scenario, stored in a structured form after the scenario information has been broken down and extracted. For example, in the case of a gas explosion accident, the scenario information in the accident report might contain information such as gas accumulation, ignition source, coal dust explosiveness, coal seam dip angle, coal seam thickness, number of deaths, and number of injuries. However, this information might be scattered throughout the accident report file. In contrast, the scenario elements, after extracting the scenario information, store information about gas accumulation, ignition source, coal dust explosiveness, and coal seam dip angle in text form, and information about coal seam dip angle, coal seam thickness, number of deaths, and number of injuries in numerical form. Furthermore, all of this information is stored in a structured (regular) manner.

[0017] In this embodiment, the scenario information obtained based on historical incidents is referred to as "historical scenario information".

[0018] Specifically, in this step, historical accident cases from both domestic and international sources can be collected through accident databases such as the National Emergency Management Department's accident case database and the databases of safety supervision systems in various provinces and cities. Simultaneously, connections are made with safety monitoring platforms in key industries (chemical, energy, transportation, etc.) to obtain real-time accident reports. The historical context information contained in the collected historical accidents undergoes preliminary screening. For example, an improved SimHash algorithm is used to calculate the similarity of accident cases. A master case retention mechanism is triggered when the similarity threshold of historical accident cases is ≥85%, automatically merging duplicate records and thus eliminating recurring historical accident cases. Simultaneously, field integrity verification is implemented for unstructured data. By establishing a field integrity verification matrix, key data such as accident time, location, type, and number of casualties are set as necessary parameters. Missing information in this area is considered invalid and filtered out. A data quality monitoring mechanism for format correction is also established.

[0019] After obtaining the filtered historical context information, word segmentation technology can be used to parse the historical context information, extract the key information of the historical accident scenarios, and then convert the unstructured text of the historical context information into formatted data storage such as JSON or database format through structured processing to facilitate subsequent data retrieval.

[0020] For example, after initial cleaning and screening, historical accident scenario information is parsed using word segmentation, part-of-speech tagging, and named entity recognition technologies. Simultaneously, a word segmentation optimization strategy is implemented: for chemical accidents, priority is given to identifying chemical names (e.g., styrene, chlorine); for construction accidents, engineering terms (e.g., scaffolding, tower crane) are prioritized. A dedicated part-of-speech tag set for accident scenarios is constructed, categorized into general and specialized text. Key accident scenario information is extracted, including the accident object (e.g., equipment name, chemical name), causative factors (e.g., operational errors, equipment malfunctions), consequences (e.g., number of casualties, economic losses), and time and location (e.g., accident time, GPS coordinates). After converting unstructured text to JSON format through structured processing, data validation is performed to ensure no missing parameters and correct data type verification before database storage.

[0021] Step 120: Extract multiple historical scenario elements from the filtered historical scenario information corresponding to historical accidents.

[0022] The definition of contextual elements has been explained in the previous step. Contextual elements extracted from historical contextual information are called "historical contextual elements".

[0023] Specifically, deep learning models based on convolutional neural networks (CNNs) can be used to extract historical scene elements. CNNs can be designed as multi-layered network architectures, trained using labeled original accident datasets. The trained model extracts and removes noise from historical accident scene elements in different accident domains. During model training, a performance optimization mechanism is established, employing training strategies such as dataset partitioning and optimized configuration. Dynamic parameter optimization is used to adaptively adjust parameters, thereby continuously improving the model's reliability.

[0024] In some embodiments, the scenario element dimensions corresponding to multiple historical scenario elements extracted from the filtered historical scenario information corresponding to historical accidents may include: accident object, cause of disaster, consequences of disaster, and disaster relief subject; The accident object is used to describe the accident itself and its objective characteristics. The contextual elements of the accident object include spatial information and temporal information of the accident type, as well as the basic overview of the accident scene. The causes of disaster are used to describe the influencing factors of the occurrence and development of accident risks. The scenario elements of the causes of disaster include the causes of the accident and the causes of its development. The disaster consequences are used to describe the objects affected by the accident risk. The scenario elements of the disaster consequences include the degree of damage to the accident carriers, including on-site workers, ventilation facilities, and tunnels, as well as the secondary accidents caused by the accident. The disaster relief entity is used to describe the emergency response process and results in response to an accident. The scenario elements of the disaster relief entity include the resources invested in the emergency activities, the participating emergency organizations, the technical measures taken, and the resulting response.

[0025] The following uses a gas explosion accident in the coal mining industry as an example to illustrate this. Regarding the accident object, the description of the accident object refers to the information and objective characteristics of the accident itself, such as spatial information, temporal information, and mine background. It can be divided into two parts: basic accident information and mine background information. These two parts can be further subdivided into secondary scenario elements, including time, location, accident level, mine type, ventilation method, total intake air volume, and total return air volume. Regarding the cause of the disaster, the cause of the gas explosion in a coal mine represents the factors influencing the generation and development of the risk, that is, the direct cause of the gas explosion. It can be divided into two parts: the hazard cause and the hazard environment. These two parts can be further subdivided into secondary scenario elements, including gas accumulation, ignition source, coal dust explosiveness, coal spontaneous combustion tendency, coal seam thickness, and coal seam dip angle. Regarding the consequences of the disaster, the consequences of a gas explosion refer to the object of the risk's impact, reflecting the degree to which it suffers adverse effects, such as the degree of damage to on-site workers, ventilation facilities, roadways, and other structures, and whether secondary accidents occur. The consequences of a gas explosion include the affected entities, economic losses, and damage and pollution to objects. Regarding the disaster response entities, the description of the emergency response process and results for a gas explosion accident encompasses the resource input, organizational and technical means employed to handle the emergency, primarily consisting of three aspects: emergency personnel, emergency supplies, and emergency response measures. The emergency personnel element for a gas explosion accident includes the specific departments and rescue forces involved in the rescue, such as rescue teams and experts.

[0026] This process involves initial cleaning of the collected raw accident scenario information to remove duplicate, invalid, or incorrectly formatted data. Simultaneously, a scenario information quality monitoring mechanism is established to ensure that the collected historical scenario elements meet the requirements for subsequent processing.

[0027] Step 130: Divide the historical scenario information into two stages: internal handling and external rescue, and form a modular system for emergency response. According to the modular system, use the extracted historical scenario elements as key entities to construct a historical accident map.

[0028] Specifically, historical accident scenario information can be vertically divided into two stages: internal handling and external rescue, forming a modular emergency response system. This system includes four modules: accident initiation, internal handling, external rescue, and emergency termination. Within each of these four modules, several secondary tasks are horizontally divided, clearly defining the elements such as people, objects, equipment, fixed actions, and flexible actions within each task as key entities in the historical accident graph. Each secondary task includes multiple key actions and the associated states between tasks, establishing a task dependency graph. After constructing the key entities, the reusability and independence of each entity module are analyzed to obtain the causal relationships and influences between parts. A graph optimization mechanism based on reusability and independence is designed to organically combine the modules, organizing these entities and relationships in graph form to ultimately form a historical accident graph, thereby establishing a graph structure representation for large-scale accident data. Figure 1b The image shown is a schematic diagram of a historical accident map.

[0029] Step 140: The historical scenario elements in the historical accident map are grouped using a knowledge graph-based multimodal clustering analysis algorithm; the grouped historical scenario elements are represented by feature vectors to construct the ontological features, disaster chain features and emergency response features of each historical accident; and dynamic weighted spectrum clustering and scenario clustering are used to identify the commonalities and differences of different categories of historical accident scenarios. Step 150: Based on the identification results, construct and optimize the scenario sample library and the emergency rescue plan sample library. The scenario sample library contains historical scenario elements of historical accident scenarios, and the emergency rescue plan sample library contains emergency rescue elements of emergency rescue plans corresponding to the historical accident scenarios.

[0030] Specifically, after forming a historical accident map, a knowledge graph-based multimodal clustering analysis algorithm can be used to group the scenarios. Multidimensional features are then used to extract ontological features, hazard chain features, and emergency response features for each historical accident. These features are represented by feature vectors. Dynamic weighted spectral clustering and scenario cluster optimization are employed to identify the commonalities and differences among different accident scenarios. An adaptive adjustment of weights is performed through a dynamic optimization mechanism, thereby constructing and optimizing a scenario sample library and an emergency rescue plan sample library. The scenario sample library contains historical scenario elements of historical accident scenarios, while the emergency rescue plan sample library contains emergency rescue elements of emergency rescue plans corresponding to the historical accident scenarios.

[0031] The knowledge graph-based accident scenario construction method provided in this application extracts multiple historical scenario elements from historical scenario information selected from historical accident cases; divides the historical scenario information into two stages: internal handling and external rescue, forming a modular system for emergency response; constructs a historical accident graph using the extracted historical scenario elements as key entities according to the modular system; groups the historical scenario elements in the historical accident graph using a knowledge graph-based multimodal clustering analysis algorithm; expresses the features of the grouped historical scenario elements in the form of feature vectors, constructing the ontology features, disaster chain features, and emergency response features of each historical accident; and identifies the commonalities and differences of different categories of historical accident scenarios using dynamic weighted spectral clustering and scenario clustering; and constructs and optimizes a scenario sample library and an emergency rescue plan sample library based on the identification results. The scenario sample library contains historical scenario elements of historical accident scenarios, and the emergency rescue plan sample library contains emergency rescue elements of emergency rescue plans corresponding to historical accident scenarios, thereby automating the construction and updating of the scenario sample library and the emergency rescue plan sample library.

[0032] Example 2 Figure 2 This is a flowchart illustrating a method for handling accident scenarios according to an embodiment of this disclosure. Figure 2 As shown, the method may specifically include the following steps S210-S240.

[0033] Step S210: Extract multiple target scenario elements from the target scenario information corresponding to the target accident.

[0034] The definitions of "accident" and "scenario information" can be found in the aforementioned embodiments. The scenario information corresponding to the target accident is called "target scenario information," and the scenario elements extracted from the target scenario information are called "target scenario elements."

[0035] Similar to the description in Example 1, the target scenario elements also include the same four dimensions: accident object, cause of disaster, consequences of disaster, and disaster relief subject. The following uses a gas explosion accident in the coal mining industry as an example for specific illustration. Regarding the accident object, the description of the coal mine gas explosion accident object refers to the information and objective characteristics of the accident itself, such as spatial information, temporal information, and mine background. It can be divided into two parts: basic accident information and mine background information. These two parts can be further refined into secondary scenario elements, including time, location, accident level, mine type, ventilation method, total intake air volume, and total return air volume. Regarding the cause of disaster, the cause of a coal mine gas explosion accident represents the factors influencing the generation and development of risk, that is, the direct cause of the gas explosion. It can be divided into two parts: risk-causing factors and risk-causing environment. These two parts can be further refined into secondary scenario elements, including gas accumulation, ignition source, coal dust explosiveness, coal spontaneous combustion tendency, coal seam thickness, and coal seam dip angle. Regarding the consequences of a gas explosion, these consequences refer to the objects affected by the risk, reflecting the degree to which they suffer adverse effects. Examples include the extent of damage to on-site workers, ventilation facilities, tunnels, and other structures, as well as whether secondary accidents occur. The consequences of a gas explosion include the affected entities, economic losses, and damage and pollution to objects. Regarding the disaster response entities, the disaster response entities in a gas explosion describe the process and results of emergency response actions. This includes the resource input required to handle the emergency, the emergency organization and technical means employed, and mainly consists of three aspects: emergency personnel, emergency supplies, and emergency response measures. The emergency personnel element in a gas explosion includes the specific departments and rescue forces involved in the rescue, such as rescue teams and experts.

[0036] The process of acquiring target scenario information is similar to that of acquiring historical scenario information. It involves initially cleaning the collected raw target accident scenario information to remove duplicate, invalid, or incorrectly formatted data. Simultaneously, a scenario information quality monitoring mechanism is established to ensure that the final collected target scenario information meets the requirements for subsequent processing.

[0037] The extraction of target context elements is achieved through word segmentation and keyword extraction from a corpus composed of target context information. Before performing word segmentation, a custom dictionary and / or stop dictionary specifically for accident analysis are constructed. The custom dictionary includes proper nouns that need to be counted in word segmentation, while the stop dictionary includes words and symbols that do not need to be counted in word segmentation, thereby improving the efficiency and accuracy of word segmentation and keyword extraction.

[0038] The custom dictionary and the deactivated dictionary can be adjusted and updated to further improve the efficiency and accuracy of word segmentation and keyword extraction.

[0039] This involves constructing a deep learning model based on a convolutional neural network (CNN), training it using a labeled original accident dataset, and training the model to extract accident scenario elements and eliminate noise. A performance optimization mechanism is established to continuously improve the model's reliability through dynamic parameter optimization. The constructed deep learning model is then used to extract target scenario elements.

[0040] Step S220: Extract the scenario state S, initial disaster-causing factor C, disaster consequence state I, and emergency rescue activities E of the target accident from multiple target scenario elements to form scenario nodes and their variables in the evolution of the target accident, obtain the evolution process of the target accident, and thus integrate the emergency rescue process in the evolution process of the target accident to obtain the target accident scenario.

[0041] Here, the scenario state S includes the initial scenario state S0 and subsequent scenario states S1, S2, S3, ..., where the initial scenario state S0 is used to indicate the scenario state at the time of the accident under the influence of the initial disaster-causing factor C, and the subsequent scenario states S1, S2, S3, ... are different scenario states that evolve under the combined influence of disaster consequences I and emergency rescue activities E.

[0042] The initial hazard factor C can include factors that, under natural or man-made conditions, may negatively impact human life, property, and various aspects of production and daily life, thereby leading to a sudden accident. For example, in a coal mine gas explosion accident, the two main factors leading to the accident are gas accumulation and ignition sources. The simultaneous presence of both will result in a gas explosion.

[0043] Emergency rescue activity E is used to indicate the emergency rescue plan (also known as the "response plan") to be taken in response to the situation of disaster consequence I. It is a measure taken to control the scope of the accident, reduce the degree of harm of the accident, and reduce the losses caused by it. For example, measures taken by mine rescue team members to repair ventilation systems, clear roadways, and close and extinguish fires.

[0044] Step S230: Calculate the similarity between the target scenario element in the target accident scenario and the historical scenario element in the scenario sample library constructed based on the knowledge graph-based accident scenario construction method in Embodiment 1. The weight of each scenario element in the similarity calculation process is determined by a weight adjustment method based on the attributes of each scenario element in the accident scenario.

[0045] The attributes of a context element may include at least one of the following: numerical attributes, conceptual attributes, and fuzzy attributes.

[0046] Step S240: Compare the calculated similarity with a preset similarity threshold. If the similarity is greater than or equal to the preset similarity threshold, output the emergency rescue plan corresponding to the historical accident scenario with the similarity from the emergency rescue plan sample library as the matching plan for the target accident.

[0047] The above embodiments of this disclosure first extract target scenario elements from the target scenario information of the target accident, construct a target accident scenario based on the extracted target scenario elements, obtain the evolution process of the target accident, and then calculate the similarity between the target scenario elements in the target accident scenario and the historical scenario elements of each historical accident scenario in the scenario sample library. By adding an adjustment factor (used to adjust the weight of each scenario element) to the calculation of the similarity when calculating the similarity, the effectiveness and rationality of the similarity calculation are further improved, thereby improving the accuracy of scenario element matching. Correspondingly, the reliability of the obtained matching scheme can be effectively improved, and the accident handling capability can be greatly improved.

[0048] Practical experience has shown that calculating the weights of scenario elements using conventional methods, such as the entropy method, still falls short in terms of accuracy when matching scenario elements. Therefore, this solution proposes adjusting and determining the weights of scenario elements based on their characteristics using a weight adjustment method. This further improves the accuracy of scenario element matching, making the resulting matching schemes more reliable.

[0049] Example 3 According to the embodiments of this disclosure, in step S220 above, extracting the scenario state S, initial disaster-causing factor C, disaster consequence state I, and emergency rescue activity E of the target accident from multiple target scenario elements to form scenario nodes and their variables in the evolution of the target accident, and obtaining the evolution process of the target accident, thereby integrating the emergency rescue process in the evolution process of the target accident to obtain the target accident scenario processing process includes: Among multiple scenario elements, the key scenario elements that play a dominant role in the scenario evolution of the target accident are identified. Based on the accident scenario state S, initial disaster-causing factor C, disaster consequence situation I, and emergency rescue activities extracted from the key scenario elements, scenario nodes and their variables in the evolution of the target accident are formed. The evolution process of the target accident is obtained, and an accident scenario evolution model is obtained based on the historical accident map. A dynamic Bayesian network is used to introduce a multi-factor coupling mechanism to decompose and classify each accident scenario, analyze each scenario element, clarify the content of accident scenario construction, scenario parameters, and granular elements of each module, and classify scenario element categories. For example, dynamic nodes that are prone to change, such as weather, personnel behavior, and equipment failure, are identified. The differences and changes of dynamic nodes in the accident scenario are input into the scenario evolution model to reduce their impact on scenario evolution.

[0050] like Figure 3 As shown, the specific processing procedure of step S220 above includes the following steps (steps S221~223).

[0051] Step S221: Construct scenario nodes based on scenario elements.

[0052] To predict and analyze the scenario evolution of a target accident, scenario elements can be treated as scenario nodes, and a diagram of the accident evolution process can be constructed using Bayesian network modeling tools. Typically, scenario elements may be caused by several different factors, and the causes of these elements can be represented by the variables of the scenario nodes (also called "scenario node variables"). For example, a power outage is a cause of gas accumulation.

[0053] In an example of a coal mine production scenario, under the initial scenario state, gas accumulation, ignition source, and oxygen are considered as initial disaster-causing factors. Under the gas accumulation scenario node, abnormal outbursts, airflow short circuits, and power outages are scenario node variables. Under the ignition source scenario node, friction, spontaneous combustion of coal, and blasting are scenario node variables. The scenario elements affecting the evolution of a coal mine explosion accident are connected according to a scenario interaction diagram to determine the interaction diagram under an accident scenario. For example... Figure 4 As shown, a simple interaction diagram can include: scenario states (S0, S1, S2), initial causative factor (C), disaster consequence situation (I), and emergency rescue activities (E). In this example, the initial causative factor C is gas accumulation; the disaster consequence situation I is the roadway; scenario state S0 is excessively high gas concentration; emergency rescue activity E0 is cutting off power and restoring ventilation; after taking emergency rescue activities, the scenario states change to S1 ventilation equipment returns to normal and S2 gas concentration decreases.

[0054] After constructing scenario nodes based on scenario elements, several interaction graphs can be connected using directed line segments according to the accident evolution path and the causal and dependency relationships between scenario elements to construct a Bayesian network structure diagram of the accident. A dynamic Bayesian network is adopted to introduce a multi-factor coupling mechanism, decomposing and classifying each scenario, analyzing each scenario element, clarifying the content of the accident scenario construction, scenario parameters, and the granular elements of each module, and classifying scenario elements into categories, such as identifying dynamic nodes that are prone to change, such as weather, personnel behavior, and equipment failure. The differences and changes of dynamic nodes in the accident scenario are input into the scenario evolution model to reduce their impact on scenario evolution. In coal mine accident cases, the causal relationships between scenario elements typically include: explosion leading to fire, power outage leading to gas accumulation, etc. Dependencies can be emergency measures relied upon under specific scenarios, such as alarms being triggered at the gas explosion node. Figure 5As shown in the Bayesian network structure diagram, sub-scenario S represents the scenario state, C represents the initial causative factor, I represents the disaster consequence state, and E represents the emergency rescue activity. The accident evolution process can include several stages such as accident occurrence, evolution, and disappearance. Specifically, the initial causative factor C is gas accumulation, scenario state S0 is excessively high gas concentration, emergency rescue activity E includes cutting off power, restoring ventilation, evacuating personnel, and drilling for extraction, disaster consequence state I1 is ventilation equipment (after emergency rescue activities, scenario state S1: ventilation equipment returns to normal); disaster consequence state I2 is the roadway (after emergency rescue activities, scenario state S2: gas concentration decreases); disaster consequence state I3 is personnel (after emergency rescue activities, scenario state S3: personnel evacuate to a safe area); disaster consequence state I4 is the tunneling face (after emergency rescue activities, scenario state S4: the rate of gas accumulation slows down). Subsequent emergency rescue activities are carried out according to the new scenario state until the accident evolves to the disappearance stage.

[0055] Step S222: Based on the relationships between various scenario nodes, construct a Bayesian network structure diagram of the target accident.

[0056] Specifically, a Bayesian network structure diagram can be constructed using the FCI (Fast Causal Inference) algorithm and a Bayesian network modeling tool. By analyzing the causal and dependency relationships between scenario nodes, the FCI algorithm performs structure learning, determining the interaction relationships between scenario nodes to construct a Bayesian network structure diagram of the coal mine accident scenario. Specifically, in this embodiment, the FCI algorithm is introduced into the Bayesian network modeling tool, replacing the original algorithm for handling causal relationships, resulting in more accurate causal relationship identification during the Bayesian network modeling tool's generation process. During scenario evolution, there are not only progressive causal relationships between sub-scenario elements at different levels, but also mutually influential causal relationships between sub-scenario elements at the same level, such as S1, S2, S3, and S4. Scenario elements extracted from the accident scenario information are input into the Bayesian network modeling tool to generate the accident's Bayesian network structure diagram.

[0057] Bayesian networks are probabilistic graphical models that represent random variables and their conditional probabilities through nodes and edges, effectively handling uncertainty. The core idea of ​​Bayesian networks is to use Bayes' theorem to update the posterior probability of an event; the network dynamically adjusts its internal probability distribution as new evidence emerges. GENIE (Generative Network Interface for Everyone) is a tool for building, learning, and exploring Bayesian networks, providing a user-friendly graphical interface that allows users without programming backgrounds to easily create and analyze complex probabilistic models. GENIE supports various types of Bayesian network architectures, including Dynamic Bayesian Networks (DBNs) and Markov Random Fields.

[0058] Step S223: Determine the probability distribution of different nodes in the scenario nodes in the Bayesian network structure diagram, determine which of the scenario nodes are emergency critical nodes based on the probability distribution of different scenario nodes, and compare the emergency critical nodes to obtain the results.

[0059] Specifically, the Bayesian network modeling tool GENIE can be used to calculate the node probability distribution of scenario nodes in the Bayesian network structure diagram, and then emergency critical nodes can be identified from the scenario nodes based on the node probability distribution. Emergency critical nodes in the Bayesian network structure diagram correspond to key scenarios that are relatively important in the accident evolution process. For example, several nodes with relatively high probabilities in the Bayesian network structure diagram can be identified as emergency critical nodes.

[0060] After obtaining the node probability distribution in the Bayesian network structure graph of historical accidents, emergency key nodes are determined based on the node probability magnitude. Emergency key nodes refer to nodes with higher probabilities in the generated Bayesian network structure graph. These emergency key nodes correspond to key scenarios, which are then prioritized in the emergency response process. Emergency priorities can be generated based on these key scenarios. By analyzing the probability changes of the corresponding emergency key nodes and other nodes before and after them, countermeasures are determined and used as emergency priorities, guiding emergency rescue work. Specifically, when calculating the similarity between the target accident scenario and historical accident scenarios, emphasis is placed on scenario elements related to emergency priorities, increasing the weight of these elements in the similarity evaluation, and obtaining comparison results based on these weights.

[0061] In one implementation, determining the node probability distribution of the scenario nodes in the Bayesian network structure graph in step S223 above may include: Divide multiple scenario nodes into root node and secondary nodes; The prior probability of the root node in the Bayesian network structure graph is determined based on deep learning algorithms. The conditional probability of a secondary node in a Bayesian network structure graph is determined based on the Laplace smoothing algorithm, where a secondary node is a node with a parent node. The node probability distribution is obtained based on the prior probability and conditional probability, using the joint probability formula.

[0062] In the Bayesian network structure diagram, the root node refers to the initial node of the Bayesian network that does not have a parent node, such as nodes like gas accumulation or fire source; the secondary node refers to the network node with a parent node, such as nodes like roof collapse or fire.

[0063] Prior probability refers to the probability of a root node. It is determined by statistically analyzing the node states (T / F, true / false) in each incident case to form a state matrix. Then, using a deep learning algorithm within the Bayesian network modeling tool GENIE, the network structure is optimized to determine the prior probability of the root nodes. For historical incidents, all corresponding data in the database are input into the deep learning algorithm to determine the prior probability of each root node. For the current target incident, the prior probability of the target incident's root nodes is estimated using a deep learning algorithm.

[0064] Conditional probability refers to the probability of a secondary node. The Bayesian network modeling tool GENIE introduces the Laplace smoothing algorithm to replace the original algorithm for calculating the conditional probability of secondary nodes, making the calculated probability closer to the true value. Specifically, before calculating the conditional probability, the frequency of each secondary node is first counted. If the frequency is zero, it is incremented by 1; otherwise, the original value is retained. Then, the algorithm in GENIE is used to calculate the conditional probability of the secondary node. By using Laplace smoothing to adjust the frequency of secondary nodes, the frequency of secondary nodes is ensured to be both present and meaningful, guaranteeing that the probability of each state is greater than 0, thereby optimizing the conditional probability of the nodes.

[0065] After determining the prior probability of the root node and the conditional probability of the secondary nodes, the Bayesian network modeling tool GENIE is used for calculation and analysis. Specifically, the joint probability formula is used, and the Locality Sensitive Analysis (LSA) algorithm is incorporated for optimization. The probability of each scenario node is calculated using the joint probability formula, thus obtaining the node probability distribution. The joint probability formula is as follows:

[0066] in, Represents the probability distribution of nodes. Representation and node The corresponding conditional probability distribution function, Represents a node The parent node, Represents a node.

[0067] Local Sensitivity Analysis (LSA) is a method used to evaluate the sensitivity of a model's output to small perturbations in the input data. It is often used as a supplementary tool to analyze and improve model performance. LSA can optimize parts of the following algorithms: Support Vector Machines (SVM), Recurrent Neural Networks (RNN), Convolutional Neural Networks (CNN), Conditional Random Fields (CRF), and Transformers. An LSA module can be developed based on the LSA algorithm and imported into the GENIE tool. The LSA module's functions include accepting model input and output data from the GENIE tool, calculating sensitivity metrics, and then performing data training optimization to assist and enhance the performance of existing algorithms within the GENIE tool. The existing GENIE tool suffers from high model complexity and large computational data requirements. Through optimization using the Local Sensitivity Analysis (LSA) algorithm, insensitive features are identified, and their weights can be removed or reduced during training, reducing model complexity, accelerating training and prediction, reducing the risk of overfitting, and overall better adapting to changes in data distribution.

[0068] The above process is used to process scenario information from historical accidents to obtain node probability distributions. The above process can be repeated for target accidents to obtain node probability distributions for target accidents. Based on the node probability distributions, emergency critical nodes can be identified from scenario nodes, which can provide guidance for emergency rescue work.

[0069] like Figure 6 As shown, a state matrix is ​​first formed by statistically analyzing the states (T / F) of nodes in each accident case. The accident scenario nodes are divided into root nodes and secondary nodes. The prior probability of the root node is determined using a deep learning algorithm; Laplace smoothing is introduced to determine the conditional probability of the secondary nodes; then, the probability of each node is calculated using the joint probability formula, with Local Sensitivity Analysis (LSA) algorithm incorporated for optimization, to identify critical emergency nodes and generate emergency key points as the focus of the emergency response process.

[0070] Example 4 According to the embodiments of this disclosure, in step 230 above, based on the attributes (including at least one of numerical attributes, conceptual attributes, and fuzzy attributes) of each scenario element in the accident scenario, the weight of each scenario element in the similarity calculation process is determined by a weight adjustment method, including: calculating the weight of each scenario element using the entropy method.

[0071] Entropy-based weighting is a method for determining the weights of indicators. Its basic principle is based on the concept of information entropy. It assesses the degree of dispersion of each indicator by calculating its entropy value, thereby determining the weight. The core of entropy-based weighting lies in using information entropy to measure the degree of disorder in a system. The smaller the information entropy value, the greater the dispersion of the indicator, meaning the greater its weight in the overall evaluation. Specifically, the greater the difference in a particular indicator, the smaller its entropy value, the more information it carries, and therefore the greater its weight.

[0072] To measure the weight of each scenario element in the similarity evaluation, an entropy method is used to construct a data matrix. The constructed scenario sample library is then formed into a data matrix consisting of m samples and n indicators, as shown in the following formula:

[0073] In the above formula: This represents the value of the m-th indicator in the n-th sample. For example, if a scenario element in the scenario sample library is "explosion," its indicator could be the concentration of carbon dioxide, which could be extracted from accident record text.

[0074] After constructing the data matrix, the indicators for each scenario element are standardized. Since the attributes, units, and orders of magnitude of each scenario element may differ, it is necessary to standardize the indicators and adjust the original data range to reduce the impact of these differences on the evaluation results. Zoom to Within the interval. The formula for standardization is as follows: Positive indicators:

[0075] Negative indicators:

[0076] In the above formula: For the first The maximum value of the indicator, For the first The minimum value of the indicator. These are standardized values. If a larger value for the indicator is better, then the first formula should be used; if a smaller value for the indicator is better, then the second formula should be used.

[0077] After standardization, the results are normalized to construct a data weight matrix. The data weight matrix refers to the matrix used to calculate the weight of the first data point. The first item under the indicator The proportion of each sample value This allows us to obtain the weight matrix of the data. If... If a value of 0 is found, a shifting process can be performed, which involves adding a preset value, such as 0.001, to each indicator after it has been dimensionless. The shifting process is performed after calculation using the following formula:

[0078]

[0079] Then, based on the weight matrix, the index entropy value and information utility value are determined.

[0080] According to embodiments of this disclosure, for particularly important indicators, a penalty factor (also known as an "importance characteristic factor") can be introduced as an adjustment factor to adjust the entropy value of the relevant indicators, ensuring that these indicators receive more attention in weight allocation. That is, based on the importance determined by the characteristics of multiple scenario elements of the target incident, an importance characteristic factor is introduced to adjust the entropy value of each scenario element, so that the weight of scenario elements with higher importance is increased. This enhances the effectiveness and rationality of the similarity calculation results by adding an adjustment factor when calculating the weight of scenario elements. The formula for calculating the entropy value of the indicators is as follows:

[0081] In the above formula: This represents the index entropy value of the j-th indicator. Boltzmann's constant, . It is a penalty factor and is greater than 1.

[0082] Information utility value refers to the information utility value of an indicator, which is determined by the indicator's entropy value. The information utility value is determined by the difference between 1 and 0, and its value directly affects the weight. The larger the information utility value, the greater its importance in similarity evaluation, and the larger the weight. Calculating the information utility value... d The formula is as follows:

[0083] Based on the above information utility values, the weight value of each scenario element is calculated.

[0084] According to embodiments of this disclosure, a time weight parameter (time characteristic factor) can be introduced. Based on the occurrence time nodes of each historical accident in the scenario sample library and the characteristics of the scenario elements of the accident, combined with practical experience, the most recent data can be considered the most important. That is, for the correlation between multiple target scenario elements based on the target accident and historical scenario elements of historical accidents, a time characteristic factor is introduced to adjust the entropy value of each scenario element, so that the weight of the scenario elements with the highest correlation to the most recently occurred historical accident is increased. Thus, by adding the time characteristic factor as an adjustment factor when calculating the weight of scenario elements, the effectiveness and rationality of the similarity calculation results are enhanced. As an important basis for similarity matching, the objectivity of the weight will directly affect the accuracy of scenario matching. Calculate the weight of each scenario element. w j The formula is as follows:

[0085] in It increases with time t; d j For the first j The information utility value of each context element.

[0086] Entropy calculation, based on the inherent dispersion of data, reduces the influence of subjective judgment on evaluation results. This embodiment utilizes entropy calculation to determine the weight of each scenario element in similarity calculation, quantifying the weights of multiple factors. This helps decision-makers clarify the importance of each scenario element and improves the objectivity of the evaluation.

[0087] Meanwhile, for truly important scenario elements, this solution incorporates adjustment factors to regulate the weighting of considerations, thereby obtaining matching scenario elements more reasonably and efficiently.

[0088] Example 5 In one implementation, step 240 above, calculating the similarity between the target scenario elements in the target accident scenario and the historical scenario elements in the scenario sample library constructed by the above knowledge graph-based accident scenario construction method, may include the following steps.

[0089] (A) Calculate the structural similarity between the target scenario elements in the target accident scenario and the historical scenario elements of each historical accident scenario in the scenario sample library.

[0090] Based on the weights of the scenario elements, the structural similarity between the target accident scenario and each historical accident scenario in the scenario sample library is calculated. The structural similarity is used to indicate the similarity between the attribute types included in the target scenario elements of the target accident and the attribute types included in the historical scenario elements of the historical accident.

[0091] The attribute features of historical scenario elements in the scenario sample library may be missing. When calculating the structural similarity between the target accident scenario and historical accident scenarios, some scenarios may not be able to be matched one-to-one, and it is necessary to determine their structural similarity. Specifically, the non-empty scenario elements in the scenario sample are denoted as a set. Let the non-empty scenario elements of the target accident be denoted as a set. Take set and set The intersection and union of the set are denoted as . and Calculate the intersection separately Union The sum of the weights of the context elements is denoted as: and Next, calculate the ratio of the intersection to the union weights, and use this ratio as the structural similarity between the target accident scenario and historical accident scenarios. The formula for calculating structural similarity is as follows:

[0092] (B) Calculate the attribute similarity between the target scenario elements in the target accident scenario and the historical scenario elements of each historical accident scenario in the scenario sample library.

[0093] Based on the weights of scenario elements, the attribute similarity between the target accident scenario and each historical accident scenario in the scenario sample library is calculated. The attributes of scenario elements include at least one of numerical attributes, conceptual attributes, and fuzzy attributes. These attributes distinguish the different methods used to calculate the similarity of each scenario element. Conceptual attributes refer to the features of scenario elements described using words or symbols; fuzzy attributes refer to the features of scenario elements described using fuzzy variables with interval values. Attribute similarity refers to the fact that different contextual elements have different attributes, and the similarity calculation methods also differ. This disclosure provides context similarity calculation methods for numerical attributes, conceptual attributes, and fuzzy attributes based on attribute categories.

[0094] Among them, numerical attribute scenario elements refer to scenario elements that can be represented by specific numerical values, such as the number of deaths, the number of injuries, and economic losses; conceptual attribute scenario elements refer to scenario elements whose characteristics can be described by words or symbols, such as classifying mine types as: high-gas mines, low-gas mines, and coal and gas outburst mines; fuzzy attribute scenario elements refer to scenario elements that are fuzzy variables without clearly defined numerical range boundaries, such as the state of a ventilation system being classified as: normal, partially damaged, and completely interrupted.

[0095] For context elements with numerical attributes, attribute similarity is calculated using a distance metric, specifically using the following Euclidean distance formula:

[0096] in, , These are the specific numerical values ​​of the scenario elements in the scenario sample library and the corresponding scenario element attributes of the target accident. This refers to the maximum value of the element attribute in that scenario. This refers to the minimum value of the element attribute in that scenario.

[0097] For scenario elements with conceptual attributes, the attribute similarity calculation method involves comparing whether they are identical. This involves pairwise comparisons of the conceptual attributes in the scenario element database and the corresponding scenario element of the target accident. If they are identical, the attribute similarity is 1; otherwise, the attribute similarity is 0. The calculation formula is as follows:

[0098] in, , These represent the specific attribute content of the scenario elements in the scenario sample library and the corresponding scenario elements of the target accident.

[0099] For context elements with fuzzy attributes, their attribute similarity can be calculated by using evaluation values ​​with linguistic variables to measure the performance of the context element. A pre-defined correspondence between context elements and their evaluation values ​​is established; these linguistic evaluation values ​​are taken from a user-defined set of linguistic value evaluations. Linguistic triangular fuzzy numbers are then used to transform fuzzy and uncertain linguistic variables into definite numerical values.

[0100] Among them, triangular fuzzy numbers are used to solve problems in uncertain environments. The concept of fuzzy sets is mainly applied to quality management and risk management. Triangular fuzzy numbers are a special type of fuzzy numbers, and their membership functions are usually represented as triangular functions, in the form A = (a, b, c), where a, b, and c represent the lower limit, the most likely value, and the upper limit, respectively.

[0101] Specifically, language sets Represents an ordered set of language evaluation values, where For the first language in this collection m If there are multiple language evaluation results, then the triangular fuzzy number of the evaluation result can be expressed as: That is, the fuzzy number of any evaluation scale in the language set is:

[0102] After representing fuzzy language elements using trigonometric functions, the similarity can be calculated by constructing a similarity function using a negative exponential function, as shown in the formula below:

[0103]

[0104]

[0105] in, and These represent the evaluation scales for the fuzzy attributes of scenario elements in the historical scenario database and the corresponding target accident scenario elements, respectively. These represent the lower bound, most likely value, and upper bound of the evaluation scale for the fuzzy attributes of scenario elements in the scenario sample library, respectively. These represent the lower bound, most likely value, and upper bound of the evaluation scale for the fuzzy attributes of the scenario elements of the target accident, respectively.

[0106] The above methods calculate the similarity of each scenario element using different approaches for numerical, conceptual, and fuzzy attributes. Based on this, the comprehensive attribute similarity between scenario elements in the scenario sample library and the corresponding scenario elements of the target accident is calculated. After obtaining the similarity results for scenario elements of each attribute type, the comprehensive similarity of all attributes is given based on the nearest neighbor algorithm. This is the attribute similarity between the target accident and historical accidents. The formula for calculating this attribute similarity is as follows:

[0107] in, For the first Each context element attribute weight The number of attributes for scenario elements in the scenario sample library. This represents the total number of attributes for the context elements.

[0108] Specifically, the K-Nearest Neighbors (KNN) algorithm is used to calculate the distance between scenario elements in the scenario sample library and each scenario element in the target accident. The K closest scenario elements to the scenario elements in the target accident are found in the scenario sample library. Then, based on the above comprehensive attribute similarity calculation formula, the calculated K closest distances are multiplied by the corresponding scenario element weights, accumulated, and then divided by the sum of the weights of each scenario element in the scenario sample library to obtain the comprehensive attribute similarity between the target accident and historical accidents in terms of accident scenarios.

[0109] (C) Based on structural similarity and attribute similarity, calculate the global similarity between the target scenario elements in the target accident scenario and the historical scenario elements in the scenario sample library constructed based on the knowledge graph.

[0110] The global similarity of context elements refers to the weighted similarity of attributes calculated by multiplying the structural similarity and attribute similarity after calculating the structural similarity and attribute similarity. The formula is as follows: .

[0111] It should be noted that when the same letter symbol appears simultaneously in the aforementioned content, its meaning only has a relatively local meaning within the most recently cited paragraph; that is, it serves only as a local parameter, not a global parameter for the entire application document. Even if the same letter symbol appears in different locations, they should not be confused. For example, this includes, but is not limited to, calculating the similarity between context elements of numerical attributes. , These are the specific numerical values ​​of the scenario elements in the scenario sample library and the corresponding scenario element attributes of the target accident; however, when calculating the similarity between scenario elements with conceptual attributes, , These represent the specific attribute content of the scenario elements in the scenario sample library and the corresponding target accident scenario elements, respectively; when calculating the evaluation scale, and These represent the evaluation scales for the fuzzy attributes of scenario elements in the scenario sample library and the corresponding target accident scenario elements, respectively.

[0112] In this embodiment of the disclosure, the similarity between the target accident and historical accidents in the scenario sample library is evaluated from both structural and attribute aspects, making the similarity evaluation results more accurate and objective. Thus, based on the similarity evaluation results, timely prediction and emergency rescue plans can be obtained for the target accident.

[0113] Example 6 In this disclosure, Figure 2 In step S240, the calculated similarity is compared with a preset similarity threshold. If the similarity is greater than or equal to the preset similarity threshold, the emergency rescue plan corresponding to the historical accident scenario with the similarity is output from the emergency rescue plan sample library as the matching plan for the target accident.

[0114] Specifically, after calculating the global similarity between each scenario element of the target accident scenario and the scenario elements of each historical accident scenario stored in the scenario sample library, the global similarity values ​​can be sorted to obtain a series of case studies of scenario elements from historical accident scenarios with different similarities. To improve emergency decision-making capabilities and identify optimal emergency organizations and measures, a similarity threshold is set. Further screening is performed, including setting a similarity threshold. It is expressed as follows:

[0115] The aforementioned similarity threshold can be a value set based on comprehensive literature data, expert experience, and historical cases.

[0116] If the similarity between the final scenario elements is greater than or equal to a preset similarity threshold, it indicates that the two accident scenarios are quite similar. In this case, the emergency rescue plan corresponding to the historical accident scenario is used as the matching plan for the target accident from the emergency rescue plan sample library. If there are multiple valid scenario elements from historical accident scenarios with a high similarity (above the threshold) to the scenario elements of the target accident scenario, a valid accident scenario element set is formed, serving as a reference for emergency decision-making for the target accident. This completes the learning process of similar accident scenario solutions based on the scenario sample library. Furthermore, based on the above matching scheme, it is not necessary to directly apply similar scenario solutions to the emergency decision-making for the target accident. The matched case scenario solutions and methods for resolving such accidents can be appropriately modified according to the actual situation to suit the current case scenario, forming an emergency decision-making plan for the current case. Therefore, by comparing and analyzing the similarity between the target accident scenario and historical accident scenarios, the subsequent evolution of the accident can be predicted in a timely manner based on the current state of the target accident, and rescue and disposal plans can be obtained, improving the accuracy of system predictions and effectively enhancing accident handling capabilities.

[0117] Those skilled in the art should understand that the “threshold” mentioned herein is a value that can be adjusted and changed with different accidents and / or at different times, and is not a fixed value that remains unchanged.

[0118] In one implementation, after comparing the calculated similarity with a preset similarity threshold, if the similarity is less than the preset similarity threshold, an additional emergency rescue plan is generated for the target accident, and the target scenario elements in the target accident scenario are added to the scenario sample library, and the additionally generated emergency rescue plan is added to the emergency rescue plan sample library.

[0119] Specifically, if the similarity between the scenario elements of the target accident scenario and the scenario elements of historical accident scenarios is less than a preset similarity threshold, it indicates that the matching degree between the two is insufficient, and the historical accidents cannot guide emergency decision-making for the target accident scenario. In this case, the target accident can be treated as a new historical accident, and the scenario elements of the new historical accident can be saved to the scenario sample library to supplement and improve the scenario sample library. At the same time, additional emergency rescue plans developed for the target accident can also be added to the emergency rescue plan sample library.

[0120] Taking coal mine production scenarios as an example, the lack of data in some coal mines reduces the completeness of the coal mine accident scenario database, thus significantly affecting the accuracy of accident prediction. Therefore, to improve prediction accuracy, it is necessary to continuously improve the scenario sample database to address the problem of missing scenario elements. This involves supplementing or reconstructing accident scenarios to improve the scenario sample database and enhance the accuracy of system predictions by enriching the data content.

[0121] In one implementation, the target incident is treated as a new historical incident, and the scenario elements of the new historical incident scenario are saved, including: The Markov chain algorithm is used to decompose the evolution process of new historical accidents, obtain a reconstructed accident dialogue process, and save the reconstructed accident dialogue process and its associated context elements.

[0122] The Markov chain algorithm is used to process scenario elements of new historical accidents, obtain the relationships between scenario elements, and output ordered and combined scenario elements. Based on this order, an accident evolution process is reconstructed, and the scenario elements of the reconstructed accident evolution process are imported into the scenario sample library. Compared with Bayesian network structure diagrams, the order of construction by Markov chains is more accurate. Based on the Markov chain algorithm, strongly correlated scenario elements can be rearranged and combined, and then imported into the historical scenario library. In summary, the Markov chain algorithm recombines causal relationships, making the scenario sample library more strongly correlated in terms of causal relationships and exhibiting more obvious correlation characteristics, further optimizing data quality.

[0123] On the other hand, new measures can be sought for new historical accidents. Bayesian network modeling tools can be used to perform correlation operations on the scenario elements of adjacent time series, so that two adjacent measures can form a correlation or constraint, reconstruct the handling measures corresponding to the scenario nodes in the accident evolution process, and save the handling measures (emergency rescue plans) corresponding to the scenario nodes to further improve the response plan library.

[0124] In one example, such as Figure 7As shown, the accident scenario processing method according to embodiments of this disclosure may include three stages: inputting the target, matching calculation, and outputting the result. Specifically: In the inputting the target stage, information about the target scenario is acquired and input into the matching calculation stage along with information from a pre-built scenario sample library (hereinafter referred to as the "scenario library"). The target scenario information specifically includes relevant information about the target scenario elements of the target accident; the scenario sample library information includes the numerical values ​​of various indicators for each scenario sample. In the matching calculation stage, based on the information from the input scenario sample library, the weight value of each scenario element is calculated using the entropy method, and then the similarity between the scenario elements of the target accident and the scenario elements of historical accidents is calculated based on the weight values. Specifically, the structural similarity and attribute similarity between the scenario elements of the target accident and the scenario elements of historical accidents are first calculated, then the global similarity is calculated based on the structural similarity and attribute similarity, and the similarity calculation result is output to the output stage. In the output stage, based on the similarity calculation results and a preset similarity threshold, if the similarity between the scenario elements of the target accident and the scenario elements of historical accidents is greater than or equal to the preset similarity threshold, then the emergency response measures (emergency rescue plans) of historical accidents are learned, and the corresponding response plans of historical accidents are used as the matching plans for the target accident; if the similarity between the target accident and historical accidents is less than the preset similarity threshold, then the target accident is treated as a new historical accident, and the relevant information of the new historical accident is entered into the scenario sample library to improve the content of the scenario sample library.

[0125] In an example of a coal mine production scenario, the scenario data processing device constructs a scenario sample library and a response library (emergency rescue plan sample library) based on a historical scenario element library. The evolution process of coal mine production accidents is obtained from historical accident scenarios, and scenario node variables are determined based on the accident evolution process and information from the scenario sample library. Then, a Bayesian network structure diagram of the coal mine accident scenario evolution is constructed based on the causal and dependency relationships between scenario nodes. Node probabilities are calculated using the Bayesian joint probability formula to determine emergency response points. This emergency response point information is input into the system for similarity calculation with historical scenario elements, thereby determining emergency measures under the coal mine production scenario. When calculating similarity, the focus is on scenario elements related to the emergency response points.

[0126] See Figure 8When a target incident occurs, the information of the target scenario is first imported. This information specifically includes details about the scenario elements of the target incident. Based on information from the historical scenario database (scenario sample database), an entropy-based weighting method is used to calculate the weight value of each scenario element. Then, based on these weight values, the global similarity between the scenario elements of the target incident and those of historical incidents is calculated. The global similarity calculation results are then judged against a preset similarity threshold. If the similarity between the scenario elements of the target incident and those of historical incidents is greater than or equal to the preset threshold, the historical measures database (emergency rescue plan sample database) is used. If the similarity is less than the preset threshold, the target incident is treated as a new historical incident, and its relevant information is entered into the historical scenario database. Correspondingly, any new emergency measures identified are also entered into the historical measures database.

[0127] Example 7 In this embodiment of the disclosure, the method for handling the above-mentioned accident scenario further includes: Collect historical scenario information from multiple historical accidents, and after noise reduction processing of the historical scenario elements extracted from the historical scenario information of multiple historical accidents, construct the historical accident scenario of each historical accident based on the noise-reduced historical scenario information. The scenario sample library is constructed by using historical scenario elements from each historical accident scenario as samples. Emergency rescue plans are generated for various historical accident scenarios, and an emergency rescue plan sample library is constructed using these emergency rescue plans as samples. The scenario sample library and the emergency rescue plan sample library are two independent sample libraries, or the scenario sample library and the emergency rescue plan sample library are integrated together as a single sample library through a correspondence relationship.

[0128] Some steps in this embodiment are similar to the processing of relevant data of the target accident in the aforementioned method embodiments, and will not be described in detail here.

[0129] In summary, the embodiments of this disclosure have at least the following technical advantages: By collecting accident case analysis, accident scenario elements are obtained to form a historical scenario element library, which in turn forms a historical scenario library (scenario sample library) and a historical measures library (emergency rescue plan sample library). Then, a Bayesian structure diagram of the accident scenario is constructed to clearly show the causal relationship between accident scenario elements, which helps to understand the development process of the accident; By combining deep learning algorithms and Laplace smoothing technology, the probability distribution of network nodes can be determined more accurately, and emergency key points can be generated, thereby improving the accuracy of emergency response; By calculating the structural similarity between the scenario elements of the target story scenario and the scenario elements of historical accidents in the scenario sample library, and by combining the attribute similarity calculated based on the weights of each scenario element determined by the entropy weight method, the global similarity is finally determined, which helps to quickly locate similar historical scenarios, thereby providing a reference for emergency response; By setting a threshold for scenario similarity, similar scenario solutions are found, and historical data is supplemented and reorganized to continuously update and optimize the scenario sample library and the emergency rescue plan sample library, thereby improving the adaptability and accuracy of the system.

[0130] Example 8 like Figure 9 As shown, this disclosure also provides a knowledge graph-based accident scenario construction system, which can be used to execute the knowledge graph-based accident scenario construction method in the above embodiments. The system includes: The data filtering module 910 is used to obtain historical accident cases, filter the historical scenario information contained in the obtained historical accident cases, remove duplicate historical accident cases, and perform field integrity verification on unstructured historical scenario information to complete the filtering of invalid historical scenario information and obtain the filtered historical scenario information. The first element extraction module 920 is used to extract multiple historical scenario elements from the filtered historical scenario information corresponding to historical accidents. The map construction module 930 is used to divide the historical scenario information into two stages: internal handling and external rescue, and form a modular system for emergency response. According to the modular system, the extracted historical scenario elements are used as key entities to construct a historical accident map. The sample optimization module 940 is used to group the historical scenario elements in the historical accident map using a knowledge graph-based multimodal clustering analysis algorithm; to express the features of the grouped historical scenario elements in the form of feature vectors, to construct the ontological features, disaster chain features and emergency response features of each historical accident, and to identify the commonalities and differences of different categories of historical accident scenarios using dynamic weighted spectrum clustering and scenario clustering. An optimization module 950 is constructed to construct and optimize a scenario sample library and an emergency rescue plan sample library based on the identification results. The scenario sample library contains historical scenario elements of historical accident scenarios, and the emergency rescue plan sample library contains emergency rescue elements of emergency rescue plans corresponding to the historical accident scenarios.

[0131] For details regarding the functions, beneficial effects, or technical problems solved by each module of the above system, please refer to the descriptions in the corresponding methods or the descriptions in the invention content; they will not be repeated here.

[0132] Example 9 like Figure 10 As shown, this disclosure also provides a system for processing accident scenario data, which can be used to execute the accident scenario data processing method in the above embodiments. The system includes: The second element extraction module 101 is used to extract multiple target scenario elements from the target scenario information corresponding to the target accident. The scenario evolution module 102 is used to extract the scenario state S, initial disaster-causing factor C, disaster consequence state I, and emergency rescue activities E of the target accident from the multiple target scenario elements to form scenario nodes and their variables in the evolution of the target accident, obtain the evolution process of the target accident, and thus integrate the emergency rescue process in the evolution process of the target accident to obtain the target accident scenario. The calculation module 103 is used to calculate the similarity between the target scenario element in the target accident scenario and the historical scenario element in the scenario sample library of the above-mentioned accident scenario construction method based on knowledge graph. The weight of each scenario element in the similarity calculation process is determined by a weight adjustment method based on the attributes of each scenario element in the accident scenario. The processing module 104 is used to compare the calculated similarity with a preset similarity threshold. If the similarity is greater than or equal to the preset similarity threshold, the emergency rescue plan corresponding to the historical accident scenario corresponding to the similarity is output from the emergency rescue plan sample library as the matching plan for the target accident.

[0133] Optionally, the calculation module 103 is specifically used to calculate the structural similarity between the target scenario elements in the target accident scenario and the historical scenario elements of each historical accident scenario in the scenario sample library; calculate the attribute similarity between the target scenario elements in the target accident scenario and the historical scenario elements of each historical accident scenario in the scenario sample library; and, based on the structural similarity and the attribute similarity, calculate the global similarity between the target scenario elements in the target accident scenario and the historical scenario elements of each historical accident scenario in the scenario sample library constructed based on the knowledge graph.

[0134] Optionally, the calculation module 103 is used to determine the weight of each scenario element in the similarity calculation process based on the attributes of each scenario element in the accident scenario, through a weight adjustment method, including: The weights of each scenario element are calculated using the entropy method; Specifically, to address the correlation between multiple target scenario elements based on the target accident and historical scenario elements of historical accidents in the scenario sample library, a time characteristic factor is introduced to adjust the entropy value of each scenario element, so that the weight of the scenario elements with the highest correlation to the most recently occurred historical accident is increased.

[0135] Optionally, the processing module 104 is further configured to collect historical scenario information of multiple historical accidents, and after denoising the historical scenario elements extracted from the historical scenario information of the multiple historical accidents, construct historical accident scenarios for each historical accident based on the denoised historical scenario information; construct the scenario sample library using the historical scenario elements in each historical accident scenario as samples; generate emergency rescue plans for each historical accident scenario, and construct the emergency rescue plan sample library using the emergency rescue plans as samples; wherein, the scenario sample library and the emergency rescue plan sample library are two independent sample libraries, or the scenario sample library and the emergency rescue plan sample library are integrated together as a single sample library through a correspondence relationship.

[0136] Optionally, the processing module 104 is further configured to compare the calculated similarity with a preset similarity threshold, and if the similarity is less than the preset similarity threshold, generate an additional emergency rescue plan for the target accident, add the target scenario elements in the target accident scenario to the scenario sample library, and add the additionally generated emergency rescue plan to the emergency rescue plan sample library.

[0137] For details regarding the functions, beneficial effects, or technical problems solved by each module of the above system, please refer to the descriptions in the corresponding methods or the descriptions in the invention content; they will not be repeated here.

[0138] Example 10 Another embodiment of the present invention relates to an electronic device, such as... Figure 11 As shown, it includes at least one processor 112; and a memory 111 communicatively connected to at least one processor 112; wherein the memory 111 stores instructions executable by at least one processor 112, the instructions being executed by at least one processor 112 to enable at least one processor 112 to execute any of the above method embodiments.

[0139] The memory 111 and processor 112 are connected via a bus, which may include any number of interconnecting buses and bridges. The bus connects various circuits of one or more processors 112 and memory 111. The bus can also connect various other circuits, such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. A bus interface provides an interface between the bus and the transceiver. The transceiver can be a single element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by processor 112 is transmitted over a wireless medium via an antenna, which further receives data and transmits it to processor 112.

[0140] Processor 112 is responsible for managing the bus and general processing, and can also provide various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory 111 can be used to store data used by processor 112 during operation.

[0141] Another embodiment of the present invention relates to a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements any of the above-described method embodiments.

[0142] Another embodiment of the present invention relates to a computer program product that, when run on a terminal device, causes an electronic device to perform any of the above-described embodiments of the method.

[0143] That is, those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0144] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0145] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0146] Those skilled in the art will understand that the above embodiments are specific examples of implementing the present invention, and in practical applications, various changes in form and detail may be made without departing from the spirit and scope of the present invention.

Claims

1. A method for handling accident scenarios, characterized in that, include: S1, extract multiple target scenario elements from the target scenario information corresponding to the target accident; S2, extract the scenario state S, initial disaster-causing factor C, disaster consequence state I, and emergency rescue activity E of the target accident from the multiple target scenario elements to form scenario nodes and their variables in the evolution of the target accident. Based on the causal and dependency relationships between the scenario nodes, construct a dynamic Bayesian network structure diagram of the accident to obtain the evolution process of the target accident. Thus, the emergency rescue process is integrated into the evolution process of the target accident to obtain the target accident scenario. In the Bayesian network structure diagram, the scenario state S includes the initial scenario state S0 and subsequent scenario states S1 and S2. i The initial scenario state S0 represents the scenario state when the accident occurs under the influence of the initial causative factor C, and the subsequent scenario state S... i The initial scenario state S0 represents the evolving situation under the influence of disaster consequence state I and emergency rescue activities E. The initial scenario state S0 includes excessively high gas concentration, and the emergency rescue activities E include cutting off power, restoring ventilation, evacuating personnel, and drilling for extraction. If the disaster consequence situation I involves ventilation equipment, then after taking emergency rescue activity E, the subsequent scenario state S in the Bayesian network structure diagram will be... i To restore ventilation equipment to normal operation; If the disaster consequence situation I is a roadway, then after taking emergency rescue activity E, the subsequent scenario state S in the Bayesian network structure diagram is... i To reduce gas concentration; If the disaster consequence situation I involves personnel, then after taking emergency rescue activities E, the subsequent scenario state S in the Bayesian network structure diagram will be... i To evacuate personnel to a safe area; If the disaster consequence situation I is the tunneling face, then after taking emergency rescue activity E, the subsequent scenario state S in the Bayesian network structure diagram is... i To slow down the rate of gas accumulation; The evolution process includes: the accident occurrence stage, the accident evolution stage, and the accident disappearance stage; S3, calculate the similarity between the target scenario element in the target accident scenario and the historical scenario element in the scenario sample library of historical accident scenarios constructed by the accident scenario construction method based on knowledge graph, wherein, based on the attributes of each scenario element in the accident scenario, the weight of each scenario element in the process of calculating the similarity is determined by a weight adjustment method. The method of determining the weight of each scenario element in the similarity calculation process by adjusting the weight includes: calculating the weight of each scenario element using the entropy method; and introducing a time characteristic factor to adjust the entropy value of each scenario element based on the correlation between multiple target scenario elements based on the target accident and historical scenario elements of historical accidents in the scenario sample library, so that the weight of the scenario elements with the highest correlation with the most recently occurred historical accident is increased. S4. The calculated similarity is compared with a preset similarity threshold. If the similarity is greater than or equal to the preset similarity threshold, the emergency rescue plan corresponding to the historical accident scenario corresponding to the similarity is output from the emergency rescue plan sample library as the matching plan for the target accident. The knowledge graph-based accident scenario construction method includes: S01, obtain historical accident cases, filter the historical scenario information contained in the obtained historical accident cases, remove duplicate historical accident cases, and perform field integrity verification on unstructured historical scenario information to complete the filtering of invalid historical scenario information and obtain the filtered historical scenario information; S02, extract multiple historical scenario elements from the filtered historical scenario information corresponding to historical accidents; S03, the historical scenario information is divided into two stages: internal handling and external rescue, and a modular system for emergency response is formed. According to the modular system, the extracted historical scenario elements are used as key entities to construct a historical accident map. S04, a multimodal clustering analysis algorithm based on knowledge graph is used to group the historical scenario elements in the historical accident graph; feature vectors are used to express the features of the grouped historical scenario elements, constructing the ontological features, disaster chain features and emergency response features of each historical accident, and dynamic weighted spectrum clustering and scenario clustering are used to identify the commonalities and differences of different categories of historical accident scenarios; S05, Based on the recognition results, construct and optimize the scenario sample library and the emergency rescue plan sample library. The scenario sample library contains historical scenario elements of historical accident scenarios, and the emergency rescue plan sample library contains emergency rescue elements of emergency rescue plans corresponding to the historical accident scenarios.

2. The method according to claim 1, characterized in that, The context element dimensions corresponding to the multiple historical context elements extracted from the filtered historical context information corresponding to historical accidents include: accident object, cause of disaster, disaster consequences, and disaster relief subject; The accident object is used to describe the accident itself and its objective characteristics. The contextual elements of the accident object include spatial information and temporal information of the accident type, as well as the basic overview of the accident scene. The causes of disaster are used to describe the influencing factors of the occurrence and development of accident risks. The scenario elements of the causes of disaster include the causes of the accident and the causes of its development. The disaster consequences are used to describe the objects affected by the accident risk. The scenario elements of the disaster consequences include the degree of damage to the accident carriers, including on-site workers, ventilation facilities, and tunnels, as well as the secondary accidents caused by the accident. The disaster relief entity is used to describe the emergency response process and results in response to an accident. The scenario elements of the disaster relief entity include the resources invested in the emergency activities, the participating emergency organizations, the technical measures taken, and the resulting response.

3. The method according to claim 1, characterized in that, The calculation of the similarity between the target scenario elements in the target accident scenario and the historical scenario elements in the scenario sample library constructed by the knowledge graph-based accident scenario construction method includes: Calculate the structural similarity between the target scenario elements in the target accident scenario and the historical scenario elements of each historical accident scenario in the scenario sample library; Calculate the attribute similarity between the target scenario elements in the target accident scenario and the historical scenario elements of each historical accident scenario in the scenario sample library; Based on the structural similarity and the attribute similarity, the global similarity between the target scenario elements in the target accident scenario and the historical scenario elements of each historical accident scenario in the scenario sample library is calculated.

4. The method according to claim 1, characterized in that, The method further includes: Collect historical scenario information of multiple historical accidents, and after noise reduction processing of the historical scenario elements extracted from the historical scenario information of multiple historical accidents, construct the historical accident scenario of each historical accident based on the noise-reduced historical scenario information. The scenario sample library is constructed by using historical scenario elements from each historical accident scenario as samples. Emergency rescue plans are generated for each historical accident scenario, and an emergency rescue plan sample library is constructed using these emergency rescue plans as samples. The scenario sample library and the emergency rescue plan sample library are two independent sample libraries, or the scenario sample library and the emergency rescue plan sample library are integrated together as a single sample library through a correspondence relationship.

5. The method according to claim 1, characterized in that, After comparing the calculated similarity with a preset similarity threshold, the method further includes: If the similarity is less than the preset similarity threshold, an additional emergency rescue plan is generated for the target accident, and the target scenario elements in the target accident scenario are added to the scenario sample library, as well as the additionally generated emergency rescue plan is added to the emergency rescue plan sample library.

6. A knowledge graph-based accident scenario construction system that applies the accident scenario processing method of any one of claims 1 to 5, characterized in that, include: The data filtering module is used to obtain historical accident cases, filter the historical scenario information contained in the obtained historical accident cases, remove duplicate historical accident cases, and perform field integrity verification on unstructured historical scenario information to filter invalid historical scenario information and obtain the filtered historical scenario information. The first element extraction module is used to extract multiple historical scenario elements from the filtered historical scenario information corresponding to historical accidents. The map construction module is used to divide the historical scenario information into two stages: internal handling and external rescue, and form a modular system for emergency response. According to the modular system, the extracted historical scenario elements are used as key entities to construct a historical accident map. The sample optimization module is used to group the historical scenario elements in the historical accident map using a knowledge graph-based multimodal clustering analysis algorithm; it uses feature vectors to express the features of the grouped historical scenario elements, constructs the ontological features, disaster chain features and emergency response features of each historical accident, and uses dynamic weighted spectrum clustering and scenario clustering to identify the commonalities and differences of different categories of historical accident scenarios; An optimization module is constructed to build and optimize a scenario sample library and an emergency rescue plan sample library based on the identification results. The scenario sample library contains historical scenario elements of historical accident scenarios, and the emergency rescue plan sample library contains emergency rescue elements of emergency rescue plans corresponding to the historical accident scenarios.

7. A system for handling accident scenarios using the method for handling any one of claims 1 to 5, characterized in that, include: The second element extraction module is used to extract multiple target scenario elements from the target scenario information corresponding to the target accident. The scenario evolution module is used to extract the scenario state S, initial disaster-causing factor C, disaster consequence state I, and emergency rescue activities E of the target accident from the multiple target scenario elements to form scenario nodes and their variables in the evolution of the target accident, obtain the evolution process of the target accident, and thus integrate the emergency rescue process in the evolution process of the target accident to obtain the target accident scenario. The calculation module is used to calculate the similarity between the target scenario elements in the target accident scenario and the historical scenario elements in the scenario sample library of historical accident scenarios constructed by the accident scenario construction system based on knowledge graph. The weight of each scenario element in the similarity calculation process is determined by a weight adjustment method based on the attributes of each scenario element in the accident scenario. The processing module is used to compare the calculated similarity with a preset similarity threshold. If the similarity is greater than or equal to the preset similarity threshold, the emergency rescue plan corresponding to the historical accident scenario corresponding to the similarity is output from the emergency rescue plan sample library as the matching plan for the target accident. The knowledge graph-based accident scenario construction system includes: The data filtering module is used to obtain historical accident cases, filter the historical scenario information contained in the obtained historical accident cases, remove duplicate historical accident cases, and perform field integrity verification on unstructured historical scenario information to filter invalid historical scenario information and obtain the filtered historical scenario information. The first element extraction module is used to extract multiple historical scenario elements from the filtered historical scenario information corresponding to historical accidents. The map construction module is used to divide the historical scenario information into two stages: internal handling and external rescue, and form a modular system for emergency response. According to the modular system, the extracted historical scenario elements are used as key entities to construct a historical accident map. The sample optimization module is used to group the historical scenario elements in the historical accident map using a knowledge graph-based multimodal clustering analysis algorithm; it uses feature vectors to express the features of the grouped historical scenario elements, constructs the ontological features, disaster chain features and emergency response features of each historical accident, and uses dynamic weighted spectrum clustering and scenario clustering to identify the commonalities and differences of different categories of historical accident scenarios; An optimization module is constructed to build and optimize a scenario sample library and an emergency rescue plan sample library based on the identification results. The scenario sample library contains historical scenario elements of historical accident scenarios, and the emergency rescue plan sample library contains emergency rescue elements of emergency rescue plans corresponding to the historical accident scenarios.

Images