Method, system, device and medium for accident analysis based on large model knowledge enhancement

Abstract

The application relates to an accident analysis method, system, device and medium based on large model knowledge enhancement, which comprises the following steps: inputting obtained safety accident text data into a pre-constructed large language model to extract the location and production link of a safety accident; extracting entities and personnel operation behaviors in the location and production link by using the large language model; performing mixed knowledge retrieval according to the extracted entities to obtain an entity knowledge set; classifying and reordering the obtained entity knowledge set according to a pre-established knowledge standard, the location and production link and the personnel operation behaviors to obtain an entity knowledge list after reordering; and inputting prompt words composed of the safety accident text, the location and production link, the entity knowledge list after reordering and a safety accident cause request into the large language model to generate a safety accident cause analysis text. The application can be widely applied to the technical field of knowledge enhancement and reordering in computer natural language processing, especially in the field of industrial safety production.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of knowledge enhancement and reordering in computer natural language processing, and specifically relates to the field of industrial safety production, and particularly relates to an accident analysis method, system, device and medium based on large model knowledge enhancement. BACKGROUND

[0002] Industrial safety is one of the important factors to ensure normal production, and it is crucial to quickly analyze the causes of sudden abnormal conditions to take effective emergency measures.

[0003] Although the existing large language model has strong language generation capability, it has the risk of "illusion" for the analysis of a certain result in a specific field, and lacks the ability to dynamically evaluate the safety situation of key entities in the context, which leads to two typical defects in the accident cause generated by the model: one is difficult to accurately locate the key cause element, and the other is even if the key cause is identified, it cannot correctly evaluate its importance weight. These limitations seriously restrict the practical application value of large language models in the field of industrial safety: on the one hand, due to the misjudgment or omission of core causes, the best disposal opportunity may be missed; on the other hand, improper weight allocation will lead the safety measures formulated subsequently to deviate from the real risk control focus, making it difficult for this advanced technology to play its due role in decision support in key links such as preventive maintenance and emergency response. SUMMARY

[0004] In view of the above problems, the purpose of the present application is to provide an accident analysis method, system, device and medium based on large model knowledge enhancement, to overcome the defects of the existing large language model in the analysis of industrial safety accidents, to solve the two core problems of inaccurate positioning of key causes and improper evaluation of importance, and to make the large language model technology play its due role in decision support in the field of industrial safety.

[0005] To achieve the above purpose, the present application adopts the following technical solutions:

[0006] In the first aspect, the present application provides an accident analysis method based on large model knowledge enhancement, comprising the following steps:

[0007] Input the obtained safety accident text data into the pre-constructed large language model, and extract the location and production link of the safety accident;

[0008] According to the extracted location and production link, the entity and personnel operation behavior in the location and production link are extracted by using the large language model;

[0009] According to the extracted entity, a mixed knowledge retrieval is performed to obtain an entity knowledge set:

[0010] According to the pre-established knowledge standard, the place and the production link and the personnel operation behavior, the obtained entity knowledge set is classified, and the classified entity knowledge is reordered to obtain an entity knowledge list after reordering;

[0011] According to the safety accident text, the place and the production link, the reordered entity knowledge list and the safety accident reason request, a prompt word is constituted, and a large language model is input to generate a safety accident reason analysis text.

[0012] Further, the safety accident text data obtained is input into a large language model constructed in advance to extract the place and the production link of the safety accident, comprising:

[0013] A prompt word for extracting the place and the production link is constructed, which is represented as:

[0014] {First-Prompt}={Accident}+{First-Query}

[0015] Wherein, {First-Prompt} represents the prompt word for extracting the place and the production link; {Accident} represents the safety accident text; {First-Query} represents the request for indicating the place and the production link of the safety accident;

[0016] The constructed prompt word for extracting the place and the production link is input into a large language model to output the place and the production link of the safety accident.

[0017] Further, according to the extracted place and the production link, the entity and the personnel operation behavior in the place and the production link are extracted by using a large language model, comprising:

[0018] According to the extracted place and the production link, a prompt word for extracting the entity and the personnel operation behavior is constructed respectively;

[0019] Wherein, the prompt word for extracting the entity is represented as:

[0020] {Second-Prompt}={Accident}+{Routine}+{Second-Query}

[0021] The prompt word for extracting the personnel operation behavior is represented as:

[0022] {Third-Prompt}={Accident}+{Third-Query}

[0023] {Second-Prompt} and {Third-Prompt} represent prompt words for extracting entities and personnel operation behaviors respectively; {Accident} represents a safety accident text; {Routine} represents a place and a production link; {Second-Query} represents a request for extracting entities in the place and the production link in the safety accident text; and {Third-Query} represents a request for extracting personnel operation behaviors in the safety accident text.

[0024] The constructed prompt words for extracting entities and personnel operation behaviors are respectively input into a large language model to obtain a preset number of entities and personnel operation behaviors.

[0025] Further, the mixed knowledge retrieval according to the extracted entities obtains an entity knowledge set, including:

[0026] The entity is input into a search engine as a request to crawl a descriptive and functional entity knowledge set of the entity.

[0027] If the entity knowledge fails to be crawled in the previous step, the entity is used to construct a prompt word, which is input into a large language model to obtain an entity knowledge set.

[0028] Further, the obtained entity knowledge set is classified according to a pre-established knowledge standard, a place, a production link and a personnel operation behavior, and the classified entity knowledge is reordered to obtain a reordered entity knowledge list, including:

[0029] The obtained entity knowledge set is classified based on a pre-established knowledge standard and a personnel operation behavior.

[0030] The classified entity knowledge is reordered to generate a reordered entity knowledge list.

[0031] Further, the obtained entity knowledge set is classified based on a pre-established knowledge standard and a personnel operation behavior, including:

[0032] The knowledge standard is established to divide the entity knowledge set into three categories of relevant valid knowledge, irrelevant knowledge and relevant knowledge.

[0033] Based on the established knowledge standard, a place, a production link and a personnel operation behavior, an entity knowledge classification prompt word is constructed.

[0034] The entity knowledge classification prompt word is:

[0035] {Fourth-Prompt} = {Knowledge-Norm} + {ENT1-Knowledge1} + {ENT2-Knowledge2} + {ENT3-Knowledge3} + … + {ENTK Knowledge K}+{Routine}+{Action}+{Fourth-Query}

[0036] Wherein, {Fourth-Prompt} represents an entity knowledge classification prompt word; {Knowledge-Norm} represents a knowledge standard; {Routine} represents a place and a production link; {ENT1-Knowledge1}, …, {ENT K Knowledge K} represents an entity knowledge; {Action} represents a personnel operation behavior; {Fourth-Query} represents a request for classifying according to a knowledge standard;

[0037] The entity knowledge classification prompt word is input into a large language model, and classified entity knowledge is output.

[0038] Further, the classified entity knowledge is reordered, and a reordered entity knowledge list is generated, comprising:

[0039] A prompt word for reordering entity knowledge is constructed, and is represented as:

[0040] {Fifth-Prompt}={Classified-Knowledge}+{Fifth-Query}

[0041] Wherein, {Fifth-Prompt} represents a prompt word for reordering entity knowledge; {Classified-Knowledge} represents classified entity knowledge; {Fifth-Query} represents a request for sorting knowledge in a manner of sorting relevant knowledge in front, sorting irrelevant knowledge in back, and removing irrelevant knowledge;

[0042] The constructed prompt word for reordering entity knowledge is input into a large language model, and a reordered entity knowledge list is output.

[0043] In a second aspect, the present application provides an accident analysis system based on large model knowledge enhancement, comprising:

[0044] A place process extraction module is configured to input acquired safety accident text data into a pre-constructed large language model, and extract a place and a production link of a safety accident;

[0045] An entity and task operation extraction module is configured to extract, according to the extracted place and production link, an entity and a personnel operation behavior in the place and production link by using a large language model;

[0046] The mixed knowledge retrieval module is configured to perform mixed knowledge retrieval based on the extracted entity to obtain an entity knowledge set.

[0047] The mixed knowledge reordering module is configured to classify the obtained entity knowledge set based on a pre-established knowledge standard, location, production link and personnel operation behavior, reorder the classified entity knowledge, and obtain a reordered entity knowledge list.

[0048] The output module is configured to construct a prompt word based on the safety accident text, location, production link, reordered entity knowledge list and safety accident cause request, input a large language model, and generate a safety accident cause analysis text.

[0049] In a third aspect, the present application provides a computer readable storage medium storing one or more programs, the one or more programs including instructions that, when executed by a computing device, cause the computing device to perform the large model knowledge enhanced accident analysis method.

[0050] In a fourth aspect, the present application provides a computing device including one or more processors and a memory, the memory storing one or more programs configured to be executed by the one or more processors, the one or more programs including instructions for performing the large model knowledge enhanced accident analysis method.

[0051] The present application has the following advantages due to the above technical solutions:

[0052] The present application introduces a mixed knowledge retrieval part and a mixed knowledge reordering part, so that the large language model can combine internal knowledge and search engine retrieved knowledge. Compared with the general large language model automatic generation method, the present application can effectively reduce model hallucination, and improve the mining ability and inference ability of the large language model in analyzing the cause of the safety accident.

[0053] Therefore, the present application can be widely applied to the field of knowledge enhancement and reordering technology in computer natural language processing. BRIEF DESCRIPTION OF DRAWINGS

[0054] Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The accompanying drawings are included to provide a description of the preferred embodiments and are not meant to limit the present application. Throughout the drawings, the same reference numerals are used for the same components. In the drawings:

[0055] Figure 1 is a flowchart of the large model knowledge enhanced accident analysis method provided by the embodiments of the present application. DETAILED DESCRIPTION

[0056] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions of the embodiments of the present application will be described clearly and completely below with reference to the drawings of the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the described embodiments of the present application, all other embodiments obtained by a person of ordinary skill in the art belong to the scope of protection of the present application.

[0057] It should be noted that the terms used herein are only intended to describe specific embodiments, and are not intended to limit the exemplary embodiments according to the present application. As used herein, the singular form is intended to include the plural form unless the context clearly indicates otherwise, and it should also be understood that when the terms "comprise" and / or "include" are used in the specification, there is a presence of a feature, step, operation, device, component and / or combination thereof.

[0058] In some embodiments of the present application, a large model knowledge enhanced accident analysis method is provided, which is divided into a place and production link extraction part, an entity and personnel operation behavior extraction part, a mixed knowledge retrieval part, a mixed knowledge reordering part and a structured prompt word part. The present application provides entity knowledge with strong correlation with safety accidents after screening to reduce the language model illusion risk and enhance the ability of the language model to mine accident causes.

[0059] Correspondingly, in some other embodiments of the present application, a large model knowledge enhanced accident analysis system, device and medium are provided.

[0060] Embodiment 1

[0061] As shown in Figure 1 , the present application provides a large model knowledge enhanced accident analysis method, which comprises the following steps:

[0062] 1) input the obtained safety accident text data into a pre-constructed large language model to extract the place and production link of the safety accident;

[0063] 2) according to the extracted place and production link, use the large language model to extract the entity and personnel operation behavior in the place and production link;

[0064] 3) perform mixed knowledge retrieval according to the extracted entity to obtain an entity knowledge set:

[0065] 4) classify the obtained entity knowledge set according to the pre-established knowledge standard, place and production link and personnel operation behavior, and reorder the classified entity knowledge to obtain an entity knowledge list after reordering;

[0066] 5) According to the safety accident text, the place and production link, the reordered entity knowledge list and the safety accident reason request constitute the prompt word, input the large language model to generate the safety accident reason analysis text.

[0067] Further, in the above step 1), the following steps are included:

[0068] 1.1) Construct the prompt word for extracting the place and production link.

[0069] In this embodiment, the constructed prompt word for extracting the place and production link contains the safety accident text and the request (“point out the safety accident place and production link”). Assuming that the safety accident text is recorded as: {Accident}, the prompt word for extracting the place and production link is recorded as: {First-Prompt}; the request is recorded as: {First-Query}, then the prompt word for extracting the place and production link is expressed as:

[0070] {First-Prompt}={Accident}+{First-Query}

[0071] 1.2) Input the constructed prompt word for extracting the place and production link into the large language model, and output the safety accident place and production link.

[0072] Further, in the above step 2), the following steps are included:

[0073] 2.1) According to the extracted place and production link, construct the prompt word for extracting entity and personnel operation behavior respectively.

[0074] In this embodiment, the constructed prompt word for extracting entity contains the safety accident text, the place and production link and the request (“please extract the entity in the place and production link in the safety accident text”). Assuming that the prompt word for extracting entity is recorded as: {Second-Prompt}, the request is recorded as: {Second-Query}, and the place and production link is recorded as: {Routine}; then the prompt word for extracting entity is expressed as:

[0075] {Second-Prompt}={Accident}+{Routine}+{Second-Query}

[0076] The constructed prompt word for obtaining personnel operation behavior contains the safety accident text and the request (“please extract the operation of the character in the safety accident text”), assuming that the request is recorded as: {Third-Query}, then the prompt word for obtaining personnel operation behavior is expressed as:

[0077] {Third-Prompt}={Accident}+{Third-Query}

[0078] 2.2) Input the constructed prompt words of the extracted entities and personnel operation behaviors into the large language model respectively to obtain a preset number of entities and personnel operation behaviors.

[0079] In this embodiment, a constant K is set, representing the extracted entity of the application, referred to as Entity, i.e. ENT. After inputting the prompt words of the extracted entities into the large language model, the large language model outputs K entities, denoted as: {ENT1, ENT2, …, ENT K}; and after inputting the prompt words of the extracted personnel operation behaviors into the large language model, the large language model outputs the personnel operation behaviors, denoted as {Action}.

[0080] Further, in the above step 3), the following steps are included:

[0081] 3.1) Input the entity into the search engine as a request to crawl the descriptive and functional entity knowledge set of the entity;

[0082] 3.2) If the crawling of the entity knowledge fails in step 3.1), construct a prompt word with the entity and input it into the large language model to obtain the entity knowledge set.

[0083] In this embodiment, the entity knowledge set is denoted as: {ENT1-Knowledge1}, {ENT2-Knowledge2}, {ENT3-Knowledge3}, …, {ENT K -Knowledge K}.

[0084] Further, in the above step 4), the following steps are included:

[0085] 4.1) Classify the obtained entity knowledge set based on the pre-established knowledge standard and personnel operation behavior.

[0086] 4.2) Reorder the classified entity knowledge to generate a reordered entity knowledge list.

[0087] Further, in the above step 4.1), the following steps are included:

[0088] 4.1.1) Establish a knowledge standard.

[0089] In this embodiment, the knowledge standard is denoted as { Knowledge-Norm}, which is used to classify the entity knowledge into the following three categories:

[0090] ① Relevant knowledge: entity knowledge applicable to the site and production link and indicating personnel operation behavior failure;

[0091] ② Irrelevant knowledge: entity knowledge not applicable to the site and production link.

[0092] ③ Relevant knowledge: the remaining knowledge after classifying relevant and irrelevant knowledge.

[0093] 4.1.2) Based on the established knowledge standard, location and production link, and personnel operation behavior, build entity knowledge classification prompt words.

[0094] In this embodiment, the built entity knowledge classification prompt words are denoted as {Fourth-Prompt}, which includes knowledge standard, entity knowledge, location and production link, personnel operation behavior and request (“Please classify according to knowledge standard”), and is expressed as:

[0095] {Fourth-Prompt}={ Knowledge-Norm}+{ENT1-Knowledge1}+{ENT2-Knowledge2}+{ENT3-Knowledge3}+…+{ENT K -Knowledge K}+{Routine}+{Action}+{Fourth-Query}。

[0096] 4.1.3) Input the entity knowledge classification prompt words into the large language model, and output the classified entity knowledge.

[0097] In this embodiment, the classified entity knowledge is denoted as {Classified-Knowledge}.

[0098] Further, in the above step 4.2), the following steps are included:

[0099] 4.2.1) Build entity knowledge reordering prompt words.

[0100] In this embodiment, the built entity knowledge reordering prompt words are denoted as {Fifth-Prompt}, which includes classified entity knowledge {Classified-Knowledge} and request (“Please sort the relevant knowledge in front, the irrelevant knowledge behind, and remove the irrelevant knowledge”).

[0101] {Fifth-Prompt}={Classified-Knowledge}+{Fifth-Query}

[0102] 4.2.2) Input the built entity knowledge reordering prompt words into the large language model, and output the reordered entity knowledge list.

[0103] In this embodiment, the reordered entity knowledge is denoted as: {Reordered-Knowledge}.

[0104] Further, in the above step 5), the following steps are included:

[0105] 5.1) Construct the prompt of the safety accident analysis text.

[0106] In this embodiment, the constructed prompt of the safety accident analysis text is denoted as {Sixth-Prompt}, which includes the safety accident text, the location and production link, the reordered entity knowledge, and the request ("Please analyze the accident cause as comprehensively as possible according to the knowledge"), denoted as {Sixth-Query}, and the prompt is expressed as:

[0107] {Sixth-Prompt}={Accident}+{Rountine}+{Reordered-Knowledge}+{Sixth-Query}

[0108] 5.2) Input the prompt of the safety accident analysis text into the large language model to output the safety accident cause analysis text.

[0109] In this embodiment, the safety accident cause analysis text is denoted as: {Accident-Analysis}.

[0110] Embodiment 2

[0111] This embodiment illustrates the advantages of the accident analysis method based on large model knowledge enhancement proposed by the present application over the general large language model automatic generation of safety accident report method in a statistical sense.

[0112] The safety accident text for testing the module has a total of 200 safety accident texts, and the score uses BertScore. BertScore is a semantic evaluation index based on a pre-trained language model, which calculates the similarity between generated text and reference text in the BERT embedding space, and comprehensively measures the quality of text generation from three dimensions of precision, recall and F1 value, which can more accurately reflect the performance of generated text in semantic fidelity, content integrity and overall consistency.

[0113] Table 1 Comparison of basic prompts of the present application and general large language models

[0114]

[0115] As can be seen from Table 1, the basic prompt words of the method of the application add the most basic format requirements, such as "please analyze the causes of the safety accident as comprehensively as possible", and the overall report quality score is obviously higher than that of the generated report using only the same general large language model. This shows that the method of the application can reduce model hallucinations while improving the mining and inference capabilities of the large language model in analyzing the causes of the safety accident.

[0116] Embodiment 3

[0117] The above embodiment 1 provides an accident analysis method based on large model knowledge enhancement, and correspondingly, the present embodiment provides an accident analysis system based on large model knowledge enhancement. The system provided in the present embodiment can implement the accident analysis method based on large model knowledge enhancement of embodiment 1. The system can be implemented by software, hardware or a combination of software and hardware. For example, the system can include integrated or separate functional modules or functional units to perform the corresponding steps in the methods of embodiment 1. Since the system of the present embodiment is basically similar to the method embodiment, the description process of the present embodiment is relatively simple, and the related parts can be referred to the part of the description of embodiment 1. The system embodiment provided in the present embodiment is only illustrative.

[0118] The accident analysis system based on large model knowledge enhancement provided in the present embodiment comprises:

[0119] A location and production link extraction module for inputting the obtained safety accident text data into a pre-constructed large language model to extract the location and production link of the safety accident;

[0120] An entity and task operation extraction module for extracting entities and personnel operation behaviors in the location and production link using the large language model according to the extracted location and production link;

[0121] A mixed knowledge retrieval module for performing mixed knowledge retrieval according to the extracted entities to obtain an entity knowledge set;

[0122] A mixed knowledge reordering module for classifying the obtained entity knowledge set according to the pre-established knowledge standard, location and production link and personnel operation behavior, and reordering the classified entity knowledge to obtain a reordered entity knowledge list;

[0123] An output module for constructing prompt words according to the safety accident text, location and production link, reordered entity knowledge list and safety accident cause request, and inputting the prompt words into the large language model to generate a safety accident cause analysis text.

[0124] Embodiment 4

[0125] The embodiment provides a processing device corresponding to the accident analysis method based on large model knowledge enhancement provided in the embodiment 1, and the processing device can be a processing device for a client, such as a mobile phone, a notebook computer, a tablet computer, a desktop computer, etc., to execute the method in the embodiment 1.

[0126] The processing device includes a processor, a memory, a communication interface and a bus, the processor, the memory and the communication interface are connected through the bus to complete the communication between each other. The memory stores a computer program capable of running on the processor, and the processor executes the computer program to execute the accident analysis method based on large model knowledge enhancement provided in the embodiment 1.

[0127] Preferably, the memory can be a high-speed random access memory (RAM: Random Access Memory), and can also include a non-volatile memory, such as at least one disk memory.

[0128] Preferably, the processor can be a central processing unit (CPU), a digital signal processor (DSP) and various types of general-purpose processors, which are not limited here.

[0129] Embodiment 5

[0130] The accident analysis method based on large model knowledge enhancement in the embodiment 1 can be specifically implemented as a computer program product, and the computer program product can include a computer readable storage medium, which is loaded with computer readable program instructions for executing the accident analysis method based on large model knowledge enhancement described in the embodiment 1.

[0131] The computer readable storage medium can be a tangible device that maintains and stores instructions for use by an instruction execution device. The computer readable storage medium can be, for example but not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any combination of the above.

[0132] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer usable program code.

[0133] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flow or blocks. Figure 1 one or more flow or blocks.

[0134] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart block or blocks. Figure 1 one or more flow or blocks. Figure 1 one or more flow or blocks.

[0135] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flow or blocks. Figure 1 one or more flow or blocks.

[0136] Finally, it should be noted that the above-mentioned embodiments are merely used to illustrate the technical solutions of the present application, rather than limit the technical solutions of the present application. Although the present application has been described in detail with reference to the above-mentioned embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalent replaced without departing from the spirit and scope of the present application, and any modification or equivalent replacement should be covered in the protection scope of the claims of the present application.

Claims

1. A large model knowledge enhancement-based accident analysis method, characterized in that, The method comprises the following steps: inputting the obtained safety accident text data into a pre-constructed large language model, extracting the location and production link of the safety accident; extracting entities and personnel operation behaviors in the location and production link according to the extracted location and production link by using the large language model; performing mixed knowledge retrieval according to the extracted entities to obtain an entity knowledge set: classifying the obtained entity knowledge set according to the pre-established knowledge standard, location, production link and personnel operation behavior, and reordering the classified entity knowledge to obtain a reordered entity knowledge list; constructing a prompt word according to the safety accident text, location, production link, reordered entity knowledge list and safety accident cause request, inputting the prompt word into the large language model to generate a safety accident cause analysis text; the mixed knowledge retrieval according to the extracted entities to obtain the entity knowledge set comprises: inputting the entity into a search engine as a request to crawl descriptive and functional entity knowledge of the entity; if the entity knowledge fails to be crawled, constructing a prompt word with the entity and inputting the prompt word into the large language model to obtain the entity knowledge set.

2. The method of claim 1, wherein, the inputting of the obtained safety accident text data into the pre-constructed large language model to extract the location and production link of the safety accident comprises: constructing a prompt word for extracting the location and production link, which is represented as: {First-Prompt}={Accident}+{First-Query} wherein, {First-Prompt} represents the prompt word for extracting the location and production link; {Accident} represents the safety accident text; and {First-Query} represents a request for indicating the location and production link of the safety accident; inputting the constructed prompt word for extracting the location and production link into the large language model to output the location and production link of the safety accident.

3. The method of claim 1, wherein, the extracting of entities and personnel operation behaviors in the location and production link according to the extracted location and production link by using the large language model comprises: constructing prompt words for extracting entities and personnel operation behaviors respectively according to the extracted location and production link; wherein, the prompt word for extracting entities is represented as: {Second-Prompt}={Accident}+{Routine}+{Second-Query} the prompt word for extracting personnel operation behaviors is represented as: {Third-Prompt}={Accident}+{Third-Query} wherein, {Second-Prompt} and {Third-Prompt} represent the prompt words for extracting entities and personnel operation behaviors respectively; {Accident} represents the safety accident text; {Routine} represents the location and production link; {Second-Query} represents a request for extracting entities in the location and production link in the safety accident text; and {Third-Query} represents a request for extracting personnel operation behaviors in the safety accident text; inputting the constructed prompt words for extracting entities and personnel operation behaviors into the large language model respectively to obtain a preset number of entities and personnel operation behaviors. 4.The method of claim 1, wherein, The obtained entity knowledge set is classified based on the pre-established knowledge standard, location, production link and personnel operation behavior, and the classified entity knowledge is reordered to obtain an entity knowledge list after reordering, including: The obtained entity knowledge set is classified based on the pre-established knowledge standard and personnel operation behavior; The classified entity knowledge is reordered to generate an entity knowledge list after reordering.

5. The method of claim 4, wherein, The obtained entity knowledge set is classified based on the pre-established knowledge standard and personnel operation behavior, including: Establish a knowledge standard for classifying entity knowledge into three categories: relevant valid knowledge, irrelevant knowledge and related knowledge; Based on the established knowledge standard, location, production link and personnel operation behavior, construct entity knowledge classification prompt words; The entity knowledge classification prompt words are: {Fourth-Prompt}={Knowledge-Norm}+{ENT1-Knowledge1}+{ENT2-Knowledge2}+{ENT3-Knowledge3}+…+{ENT K -Knowledge K}+{Routine}+{Action}+{Fourth-Query} Wherein, {Fourth-Prompt} represents an entity knowledge classification prompt word; {Knowledge-Norm} represents a knowledge standard; {Routine} represents a place and a production link; {ENT1-Knowledge1}, …, {ENT K -Knowledge K} represents an entity knowledge; {Action} represents a personnel operation behavior; {Fourth-Query} represents a request according to a knowledge standard classification; The entity knowledge classification prompt words are input into a large language model to output the classified entity knowledge.

6. The method of claim 1, wherein, The classified entity knowledge is reordered to generate an entity knowledge list after reordering, including: Construct an entity knowledge reordering prompt word, represented as: {Fifth-Prompt}={Classified-Knowledge}+{Fifth-Query} Where {Fifth-Prompt} represents the entity knowledge reordering prompt word; {Classified-Knowledge} represents the classified entity knowledge; {Fifth-Query} represents a request for knowledge sorting in the manner of sorting relevant valid knowledge first, related knowledge second and irrelevant knowledge last; The constructed entity knowledge reordering prompt word is input into a large language model to output an entity knowledge list after reordering.

7. An accident analysis device based on large model knowledge enhancement, characterized by, Including: A location and production link extraction module for inputting the obtained safety accident text data into a pre-constructed large language model to extract the location and production link of the safety accident; An entity and task operation extraction module for extracting entities and personnel operation behaviors in the location and production link using a large language model based on the extracted location and production link; A mixed knowledge retrieval module for retrieving mixed knowledge based on the extracted entities to obtain an entity knowledge set; A mixed knowledge reordering module for classifying the obtained entity knowledge set based on the pre-established knowledge standard, location, production link and personnel operation behavior, and reordering the classified entity knowledge to obtain an entity knowledge list after reordering; An output module for inputting the prompt word composed of the safety accident text, location and production link, entity knowledge list after reordering and safety accident cause request into a large language model to generate a safety accident cause analysis text; The obtained entity knowledge set is classified based on the pre-established knowledge standard and personnel operation behavior, including: Enter the entity into the search engine as a request to crawl the descriptive and functional entity knowledge of the entity; If the entity knowledge retrieval fails, construct a prompt word with the entity and input it into a large language model to obtain an entity knowledge set.

8. A computer-readable storage medium storing one or more programs, the one or more programs comprising instructions that when executed by a computer cause the computer to perform a method of any of claims 1-7. The one or more programs include instructions that, when executed by a computing device, cause the computing device to perform any of the methods of claims 1-6.

9. A computing device, comprising: Including: One or more processors and memory having stored therein one or more programs configured to, working with the one or more processors, cause performance of any of the methods of claims 1-6.

Images