Emergency plan generation method for sudden biological safety event based on large model

CN119005672APending Publication Date: 2024-11-22CHINESE PEOPLE'S PUBLIC SECURITY UNIVERSITY
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202410907153.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-08
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

The traditional text-form biosafety incident emergency plans are difficult to be updated and adjusted quickly, resulting in low emergency response efficiency and ineffective response to emergencies.

Method used

The emergency plan generation method based on the big model is adopted, and the training set is constructed by collecting and cleaning relevant information, and the emergency plan is generated using the Transformer model, combining real-time information and user input to generate reliable emergency plan.

Benefits of technology

Ensure that emergency plans are practical and real-time, can respond to emergencies quickly and efficiently, and facilitate adjustment of plans through regular updates and prompt word templates.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119005672A_ABST
    Figure CN119005672A_ABST
Patent Text Reader

Abstract

The invention relates to a method for generating an emergency plan of an emergent biological safety event based on a large model, and the method comprises the steps: collecting the related information of the emergent biological safety event and the emergency plan, and generating a training set; training a Transform model based on the training set to construct an emergency plan generation model; constructing a cue word template for a user to perform model input; constructing an emergency plan data vector library by adopting a text vectorization technology; and calculating the similarity between the model input of the user and the emergency plan data content, extracting the emergency plan data content with the highest correlation based on the similarity, and inputting the emergency plan data content, the 3S information and the model input of the user into a large model to generate a complete emergency plan. By adopting the method disclosed by the invention, on the basis of a digital emergency plan knowledge representation system, the emergency workflow is analyzed, the characteristic information of the emergency event is identified, and a feasible emergency plan is provided for a decision maker by virtue of a large model system and in combination with objective information of the event.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of emergency plan generation for sudden biosafety incidents, and in particular relates to a method for generating emergency plan for sudden biosafety incidents based on a large model. Background Art

[0002] With the acceleration of globalization and the rapid development of biotechnology, the risk of sudden biosafety incidents has increased significantly. Once a sudden biosafety incident occurs, it will pose a serious threat to public safety and social stability. For example, infectious diseases spread rapidly between people, which may cause widespread social impact in a short period of time and seriously endanger public health and life safety. Biological attacks are more hidden and sudden, usually occurring without warning, causing serious impact on public safety. The complexity and urgency of these biosafety incidents make the speed of generating biosafety incident emergency plans particularly important.

[0003] The digitization and operability of traditional text-based emergency plans are low, making it difficult to meet the actual situation and inconvenient for self-inspection, modification and improvement. This lagging emergency plan reduces the efficiency of emergency plan generation, and it is often difficult to respond to sudden biosafety incidents quickly and efficiently, bringing huge challenges to the handling of biosafety incidents. Effective emergency response requires timely and accurate formulation and implementation of plans, and at the same time has a certain life cycle. With the evolution of the closed loop of emergency management "early warning prediction-plan preparation-corresponding disposal-assessment and recovery", emergency plans should also go through a cycle of revision, learning, execution, comparison and improvement to minimize the negative impact of the incident. Therefore, there is an urgent need to invent an effective description of emergency plans in text form, and to make timely adjustments as the incident evolves and develops, close to the actual situation, and to quickly display the updated plan to relevant personnel. Summary of the invention

[0004] In view of the defects existing in the prior art, the purpose of the present invention is to provide a method for generating emergency plans for sudden biosafety incidents based on a large model. On the basis of a digital emergency plan knowledge representation system, the emergency workflow is analyzed to identify the characteristic information of the emergency. With the help of a large model system and combined with objective information of the event, a feasible emergency plan is provided to decision makers.

[0005] To achieve the above purpose, the technical solution adopted by the present invention is: a method for generating an emergency plan for sudden biosafety incidents based on a large model, the method comprising the following steps:

[0006] S1. Collect information related to sudden biosafety incidents and their emergency response plans to generate a training set;

[0007] S2. Use the Transformer model as the infrastructure and train the Transformer model based on the training set to build an emergency plan generation model.

[0008] S3, constructing a prompt word template for users to input the model;

[0009] S4. Use text vectorization technology to build an emergency plan data vector library based on the collected information related to sudden biosafety incidents and their emergency plans;

[0010] S5. Calculate the similarity between the user's model input and the emergency plan data content. Based on the similarity, extract the emergency plan data content with the highest relevance, and input it into the big model together with the real-time collected spatial information and the user's model input to generate a complete emergency plan.

[0011] Further, step S1 includes the following sub-steps:

[0012] S11. Use information collection technology to obtain information related to sudden biosafety incidents and their emergency response plans;

[0013] S12. Through text cleaning, remove irrelevant information, invalid laws, and outdated technologies, retain key disposal information, and ensure that the emergency plan can cover various emergencies and is practical;

[0014] S13. The collected and cleaned emergency plan data is stored in the database, and a training set is generated for the emergency plan. After the training set is constructed, new data is pulled regularly to achieve real-time update of information related to sudden biosafety incidents and their emergency plans.

[0015] Furthermore, in step S11, the ways of obtaining relevant information on sudden biosafety incidents and their emergency response plans include historical cases, policy documents, professional literature and news reports.

[0016] Furthermore, the key disposal information in step S12 includes a disposal subject, a disposal object and a disposal method.

[0017] Further, step S2 includes the following sub-steps:

[0018] S21. The Transformer model is used as the basic structure. The Transformer model consists of two parts: the encoder and the decoder. The encoder is composed of several encoding layers, each of which contains a multi-head self-attention mechanism and a feedforward neural network.

[0019] S22. Introduce the Luong attention mechanism in the attention layer of each encoder and decoder to calculate the attention weight between the encoder and decoder;

[0020] S23. Train the Transformer model based on the training set to obtain an emergency plan generation model.

[0021] Furthermore, the design of the prompt word template in step S3 requires the user to input a model input including the region to which the sudden biosafety incident belongs, environmental conditions, transmission routes, impact range, pathogens and harmful substance characteristics.

[0022] Furthermore, in step S4, the collected emergency plans for sudden biosafety incidents are used as model input, and the vectors output by the text embedding model are stored in a vector database to construct an emergency plan data vector library.

[0023] Further, step S5 includes the following sub-steps:

[0024] S51. Use automatic information collection technology to update on-site information and spatial positioning in real time, and conduct corresponding spatial analysis;

[0025] S52, calculating the similarity between the feature vector corresponding to the model input content input by the user and the emergency plan data vector;

[0026] S53. Based on the vector similarity comparison result, the emergency plan data vector with the highest correlation is taken out and input into the big model together with the collected information and user input content to generate a complete emergency plan.

[0027] Further, the similarity score calculation formula in step S52 is as follows:

[0028]

[0029] Among them, Vec(D) represents the emergency plan data vector output after the information related to sudden biosafety incidents and their emergency plans are input into the text embedding model;

[0030] Vec(Q) represents the text vector output after the user's model input content is input into the text embedding model.

[0031] The beneficial technical effect of the present invention is that: the present invention discloses a method for generating an emergency plan for sudden biosafety incidents based on a large model, which removes irrelevant information, invalid laws, and obsolete technologies by performing text cleaning on the relevant information of sudden biosafety incidents and their emergency plans, and retains key disposal information such as disposal subjects, disposal objects, and disposal methods, so as to ensure that the emergency plan can cover various emergencies and has practical operability. By using the training set constructed by obtaining relevant information of sudden biosafety incidents and their emergency plans to train the model, it can be ensured that when a sudden biosafety incident occurs, the model can quickly output a reliable emergency plan, laying the foundation for quickly and efficiently responding to sudden biosafety incidents.

[0032] After building the training set, new data is pulled regularly to update the relevant information of sudden biosafety incidents and their emergency plans in real time, ensuring the accuracy and real-time performance of the model. By building a prompt word template for users to input the model, it is convenient to modify and adjust the emergency plan, so that the model output can be more in line with the actual situation. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 The present invention is a flowchart of a method for generating an emergency plan for sudden biosafety incidents based on a large model, as shown in the first embodiment of the present invention. DETAILED DESCRIPTION

[0034] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0035] Embodiment 1

[0036] like Figure 1 As shown, an embodiment of the present invention provides a method for generating an emergency plan for sudden biosafety incidents based on a large model, the method comprising the following steps:

[0037] S1. Collect information related to sudden biosafety incidents and their emergency response plans to generate a training set;

[0038] Step S1 includes the following three sub-steps:

[0039] S11. Use information collection technology to obtain relevant information on sudden biosafety incidents and their emergency response plans. The ways to obtain relevant information on sudden biosafety incidents and their emergency response plans include historical cases, policy documents, professional literature, and news reports.

[0040] S12. Through text cleaning, remove irrelevant information, invalid laws, and outdated technologies, retain key disposal information such as disposal subject, disposal object, and disposal method, to ensure that the emergency plan can cover various emergencies and is practical.

[0041] S13. Use the ElasticSearch document database to store the collected and cleaned data in the database in json format, and build an emergency plan to generate a training set. After building the training set, regularly pull new data to achieve real-time updates of information related to sudden biosafety incidents and their emergency plans.

[0042] S2. Use the Transformer model as the infrastructure, train the Transformer model based on the training set, and build an emergency plan generation model.

[0043] Step S2 includes the following three sub-steps:

[0044] S21. The Transformer model is used as the basic structure. The Transformer model consists of two parts: the encoder and the decoder. The encoder is composed of several encoding layers, each of which contains a multi-head self-attention mechanism and a feedforward neural network. The calculation formula of the self-attention mechanism is as follows:

[0045]

[0046] Among them, Q, K, V represent the matrices of query, key and value respectively, d k Represents the dimension of the attention head.

[0047] S22. The Luong attention mechanism is introduced in the attention layer of each encoder and decoder to calculate the attention weight between the encoder and decoder. The attention calculation formula is as follows:

[0048]

[0049] Among them, score(s t ,h i ) is obtained by calculating the dot product of the decoder hidden state and the encoder hidden state.

[0050] S23. Train the Transformer model based on the training set to obtain an emergency plan generation model.

[0051] S3. Construct a prompt word template for users to input the model.

[0052] In order to ensure the normalization and standardization of the generated emergency plans, a prompt word template is designed to require users to enter information such as the region, environmental conditions, transmission routes, impact range, pathogens or harmful substances characteristics of the sudden biosafety incident. The specific input content is as follows:

[0053] The region refers to the specific location where the incident occurred, such as "XX Province", "XX City", "XX County", etc.;

[0054] Environmental conditions refer to the specific environment where the event occurred, such as "enterprise factory", "public transportation", "residential shopping mall", etc.;

[0055] Transmission route refers to the way the event is spread or diffused, such as "air transmission", "contact transmission", "water source transmission", etc.;

[0056] The scope of impact refers to the scope of losses and the degree of impact caused by the incident, such as "casualties", "property losses", "environmental pollution", etc.

[0057] Pathogen or harmful substance characteristics refer to the material properties that caused the incident, such as "viral infection", "food poisoning" and "hazardous chemical leakage".

[0058] S4. Use text vectorization technology to build an emergency plan data vector library based on the collected information related to sudden biosafety incidents and their emergency plans;

[0059] Step S4 includes using text vectorization technology to vectorize the data collected in the emergency plan for sudden biosafety incidents, using Faiss as the vector database, and the high-dimensional vector compression method IVFPQ as the search index.

[0060] The specific method to vectorize the data collected in the emergency plan for sudden biosafety incidents is to use text2vec-base-chines as a text embedding model, store the vectors output by the text embedding model into a vector database, and build an emergency plan data vector library so that more professional information can be introduced into it during the plan generation stage.

[0061] S5. Calculate the similarity between the user's model input and the vector library content. Based on the similarity, extract the most relevant content and input it into the big model together with the "3S" information and the user's model input to generate a complete emergency plan.

[0062] Step S5 includes the following three sub-steps:

[0063] S51. Use the "3S" information automatic collection technology, in which RS and GPS provide GIS with real-time updated field information and spatial positioning, and GIS performs corresponding spatial analysis to extract valid information from the data collected by the "3S" system.

[0064] S52. Calculate the similarity between the feature vector corresponding to the user's model input content and the feature vector corresponding to the information related to the sudden biosafety incident and its emergency plan. The similarity score algorithm formula is as follows:

[0065]

[0066] Among them, Vec(D) represents the text vector output after the information related to sudden biosafety incidents and their emergency plans are input into the text embedding model;

[0067] Vec(Q) represents the text vector output after the user's model input content is input into the text embedding model.

[0068] S53. Based on vector similarity, the most relevant information related to sudden biosafety incidents and the feature vectors corresponding to their emergency plans are extracted and input into the big model together with the information collected by the "3S" system and the user input content to generate a complete emergency plan.

[0069] It can be seen from the above embodiments that the method for generating emergency plans for sudden biosafety incidents based on a large model disclosed in the present invention performs text cleaning on the information related to sudden biosafety incidents and their emergency plans, removes irrelevant information, invalid laws, and outdated technologies, and retains key disposal information such as disposal subjects, disposal objects, and disposal methods, so as to ensure that the emergency plan can cover various emergencies and has practical operability. By using the training set constructed by obtaining relevant information about sudden biosafety incidents and their emergency plans to train the model, it can be ensured that when a sudden biosafety incident occurs, the model can quickly output a reliable emergency plan, laying the foundation for quickly and efficiently responding to sudden biosafety incidents.

[0070] The method described in the present invention is not limited to the embodiments described in the specific implementation manner. Those skilled in the art may derive other implementation manners based on the technical solution of the present invention, which also fall within the technical innovation scope of the present invention.

Claims

1. A method for generating an emergency plan for sudden biosafety incidents based on a large model, the method comprising the following steps: S1. Collect information related to sudden biosafety incidents and their emergency response plans to generate a training set; S2. Use the Transformer model as the infrastructure and train the Transformer model based on the training set to build an emergency plan generation model. S3, constructing a prompt word template for users to input the model; S4. Use text vectorization technology to build an emergency plan data vector library based on the collected information related to sudden biosafety incidents and their emergency plans; S5. Calculate the similarity between the user's model input and the emergency plan data content. Based on the similarity, extract the emergency plan data content with the highest relevance, and input it into the big model together with the real-time collected spatial information and the user's model input to generate a complete emergency plan.

2. A method for generating an emergency plan for sudden biosafety incidents based on a large model as claimed in claim 1, characterized in that: Step S1 includes the following sub-steps: S11. Use information collection technology to obtain information related to sudden biosafety incidents and their emergency response plans; S12. Through text cleaning, remove irrelevant information, invalid laws, and outdated technologies, retain key disposal information, and ensure that the emergency plan can cover various emergencies and is practical; S13. The collected and cleaned emergency plan data is stored in the database, and a training set is generated for the emergency plan. After the training set is constructed, new data is pulled regularly to achieve real-time update of information related to sudden biosafety incidents and their emergency plans.

3. A method for generating an emergency plan for sudden biosafety incidents based on a large model as claimed in claim 2, characterized in that: In step S11, the methods for obtaining relevant information on sudden biosafety incidents and their emergency response plans include historical cases, policy documents, professional literature, and news reports.

4. A method for generating an emergency plan for sudden biosafety incidents based on a large model as claimed in claim 2, characterized in that: The key disposal information in step S12 includes the disposal subject, disposal object and disposal method.

5. A method for generating an emergency plan for sudden biosafety incidents based on a large model as claimed in claim 1, characterized in that: Step S2 includes the following sub-steps: S21. The Transformer model is used as the basic structure. The Transformer model consists of two parts: the encoder and the decoder. The encoder is composed of several encoding layers, each of which contains a multi-head self-attention mechanism and a feedforward neural network. S22. Introduce the Luong attention mechanism in the attention layer of each encoder and decoder to calculate the attention weight between the encoder and decoder; S23. Train the Transformer model based on the training set to obtain an emergency plan generation model.

6. The method for generating an emergency plan for sudden biosafety incidents based on a large model as claimed in claim 1, characterized in that: The model inputs that the design prompt word template described in step S3 requires the user to input include the region to which the sudden biosafety incident belongs, environmental conditions, transmission routes, impact range, pathogens and harmful substance characteristics.

7. The method for generating an emergency plan for sudden biosafety incidents based on a large model as claimed in claim 1, characterized in that: In step S4, the collected emergency plans for sudden biosafety incidents are used as model input, the vectors output by the text embedding model are stored in a vector database, and an emergency plan data vector library is constructed.

8. A method for generating an emergency plan for sudden biosafety incidents based on a large model as claimed in claim 1, characterized in that: Step S5 includes the following sub-steps: S51. Use automatic information collection technology to update on-site information and spatial positioning in real time, and conduct corresponding spatial analysis; S52, calculating the similarity between the feature vector corresponding to the model input content input by the user and the emergency plan data vector; S53. Based on the vector similarity comparison result, the emergency plan data vector with the highest correlation is taken out and input into the big model together with the collected information and user input content to generate a complete emergency plan.

9. A method for generating an emergency plan for sudden biosafety incidents based on a large model as claimed in claim 8, characterized in that: The similarity score calculation formula in step S52 is as follows: Among them, Vec(D) represents the emergency plan data vector output after the information related to sudden biosafety incidents and their emergency plans are input into the text embedding model; Vec(Q) represents the text vector output after the user's model input content is input into the text embedding model.

Citation Information

Cited By

  • Biological safety risk scheme prediction generation method and system based on knowledge graph

    CN120450422B