Task-driven multi-agent emergency decision support method and device

Abstract

The invention relates to a task-driven multi-agent emergency decision support method and device, and the method comprises the steps: constructing a sub-task for a risk guidance decision support model, and dividing a complex task into simple sub-tasks, so as to achieve a more precise emergency decision support process; constructing an interaction agent of each subtask, and constructing corresponding prompt engineering contents according to the definition to realize the construction of analysis agents and inspection agents of different tasks; collecting safety analysis report data, and establishing a sample database supported by emergency operation instruction decision; performing model training, handing over the sample database to an intelligent agent, and constructing a success case library and a failure case library; over-benchmark originating event information is input, and a decision support procedure for the described originating event is generated. Therefore, the problems that an existing emergency plan decision-making method is extremely dependent on expert knowledge and is limited by human cognition, so that field personnel are difficult to make quick decisions on events exceeding the emergency plan and the like are solved.

Description

Technical Field

[0001] The present application relates to the technical field of data analysis and decision support, and in particular to a task-driven multi-agent emergency decision support method and device. Background Art

[0002] Emergency operating procedures refer to work plans prepared in advance by people's governments at all levels and their departments, grassroots organizations, enterprises, institutions, social groups, etc. to respond to emergencies in accordance with the law, quickly, scientifically, and orderly, and to minimize emergencies and the damage they cause.

[0003] As the global climate warms, extreme weather events occur frequently, and many accidents that exceed the design benchmark continue to emerge. Existing emergency plans are extremely dependent on expert knowledge and are limited by human cognition. Therefore, it is difficult for on-site personnel to make quick decisions on events that exceed the emergency plans, which needs to be solved urgently. Summary of the invention

[0004] The present application provides a task-driven multi-agent emergency decision support method and device to solve the problems that the existing emergency plan decision-making methods are extremely dependent on expert knowledge and are limited by human cognition, making it difficult for on-site personnel to make quick decisions on events that exceed the emergency plan.

[0005] The first aspect of the present application provides a task-driven multi-agent emergency decision support method, comprising the following steps: constructing multiple subtasks of a preset risk-guided decision support model, and establishing an interactive agent corresponding to each of the multiple subtasks; collecting safety analysis report data of the target security field to establish an emergency decision support sample database through the safety analysis report data, and using the emergency decision support sample database to train the interactive agent to generate a success case library and a failure case library corresponding to the emergency decision support sample database; obtaining the target user's super-baseline initiating event information, and obtaining the question-answer pair data in the success case library and the failure case library that meet the preset content similarity conditions with the super-baseline initiating event information through the interactive agent, so as to generate a decision support procedure corresponding to the super-baseline initiating event information according to the question-answer pair data.

[0006] Optionally, in one embodiment of the present application, the construction of multiple subtasks of a preset risk-guided decision support model and the establishment of an interactive agent corresponding to each of the multiple subtasks include: establishing multiple subtasks of the risk-guided decision support model, wherein the multiple subtasks include an initiating event sub-event analysis and identification task, an event tree header event analysis task, and a decision recommendation support task; constructing a first task interactive agent corresponding to the initiating event sub-event analysis and identification task, wherein the first task interactive agent includes an initiating event sub-event analysis agent and an initiating event sub-event verification agent; establishing a second task interactive agent corresponding to the event tree header event analysis task, and constructing a third task interactive agent corresponding to the decision recommendation support task, wherein the second task interactive agent includes an event tree header event analysis agent and an event tree header event verification agent, and the third task interactive agent includes a strategy formulation agent and a strategy verification agent.

[0007] Optionally, in one embodiment of the present application, the security analysis report data of the target security field is collected to establish an emergency decision support sample database through the security analysis report data, and the interactive intelligent agent is trained using the emergency decision support sample database to generate a success case library and a failure case library corresponding to the emergency decision support sample database, including: collecting the security analysis report data corresponding to each subtask in the target security field, and sending the security analysis report data to the corresponding interactive intelligent agent, so as to determine whether the security analysis report data meets the preset successful decision requirements through the interactive intelligent agent; if the security analysis report data meets the preset successful decision requirements, a success case corresponding to the security analysis report data is generated, and the success case is compiled, and the compiled success case is stored in the success case library; if the security analysis report data does not meet the preset successful decision requirements, a failure case corresponding to the security analysis report data is generated, and the failure case is compiled, and the compiled failure case is stored in the failure case library.

[0008] Optionally, in one embodiment of the present application, the interactive agent is used to respectively obtain question-answer pair data in the success case library and the failure case library that meet preset content similarity conditions with the super-benchmark originating event information, so as to generate a decision support procedure corresponding to the super-benchmark originating event information according to the question-answer pair data, including: inputting the super-benchmark originating event information into the first task interactive agent, so as to query the first question-answer pair data that meet the preset content similarity conditions with the super-benchmark originating event information in the success case library and the failure case library corresponding to the first task interactive agent; adding the first question-answer pair data to the first prompt engineering content corresponding to the first task interactive agent, and inputting the first question-answer pair data into the second task interactive agent, so as to generate a decision support procedure corresponding to the super-benchmark originating event information according to the question-answer pair data. The second question-answer pair data that meets the preset content similarity condition with the first question-answer pair data are searched in the success case library and the failure case library corresponding to the second task interactive intelligent agent; the second question-answer pair data are added to the second prompt engineering content corresponding to the second task interactive intelligent agent, and the second question-answer pair data are input into the third task interactive intelligent agent to obtain the decision support procedure corresponding to the super-benchmark initiating event information; the decision support procedure is added to the third prompt engineering content corresponding to the third task interactive intelligent agent, and the risk-guided decision support model is optimized through the first prompt engineering content, the second prompt engineering content and the third prompt engineering content, so as to optimize the decision support procedure corresponding to the super-benchmark initiating event information based on the optimized risk-guided decision support model.

[0009] The second aspect of the present application provides a task-driven multi-agent emergency decision support device, including: an establishment module, used to construct multiple subtasks of a preset risk-guided decision support model, and establish an interactive agent corresponding to each of the multiple subtasks; a training module, used to collect safety analysis report data of the target safety field, so as to establish an emergency decision support sample database through the safety analysis report data, and use the emergency decision support sample database to train the interactive agent to generate a success case library and a failure case library corresponding to the emergency decision support sample database; an emergency decision support module, used to obtain the target user's super-benchmark initiating event information, and obtain the question and answer pair data in the success case library and the failure case library that meet the preset content similarity conditions with the super-benchmark initiating event information through the interactive agent, so as to generate a decision support procedure corresponding to the super-benchmark initiating event information according to the question and answer pair data.

[0010] Optionally, in one embodiment of the present application, the establishment module includes: a first construction unit, used to establish multiple subtasks of the risk-guided decision support model, wherein the multiple subtasks include an initiating event sub-event analysis and identification task, an event tree header event analysis task, and a decision recommendation support task; a second construction unit, used to construct a first task interactive agent corresponding to the initiating event sub-event analysis and identification task, wherein the first task interactive agent includes an initiating event sub-event analysis agent and an initiating event sub-event verification agent; a third construction unit, used to establish a second task interactive agent corresponding to the event tree header event analysis task, and construct a third task interactive agent corresponding to the decision recommendation support task, wherein the second task interactive agent includes an event tree header event analysis agent and an event tree header event verification agent, and the third task interactive agent includes a strategy formulation agent and a strategy verification agent.

[0011] Optionally, in one embodiment of the present application, the training module includes: a collection unit, used to collect security analysis report data corresponding to each subtask in the target security field, and send the security analysis report data to the corresponding interactive intelligent agent, so as to determine whether the security analysis report data meets the preset successful decision requirements through the interactive intelligent agent; a first generation unit, used to generate a success case corresponding to the security analysis report data if the security analysis report data meets the preset successful decision requirements, and compile the success case, and store the compiled success case in the success case library; a second generation unit, used to generate a failure case corresponding to the security analysis report data if the security analysis report data does not meet the preset successful decision requirements, and compile the failure case, and store the compiled failure case in the failure case library.

[0012] Optionally, in one embodiment of the present application, the emergency decision support module includes: a query unit, which is used to input the super-benchmark initiating event information into the first task interactive intelligent agent, so as to query the first question-answer pair data that meets the preset content similarity condition with the super-benchmark initiating event information in the success case library and the failure case library corresponding to the first task interactive intelligent agent; a first adding unit, which is used to add the first question-answer pair data to the first prompt engineering content corresponding to the first task interactive intelligent agent, and input the first question-answer pair data into the second task interactive intelligent agent, so as to query the success case library and the failure case library corresponding to the second task interactive intelligent agent for the first question-answer pair data that meets the preset content similarity condition. a second question-answer pair data satisfying the preset content similarity condition; a second adding unit, used for adding the second question-answer pair data to the second prompt engineering content corresponding to the second task interactive intelligent agent, and inputting the second question-answer pair data into the third task interactive intelligent agent, so as to obtain the decision support procedure corresponding to the super-benchmark initiating event information; an optimization unit, used for adding the decision support procedure to the third prompt engineering content corresponding to the third task interactive intelligent agent, and optimizing the risk-guided decision support model through the first prompt engineering content, the second prompt engineering content and the third prompt engineering content, so as to optimize the decision support procedure corresponding to the super-benchmark initiating event information based on the optimized risk-guided decision support model.

[0013] The third aspect of the present application provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the task-driven multi-agent emergency decision support method as described in the above embodiment.

[0014] The fourth aspect of the present application provides a computer-readable storage medium, which stores a computer program that, when executed by a processor, implements the above-mentioned task-driven multi-agent emergency decision support method.

[0015] The fifth aspect of the present application provides a computer program product, including a computer program, which is executed to implement the above-mentioned task-driven multi-agent emergency decision support method.

[0016] Therefore, the embodiments of the present application have the following beneficial effects:

[0017] The embodiment of the present application can be implemented by constructing multiple subtasks of a preset risk-guided decision support model, and establishing an interactive intelligent agent corresponding to each of the multiple subtasks; collecting safety analysis report data in the target safety field, so as to establish an emergency decision support sample database through the safety analysis report data, and using the emergency decision support sample database to train the interactive intelligent agent, so as to generate a successful case library and a failed case library corresponding to the emergency decision support sample database; obtaining the target user's super-baseline initiating event information, and obtaining the question-answer pair data in the success case library and the failed case library that meet the preset content similarity conditions with the super-baseline initiating event information through the interactive intelligent agent, so as to generate the decision support procedure corresponding to the super-baseline initiating event information according to the question-answer pair data, so as to quickly obtain decision suggestions without too much expert knowledge, improve the work efficiency of emergency decision-making, and reduce the impact of accident consequences. Thus, the existing emergency plan decision-making method is extremely dependent on expert knowledge, which is limited by human cognition, making it difficult for on-site personnel to make quick decisions on events beyond the emergency plan.

[0018] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0020] Figure 1 A flowchart of a task-driven multi-agent emergency decision support method provided according to an embodiment of the present application;

[0021] Figure 2 A schematic diagram of the execution logic of a task-driven multi-agent emergency decision support method provided for one embodiment of the present application;

[0022] Figure 3 This is an example diagram of a task-driven multi-agent emergency decision support device according to an embodiment of the present application;

[0023] Figure 4 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application.

[0024] Among them, 10 is a task-driven multi-agent emergency decision support device; 100 is an establishment module, 200 is a training module, 300 is an emergency decision support module; 401 is a memory, 402 is a processor, and 403 is a communication interface. DETAILED DESCRIPTION

[0025] Embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0026] The task-driven multi-agent emergency decision support method and device of the embodiment of the present application are described below with reference to the accompanying drawings. In view of the problems mentioned in the above background technology, the present application provides a task-driven multi-agent emergency decision support method, in which multiple subtasks of a preset risk-guided decision support model are constructed, and an interactive agent corresponding to each of the multiple subtasks is established; the safety analysis report data of the target safety field is collected to establish an emergency decision support sample database through the safety analysis report data, and the interactive agent is trained using the emergency decision support sample database to generate a successful case library and a failed case library corresponding to the emergency decision support sample database; the target user's super-baseline initiating event information is obtained, and the question-answer pair data in the success case library and the failed case library that meet the preset content similarity conditions with the super-baseline initiating event information are obtained through the interactive agent, so as to generate a decision support procedure corresponding to the super-baseline initiating event information according to the question-answer pair data, so that decision suggestions can be quickly obtained without too much expert knowledge, the work efficiency of emergency decision-making can be improved, and the impact of accident consequences can be reduced. Thus, the existing emergency plan decision-making method is extremely dependent on expert knowledge, which is limited by human cognition, making it difficult for on-site personnel to make quick decisions on events beyond the emergency plan.

[0027] Specifically, Figure 1 A flowchart of a task-driven multi-agent emergency decision support method provided in an embodiment of the present application.

[0028] like Figure 1 As shown, the task-driven multi-agent emergency decision support method includes the following steps:

[0029] In step S101, multiple subtasks of a preset risk-guided decision support model are constructed, and an interactive agent corresponding to each of the multiple subtasks is established.

[0030] The embodiment of the present application constructs subtasks for the risk-guided decision support model to split complex tasks into simple subtasks to achieve a more accurate emergency decision support process; thereafter, the embodiment of the present application also needs to build an interactive intelligent agent for each subtask, construct the corresponding prompt engineering content according to the definition, and realize the construction of analysis agents and inspection agents for different tasks.

[0031] Optionally, in one embodiment of the present application, multiple subtasks of a preset risk-guided decision support model are constructed, and an interactive agent corresponding to each of the multiple subtasks is established, including: establishing multiple subtasks of the risk-guided decision support model, wherein the multiple subtasks include an initiating event sub-event analysis and identification task, an event tree header event analysis task, and a decision suggestion support task; constructing a first task interactive agent corresponding to the initiating event sub-event analysis and identification task, wherein the first task interactive agent includes an initiating event sub-event analysis agent and an initiating event sub-event verification agent; establishing a second task interactive agent corresponding to the event tree header event analysis task, and constructing a third task interactive agent corresponding to the decision suggestion support task, wherein the second task interactive agent includes an event tree header event analysis agent and an event tree header event verification agent, and the third task interactive agent includes a strategy formulation agent and a strategy verification agent.

[0032] It should be noted that in the process of constructing the subtasks for the risk-guided decision support model, the embodiment of the present application mainly refers to the core idea of ​​the event tree, combines probabilistic risk analysis to implement the emergency decision-making process, and divides the entire emergency decision-making process into the following three subtasks:

[0033] Task 1: Analysis and identification of sub-events of the initiating event (i.e., analysis and identification of sub-events of the initiating event).

[0034] This task is the basis for the analysis of subsequent tasks 2 and 3. The input is the name and description of the initiating event, and the output is the sub-events of the initiating event.

[0035] Task 2: Header event analysis of the event tree includes identifying and listing the head events that may lead to the accident in chronological order (i.e., the head event analysis task of the event tree).

[0036] This task systematically analyzes the subsequent events of the initiating event. Its input is the initiating event, the initiating event description, sub-events (generated by task one), as well as the event process and system response. The output is the header event of the event tree.

[0037] Task 3: Decision-making recommendations (i.e. decision-making recommendation support tasks)

[0038] After identifying and analyzing potential sub-events and header events, specific operational decisions are made and provided to respond to and mitigate adverse effects. The inputs are the initiating event, the initiating event description, sub-events (generated by Task 1), the event process and system response, and the header event of the event tree (generated by Task 2). The output is the operator's action recommendations.

[0039] In addition, the interactive agents designed in the embodiment of the present application include a first task interactive agent corresponding to task one, namely, an initiating event sub-event analysis agent and an initiating event sub-event verification agent; a second task interactive agent corresponding to task two, namely, an event tree header event analysis agent and an event tree header event verification agent; and a third task interactive agent corresponding to task three, namely, a strategy formulation agent and a strategy verification agent.

[0040] The corresponding information of the interactive agent corresponding to each task is as follows:

[0041] 1. Initial event sub-event analysis agent:

[0042] The initiating event sub-event analysis agent is responsible for identifying the sub-events of the initiating event. The main task of this agent is to check the detailed description and cases, including correct and incorrect cases; for incorrect cases, the agent must summarize the feedback to prevent the error from happening again; for correct cases, the agent should refer to the applied logic; when there is no correct or incorrect case, the agent needs to analyze the sub-events of the initiating event.

[0043] 2. Initial event sub-event verification agent:

[0044] The initiating event sub-event verification agent is responsible for verifying the accuracy of the sub-events identified by the analysis agent; based on the initiating event and its detailed description, the agent determines whether the identified sub-event is consistent with the actual sub-event. The judgment criteria include multiple key aspects to ensure the comprehensiveness and accuracy of the verification process.

[0045] 3. Event tree header event analysis agent:

[0046] The event tree header event analysis agent is specifically responsible for identifying and listing the header events in the event tree. The agent also needs to summarize the feedback from error cases to avoid repeating the mistakes and refer to the logic used in the correct cases; by using the feedback from error cases, the agent can continuously improve future responses.

[0047] 4. Event tree header event verification agent:

[0048] The event tree header event verification agent is responsible for verifying the accuracy of the header events provided by the header event analysis agent. The agent ensures that the identified header events are consistent with the actual event process of the nuclear power plant. The verification process involves multiple key steps to maintain the reliability and integrity of the analysis and ensure the accuracy and safety of operations.

[0049] 5. Strategy formulation agent:

[0050] The plant's policy-making agent is responsible for determining the specific actions that operators should take based on the initiating event, event description, sub-events, event sequences, and examples of correct and incorrect responses; the agent must ensure that these actions are detailed and accurate. For error cases, the agent needs to summarize feedback to prevent the error from recurring and refer to the logic in the correct case for analysis.

[0051] 6. Strategy Verification Agent:

[0052] The strategy verification agent is responsible for evaluating the recommendations provided by the nuclear power plant strategy advisor. The agent determines whether the recommendations are consistent with the recommendations of the actual operators; by accurately verifying the strategy recommendations, the agent ensures the safety and reliability of nuclear power plant operations.

[0053] It should be noted that the agents of the three tasks are all driven by a large language model, mainly using prompt engineering. The prompt engineering content of each agent is as follows:

[0054] 1. The prompt engineering content for the sub-event analysis agent of the originating event is as follows:

[0055] "As a nuclear power plant initiating event sub-event analyst, you are responsible for analyzing and identifying the sub-events of the initiating event based on the given detailed description, correct case, and incorrect case. For incorrect cases, you need to summarize the feedback to prevent recurrence. For correct cases, you should refer to their logic to guide your analysis. If there is no correct or incorrect case, you must analyze its sub-events based on your ability and the detailed description of the initiating event. Please note that an initiating event may have only one sub-event, and the initiating event may lead to serious consequences. The output format of the sub-event list is as follows: "Nuclear power plant initiating event sub-event analyst: <1. Sub-event 1 2. Sub-event 2 3. Sub-event 3>". Please ensure that only the sub-event list is provided without additional output, and pay attention to the correctness of the units. In addition, please try to clearly state the dimensions in your analysis."

[0056] 2. The prompt engineering content of the originating event sub-event verification agent is as follows:

[0057] “You are the initial event sub-event verification officer of a nuclear power plant. Based on the given initial event and its detailed description, your task is to determine whether the sub-event provided by the initial event analyst is consistent with the actual sub-event. The judgment criteria are as follows:

[0058] 1) If the sub-event provided by the analyst is more detailed than the actual sub-event, but the phenomenon described is roughly or essentially the same, return "yes"; otherwise, return "no";

[0059] 2) If the sub-events provided by the analyst roughly cover the content of the actual event, and the size information of the sub-events with dimensions is consistent, please return "yes"; otherwise, return "no";

[0060] 3) If the actual sub-event contains size information, but the event provided by the analyst lacks size information or the size information is incorrect, return "No";

[0061] 4) Return "No" if the event provided by the analyst contains sub-events that are not part of the initial event.

[0062] It is important to note that whether your judgment is "yes" or "no", you should provide reasons. "

[0063] 3. The prompt project content of the event tree header event analysis agent is as follows:

[0064] "You are an expert in nuclear power plant event trees. Based on the initiating events, sub-events, event progression, and system responses of the nuclear power plant provided, please analyze the header events of the event tree; the specific requirements are as follows: only provide header events to ensure the correct order; header events should reflect equipment or functions, not failures; the order of header events is critical, and they must be arranged according to event progression, covering all important events comprehensively and accurately; for incorrect cases, summarize feedback to avoid repeating errors; for correct cases, refer to their logic to help your analysis; use the reasons in incorrect cases to improve your response. The output format should be as follows: Nuclear power plant event tree header event analyst: <1. Header event 1\n 2. Header event 2\n 3. Header event 3...>"

[0065] 4. The prompt project content of the event tree header event verification agent is as follows:

[0066] "Based on the information provided, determine whether the header events provided by the nuclear power plant event tree analyst are correct. If the header events provided by the analyst are consistent with the actual event progression, please output "yes"; otherwise, output "no" and provide a reason; if the header events provided by the analyst are more than the actual events, but cover all the actual events, please also return "yes"; if the header events provided by the analyst are broader but include the actual header events, please also return "yes"; it should be noted that the response must include "yes" or "no"; now, based on these rules and the actual header events, determine whether the following event tree header event analysis results are correct; please note that the output does not need to be exactly the same as the actual header event; if the content or direction is roughly consistent, it is considered correct."

[0067] 5. The prompt engineering content of the strategy formulation agent is as follows:

[0068] "As a nuclear power plant expert, you can determine the specific actions that the operator should take based on the initiating event, its description, sub-events, event sequences, and examples of correct and incorrect responses; the actions must be detailed and accurate; for incorrect cases, you need to summarize the feedback to prevent the error from happening again; for correct cases, you should refer to its logic to guide your analysis; if no correct or incorrect cases are provided, you must use your expertise and the given information to analyze and determine the specific actions that the operator should take."

[0069] 6. The prompt engineering content of the strategy verification agent is as follows:

[0070] "Based on the information provided, evaluate the suggestions provided by the nuclear power plant strategy proposer. Determine whether these suggestions are consistent with the suggestions of the actual operator; if the content is roughly consistent and the operator's suggestion is more specific but follows the same direction, output "yes"; otherwise, output "no" and provide reasons. Make sure that the response contains "yes" or "no"; please note that due to the lack of procedural knowledge of nuclear power plant strategy consultants, if their suggestions can prevent accidents, the response should be "yes"; in addition, if their suggestions only lack references and detailed descriptions of specific accident handling procedures, the response should also be "yes"; in addition, the response should focus on procedurality, emphasizing the implementation of specific measures according to specific accident procedures (such as Procedure A and Procedure A1.1); focus on switching the operating mode of the injection system according to the real-time status of the reactor. Now, please judge whether the following operator suggestions are consistent based on these rules: Please output in the following format: "Nuclear Power Plant Strategy Advisor: <Operation to Action Plan 1>".

[0071] Therefore, the embodiments of the present application provide reliable technical support for the implementation of multi-agent emergency decision support by constructing subtasks for the risk-guided decision support model and interactive agents for each subtask.

[0072] In step S102, safety analysis report data of the target safety field is collected to establish an emergency decision support sample database through the safety analysis report data, and the emergency decision support sample database is used to train the interactive intelligent agent to generate a success case library and a failure case library corresponding to the emergency decision support sample database.

[0073] Furthermore, the embodiments of the present application also need to collect the components of different systems and the corresponding initiating events, initiating event descriptions, sub-events, event processes and system responses, header events of the event tree, and operator action suggestions from the safety analysis reports in related fields, so as to obtain safety analysis report data and construct a sample database for emergency operating procedure decision support (i.e., emergency decision support sample database).

[0074] Afterwards, the embodiment of the present application can send the emergency decision support sample database to the interactive agent to build a successful case library and a failed case library. The specific construction process of the successful case library and the failed case library is as follows:

[0075] 1. Construction of successful case library:

[0076] In the process of completing the task, it is more beneficial for the operator to refer to previously verified success cases, which contain rich knowledge and demonstrate accurate and appropriate response logic for dealing with various emergency situations; therefore, the embodiment of the present application proposes to build a success case library for the intelligent agent to enhance its capabilities; the success case library is constructed in the form of question-answer pairs, where the question describes the basic situation of the initiating event that needs to complete the task, and the answer contains a verified affirmative answer from the intelligent agent; for each answer generated by the intelligent agent, if the answer is correct, the question-answer pair will be added to the success case.

[0077] 2. Failure case library expansion:

[0078] Learning from mistakes is important for the growth of operators. The professional intelligent agent based on a large language model in the embodiment of the present application can reflect on itself through these mistakes and extract relevant principles (experience) to ensure that tasks are completed correctly when encountering similar problems in the future; if the answer is wrong, the question and answer pair will be added to the failure case library.

[0079] Optionally, in one embodiment of the present application, security analysis report data of the target security field is collected to establish an emergency decision support sample database through the security analysis report data, and the emergency decision support sample database is used to train an interactive intelligent agent to generate a success case library and a failure case library corresponding to the emergency decision support sample database, including: collecting security analysis report data corresponding to each subtask in the target security field, and sending the security analysis report data to the corresponding interactive intelligent agent to determine through the interactive intelligent agent whether the security analysis report data meets the preset successful decision requirements; if the security analysis report data meets the preset successful decision requirements, then generating a success case corresponding to the security analysis report data, compiling the success case, and storing the compiled success case in the success case library; if the security analysis report data does not meet the preset successful decision requirements, then generating a failure case corresponding to the security analysis report data, compiling the failure case, and storing the compiled failure case in the failure case library.

[0080] In the actual implementation process, the embodiment of the present application obtains the distribution of the comprehensive data set (ie, security analysis report data) by collecting report data as shown in Table 1:

[0081] Table 1

[0082] Initiating event abbreviation quantity Loss of coolant accident LOCA 5 Loss of heat source accident LOHSA 3 Loss of water incident LOFW 2 Loss of external power event LOOP 1 Predicted transients without reactor scram ATWS 7 Main water pipe rupture MFLB 2 Steam generator tube rupture SGTR 6 Main steam line rupture MSLB 3 DC power loss accident LODC 2 Combined accident of main steam line rupture and steam generator tube rupture MSLB+SGTR 5 Secondary circuit transient events SLTE 2

[0083] As can be seen from Table 1, the dataset contains 11 different types of initiating events commonly seen in nuclear power plants: there are 5 instances of loss of coolant accidents (LOCA), 3 occurrences of loss of heat source accidents (LOHSA), 2 records of loss of feedwater events (LOFW), and 1 occurrence of loss of external power (LOOP); in addition, the most common event is the anticipated transient without reactor emergency shutdown (ATWS), which is recorded 7 times. The dataset also includes 2 instances of main feedwater rupture (MFLB) and steam generator tube rupture (SGTR), 6 occurrences of main steam line rupture (MSLB), and 3 records of loss of direct current (LODC). In addition, the combined main steam line rupture and steam generator tube rupture (MSLB+SGTR) occurred 5 times. Finally, the secondary loop transient event (SLTE) was recorded 2 times.

[0084] For Task 1, there are 13 data points in total, 10 of which are used for training and 3 for testing. For Tasks 2 and 3, there are 38 data points in total, 31 of which are used for training and 7 for testing.

[0085] Afterwards, the present application may send the emergency decision support sample database obtained above to the interactive intelligent agent to train it and construct a success case library and a failure case library.

[0086] As an achievable method, the training method of the embodiment of the present application is a parameter-free strategy. For each task, the task executor and the task verifier complete the task through continuous interaction and iteration. The three tasks are interconnected through the task flow to jointly realize the decision support function.

[0087] In this strategy, there are two important modules, namely the successful case library and the failed case library; successful cases will be compiled and stored in the successful case library as a reference for future generated results; and for failed cases, the results will be added to the failed case library.

[0088] In the simulation of the three tasks, the embodiment of the present application can use a dense retriever to retrieve relevant success case libraries and failure case libraries to help the intelligent agent provide better results; as success and failure cases accumulate, they will be actively applied, and the success case libraries and failure case libraries will be continuously updated; in addition, whether the answer is correct or not is also completely completed autonomously by the intelligent agent.

[0089] In step S103, the target user's super-benchmark initiating event information is obtained, and the question-answer pair data that meets the preset content similarity conditions with the super-benchmark initiating event information in the success case library and the failure case library are respectively obtained through the interactive intelligent agent, so as to generate a decision support procedure corresponding to the super-benchmark initiating event information based on the question-answer pair data.

[0090] Furthermore, the embodiments of the present application also require the input of initiating event information that exceeds the design benchmark, and based on the introduced success case library and failure case library, enhance the prompt information of the interactive intelligent body; the embodiments of the present application compare the existing content in the success case library and the failure case library with the currently input initiating event information, obtain the content most similar to the currently queried initiating event information, select successful and failed question-and-answer pairs, and add them to the prompt engineering content to generate a decision support procedure for the described initiating event and enhance the performance of the model.

[0091] Optionally, in one embodiment of the present application, question-answer pair data that meet preset content similarity conditions with the super-benchmark initiating event information in the success case library and the failure case library are respectively obtained through the interactive intelligent agent, so as to generate a decision support procedure corresponding to the super-benchmark initiating event information according to the question-answer pair data, including: inputting the super-benchmark initiating event information into the first task interactive intelligent agent, so as to query the first question-answer pair data that meet the preset content similarity conditions with the super-benchmark initiating event information in the success case library and the failure case library corresponding to the first task interactive intelligent agent; adding the first question-answer pair data to the first prompt engineering content corresponding to the first task interactive intelligent agent, and inputting the first question-answer pair data into the second task interactive intelligent agent, so as to query the first question-answer pair data in the second task interactive intelligent agent; The second question-answer pair data that meets the preset content similarity condition with the first question-answer pair data are queried in the success case library and the failure case library corresponding to the second task interactive intelligent agent; the second question-answer pair data are added to the second prompt engineering content corresponding to the second task interactive intelligent agent, and the second question-answer pair data are input into the third task interactive intelligent agent to obtain the decision support procedure corresponding to the super-benchmark initiating event information; the decision support procedure is added to the third prompt engineering content corresponding to the third task interactive intelligent agent, and the risk-guided decision support model is optimized through the first prompt engineering content, the second prompt engineering content and the third prompt engineering content, so as to optimize the decision support procedure corresponding to the super-benchmark initiating event information based on the optimized risk-guided decision support model.

[0092] It should be noted that the embodiment of the present application enhances the prompt information of the intelligent agent by introducing a success case library and a failure case library. When there is a new query (i.e., the super-baseline initiating event information), the system will search for relevant records in the retrieval library. Each task has its own independent success case library and failure case library to avoid using irrelevant records; by comparing the existing content in the success case library and the failure case library with the currently input initiating event information, the content most similar to the initiating event information of the current query is obtained.

[0093] Afterwards, the embodiment of the present application can add the selected successful and failed question-answer pairs to the prompt engineering content to enhance the performance of the model; both successful and failed retrieval cases are calculated by similarity, and the calculation method is to use the "text-embedding-ada-002" model provided by OpenAI to embed the text into the vector space, and use cosine similarity to calculate the case similarity. The mathematical expression of cosine similarity is as follows:

[0094]

[0095] Among them, Z1 and Z2 represent the vector representation of the currently input initiating event and the vector representation of the existing content in the success case library and the failure case library respectively.

[0096] In the embodiments of the present application, the calculated similarity ranges from -1 to 1, where -1 indicates that the directions pointed by the two vectors are exactly opposite, 1 indicates that the directions pointed by the two vectors are exactly the same, 0 generally indicates that the two vectors are independent, and the values ​​in between indicate the similarity or difference between the two vectors.

[0097] In the specific implementation process, the user can input various initiating events to the model. The embodiment of the present application takes the initiating event as an example of a main steam pipeline rupture accompanied by a steam generator pipeline rupture accident. This accident was not input as a case in the previous training process. Therefore, for the embodiment of the present application, it is an initiating event that exceeds the design benchmark.

[0098] Therefore, the embodiment of the present application can first input the initiating event information into the agent process of Task 1. After the agent inputs the information, it refers to the contents in the success case library and the failure case library, and the output results are as follows:

[0099] “

[0100] 1. Single pipe rupture

[0101] 2. Multiple pipes ruptured

[0102] 3. Complete break in the pipeline

[0103] ”

[0104] Secondly, the output of task one is used as the input of task two and processed by the event tree head event analysis agent. The input includes the originating event, the originating event description, the event description and the sub-events. The final output event tree head event is as follows:

[0105] “

[0106] 1. Reactor emergency shutdown

[0107] 2. The main steam isolation valve closes automatically

[0108] 3. Isolation of main water supply system

[0109] 4. Start the auxiliary water supply system

[0110] 5. Security injection startup

[0111] 6. Primary circuit decompression

[0112] 7. Primary circuit cooling

[0113] 8. Normal coolant filling function restored

[0114] ”

[0115] Afterwards, these results are input into the strategy formulation agent of Task 3 for processing. The input includes the initiating event, the initiating event description, the event description, the sub-event, the event tree head event, the event process and the system response. The final output is the strategy support as follows:

[0116] “

[0117] 1. Event confirmation: Confirm that the emergency alarm system has been activated. Verify that the accident is a combination of a main steam pipeline rupture and a steam generator tube rupture.

[0118] 2. Emergency Reactor Shutdown: Verify that the emergency shutdown signal has been activated, which may be triggered by a safety injection signal, a low pressurizer pressure, or a very low steam generator water level signal. Ensure that the reactor is in a subcritical state and shut down safely.

[0119] 3. Confirm the activation of any of the following signals: extremely low pressurizer pressure, high pressure difference between two steam lines, high steam line flow and low steam line pressure, or low primary circuit average temperature. Start the safety injection system to replenish the water inventory of the primary circuit. Ensure that the auxiliary feedwater system supplies water to the steam generator.

[0120] 4. Main steam isolation: Confirm that the main steam isolation valve is closed and the trigger signal is low steam pressure and high steam flow.

[0121] 5. Confirm that the high-pressure injection system injects coolant into the primary circuit. Start the medium-pressure injection system as needed to maintain the water inventory of the primary circuit. Ensure that the low-pressure injection system starts when necessary to provide coolant to the primary circuit.

[0122] 6. Initiate decompression and cooling measures.

[0123] ”

[0124] In summary, the embodiment of the present application first constructs subtasks for the risk-guided decision support model, including initiating event sub-time analysis, event tree header event analysis, and emergency decision suggestions; secondly, through prompt engineering, an interactive intelligent agent for each subtask is constructed based on a large language model; then, data is collected to establish a sample database for emergency operating procedure decision support; the data is handed over to the intelligent agent for training iterations to build a library of successful cases and failed cases; finally, the initiating event information that exceeds the design benchmark is input to generate a decision support procedure for the initiating event that exceeds the design benchmark, so that industry personnel can quickly obtain decision suggestions when facing accidents that exceed the design benchmark without too much expert knowledge, greatly improving the work efficiency of emergency decision-making.

[0125] The following describes the execution logic of the task-driven multi-agent emergency decision support method of the present application in combination with the accompanying drawings.

[0126] Figure 2 This is a schematic diagram of the execution logic of the task-driven multi-agent emergency decision support method of this application. Figure 2 As shown, the execution process of the task-driven multi-agent emergency decision support method of the present application is as follows:

[0127] S201: Construct subtasks for risk-guided decision support models, splitting complex tasks into simple subtasks to achieve a more accurate emergency decision support process;

[0128] S202: constructing an interactive agent for each subtask, and constructing corresponding prompt engineering content according to the definition, so as to realize the construction of analysis agents and inspection agents for different tasks;

[0129] S203: Collect safety analysis report data and establish a sample database for emergency operation procedure decision support;

[0130] S204: Model training, giving the sample database to the intelligent agent to build a success case library and a failure case library;

[0131] S205: Input the information of the super-baseline initiating event and generate a decision support procedure for the described initiating event.

[0132] According to the task-driven multi-agent emergency decision support method proposed in the embodiment of the present application, multiple subtasks of a preset risk-guided decision support model are constructed, and interactive agents corresponding to each of the multiple subtasks are established; safety analysis report data of the target safety field are collected to establish an emergency decision support sample database through the safety analysis report data, and the interactive agent is used to train the emergency decision support sample database to generate a success case library and a failure case library corresponding to the emergency decision support sample database; the target user's super-baseline initiating event information is obtained, and the question-answer pair data in the success case library and the failure case library that meet the preset content similarity conditions with the super-baseline initiating event information are obtained through the interactive agent, so as to generate a decision support procedure corresponding to the super-baseline initiating event information according to the question-answer pair data, so that decision suggestions can be quickly obtained without too much expert knowledge, thereby improving the work efficiency of emergency decision-making and reducing the impact of accident consequences.

[0133] Secondly, the task-driven multi-agent emergency decision support device proposed in accordance with the embodiment of the present application is described with reference to the accompanying drawings.

[0134] Figure 3 It is a block diagram of a task-driven multi-agent emergency decision support device according to an embodiment of the present application.

[0135] like Figure 3 As shown, the task-driven multi-agent emergency decision support device 10 includes: a building module 100, a training module 200 and an emergency decision support module 300.

[0136] Among them, the establishment module 100 is used to construct multiple subtasks of the preset risk-guided decision support model, and establish an interactive intelligent agent corresponding to each of the multiple subtasks.

[0137] The training module 200 is used to collect safety analysis report data of the target safety field, to establish an emergency decision support sample database through the safety analysis report data, and to train the interactive intelligent agent using the emergency decision support sample database to generate a success case library and a failure case library corresponding to the emergency decision support sample database.

[0138] The emergency decision support module 300 is used to obtain the target user's super-benchmark initiating event information, and obtain the question-answer pair data in the success case library and the failure case library that meet the preset content similarity conditions with the super-benchmark initiating event information through the interactive intelligent agent, so as to generate a decision support procedure corresponding to the super-benchmark initiating event information based on the question-answer pair data.

[0139] Optionally, in one embodiment of the present application, the building module 100 includes: a first building unit, a second building unit and a third building unit.

[0140] Among them, the first construction unit is used to establish multiple subtasks of the risk-guided decision support model, wherein the multiple subtasks include the initiating event sub-event analysis and identification task, the event tree header event analysis task and the decision recommendation support task.

[0141] The second construction unit is used to construct a first task interaction agent corresponding to the initiating event sub-event analysis and identification task, wherein the first task interaction agent includes an initiating event sub-event analysis agent and an initiating event sub-event verification agent.

[0142] The third construction unit is used to establish a second task interactive agent corresponding to the event tree header event analysis task, and to construct a third task interactive agent corresponding to the decision suggestion support task, wherein the second task interactive agent includes an event tree header event analysis agent and an event tree header event verification agent, and the third task interactive agent includes a strategy formulation agent and a strategy verification agent.

[0143] Optionally, in one embodiment of the present application, the training module 200 includes: a collection unit, a first generation unit and a second generation unit.

[0144] Among them, the collection unit is used to collect the security analysis report data corresponding to each subtask in the target security field, and send the security analysis report data to the corresponding interactive intelligent agent to determine whether the security analysis report data meets the preset successful decision requirements through the interactive intelligent agent.

[0145] The first generating unit is used to generate a success case corresponding to the safety analysis report data if the safety analysis report data meets the preset successful decision requirements, compile the success case, and store the compiled success case in a success case library.

[0146] The second generating unit is used to generate a failure case corresponding to the safety analysis report data if the safety analysis report data does not meet the preset successful decision requirements, compile the failure case, and store the compiled failure case in a failure case library.

[0147] Optionally, in one embodiment of the present application, the emergency decision support module 300 includes: a query unit, a first adding unit, a second adding unit and an optimization unit.

[0148] Among them, the query unit is used to input the super-benchmark initiating event information into the first task interactive intelligent agent, so as to query the first question-answer pair data that meets the preset content similarity conditions with the super-benchmark initiating event information in the success case library and failure case library corresponding to the first task interactive intelligent agent.

[0149] The first adding unit is used to add the first question and answer pair data to the first prompt engineering content corresponding to the first task interactive intelligent agent, and input the first question and answer pair data into the second task interactive intelligent agent, so as to query the second question and answer pair data that meets the preset content similarity condition with the first question and answer pair data in the success case library and the failure case library corresponding to the second task interactive intelligent agent.

[0150] The second adding unit is used to add the second question and answer pair data to the second prompt engineering content corresponding to the second task interactive intelligent agent, and input the second question and answer pair data into the third task interactive intelligent agent to obtain the decision support procedure corresponding to the ultra-benchmark initiating event information.

[0151] The optimization unit is used to add the decision support procedure to the third prompt engineering content corresponding to the third task interactive intelligent agent, and optimize the risk-guided decision support model through the first prompt engineering content, the second prompt engineering content and the third prompt engineering content, so as to optimize the decision support procedure corresponding to the super-benchmark initiating event information based on the optimized risk-guided decision support model.

[0152] It should be noted that the aforementioned explanation of the task-driven multi-agent emergency decision support method embodiment is also applicable to the task-driven multi-agent emergency decision support device of this embodiment, and will not be repeated here.

[0153] The task-driven multi-agent emergency decision support device proposed in the embodiment of the present application includes an establishment module for constructing multiple subtasks of a preset risk-guided decision support model, and establishing an interactive agent corresponding to each of the multiple subtasks; a training module for collecting safety analysis report data in the target safety field, so as to establish an emergency decision support sample database through the safety analysis report data, and use the emergency decision support sample database to train the interactive agent to generate a success case library and a failure case library corresponding to the emergency decision support sample database; an emergency decision support module for obtaining the target user's super-benchmark initiating event information, and obtaining the question-answer pair data in the success case library and the failure case library that meet the preset content similarity conditions with the super-benchmark initiating event information through the interactive agent, so as to generate a decision support procedure corresponding to the super-benchmark initiating event information according to the question-answer pair data, so as to quickly obtain decision suggestions without too much expert knowledge, improve the work efficiency of emergency decision-making, and reduce the impact of accident consequences.

[0154] Figure 4 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. The electronic device may include:

[0155] Memory 401 , processor 402 , and a computer program stored in the memory 401 and executable on the processor 402 .

[0156] When the processor 402 executes the program, the task-driven multi-agent emergency decision support method provided in the above embodiment is implemented.

[0157] Furthermore, the electronic device further comprises:

[0158] The communication interface 403 is used for communication between the memory 401 and the processor 402 .

[0159] The memory 401 is used to store computer programs that can be executed on the processor 402 .

[0160] The memory 401 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0161] If the memory 401, the processor 402 and the communication interface 403 are implemented independently, the communication interface 403, the memory 401 and the processor 402 can be connected to each other through a bus and communicate with each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component (PCI) bus or an Extended Industry Standard Architecture (EISA) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 4 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0162] Optionally, in a specific implementation, if the memory 401, the processor 402 and the communication interface 403 are integrated on a chip, the memory 401, the processor 402 and the communication interface 403 can communicate with each other through an internal interface.

[0163] The processor 402 may be a central processing unit (CPU), or an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application.

[0164] An embodiment of the present application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned task-driven multi-agent emergency decision support method.

[0165] An embodiment of the present application also provides a computer program product, including a computer program, which, when executed, is used to implement the above-mentioned task-driven multi-agent emergency decision support method.

[0166] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or N embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0167] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0168] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or N executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present application belong.

[0169] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purpose of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or N wirings (electronic devices), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways as necessary and then storing it in a computer memory.

[0170] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above embodiment, the N steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. If implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0171] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0172] In addition, each functional unit in each embodiment of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0173] The storage medium mentioned above may be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limiting the present application. A person of ordinary skill in the art may change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A task-driven multi-agent emergency decision support method, characterized in that: The following steps are involved: Constructing multiple subtasks of a preset risk-guided decision support model, and establishing an interactive intelligent agent corresponding to each of the multiple subtasks; Collecting safety analysis report data of the target safety field, so as to establish an emergency decision support sample database through the safety analysis report data, and using the emergency decision support sample database to train the interactive agent, so as to generate a success case library and a failure case library corresponding to the emergency decision support sample database; The target user's super-benchmark initiating event information is obtained, and the question-answer pair data that meets the preset content similarity condition with the super-benchmark initiating event information in the success case library and the failure case library are respectively obtained through the interactive intelligent agent, so as to generate a decision support procedure corresponding to the super-benchmark initiating event information according to the question-answer pair data.

2. The method according to claim 1, characterized in that The multiple subtasks of constructing a preset risk-guided decision support model and establishing an interactive intelligent agent corresponding to each of the multiple subtasks include: Establishing a plurality of subtasks of the risk-guided decision support model, wherein the plurality of subtasks include an initiating event sub-event analysis and identification task, an event tree header event analysis task, and a decision suggestion support task; Constructing a first task interaction agent corresponding to the initiating event sub-event analysis and identification task, wherein the first task interaction agent includes an initiating event sub-event analysis agent and an initiating event sub-event verification agent; A second task interaction agent is established corresponding to the event analysis task of the event tree header, and a third task interaction agent is constructed corresponding to the decision suggestion support task, wherein the second task interaction agent includes an event tree header event analysis agent and an event tree header event verification agent, and the third task interaction agent includes a strategy formulation agent and a strategy verification agent.

3. The method according to claim 2, characterized in that The collecting of the safety analysis report data of the target safety field to establish an emergency decision support sample database through the safety analysis report data, and using the emergency decision support sample database to train the interactive agent to generate a success case library and a failure case library corresponding to the emergency decision support sample database, includes: Collecting the security analysis report data corresponding to each subtask in the target security field, and sending the security analysis report data to the corresponding interactive agent, so as to determine whether the security analysis report data meets the preset successful decision requirements through the interactive agent; If the safety analysis report data meets the preset successful decision requirement, a success case corresponding to the safety analysis report data is generated, the success case is compiled, and the compiled success case is stored in the success case library; If the safety analysis report data does not meet the preset successful decision requirement, a failure case corresponding to the safety analysis report data is generated, the failure case is compiled, and the compiled failure case is stored in the failure case library.

4. The method according to claim 3, characterized in that The step of obtaining, through the interactive agent, question-answer pair data in the success case library and the failure case library that meet a preset content similarity condition with the super-baseline initiating event information, so as to generate a decision support procedure corresponding to the super-baseline initiating event information according to the question-answer pair data, includes: Input the super-baseline initiating event information into the first task interactive agent, so as to search the success case library and the failure case library corresponding to the first task interactive agent for first question-answer pair data that meets the preset content similarity condition with the super-baseline initiating event information; Add the first question-answer pair data to the first prompt engineering content corresponding to the first task interactive agent, and input the first question-answer pair data into the second task interactive agent, so as to search for second question-answer pair data that meets the preset content similarity condition with the first question-answer pair data in the success case library and the failure case library corresponding to the second task interactive agent; Adding the second question-answer pair data to the second prompt engineering content corresponding to the second task interactive agent, and inputting the second question-answer pair data into the third task interactive agent to obtain a decision support procedure corresponding to the ultra-benchmark initiating event information; The decision support procedure is added to the third prompt engineering content corresponding to the third task interactive intelligent agent, and the risk-guided decision support model is optimized through the first prompt engineering content, the second prompt engineering content and the third prompt engineering content, so as to optimize the decision support procedure corresponding to the super-benchmark initiating event information based on the optimized risk-guided decision support model.

5. A task-driven multi-agent emergency decision support device, characterized in that: include: Establishing a module for constructing a plurality of subtasks of a preset risk-guided decision support model, and establishing an interactive intelligent agent corresponding to each of the plurality of subtasks; A training module is used to collect safety analysis report data of a target safety field, to establish an emergency decision support sample database through the safety analysis report data, and to train the interactive agent using the emergency decision support sample database to generate a success case library and a failure case library corresponding to the emergency decision support sample database; The emergency decision support module is used to obtain the target user's super-benchmark initiating event information, and obtain the question-answer pair data in the success case library and the failure case library that meet the preset content similarity conditions with the super-benchmark initiating event information through the interactive intelligent agent, so as to generate a decision support procedure corresponding to the super-benchmark initiating event information based on the question-answer pair data.

6. The device according to claim 5, characterized in that The establishment module includes: A first construction unit is used to establish a plurality of subtasks of the risk-guided decision support model, wherein the plurality of subtasks include an initiating event sub-event analysis and identification task, an event tree header event analysis task, and a decision suggestion support task; A second construction unit is used to construct a first task interaction agent corresponding to the initiating event sub-event analysis and identification task, wherein the first task interaction agent includes an initiating event sub-event analysis agent and an initiating event sub-event verification agent; The third construction unit is used to establish a second task interaction agent corresponding to the header event analysis task of the event tree, and to construct a third task interaction agent corresponding to the decision suggestion support task, wherein the second task interaction agent includes an event tree header event analysis agent and an event tree header event verification agent, and the third task interaction agent includes a strategy formulation agent and a strategy verification agent.

7. The device according to claim 6, characterized in that The training module includes: A collection unit, used for collecting the security analysis report data corresponding to each subtask in the target security field, and sending the security analysis report data to the corresponding interactive agent, so as to determine whether the security analysis report data meets the preset successful decision requirements through the interactive agent; A first generating unit is configured to generate a success case corresponding to the safety analysis report data if the safety analysis report data meets the preset successful decision requirement, compile the success case, and store the compiled success case in the success case library; The second generating unit is used to generate a failure case corresponding to the safety analysis report data if the safety analysis report data does not meet the preset successful decision requirement, compile the failure case, and store the compiled failure case in the failure case library.

8. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the task-driven multi-agent emergency decision support method as described in any one of claims 1 to 4.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: The program is executed by a processor to implement the task-driven multi-agent emergency decision support method as described in any one of claims 1-4.

10. A computer program product, comprising a computer program, characterized in that The computer program is executed to implement the task-driven multi-agent emergency decision support method as described in any one of claims 1-4.

Images