Method and device for simulating test on medical risk prevention and control system

Abstract

This specification provides a method and apparatus for simulation testing of a medical risk prevention and control system. In the simulation testing method, a main intelligent agent generates a task sequence based on user instructions related to medical risk prevention and control. This task sequence includes a task to generate medical risk samples and a task to deploy them. The main intelligent agent sends a first prompt word corresponding to the generation task to a first sub-intelligent agent. Based on the first prompt word, the first sub-intelligent agent searches a medical knowledge base and outputs several risk parameter pairs to form a medical risk sample. The main intelligent agent generates a second prompt word corresponding to the deployment task based on the medical risk sample and sends the second prompt word to a second sub-intelligent agent. Based on the second prompt word, the second sub-intelligent agent deploys the medical risk sample into the simulation system corresponding to the medical risk prevention and control system, obtaining the disposal result of the medical risk sample, which is used to determine the test result for the medical risk prevention and control system.

Description

Technical Field

[0001] This specification relates to the field of computer technology, and in particular to a method and apparatus for simulation testing of medical risk prevention and control systems. Background Technology

[0002] In the field of internet healthcare, such as online drug purchases and online consultations, ensuring medication safety and transaction compliance are crucial cornerstones. Currently, the industry mainly adopts a traditional risk control model that relies on historical real risk events (negative feedback) for passive response and iterative optimization.

[0003] However, due to the characteristics of online medical risks—namely, their "sparseness" (low probability of actual occurrence) and "long tail" (complex and diverse scenarios)—this traditional model suffers from limitations in verification and control efficiency when addressing these risk characteristics. Therefore, a method is needed to proactively and efficiently simulate and test medical risk control systems. Summary of the Invention

[0004] This specification describes one or more embodiments of a method for simulating and testing a medical risk prevention and control system, which can proactively and efficiently perform simulation and testing on the medical risk prevention and control system.

[0005] Firstly, a method for simulation testing of a medical risk prevention and control system is provided, including:

[0006] The main intelligent agent generates a task sequence based on user instructions related to medical risk prevention and control; the task sequence includes tasks for generating and deploying medical risk samples.

[0007] The main agent sends the first prompt word corresponding to the generated task to the first sub-agent;

[0008] The first sub-agent searches the medical knowledge base based on the first prompt word and outputs several risk parameter pairs to form a medical risk sample.

[0009] The main intelligent agent generates a second prompt word corresponding to the delivery task based on the medical risk sample, and sends the second prompt word to the second sub-intelligent agent;

[0010] The second sub-agent, based on the second prompt word, sends the medical risk sample to the simulation system corresponding to the medical risk prevention and control system to obtain the handling result of the medical risk sample, which is used to determine the test result for the medical risk prevention and control system.

[0011] Secondly, a device for simulation testing of a medical risk prevention and control system is provided, comprising:

[0012] The generation unit is used by the main intelligent agent to generate a task sequence based on user instructions related to medical risk prevention and control; the task sequence includes the task of generating medical risk samples and the task of distributing them.

[0013] A sending unit is used to send the first prompt word corresponding to the generated task from the main intelligent agent to the first sub-intelligent agent;

[0014] The retrieval unit is used by the first sub-intelligent agent to retrieve the medical knowledge base based on the first prompt word and output several risk parameter pairs to form a medical risk sample.

[0015] The generation unit is further configured to generate a second prompt word corresponding to the delivery task based on the medical risk sample by the main intelligent agent, and send the second prompt word to the second sub-intelligent agent;

[0016] The delivery unit is used by the second sub-agent to deliver the medical risk sample to the simulation system corresponding to the medical risk prevention and control system based on the second prompt word, and to obtain the treatment result of the medical risk sample, which is used to determine the test result for the medical risk prevention and control system.

[0017] Thirdly, a computer storage medium is provided on which a computer program is stored, which, when executed in a computer, causes the computer to perform the method of the first aspect.

[0018] Fourthly, a computing device is provided, including a memory and a processor, wherein the memory stores executable code, and the processor executes the executable code to implement the method of the first aspect.

[0019] The method for simulation testing of a medical risk prevention and control system provided in one or more embodiments of this specification uses a main intelligent agent to coordinate multiple sub-intelligent agents with specialized functions to convert user instructions related to medical risk prevention and control into task sequences and produce test results. This enables proactive and efficient simulation testing of the medical risk prevention and control system, thereby effectively supporting the verification and optimization of the medical risk prevention and control system before the actual occurrence of risks. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments in this specification, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a schematic diagram illustrating an implementation scenario of one embodiment disclosed in this specification;

[0022] Figure 2 An interactive diagram illustrating a method for simulation testing of a medical risk prevention and control system according to an embodiment of this specification;

[0023] Figure 3 This diagram illustrates an apparatus for simulating a medical risk control system according to one embodiment of this specification. Detailed Implementation

[0024] The solution provided in this specification will now be described with reference to the accompanying drawings.

[0025] The existing risk prevention and control methods mainly include the following:

[0026] First, manual regression testing. This method involves risk control strategists manually constructing a small number of test cases based on their experience and validating them in a test or simulation environment. Its drawbacks are: extremely limited test case coverage, making it difficult to handle complex and ever-changing combined risk scenarios; high labor costs, hindering high-frequency testing; and excessive reliance on personal experience, which can easily lead to overlooked risks.

[0027] Second, backtesting based on historical cases. This method re-injects real-world risk cases from history into the system to test its interception effectiveness. Its drawbacks are: it can only verify "known risks" and cannot identify "unknown risks" that have not yet occurred; furthermore, because it cannot effectively generalize and derive from the cases, the testing scope is limited to the existing set of historical cases.

[0028] Third, simple rule-based or script-based automated testing. This method involves writing fixed scripts to periodically execute a batch of pre-set test data. Its disadvantages include: high script maintenance costs and insufficient flexibility; frequent script adjustments required if the test scenario changes or the strategy is upgraded; and a lack of intelligence, failing to automatically generate new test scenarios with business significance.

[0029] Fourth, the single risk generation model. This method uses a generative model to generate risk data samples in batches. Its disadvantages are: poor diversity and controllability of generated samples, making it difficult to accurately cover specific risk labels; lack of a closed-loop verification process with the business chain, and the inability to obtain real-time feedback and iterative optimization of sample quality in the simulation environment; and overall, it is still an "isolated generator" rather than a "collaborative training system".

[0030] To address this, this solution proposes a method where a master agent coordinates multiple sub-agents with specialized roles to convert user instructions related to medical risk prevention and control into task sequences and generate test results. This enables proactive and efficient simulation testing of the medical risk prevention and control system, effectively supporting the verification and optimization of the system before actual risks occur.

[0031] Figure 1 This is a schematic diagram illustrating an implementation scenario of one of the embodiments disclosed in this specification. Figure 1 In this system, the main agent Magent receives user instructions related to medical risk prevention and control, and transforms these instructions into a task sequence. This task sequence includes tasks for generating and deploying medical risk samples. Magent then sends generation instructions to Sagent1, the sub-agent executing the generation task. Sagent1 then searches the medical knowledge base to obtain several risk parameter pairs. Based on these risk parameter pairs, multiple medical risk samples are generated. Magent then sends deployment instructions to Sagent2, the sub-agent executing the deployment task. Sagent2 deploys each medical risk sample into the simulation system corresponding to the medical risk prevention and control system, obtaining the handling result of the medical risk samples. Finally, based on this handling result, the test results for the medical risk prevention and control system are determined.

[0032] Figure 2 This diagram illustrates an interactive method for simulation testing of a medical risk prevention and control system according to one embodiment of this specification. Figure 2 As shown, the method may include:

[0033] In step S202, the main intelligent agent Magent receives user instructions related to medical risk prevention and control.

[0034] In one example, the user instruction described above is used to instruct the generation of medical risk samples for prescription drug purchase scenarios. For instance, the user instruction could be: "Generate 200 medical risk samples for prescription drug purchases where the patient's physiological state does not match the medication, and conduct risk control tests in the prescription submission scenario."

[0035] Of course, in practice, medical risk samples can also be generated for other scenarios (such as submitting follow-up visit certificates), and this manual does not limit this.

[0036] In other words, in this solution, simulation testing of the medical risk prevention and control system can be started with a single sentence, which greatly lowers the barrier to entry and makes continuous, high-frequency proactive testing possible.

[0037] In step S204, the main intelligent agent Magent generates a task sequence according to user instructions.

[0038] Specifically, the main agent Magent can understand the intent of user commands and perform task planning based on the intent understanding results to obtain task sequences and scheduling logic; the scheduling logic instructs the sub-agents corresponding to a single task.

[0039] The aforementioned intent understanding includes identifying one or more of the risk category, target business scenario, and sample number in the user instruction.

[0040] For the user instructions in the aforementioned example, the identified risk category was: "physiological state does not match the medication", the target business scenario was: "prescription purchase", and the sample size was: "200".

[0041] It should be understood that in practice, the above-mentioned risk categories can also include: "the medication in use does not match the new prescription", "the underlying disease does not match the medication", and so on.

[0042] In this scheme, the above task sequence includes a generation task followed by a deployment task. In one example, the generation task describes generating a specified number of medical risk samples with a specified risk category and conforming to a specified business scenario. The deployment task describes deploying these medical risk samples into the simulation system of the medical risk prevention and control system.

[0043] The simulation system here can be a digital twin environment for business processes built on infrastructure such as new bus and CCT simulation. This simulation system is used to simulate real online medical business processes (such as prescription submission and consultation dialogue) and load all current online risk control strategies.

[0044] In step S206, the main agent Magent generates a prompt word prompt1 corresponding to the generation task and sends the prompt word prompt1 to the sub-agent Sagent1 corresponding to the generation task.

[0045] The prompt word "prompt1" mentioned above can also be called a generation instruction.

[0046] In one example, the prompt word prompt1 may include: the role of the model (e.g., "You are a medical risk parameter generating agent"), instruction description (e.g., "Your task is to retrieve relevant knowledge such as drug contraindications, diagnostic rules, and risk rules from the medical knowledge base based on the input risk category, and generate logically consistent structured parameter pairs that can be directly used for risk simulation"), reference examples, input data, and output format (e.g., JSON format).

[0047] The input data mentioned above may include, but is not limited to, "risk category", "business scenario" and "number of samples".

[0048] In step S208, the sub-agent Sagent1 (also known as the risk-generating agent) searches the medical knowledge base based on the prompt word prompt1 and outputs several risk parameter pairs.

[0049] In one example, the aforementioned medical knowledge base stores and maintains domain-specific knowledge in a structured format, including but not limited to: drug SPU information (including contraindications), diagnostic coding library, medical knowledge graph, historical risk case library, and risk control strategy rule definitions, etc.

[0050] Specifically, the sub-agent Sagent1 retrieves target knowledge related to the risk category from the medical knowledge base based on the risk category included in the prompt word prompt1, and combines several risk parameter pairs that conform to medical logic based on the retrieved target knowledge.

[0051] Of course, in practice, these risk parameter pairs can also be combined in other ways, and this manual does not limit this.

[0052] In one example, the aforementioned risk parameters could be: "Physiological state: lactation, drug: tetracycline, contraindication reason: may affect infant bone and teeth development." In another example, the aforementioned risk parameters could be: "Underlying disease: severe liver injury, drug: acetaminophen, contraindication reason: increased risk of hepatotoxicity."

[0053] It should be noted that this solution, by searching the medical knowledge base, can ensure that the generated medical risk samples have strict business relevance and medical logical correctness, fundamentally solving the problem of uncontrollable and invalid output results from random or general generation models.

[0054] The aforementioned risk parameter pairs can be referred to as core parameters. After obtaining these core parameters, they can be completed in the following two ways:

[0055] First, multiple auxiliary parameters corresponding to various event types can be pre-configured. These auxiliary parameters, corresponding to a specific event category, may include, but are not limited to, simulated user IDs, timestamps, and non-risk-related product information. When completing the core parameters, the corresponding event identifier can be determined first, then the event category can be determined based on that identifier, and finally, the aforementioned risk parameter pairs can be completed based on the auxiliary parameters corresponding to that event category.

[0056] Second, intelligent agents are used to complete the above-mentioned risk parameter pairs, thereby generating multiple medical risk samples.

[0057] Specifically, sub-agent Sagent1 generates the prompt word prompt2 and sends prompt2 to sub-agent Sagent3 (also known as the sample generation agent). Sub-agent Sagent3 completes the data for several risk parameter pairs and constructs a medical risk sample that conforms to the data structure of the target business scenario based on the completed risk parameter pairs.

[0058] In one example, the prompt word prompt2 may include: the role of the model (e.g., "You are a medical risk parameter completion agent"), the instruction description (e.g., "Your task is to infer and complete the missing parameters based on the incomplete input risk parameter pairs, combined with the medical knowledge base and business logic, so that the completed risk parameter pairs can be directly used to construct downstream medical risk samples"), reference examples, input data, and output format (e.g., JSON format).

[0059] The input data mentioned above may include several risk parameter pairs and target business scenarios.

[0060] In step S210, the main agent Magent generates a prompt word prompt3 corresponding to the delivery task based on the medical risk sample, and sends the prompt word prompt3 to the sub-agent Sagent2.

[0061] The prompt word "prompt3" mentioned above can also be called a delivery instruction.

[0062] In one example, the prompt word prompt3 may include: the role of the model (e.g., "You are a professional medical risk prevention and control simulation test engineer"), the instruction description (e.g., "Your task is to accurately deploy medical risk samples into the simulation system for regression testing"), reference examples, input data, and output format (e.g., JSON format).

[0063] The input data mentioned above may include, but is not limited to, "medical risk samples".

[0064] In step S212, the sub-agent Sagent2 (also known as the sample delivery agent) delivers the medical risk sample to the simulation system corresponding to the medical risk prevention and control system based on the prompt word prompt3, and obtains the disposal result of the medical risk sample.

[0065] In one embodiment, medical risk samples can be deployed to the simulation system at a controlled query rate through a dedicated testing channel.

[0066] After receiving a medical risk sample, the simulation system can execute risk judgment logic that is completely consistent with actual business operations and output the handling result of interception or passage.

[0067] As can be seen, this solution provides a safe and realistic "battleground," making it possible to conduct high-intensity stress tests on medical risk prevention and control systems under zero-risk conditions. Furthermore, by deploying medical risk samples into the simulation system, this solution ensures that test traffic is separated from real business traffic, thus achieving "training without affecting real-world application."

[0068] Additionally, the above task sequence may also include analysis tasks, so the main agent Magent may also generate a prompt word prompt4 corresponding to the analysis task and send the prompt word prompt4 to the sub-agent Sagent4 (also known as the test analysis agent) corresponding to the analysis task. The sub-agent Sagent4 performs analysis processing based on the prompt word prompt4 and obtains the test results for the medical risk prevention and control system based on the analysis processing results.

[0069] In one example, the prompt word prompt4 may include: the role the model plays (e.g., "You are an experienced risk control strategy analyst"), the instruction description (e.g., "Your task is to conduct an in-depth review of the system being tested"), a reference example, input data, a description of the analysis task, and the output format (e.g., JSON format).

[0070] The input data mentioned above may include the treatment results of each medical risk sample, etc.

[0071] In addition, the above analysis task description may include one or more of the following: calculating the effectiveness indicators of the medical risk prevention and control system (e.g., recall rate, false positive rate, etc.); performing automatic attribution analysis on unblocked medical risk samples, diagnosing the root causes of prevention and control failures, and generating a list of missed sample diagnoses; generating structured test reports; and automatically creating issue tickets on the collaboration platform for tracking problem remediation.

[0072] The structured test report mentioned above can be seen as the test results for the medical risk prevention and control system.

[0073] In summary, the above-mentioned analysis and processing in this plan constitute a complete operational closed loop of "problem discovery -> cause identification -> task assignment -> solution verification".

[0074] In summary, this solution achieves the following business processes through collaboration between the main intelligent agent and sub-intelligent agents:

[0075] First, the requirement parsing flow: user command -> main intelligent agent.

[0076] Second, the sample generation flow: main agent -> risk generation agent -> medical knowledge base -> generate core parameters -> sample generation agent -> produce complete medical risk samples.

[0077] Third, the attack and defense execution flow: main intelligent agent -> sample delivery intelligent agent -> deliver medical risk samples to the simulation system -> simulation system processes and returns results.

[0078] Fourth, analyze the closed-loop flow: main agent -> test and analyze agent -> handling results -> analysis and diagnosis -> generate reports and work orders -> output layer.

[0079] The unique process design described above can bring about the following innovations:

[0080] 1. Innovation of Medical Risk Prevention and Control Testing Architecture Based on Multi-Agent Collaboration: It is the first to create an automated testing framework in which a main agent uniformly schedules and multiple professional sub-agents such as risk generation, event delivery, and result analysis cooperate, realizing full-process intelligentization from natural language instructions to test reports.

[0081] 2. A precise risk sample generation method guided by domain knowledge integration: This method proposes to use a structured medical knowledge base (drugs, diagnoses, risk rules) as external knowledge to guide the AI ​​agent to generate high-fidelity, highly relevant medical risk samples that conform to business logic. This solves the problems of low quality and uncontrollable samples generated by random generation or single models.

[0082] 3. A fully automated closed-loop approach for risk control testing: A dedicated test traffic delivery channel and intelligent analysis module were designed to achieve secure isolation of test traffic, automatic recovery of results, quantitative diagnosis of strategy effectiveness, and automatic work order processing of issues, forming a complete closed loop that drives the self-evolution of risk control strategies.

[0083] 4. Reconstruct the verification problem of the medical risk prevention and control system into a "dialogue-driven continuous attack and defense interaction process": Through a natural language interaction interface, the operation threshold of complex drills is reduced, making continuous and high-frequency active attack and defense testing a daily operation of the risk control team, thus revolutionizing the paradigm of risk control capability building.

[0084] In summary, the core innovation of this solution lies in transforming complex medical risk prevention and control testing tasks into a highly automated, interpretable, and tightly coupled standardized process of business knowledge through layered decoupling and intelligent agent collaboration. This achieves a revolutionary improvement in the efficiency, breadth, and depth of medical risk control system verification.

[0085] Furthermore, compared to existing risk control methods, this solution overcomes the following shortcomings:

[0086] First, it overcomes the limitations of "low coverage" and "reliance on experience": By combining a risk generation agent with a medical knowledge base, it can automatically generate massive, diverse, and business-logical medical risk samples on demand, covering long-tail and unknown risk combinations, thus overcoming the limitations of human experience.

[0087] Second, it overcomes the limitation of "only being able to verify known risks": AI generation capabilities can create risk scenarios that have never occurred in history but are logically possible (such as new drug combination contraindications and new voucher forgery patterns), thus enabling the testing of the ability to prevent and control unknown risks in medical risk management.

[0088] Third, it overcomes the problems of "poor flexibility and high maintenance costs": conversational interaction and automatic orchestration of intelligent agents make it extremely cost-effective to initiate a new test, eliminating the need to write and maintain fixed scripts, which greatly improves flexibility and usability.

[0089] Fourth, overcome the "lack of closed-loop and intelligent analysis": the test analysis agent realizes a fully automated closed loop from result collection, index calculation, attribution diagnosis to work order creation, transforming simulation testing from "one-time testing" into an "evolutionary engine" that drives continuous strategy iteration.

[0090] It should be understood that, after overcoming the above-mentioned shortcomings, this solution can bring about the following technical effects:

[0091] First, efficiency leap: the cycle of a single complete simulation test is shortened from several days (manual) to 10 minutes (AI-driven), improving efficiency by over 95%.

[0092] Second, depth and breadth: It can conduct systematic and saturation attack verification on specific risk domains (such as "physiological state and drug inconsistency"), and realize the quantitative measurement of risk defense strength.

[0093] Third, proactive immunity: transforming passive response into proactive discovery, it can expose and repair strategic blind spots in advance before real risks occur, significantly reducing the probability of "the first case being a major risk".

[0094] Fourth, capability solidification: Expert experience is solidified in the knowledge base and intelligent agent, providing a secure, reliable, and self-evolving proactive risk control capability foundation for Internet healthcare business.

[0095] Corresponding to the above-described method for simulating and testing a medical risk prevention and control system, one embodiment of this specification also provides an apparatus for simulating and testing a medical risk prevention and control system, such as... Figure 3 As shown, the device may include:

[0096] The generation unit 302 is used to generate a task sequence by the main intelligent agent according to user instructions related to medical risk prevention and control. The task sequence includes the task of generating medical risk samples and the task of distributing them.

[0097] The sending unit 304 is used to send the first prompt word corresponding to the generated task from the main intelligent agent to the first sub-intelligent agent;

[0098] The retrieval unit 306 is used by the first sub-agent to retrieve the medical knowledge base based on the first prompt word and output several risk parameter pairs to form a medical risk sample.

[0099] The generation unit 302 is also used to generate a second prompt word corresponding to the delivery task by the main intelligent agent based on the medical risk sample, and send the second prompt word to the second sub-intelligent agent;

[0100] The delivery unit 308 is used by the second sub-agent to deliver medical risk samples to the simulation system corresponding to the medical risk prevention and control system based on the second prompt word, so as to obtain the disposal result of the medical risk samples and determine the test result for the medical risk prevention and control system.

[0101] In one embodiment, the generating unit 302 is specifically used for:

[0102] The main intelligent agent performs intent understanding on user instructions and performs task planning based on the intent understanding results to obtain task sequences and scheduling logic. The scheduling logic at least instructs the sub-intelligent agents corresponding to a single task.

[0103] In one embodiment, the above intent understanding includes identifying one or more of the risk category, target business scenario, and sample number in the user instruction.

[0104] In one embodiment, the first prompt word indicates at least the risk category; the retrieval unit 306 is specifically used for:

[0105] The first sub-agent retrieves target knowledge related to the risk category from the medical knowledge base based on the risk category, and combines several risk parameter pairs that conform to medical logic based on the retrieved target knowledge.

[0106] The aforementioned target knowledge includes one or more of the following: drug contraindications, diagnostic rules, and risk rules.

[0107] In one embodiment, the device further includes: a completion unit 310;

[0108] The generation unit 302 is also used to generate a third prompt word by the first sub-intelligent agent and send the third prompt word to the third sub-intelligent agent. The third prompt word includes several risk parameter pairs and the target business scenario.

[0109] The completion unit 310 is used by the third sub-agent to complete the data of several risk parameters, and based on the completed risk parameter pairs, to construct a medical risk sample that conforms to the data structure of the target business scenario.

[0110] In one embodiment, the aforementioned multiple tasks further include an analysis task; the device further includes an analysis unit 312.

[0111] The generation unit 302 is further configured to generate a fourth prompt word corresponding to the analysis task by the main intelligent agent, and send the fourth prompt word to the fourth sub-intelligent agent corresponding to the analysis task, wherein the fourth prompt word includes the processing result;

[0112] Analysis unit 312 is used by the fourth sub-agent to perform analysis and processing based on the disposal results, and to obtain test results for the medical risk prevention and control system based on the analysis and processing results of the fourth sub-agent.

[0113] The aforementioned outcomes include interception or passage.

[0114] In one embodiment, the above analytical processing includes one or more of the following:

[0115] Calculate the effectiveness indicators of the medical risk prevention and control system;

[0116] Automatic attribution analysis is performed on unintercepted medical risk samples to diagnose the root causes of prevention and control failures;

[0117] Generate structured test reports;

[0118] Create issue tickets to track problem fixes.

[0119] In one embodiment, the dispensing unit 308 is specifically used for:

[0120] Medical risk samples are deployed to the simulation system through a dedicated testing channel with a controlled query rate.

[0121] The functions of each functional unit of the apparatus in the above embodiments of this specification can be implemented through the steps of the above method embodiments. Therefore, the specific working process of the apparatus provided in one embodiment of this specification will not be repeated here.

[0122] This specification provides an embodiment of an apparatus for simulating and testing medical risk prevention and control systems, which can proactively and efficiently perform simulation and testing on medical risk prevention and control systems.

[0123] According to another embodiment, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed in a computer, causes the computer to perform a combination Figure 2 The method described.

[0124] According to another embodiment, a computing device is also provided, including a memory and a processor, wherein the memory stores executable code, and when the processor executes the executable code, it implements a combination... Figure 2 The method described.

[0125] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the medium or device embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0126] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0127] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of this specification. It should be understood that the above description is only a specific embodiment of this specification and is not intended to limit the scope of protection of this specification. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solution of this specification should be included within the scope of protection of this specification.

Claims

1. A method for simulation testing of a medical risk prevention and control system, comprising: The main intelligent agent generates a task sequence based on user instructions related to medical risk prevention and control; The task sequence includes tasks for generating and deploying medical risk samples; The main agent sends the first prompt word corresponding to the generated task to the first sub-agent; The first sub-agent searches the medical knowledge base based on the first prompt word and outputs several risk parameter pairs to form a medical risk sample. The main intelligent agent generates a second prompt word corresponding to the delivery task based on the medical risk sample, and sends the second prompt word to the second sub-intelligent agent; The second sub-agent, based on the second prompt word, sends the medical risk sample to the simulation system corresponding to the medical risk prevention and control system to obtain the handling result of the medical risk sample, which is used to determine the test result for the medical risk prevention and control system.

2. The method according to claim 1, wherein, The generated task sequence includes: The main agent performs intent understanding on the user instructions and performs task planning based on the intent understanding results to obtain the task sequence and scheduling logic; the scheduling logic at least indicates the sub-agent corresponding to a single task.

3. The method according to claim 2, wherein, The intent understanding includes identifying one or more of the risk category, target business scenario, and sample number in the user instruction.

4. The method according to claim 1, wherein, The first prompt word indicates at least the risk category; The output includes several risk parameter pairs, including: The first sub-agent retrieves target knowledge related to the risk category from the medical knowledge base based on the risk category, and combines several risk parameter pairs that conform to medical logic based on the retrieved target knowledge; The target knowledge includes one or more of the following: drug contraindications, diagnostic rules, and risk rules.

5. The method according to claim 1, wherein, The medical risk sample was obtained through the following steps: The first sub-agent generates a third prompt word and sends the third prompt word to the third sub-agent; the third prompt word includes the plurality of risk parameter pairs and the target business scenario; The third sub-agent completes the data for the several risk parameters, and based on the completed pairs of risk parameters, constructs a medical risk sample that conforms to the data structure of the target business scenario.

6. The method according to claim 1, wherein, The multiple tasks also include analysis tasks; The method further includes: The main agent generates a fourth prompt word corresponding to the analysis task and sends the fourth prompt word to the fourth sub-agent corresponding to the analysis task; The fourth prompt word includes the processing result; The fourth sub-agent performs analysis and processing based on the disposal result, and obtains the test result for the medical risk prevention and control system based on the analysis and processing result of the fourth sub-agent.

7. The method according to claim 6, wherein, The results of the handling include interception or passage.

8. The method according to claim 6, wherein, The analytical processing includes one or more of the following: Calculate the performance indicators of the medical risk prevention and control system; Automatic attribution analysis is performed on unintercepted medical risk samples to diagnose the root causes of prevention and control failures; Generate structured test reports; Create issue tickets to track problem fixes.

9. The method according to claim 1, wherein, The step of deploying the medical risk sample into the simulation system corresponding to the medical risk prevention and control system includes: The medical risk samples are deployed to the simulation system through a dedicated testing channel at a controlled query rate.

10. A device for simulation testing of a medical risk prevention and control system, comprising: The generation unit is used by the main intelligent agent to generate task sequences based on user instructions related to medical risk prevention and control; The task sequence includes tasks for generating and deploying medical risk samples; A sending unit is used to send the first prompt word corresponding to the generated task from the main intelligent agent to the first sub-intelligent agent; The retrieval unit is used by the first sub-intelligent agent to retrieve the medical knowledge base based on the first prompt word and output several risk parameter pairs to form a medical risk sample. The generation unit is further configured to have the main intelligent agent generate a second prompt word corresponding to the delivery task based on the medical risk sample, and send the second prompt word to the second sub-intelligent agent; The delivery unit is used by the second sub-agent to deliver the medical risk sample to the simulation system corresponding to the medical risk prevention and control system based on the second prompt word, and to obtain the treatment result of the medical risk sample, which is used to determine the test result for the medical risk prevention and control system.

11. A computer-readable storage medium having a computer program stored thereon, wherein, When the computer program is executed in the computer, it causes the computer to perform the method of any one of claims 1-9.

12. A computing device comprising a memory and a processor, wherein, The memory stores executable code, and when the processor executes the executable code, it implements the method of any one of claims 1-9.

Images