AI simulation performance test method and system of hospital-wide clinical rehabilitation integrated system

Abstract

The invention discloses an AI simulation performance test method and system for a hospital-wide clinical rehabilitation integrated system, and the method comprises the steps: constructing a rehabilitation scene user behavior model through portrait clustering and a reinforcement learning algorithm DQN, carrying out the dynamic adjustment of a load through a time sequence prediction model LSTM and a multi-arm long-arm machine reinforcement learning strategy, carrying out the testing, and outputting a result. And reporting the performance inflection point. Therefore, the problem of load distortion caused by a fixed script is solved; dynamic load prediction and self-adaptive adjustment in a rehabilitation service scene are realized, a concurrent gradient does not need to be manually preset, a system performance inflection point is automatically identified, and the test accuracy is improved.

Description

Technical Field

[0001] This invention relates to the field of medical software performance testing technology, and in particular to an AI simulation performance testing method and system for an integrated clinical rehabilitation system for the whole hospital. Background Technology

[0002] The hospital-wide integrated clinical rehabilitation system, as a medical information platform integrating multiple departments and scenarios, covers core businesses such as rehabilitation assessment, treatment plan development, rehabilitation data management, rehabilitation expert consultation, and offline rehabilitation service appointments. Its performance and stability directly affect the efficiency of clinical rehabilitation diagnosis and treatment, patient rehabilitation experience, and the quality of medical services. Performance testing is a key means to ensure the stable operation of the system in high-concurrency scenarios (such as concentrated assessment of postoperative rehabilitation patients, surges in consultations during rehabilitation expert consultation periods, and peak periods for rehabilitation service appointments after the implementation of medical insurance reimbursement policies). The realism of the load simulation and the adaptability of the scenario are the core prerequisites for the validity of the test results.

[0003] In existing technologies, load simulation of medical information systems often uses manually written fixed-path scripts (such as login → select rehabilitation project → submit assessment form) to perform load testing through preset fixed concurrency gradients. However, when applied to a hospital-wide integrated clinical rehabilitation system, it has the following drawbacks: 1. Distortion in User Behavior Simulation in Rehabilitation Scenarios: The user groups of the integrated clinical rehabilitation system for the entire hospital have strong professional attributes and scenario-specific characteristics (such as rehabilitation physicians, rehabilitation therapists, postoperative rehabilitation patients, elderly rehabilitation patients, and rehabilitation institution managers). Their operational behaviors differ significantly (e.g., physicians operate with rigorous procedures, elderly patients have long intervals between operations, and rehabilitation institution managers export rehabilitation data in batches). Fixed scripts cannot reproduce such personalized behaviors, nor can they simulate non-deterministic operations specific to rehabilitation scenarios, such as "consulting on rehabilitation suitability during assessment scale completion," "abandoning import due to incompatible rehabilitation data formats," and "synchronizing patient treatment data between cross-hospital HIS systems and the integrated rehabilitation system." This leads to a disconnect between the test load and real clinical usage scenarios, making it impossible to discover hidden bottlenecks (such as cross-system rehabilitation data synchronization delays).

[0004] 2. Load gradient settings rely on manual experience and have poor adaptability: Testers need to preset concurrency gradients based on experience. However, the high-concurrency scenarios of the hospital-wide integrated clinical rehabilitation system are prone to sudden changes (such as concentrated appointments after the introduction of rehabilitation benefit policies or a surge in rehabilitation demand after a public health emergency). Manually preset gradients are prone to problems such as "overly coarse gradients that miss performance inflection points" or "overly fine gradients that waste resources." At the same time, existing load adjustments do not take into account the characteristics of rehabilitation business (such as the long session characteristics of rehabilitation assessment, the multi-departmental collaborative approval mechanism for rehabilitation treatment plans, and the privacy protection transmission protocol for rehabilitation data), resulting in test results that cannot reflect the system performance under real clinical business pressure.

[0005] 3. Insufficient adaptation to rehabilitation-specific scenarios: Traditional load simulation cannot dynamically adjust the load strategy according to real-time scenarios of rehabilitation business (such as a surge in consultation sessions during the consultation hours of rehabilitation experts, conflicts in the reservation of rehabilitation equipment, congestion of cross-terminal rehabilitation data query requests, and batch generation of rehabilitation assessment reports), which leads to frequent misjudgments during the test (such as misjudging the performance of the rehabilitation data query interface due to network fluctuations) or test interruptions (such as load execution failure after the resource locking logic is triggered in the batch reservation scenario).

[0006] 4. Lack of differentiated modeling for rehabilitation user groups: The behavior of the core user groups of the integrated clinical rehabilitation system for the whole hospital (rehabilitation physicians, postoperative rehabilitation patients, elderly rehabilitation patients, and rehabilitation institution managers) has not been differentiated. The unified load mode cannot cover the system performance under multiple concurrent user scenarios (such as the system pressure when rehabilitation physicians review assessment reports in batches and patients query rehabilitation plans in batches). Summary of the Invention

[0007] The purpose of this invention is to provide an AI simulation performance testing method and system for an integrated clinical rehabilitation system for the whole hospital, so as to solve one or more of the above-mentioned technical shortcomings.

[0008] To achieve this objective, the present invention adopts the following technical solution: A method for testing the AI ​​simulation performance of a hospital-wide integrated clinical rehabilitation system, comprising: S1: User behavior modeling in rehabilitation scenarios; Based on the historical access log information of the hospital's integrated clinical rehabilitation system, users were divided into multiple user groups. Extract the behavioral probability distribution for each user type group and fit the behavioral interval between each user type group or across user type groups; Simulation training is used to generate non-deterministic behavioral sequences; S2: Dynamic load prediction and adaptive scheduling for rehabilitation services; Train a time-series prediction model based on historical stress test data; Based on the user behavior model in step S1, generate the initial load according to the proportion of real user type groups; Real-time monitoring of system performance indicators and rehabilitation business scenario status; dynamic load adjustment using reinforcement learning strategies. S3: Test execution and result output; Convert the dynamic load generated in step S2 into a test script and execute the test in the test tool; After the test is completed, a test report will be generated.

[0009] In some implementations, step S1 specifically includes: S11: Data Acquisition; Collect historical access log information from the hospital's integrated clinical rehabilitation system; S12: Image clustering; Clustering algorithms were used to segment user groups into categories including: rehabilitation physicians, post-operative rehabilitation patients, elderly rehabilitation patients, and rehabilitation institution managers. S13: Behavioral pattern extraction; Extract the behavioral probability distribution for each user type group, and use a Gaussian distribution model to fit the behavioral intervals between each user type group or across user type groups. S14: Behavioral Decision-Making Training; A behavior decision engine is constructed using reinforcement learning algorithms, and simulation training is performed to form nondeterministic behavior sequences.

[0010] In some implementations, in step S11, the historical access log information includes: behavior sequence, behavior interval, terminal type, user tag and rehabilitation data interaction record; Remove abnormal data from historical access logs.

[0011] In some implementations, the clustering algorithm in step S12 is the K-means algorithm; The clustering algorithm is based on features including: operation frequency, dwell time, business path preference, and type of rehabilitation needs. In some implementations, the reinforcement learning algorithm in step S14 is the DQN algorithm; The reward function of this reinforcement learning algorithm is set to closely resemble real clinical rehabilitation behavior; The inputs to this behavioral decision engine include: the status of the rehabilitation business scenario; The scenario status includes: the online status of rehabilitation experts, the type of rehabilitation assessment scale, the status of cross-system data interaction, and the type of terminal.

[0012] In some implementations, step S2 specifically includes: S21: Training with historical data; Historical stress test data of the hospital's integrated clinical rehabilitation system were collected, and the historical stress test data were used to train a time-series prediction model; the time-series prediction model was an LSTM model. S22: Initial load generation; Based on the user behavior model in step S1, generate the initial load according to the proportion of the real user group; S23: Real-time load adjustment; The system monitors performance metrics and the status of rehabilitation services in real time using monitoring tools, including Prometheus and Grafana. The system employs a reinforcement learning strategy to dynamically adjust the number of concurrent user behaviors and request frequency, thereby dynamically adjusting the load. The reinforcement learning strategy is a multi-armed slot machine reinforcement learning strategy, and the reward function of the reinforcement learning strategy is constructed based on the priority weights of the core rehabilitation business. When the system performance indicators exceed the preset threshold for the rehabilitation scenario, the load increase is paused and the performance inflection point detection is triggered. S24: Load Optimization; If an abnormal recovery scenario is detected, the load strategy will be automatically adjusted.

[0013] In some implementations, in step S21: Historical load test data includes metrics such as: number of concurrent users, QPS, response time, CPU utilization, memory utilization, time spent querying rehabilitation data interface, response time for cross-system data synchronization, and time spent generating assessment reports. The time-series prediction model is used to fit the performance characteristics of rehabilitation-specific services, which include: batch assessment report generation, cross-HIS system data interaction, and rehabilitation equipment reservation and locking. In some implementations, in step S22: the initial load includes: the number of concurrent users for each user type and the rehabilitation-specific operation sequence.

[0014] In some implementations, in step S23: Performance metrics include: response time, error rate, CPU utilization, memory utilization, database connection pool status, and network transmission latency. The status of rehabilitation services includes: the number of online rehabilitation experts, the number of appointments for popular rehabilitation programs, the number of cross-system data interaction requests, and the number of rehabilitation assessment reports pending generation. The reward function is constructed based on "maximizing the QPS of core rehabilitation business + minimizing the response time"; Core business priorities: Rehabilitation assessment report generation > Cross-system data synchronization > Rehabilitation expert consultation > Rehabilitation program booking; The preset thresholds include: cross-system data synchronization interface response time exceeding 800ms, CPU utilization exceeding 85%, and rehabilitation assessment report generation error rate exceeding 1%.

[0015] In some implementations, in step S24: Abnormal scenarios include: a surge in rehabilitation expert consultation sessions, congestion in cross-system data interaction, scheduling conflicts for popular rehabilitation programs, and a backlog of tasks for generating batch assessment reports; The load management strategy includes: reducing the load of batch data export and prioritizing the testing of core interfaces for rehabilitation assessment.

[0016] In some implementations, step S3 specifically includes: S31: Load generation and injection; Convert the dynamic load generated in step S2 into a test script and execute the test in the test tool; S32: Real-time monitoring feedback; During the testing process, performance metrics and the status of rehabilitation business scenarios are monitored and fed back in real time to adjust load strategies. S33: Output the result; After the test is completed, output a test report; The AI ​​simulation performance testing system for the hospital-wide integrated clinical rehabilitation system includes: The rehabilitation user behavior analysis module is used to classify user groups based on historical access log information and through clustering algorithms. The rehabilitation scenario behavior decision engine module is used to fit nondeterministic behavior sequences and is trained using reinforcement learning algorithms. The rehabilitation business load prediction and scheduling module is used to predict the load using a time series prediction model and to dynamically adjust the load based on a reinforcement learning strategy. The real-time monitoring and feedback module for rehabilitation scenarios is used to collect performance indicators and the status of rehabilitation business scenarios, and feed them back to the load prediction and scheduling module and the behavior decision engine module. A compatibility adapter module is used to integrate with external testing tools and convert the generated workload into executable test scripts.

[0017] The beneficial effects of this invention are: 1. Construct a user behavior model that fits the clinical rehabilitation scenario, simulate the personalized operations and non-deterministic behaviors of rehabilitation-related users of different user types, and solve the load distortion problem caused by fixed scripts; 2. Enables dynamic load prediction and adaptive adjustment in rehabilitation business scenarios, eliminating the need for manual pre-setting of concurrency gradients, automatically identifying system performance inflection points, and balancing test sufficiency with resource consumption; 3. Dynamically adjust load strategies based on rehabilitation-specific business scenarios to improve testing accuracy; 4. Reduce the reliance on human experience in the performance testing of the hospital-wide integrated clinical rehabilitation system, adapt to the implementation of rehabilitation benefit policies and sudden high-concurrency scenarios such as public health emergencies, and ensure the real-time response and data security of the system's core business. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the steps of the AI ​​simulation performance testing method for the integrated clinical rehabilitation system of the present invention. Figure 2 This is a flowchart illustrating the steps of the AI ​​simulation performance testing method for the integrated clinical rehabilitation system of the present invention. Figure 3This is a structural diagram of the AI ​​simulation performance testing system for the integrated clinical rehabilitation system of the whole hospital of the present invention. Detailed Implementation

[0019] The present invention will now be described in further detail with reference to the accompanying drawings.

[0020] First embodiment: refer to Figure 1 and Figure 2 A method for testing the AI ​​simulation performance of a hospital-wide integrated clinical rehabilitation system, the method including: S1: User behavior modeling in rehabilitation scenarios; Based on the historical access log information of the hospital's integrated clinical rehabilitation system, users were divided into multiple user groups. Extract the behavioral probability distribution for each user type group and fit the behavioral interval between each user type group or across user type groups; Simulation training is used to form nondeterministic behavioral sequences.

[0021] S2: Dynamic load prediction and adaptive scheduling for rehabilitation services; Train a time-series prediction model based on historical stress test data; Based on the user behavior model in step S1, generate the initial load according to the proportion of real user type groups; Real-time monitoring of system performance metrics and rehabilitation service scenarios; dynamic load adjustment using reinforcement learning strategies.

[0022] S3: Test execution and result output; Convert the dynamic load generated in step S2 into a test script, execute the test in the test tool, and output a test report after the test is completed.

[0023] Therefore, this method simulates personalized operations and non-deterministic behaviors of different user types, solving the load distortion problem caused by fixed scripts; it enables dynamic load prediction and adaptive adjustment in rehabilitation business scenarios, eliminating the need for manual pre-setting of concurrency gradients, automatically identifying system performance inflection points, and balancing the sufficiency of testing with resource consumption; based on rehabilitation-specific business scenarios, it dynamically adjusts load strategies to improve testing accuracy; it reduces the reliance on human experience in the performance testing of the hospital-wide integrated clinical rehabilitation system, adapts to sudden high-concurrency scenarios such as the implementation of rehabilitation benefit policies and public health emergencies, and ensures the real-time response and data security of the system's core business.

[0024] This technology enables load simulation and stress testing optimization of medical information systems driven by AI, and is applicable to the performance testing of high-concurrency core businesses such as rehabilitation assessment scale loading, rehabilitation treatment plan formulation, cross-terminal rehabilitation data interaction, and online consultation with rehabilitation experts.

[0025] Step S1: User behavior modeling in rehabilitation scenarios, specifically including S11 to S14; S11: Data Acquisition; Collect historical access log information from the hospital's integrated clinical rehabilitation system.

[0026] The historical access log information includes: behavior sequence, behavior interval, terminal type, user tags and rehabilitation data interaction records; where behavior can also be understood as operation or operation behavior; Abnormal data in historical access logs is removed to perform data preprocessing and improve data quality.

[0027] S12: Image clustering; Clustering algorithms were used to segment user groups into categories including rehabilitation physicians, postoperative rehabilitation patients, elderly rehabilitation patients, and rehabilitation institution managers.

[0028] The clustering algorithm used is the K-means algorithm. The clustering algorithm is based on features including: operation frequency, dwell time, business path preference, and rehabilitation need type.

[0029] Classification criteria: Rehabilitation physicians: frequently review assessment reports, adhere to rigorous operating procedures, handle multiple concurrent patient inquiries, and pay close attention to data accuracy; Postoperative recovery patients: frequently check the recovery plan, have moderate intervals between procedures, frequently interact with doctors, and pay attention to recovery progress; Elderly rehabilitation patients: long operation intervals, preference for simplified operation procedures, reliance on family assistance, and low frequency of data interaction; Rehabilitation facility managers: Batch export of rehabilitation statistics data, long-term session configuration of rehabilitation projects, low-frequency high-load operations.

[0030] Therefore, by using the K-means algorithm to perform user profile clustering, users are divided into different user types. Each user type has different behavioral patterns and personalized operation characteristics, which improves the accuracy and comprehensiveness of the test.

[0031] S13: Behavioral pattern extraction; Extract the behavioral probability distribution for each user type group, and use a Gaussian distribution model to fit the behavioral intervals between each user type group or across user type groups.

[0032] Among them, the probability distribution of behaviors includes: for example, 20% of rehabilitation physicians consult with senior physicians before reviewing and evaluating reports, 15% of rehabilitation institution managers cancel operations after exporting data in batches, and 10% of postoperative patients synchronize rehabilitation training records across terminals. Behavioral intervals: For example, the average interval between actions performed by elderly users is 45 seconds, and the average interval between consultations handled by rehabilitation physicians is 10 seconds.

[0033] Therefore, the behavioral triggering patterns of each user type group can be fitted.

[0034] S14: Behavioral Decision-Making Training; A behavior decision engine is constructed using reinforcement learning algorithms, and simulation training is performed to form nondeterministic behavior sequences.

[0035] Non-deterministic behavioral sequences refer to situations where, during execution, there are multiple possible behavioral choices at one or more decision points, and the specific choice is not uniquely determined. Examples of non-deterministic behaviors in rehabilitation scenarios include rehabilitation assessment consultations, cross-system data synchronization, batch report generation, and subsequent cancellation of batch reports. The reinforcement learning algorithm used is the DQN algorithm (Deep Q-Network). The reward function of this reinforcement learning algorithm is set to closely resemble real clinical rehabilitation behavior; The inputs to this behavioral decision engine include: the status of the rehabilitation business scenario; The status of this scenario includes: the online status of rehabilitation experts, the type of rehabilitation assessment scale, the status of cross-system data interaction, and the type of terminal.

[0036] The output of the behavioral decision engine includes: the user's next action or operation, such as continuing to fill out the assessment form, consulting a doctor, abandoning the operation, or synchronizing data across systems.

[0037] Therefore, by using the DQN algorithm to form a nondeterministic behavioral sequence, the sequence closely resembles real clinical rehabilitation behavior, thus improving accuracy.

[0038] Therefore, step S1 establishes a user behavior model to simulate personalized operations and non-deterministic behaviors of different user types, solves the load distortion problem caused by fixed scripts, and improves the accuracy and comprehensiveness of the test.

[0039] User behavior modeling specific to rehabilitation scenarios: By integrating K-means clustering and DQN reinforcement learning, differentiated behavior models are constructed for core user groups (rehabilitation physicians, postoperative rehabilitation patients, etc.) of the integrated clinical rehabilitation system for the whole hospital. These models simulate specific behaviors such as rehabilitation assessment and consultation, cross-system data synchronization, etc., to solve the scenario adaptation defects of general load simulation, while taking into account the protection of medical data privacy.

[0040] Quantitative extraction of behavioral patterns in rehabilitation scenarios: Based on the operational characteristics of rehabilitation users and medical business standards, a Gaussian distribution model is used to fit the behavior or operation intervals of different user groups, and the probability distribution of rehabilitation-specific behaviors is quantitatively statistically analyzed to provide accurate data support for load simulation and ensure a high similarity between the test load and the real clinical scenario.

[0041] Step S2: Dynamic load prediction and adaptive scheduling of rehabilitation services, specifically including S21 to S24; S21: Training with historical data; Historical stress test data of the hospital's integrated clinical rehabilitation system were collected and used to train a time-series prediction model, which was an LSTM model.

[0042] The historical load test data includes metrics such as: number of concurrent users, QPS, response time, CPU utilization, memory utilization, time spent on the rehabilitation data query interface, response time for cross-system data synchronization, and time spent generating the assessment report. The time-series prediction model learns the response patterns of the system or mini-program under different loads during training, and is used to fit the performance characteristics of rehabilitation-specific services, including: batch assessment report generation, cross-HIS system data interaction, and rehabilitation equipment reservation and locking.

[0043] Therefore, by training historical load testing data with a time-series prediction model, the load can be predicted, thereby improving accuracy.

[0044] S22: Initial load generation; Based on the user behavior model in step S1, an initial load is generated according to the proportion of real user type groups.

[0045] The initial load includes: the number of concurrent users of various user types, rehabilitation-specific operation sequences, etc. The rehabilitation-specific operation sequences include: rehabilitation physicians batch reviewing and evaluating reports, rehabilitation institution managers batch exporting data, and postoperative patients centrally querying rehabilitation plans, etc. Therefore, by using the proportion of real user types as a reference, an initial load is generated to improve accuracy.

[0046] S23: Real-time load adjustment; S231: Monitor the performance indicators of the system or mini-program and the status of rehabilitation business scenarios in real time through monitoring tools.

[0047] The monitoring tools include Prometheus and Grafana. For example, Prometheus is used to capture and store data, Grafana is used to query and display the data, and Alertmanager is used to configure alerts.

[0048] Performance metrics include: response time, error rate, CPU utilization, memory utilization, database connection pool status, and network transmission latency. Response time focuses on rehabilitation data query, cross-system synchronization, and assessment report generation interfaces; The status of rehabilitation services includes: the number of online rehabilitation experts, the number of appointments for popular rehabilitation programs, the number of cross-system data interaction requests, and the number of rehabilitation assessment reports pending generation.

[0049] S232: Employ reinforcement learning strategies to dynamically adjust the number of concurrent user behaviors and request frequency, thereby dynamically adjusting the load.

[0050] Among them, the reinforcement learning strategy is the multi-armed bandit reinforcement learning strategy, and the reward function of the reinforcement learning strategy is constructed based on the priority weights of the core rehabilitation business; For example: Construct a reward function based on "maximizing the QPS of core rehabilitation business + minimizing the response time"; Core business priorities: Rehabilitation assessment report generation > Cross-system data synchronization > Rehabilitation expert consultation > Rehabilitation program booking.

[0051] S233: When the system performance indicators exceed the preset threshold of the rehabilitation scenario, pause the increase of load and trigger the performance inflection point detection.

[0052] The preset thresholds include: cross-system data synchronization interface response time exceeding 800ms, CPU utilization exceeding 85%, and rehabilitation assessment report generation error rate exceeding 1%.

[0053] Therefore, by monitoring performance indicators and the status of rehabilitation business scenarios, and by dynamically adjusting the load in conjunction with reinforcement learning strategies, the performance inflection point of the system can be accurately located.

[0054] S24: Load Optimization; If an abnormal recovery scenario is detected, the load strategy will be automatically adjusted. Among the abnormal scenarios are: a surge in rehabilitation expert consultation sessions, congestion in cross-system data interaction, scheduling conflicts for popular rehabilitation programs, and backlog of tasks for generating batch assessment reports. Load balancing strategies include reducing the load of batch data export and prioritizing testing of core interfaces for rehabilitation assessment.

[0055] Therefore, when abnormal scenarios occur, the load strategy should be adjusted in a timely manner to prioritize the protection of core test projects.

[0056] Therefore, step S2 enables dynamic load prediction and adaptive adjustment in rehabilitation business scenarios, eliminating the need for manual pre-setting of concurrency gradients, automatically identifying system performance inflection points, and balancing test sufficiency with resource consumption; it can dynamically adjust load strategies based on rehabilitation-specific business scenarios, thereby improving test accuracy.

[0057] Weighted load scheduling for rehabilitation services: Based on the priority of core rehabilitation services (assessment report generation > cross-system synchronization > expert consultation > project booking), a reinforcement learning reward function is designed. Combined with the LSTM model, the performance characteristics of long-link interfaces such as rehabilitation data query and cross-system interaction are fitted to improve the business targeting and data security of load scheduling.

[0058] Dynamic adaptive mechanism for rehabilitation scenarios: Scenario identification and load adjustment rules are designed for specific scenarios such as rehabilitation expert consultation sessions, batch assessment report generation, and cross-system data synchronization, to avoid the general load simulation ignoring the characteristics of rehabilitation business and the requirements for medical data security.

[0059] Step S3: Test execution and result output, specifically including S31 to S33; S31: Load generation and injection; Convert the dynamic load generated in step S2 into a test script and execute the test in the test tool.

[0060] The testing tools are load and performance testing tools, such as JMeter, k6, Gatling, Vegeta, Locust, etc. During testing, the entire hospital's integrated clinical rehabilitation system was injected into the testing environment, and the injection process followed medical data transmission security standards.

[0061] S32: Real-time monitoring feedback; During the testing process, the performance indicators of the system or mini-program and the status of the rehabilitation business scenario are monitored and fed back in real time, and then fed back to step S2 to adjust the load strategy and drive dynamic load adjustment.

[0062] S33: Output the result; After the test is completed, a test report will be generated.

[0063] The test report includes: performance inflection point data for rehabilitation scenarios, core business performance under concurrent user groups of various types, and bottleneck analysis specific to rehabilitation. The bottleneck analysis specifically for rehabilitation includes: cross-system rehabilitation data synchronization delays, batch assessment report generation timeouts, and conflicts in rehabilitation equipment reservation locking.

[0064] Therefore, step S3 reduces the reliance on human experience in the performance testing of the hospital-wide integrated clinical rehabilitation system, adapts to the implementation of rehabilitation benefit policies and sudden high-concurrency scenarios such as public health emergencies, and ensures the real-time response and data security of the system's core business.

[0065] Lightweight adaptation for medical systems: It supports seamless integration with the deployment architecture and data interaction protocols of the hospital's integrated clinical rehabilitation system, is compatible with mainstream stress testing tools and medical data security standards, and does not require reconstruction of existing testing processes, thereby reducing the implementation cost and compliance risks of performance testing for medical information systems.

[0066] Second embodiment: refer to Figure 3 The AI ​​simulation performance testing system for the hospital's integrated clinical rehabilitation system is used to execute the steps of the above methods; The system includes: The rehabilitation user behavior analysis module 1 is used to classify user groups based on historical access log information using a clustering algorithm. The rehabilitation scenario behavior decision engine module 2 is used to fit nondeterministic behavior sequences and is trained using a reinforcement learning algorithm; The rehabilitation business load prediction and scheduling module 3 is used to predict the load using a time series prediction model and to dynamically adjust the load based on a reinforcement learning strategy. The rehabilitation scenario real-time monitoring and feedback module 4 is used to collect system performance indicators and rehabilitation business scenario status, and feed them back to the load prediction and scheduling module and the behavior decision engine module. The compatibility adapter module 5 is used to integrate with external testing tools 6 and convert the system-generated load into executable test scripts.

[0067] This system can be deployed in three types of environments and is compatible with the technical architecture of a hospital-wide integrated clinical rehabilitation system: 1. Test terminal (client): Embedded with JMeter and other stress testing tools as a plugin, suitable for small-scale performance verification of a single functional module of the system (such as rehabilitation assessment, project appointment); 2. Application Server (Cloud Server): Deployed as an independent service, supporting large-scale concurrent load generation and distributed stress testing, adaptable to large-scale high-concurrency scenarios such as the implementation of rehabilitation and benefit policies and public health emergencies, and the deployment process complies with hospital network security standards; 3. Edge nodes (edge ​​servers close to the mini-program server): Reduce the impact of network latency on load injection, focus on ensuring the accuracy of network-sensitive business tests such as cross-system rehabilitation data synchronization and remote rehabilitation consultation, and reduce the risk of privacy leakage during data transmission.

[0068] Application Example 1: Batch generation of rehabilitation assessment reports scenario test: The system clusters user logs to identify three core user groups: "rehabilitation physicians," "rehabilitation institution managers," and "postoperative rehabilitation patients." The behavioral decision engine simulates the behaviors of 15% of rehabilitation institution managers who cancel operations after exporting rehabilitation assessment statistics in batches, and 20% of rehabilitation physicians who consult with senior physicians before reviewing assessment reports; The load scheduling module predicts the initial load based on the LSTM model, dynamically adjusts the number of concurrent users by monitoring and evaluating the response time and CPU utilization of the report generation interface in real time, and finally accurately locates the performance bottleneck of "database lock wait caused by batch report generation".

[0069] Application Example 2: Cross-system rehabilitation data synchronization high-concurrency scenario test: Clustering yielded user groups such as "rehabilitation physicians" and "postoperative rehabilitation patients," simulating the behavior of rehabilitation physicians importing patient HIS system diagnosis and treatment data in batches and postoperative patients centrally synchronizing cross-hospital rehabilitation training records. During load adjustment, a sudden increase in the response time of the cross-system data synchronization interface was detected. The load increase was automatically paused, and the inflection point of "unoptimized rehabilitation data format conversion" was located to avoid resource waste.

[0070] Beneficial technical effects: 1. Load Authenticity Adapted to Rehabilitation Clinical Scenarios: Simulating the unique behaviors and group differences of rehabilitation users, the similarity between the test load and the real-world usage scenarios of the integrated clinical rehabilitation system for the entire hospital is increased from less than 55% in traditional solutions to more than 90%. This can reveal bottlenecks specific to rehabilitation scenarios, such as cross-system data synchronization delays and batch assessment report generation timeouts. 2. Significantly improved testing efficiency: No need to manually write fixed scripts and preset concurrency gradients, test preparation time is reduced by more than 70%, the time to locate the performance inflection point in rehabilitation scenarios is reduced from hours to minutes, and resource consumption is reduced by more than 45%; 3. The test accuracy meets the needs of medical business: Through the adaptive feedback mechanism of rehabilitation scenario, test misjudgment caused by the surge of consultation sessions and data synchronization congestion is avoided. The false alarm rate is reduced from 38% in the traditional solution to below 7%, while ensuring that the test process complies with medical data security specifications. 4. Universally adaptable to rehabilitation medical scenarios: It can be flexibly adapted to various high-concurrency scenarios such as the implementation of rehabilitation benefit policies, public health emergencies, and centralized appointments, without the need for customized development for specific scenarios, thus reducing the maintenance costs and compliance risks of performance testing of the integrated clinical rehabilitation system for the whole hospital; 5. Ensure the continuity of clinical rehabilitation processes: By accurately identifying and optimizing performance bottlenecks, ensure the system operates stably in high-concurrency scenarios, reduce delays in rehabilitation diagnosis and treatment caused by system failures, and improve the quality of medical services and patient satisfaction.

[0071] The above description only discloses some embodiments of the present invention. For those skilled in the art, various modifications and improvements can be made without departing from the inventive concept of the present invention, and these all fall within the scope of protection of the invention.

Claims

1. A method for testing the AI ​​simulation performance of a hospital-wide integrated clinical rehabilitation system, comprising: S1: User behavior modeling in rehabilitation scenarios; Based on the historical access log information of the hospital's integrated clinical rehabilitation system, users were divided into multiple user groups. Extract the behavioral probability distribution for each user type group and fit the behavioral interval between each user type group or across user type groups; Simulation training is used to form non-deterministic behavioral sequences; S2: Dynamic load prediction and adaptive scheduling for rehabilitation services; Train a time-series prediction model based on historical stress test data; Based on the user behavior model in step S1, generate the initial load according to the proportion of real user type groups; Real-time monitoring of system performance indicators and rehabilitation business scenario status; dynamic load adjustment using reinforcement learning strategies. S3: Test execution and result output; Convert the dynamic load generated in step S2 into a test script and execute the test in the test tool; After the test is completed, a test report will be generated.

2. The AI ​​simulation performance testing method for the integrated clinical rehabilitation system of the whole hospital according to claim 1, characterized in that, Step S1 specifically includes: S11: Data Acquisition; Collect historical access log information from the hospital's integrated clinical rehabilitation system; S12: Image clustering; Clustering algorithms were used to segment user groups into categories including: rehabilitation physicians, post-operative rehabilitation patients, elderly rehabilitation patients, and rehabilitation institution managers. S13: Behavioral pattern extraction; Extract the behavioral probability distribution for each user type group, and use a Gaussian distribution model to fit the behavioral intervals between each user type group or across user type groups. S14: Behavioral Decision-Making Training; A behavior decision engine is constructed using reinforcement learning algorithms, and simulation training is performed to form nondeterministic behavior sequences.

3. The AI ​​simulation performance testing method for the integrated clinical rehabilitation system of the whole hospital as described in claim 2, characterized in that, In step S11, the historical access log information includes: behavior sequence, behavior interval, terminal type, user tag and rehabilitation data interaction record; Remove abnormal data from historical access logs.

4. The AI ​​simulation performance testing method for the integrated clinical rehabilitation system of the whole hospital according to claim 2, characterized in that, In step S12, the clustering algorithm is the K-means algorithm; The clustering algorithm is based on features including: operation frequency, dwell time, business path preference, and type of rehabilitation needs. In step S14, the reinforcement learning algorithm is the DQN algorithm; The reward function of this reinforcement learning algorithm is set to closely resemble real clinical rehabilitation behavior; The inputs to this behavioral decision engine include: the status of the rehabilitation business scenario; The scenario status includes: the online status of rehabilitation experts, the type of rehabilitation assessment scale, the status of cross-system data interaction, and the type of terminal.

5. The AI ​​simulation performance testing method for the integrated clinical rehabilitation system of the whole hospital according to claim 1, characterized in that, Step S2 specifically includes: S21: Training with historical data; Historical stress test data of the hospital's integrated clinical rehabilitation system were collected, and the historical stress test data were used to train a time-series prediction model; the time-series prediction model was an LSTM model. S22: Initial load generation; Based on the user behavior model in step S1, generate the initial load according to the proportion of the real user group; S23: Real-time load adjustment; The system monitors performance metrics and the status of rehabilitation services in real time using monitoring tools, including Prometheus and Grafana. The system employs a reinforcement learning strategy to dynamically adjust the number of concurrent user behaviors and request frequency, thereby dynamically adjusting the load. The reinforcement learning strategy is a multi-armed slot machine reinforcement learning strategy, and the reward function of the reinforcement learning strategy is constructed based on the priority weights of the core rehabilitation business. When the system performance indicators exceed the preset threshold for the rehabilitation scenario, the load increase is paused and the performance inflection point detection is triggered. S24: Load Optimization; If an abnormal recovery scenario is detected, the load strategy will be automatically adjusted.

6. The AI ​​simulation performance testing method for the integrated clinical rehabilitation system of the whole hospital according to claim 5, characterized in that, In step S21: Historical load test data includes metrics such as: number of concurrent users, QPS, response time, CPU utilization, memory utilization, time spent querying rehabilitation data interface, response time for cross-system data synchronization, and time spent generating assessment reports. The time-series prediction model is used to fit the performance characteristics of rehabilitation-specific services, which include: batch assessment report generation, cross-HIS system data interaction, and rehabilitation equipment reservation and locking. In step S22: The initial load includes: the number of concurrent users for each user type and the rehabilitation-specific operation sequence.

7. The AI ​​simulation performance testing method for the integrated clinical rehabilitation system of the whole hospital according to claim 5, characterized in that, In step S23: Performance metrics include: response time, error rate, CPU utilization, memory utilization, database connection pool status, and network transmission latency. The status of rehabilitation services includes: the number of online rehabilitation experts, the number of appointments for popular rehabilitation programs, the number of cross-system data interaction requests, and the number of rehabilitation assessment reports pending generation. The reward function is constructed based on "maximizing the QPS of core rehabilitation business + minimizing the response time"; Core business priorities: Rehabilitation assessment report generation > Cross-system data synchronization > Rehabilitation expert consultation > Rehabilitation program booking; The preset thresholds include: cross-system data synchronization interface response time exceeding 800ms, CPU utilization exceeding 85%, and rehabilitation assessment report generation error rate exceeding 1%.

8. The AI ​​simulation performance testing method for the integrated clinical rehabilitation system of the whole hospital according to claim 5, characterized in that, In step S24: Abnormal scenarios include: a surge in rehabilitation expert consultation sessions, congestion in cross-system data interaction, scheduling conflicts for popular rehabilitation programs, and a backlog of tasks for generating batch assessment reports; The load management strategy includes: reducing the load of batch data export and prioritizing the testing of core interfaces for rehabilitation assessment.

9. The AI ​​simulation performance testing method for the integrated clinical rehabilitation system of the whole hospital according to claim 1, characterized in that, Step S3 specifically includes: S31: Load generation and injection; Convert the dynamic load generated in step S2 into a test script and execute the test in the test tool; S32: Real-time monitoring feedback; During the testing process, performance metrics and the status of rehabilitation business scenarios are monitored and fed back in real time to adjust load strategies. S33: Output the result; After the test is completed, output a test report; The test report includes: performance inflection point data.

10. An AI simulation performance testing system for a hospital-wide integrated clinical rehabilitation system, used to perform the steps of the method described in claim 1; The system includes: The rehabilitation user behavior analysis module is used to classify user groups based on historical access log information and through clustering algorithms. The rehabilitation scenario behavior decision engine module is used to fit nondeterministic behavior sequences and is trained using reinforcement learning algorithms. The rehabilitation business load prediction and scheduling module is used to predict the load using a time series prediction model and to dynamically adjust the load based on a reinforcement learning strategy. The real-time monitoring and feedback module for rehabilitation scenarios is used to collect performance indicators and the status of rehabilitation business scenarios, and feed them back to the load prediction and scheduling module and the behavior decision engine module. A compatibility adapter module is used to integrate with external testing tools and convert the generated workload into executable test scripts.

Images