Personalized rehabilitation management method and system for diabetic patients based on knowledge graph

Abstract

The present invention relates to providing a personalized rehabilitation management method and system for diabetic patients based on a knowledge graph. By obtaining the changes in the patient's physical indicators, querying the diabetes rehabilitation knowledge graph, outputting a personalized rehabilitation plan containing quantitative execution indicators, and using the Proximal Policy Optimization reinforcement learning algorithm to monitor and evaluate the execution process of the rehabilitation plan. The method aims to improve the rehabilitation effect of diabetic patients, reduce the occurrence of complications, and improve the quality of life.

Description

Technical Field

[0001] The present invention relates to the field of smart medical technology, and specifically to a personalized rehabilitation management method and system for diabetic patients based on a knowledge graph. Background Art

[0002] Diabetes is a chronic disease. Patients need to manage and monitor their physical indicators such as blood sugar, blood pressure, weight, etc. over the long term, and implement rehabilitation plans that help alleviate the condition.

[0003] Existing methods of diabetes management and patient rehabilitation mainly rely on the doctor's experience and the patient's self-management, but due to large individual differences, traditional methods are difficult to meet the personalized needs of each patient.

[0004] In recent years, artificial intelligence technologies such as machine learning have been increasingly used in the medical field, especially the application of knowledge graphs and reinforcement learning algorithms in personalized rehabilitation management of diabetes, which has shown great potential. However, there are still some challenges, including how to effectively integrate and utilize individual patient data, how to monitor and evaluate the effectiveness of rehabilitation programs in real time, etc. Summary of the invention

[0005] The present invention mainly provides a personalized rehabilitation management method and system for diabetic patients based on knowledge graph. By obtaining the changes in the patient's physical indicators, querying the diabetes rehabilitation knowledge graph, outputting a personalized rehabilitation plan containing quantitative execution indicators, and using the Proximal Policy Optimization reinforcement learning algorithm (hereinafter referred to as the PPO algorithm) to monitor and evaluate the execution process of the rehabilitation plan. This method aims to improve the rehabilitation effect of diabetic patients, reduce the occurrence of complications, and improve the quality of life.

[0006] In order to solve the above technical problems, the present invention adopts a personalized rehabilitation management method for diabetic patients based on knowledge graph, comprising:

[0007] Obtain relevant physical indicators of diabetic patients, and pre-process and store relevant physical indicators of diabetic patients; the relevant physical indicators include: fasting blood glucose level (FBG), postprandial blood glucose level (PBG), glycosylated hemoglobin (HbA1c), glucose level, sweat pH value, microbial population, blood pressure, heart rate, electrocardiogram (EGG), weight value, body mass index (BMI), exercise calorie consumption value, blood ketone body level, serum creatinine, urine microalbumin, fundus examination record, foot examination record, neuropathy examination record;

[0008] Based on the medical literature, expert advice and clinical guideline data resources on diabetes rehabilitation, they are converted into structured entity and relationship triples to build a knowledge graph knowledge base related to diabetes rehabilitation management;

[0009] Query the knowledge graph knowledge base related to diabetes rehabilitation management, and generate personalized rehabilitation management plans based on the patient's relevant physical indicators;

[0010] Patient-related physical indicators are selected and defined as the patient's representation state, and the execution action is defined as the correction amount of the execution indicator of the personalized rehabilitation management plan. The reward function is defined according to the changes in the patient's rehabilitation status, and the PPO reinforcement learning algorithm is used to monitor and optimize the execution process of the rehabilitation plan.

[0011] Preferably, the specific method of obtaining the relevant physical indicators of the diabetic patient includes: connecting to the medical information system or laboratory information management system of the medical institution through the network, using the patient ID information to query and download the relevant physical indicators collected by the medical institution through medical examination and testing methods; collecting and uploading the relevant physical indicators of the diabetic patient through wearable monitoring equipment; collecting and uploading the relevant physical indicators of the diabetic patient through equipment at home or at community health service sites.

[0012] Preferably, the entities contained in the knowledge graph include but are not limited to: diabetes type, physical indicators, symptoms, treatment methods, optional medications, dietary management recommendations, monitoring methods, nursing measures, exercise plans, and psychological support measures.

[0013] Preferably, the relationships contained in the knowledge graph include but are not limited to: the inclusion relationship between diabetes type and physical indicators, the inclusion relationship between diabetes type and symptoms, the correlation relationship between physical indicators and symptoms, the applicability relationship of treatment methods to diabetes types, the applicability relationship of treatment methods to symptoms, the correlation relationship between treatment methods and optional drugs, the applicability relationship of dietary management suggestions to diabetes types, the applicability relationship of dietary management suggestions to physical indicators, the applicability relationship of dietary management suggestions to symptoms, the correlation relationship between dietary management suggestions and treatment methods, the applicability relationship of monitoring means to diabetes types, the applicability relationship of monitoring means to physical indicators, the applicability relationship of monitoring means to symptoms, the correlation relationship between monitoring means and treatment methods, the applicability relationship of nursing measures to diabetes types, the applicability relationship of nursing measures to physical indicators, the applicability relationship of nursing measures to symptoms, the correlation relationship between nursing measures and treatment methods, the applicability relationship of exercise plans to diabetes types, the applicability relationship of exercise plans to physical indicators, the applicability relationship of exercise plans to symptoms, the correlation relationship between exercise plans and treatment methods, the applicability relationship of psychological support measures to symptoms, and the correlation relationship between psychological support measures and treatment methods.

[0014] Preferably, the query of the knowledge graph knowledge base related to diabetes rehabilitation management generates a personalized rehabilitation management plan according to the relevant physical indicators of the patient, specifically including: automatically generating a query statement for the knowledge graph knowledge base related to diabetes rehabilitation management according to the relevant physical indicators of the diabetic patient; responding to the query statement based on the entity and relationship triples in the knowledge graph knowledge base related to diabetes rehabilitation management, and feeding back rehabilitation management related query results, the rehabilitation management related query results including: at least one of dietary management recommendations, monitoring methods, nursing measures, exercise plans, and psychological support measures; setting quantitative execution indicators for the rehabilitation management related query results according to the relevant physical indicators of the patient to form the personalized rehabilitation management plan.

[0015] Preferably, the patient characterization state is defined as:

[0016]

[0017] Among them, the symbol In turn, they represent fasting blood sugar level, postprandial blood sugar level, body mass index, heart rate, and electrocardiogram results; and the execution action is defined as:

[0018]

[0019] Among them, the symbol They represent the correction amounts for dietary management suggestions, monitoring methods, nursing measures, exercise plans, and psychological support measures respectively; the reward function is defined as:

[0020]

[0021] in , , , , Respectively represent the differences between the fasting blood glucose level, postprandial blood glucose level, body mass index, heart rate and electrocardiogram results and the ideal values; to Represents the weight of each difference value in the reward function.

[0022] Preferably, during the execution of the PPO algorithm, the strategy network is initialized Network parameters , and initialize the value network Network parameters ; According to the time step , collect trajectories of execution actions and representation states , and then by calculating the advantage function:

[0023]

[0024] The advantage function represents the time step Take action on The degree of advantage relative to the average level is is the discount factor, is the time step Instant rewards, Status Estimation of the value function under PPO; Update the network parameters of the policy network using the PPO cut objective function :

[0025]

[0026] Among them, the function Represents the time step The expected value on Indicates status The current network parameters are used Select Execute Action The probability of Indicates status Use the old network parameters Select the action to perform The probability of That is, the shear function, which means The ratio is limited to In the range of is a hyperparameter that controls the magnitude of policy updates; by maximizing the expected clipping objective function , to achieve optimal network parameters .

[0027] Preferably, the loss is calculated by the value function , Update the value network Network parameters ;function Represents the time step The expected value on Indicates status The network parameters are adopted The state value obtained, is the time step The return value; minimizing the value function loss makes the gap between the state value estimated by the value network and the actual return as small as possible.

[0028] Preferably, during the execution of the PPO algorithm, the updated policy network , it is possible to obtain the correction amount defined by the execution action according to the representation state of the patient's relevant physical indicators, and then to correct the execution indicators of the personalized rehabilitation management plan to provide the patient with a corrected personalized rehabilitation management plan.

[0029] The present invention provides a personalized rehabilitation management system for diabetic patients based on knowledge graph, comprising:

[0030] A central database is used to obtain relevant physical indicators of diabetic patients, and pre-process and store the relevant physical indicators of diabetic patients; the relevant physical indicators include: fasting blood glucose level (FBG), postprandial blood glucose level (PBG), glycosylated hemoglobin (HbA1c), glucose level, sweat pH value, microbial population, blood pressure, heart rate, electrocardiogram (EGG), weight value, body mass index (BMI), exercise calorie consumption value, blood ketone body level, serum creatinine, urine microalbumin, fundus examination record, foot examination record, neuropathy examination record;

[0031] A knowledge graph construction unit is used to convert medical literature, expert advice and clinical guideline data resources on diabetes rehabilitation into structured entity and relationship triples to build a knowledge graph knowledge base related to diabetes rehabilitation management;

[0032] A personalized rehabilitation management program generation unit is used to query the knowledge graph knowledge base related to diabetes rehabilitation management and generate a personalized rehabilitation management program based on the patient's relevant physical indicators;

[0033] The optimization execution unit is used to select patient-related physical indicators and define them as the patient's representative state, and define the execution action as the correction amount of the execution indicator of the personalized rehabilitation management plan, define the reward function according to the change of the patient's rehabilitation status, and use the PPO reinforcement learning algorithm to monitor and optimize the execution process of the rehabilitation plan.

[0034] The beneficial effects of the present invention include: through the knowledge graph, a large amount of medical literature, clinical data and expert knowledge can be integrated to provide a personalized rehabilitation plan for each patient; multiple data sources, including the patient's physical indicators, can be integrated to provide comprehensive rehabilitation suggestions; the implementation of the rehabilitation plan can be continuously optimized through the PPO algorithm to ensure the best treatment effect; the treatment plan can be adjusted according to the individual response of the patient to improve the effectiveness of the treatment and the patient's compliance. Long-term monitoring and management of patients can be carried out to prevent the occurrence of complications and promote long-term health. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 Flowchart of the personalized rehabilitation management method for diabetic patients based on knowledge graph.

[0036] Figure 2This is a framework diagram of the personalized rehabilitation management system for diabetic patients based on knowledge graph. DETAILED DESCRIPTION

[0037] The flexible joint vibration suppression method and system technical solution based on trajectory optimization control according to the present invention will be further described in detail below in conjunction with the embodiments.

[0038] like Figure 1 As shown, the personalized rehabilitation management method for diabetic patients based on knowledge graph in this embodiment includes the following steps S1-S4.

[0039] S1. Obtain relevant physical indicators of diabetic patients.

[0040] The relevant physical indicators include but are not limited to: fasting blood glucose level (FBG), postprandial blood glucose level (PBG), glycosylated hemoglobin (HbA1c), glucose level, sweat pH value, microbial population, blood pressure, heart rate, electrocardiogram (EGG), weight, body mass index (BMI), exercise calorie consumption, blood ketone body level, serum creatinine, urine microalbumin, fundus examination records, foot examination records, neuropathy examination records, etc.

[0041] The central database of the personalized rehabilitation management system for diabetic patients is responsible for obtaining the above-mentioned relevant physical indicators of diabetic patients. The specific acquisition methods include: (1) connecting to the medical information system (HIS) or laboratory information management system (LIMS) of the medical institution through the network, using the patient ID information to query and download the relevant physical indicators collected by the medical institution through medical examination and testing methods; (2) diabetic patients collect and upload the relevant physical indicators of diabetic patients through wearable monitoring devices, such as continuous blood glucose monitors, electrocardiogram monitors, clothing, shoes and socks embedded with monitoring sensors and biochips, bracelets, etc.; (3) collecting and uploading the relevant physical indicators of diabetic patients through equipment at home or at community health service sites, such as weight scales, blood pressure monitors, etc.

[0042] The central database preprocesses and stores the above-mentioned relevant physical indicators of diabetic patients. The preprocessing steps include data cleaning, outlier detection and missing data filling. The specific implementation methods are as follows: (1) Data cleaning: remove duplicate data and invalid data; for example, if the blood glucose meter uploads multiple data with the same timestamp, only one is retained; (2) Outlier detection: use statistical methods to detect outliers, such as blood glucose values ​​that are beyond the normal range; outliers can be determined by setting a threshold range; (3) Missing data filling: use interpolation methods or prediction models based on historical data to fill missing data points.

[0043] Through step S1, various relevant physical indicators of diabetic patients can be comprehensively and accurately obtained and managed, providing a reliable data basis for subsequent personalized rehabilitation management plans.

[0044] S2. Construct a knowledge graph knowledge base related to diabetes rehabilitation management.

[0045] In this step, a large amount of data resources such as medical literature, expert advice and clinical guidelines on diabetes rehabilitation are collected, and the text data of these data resources are converted into structured entity and relationship triplets through natural language processing (NLP) technology to form a knowledge graph knowledge base related to diabetes rehabilitation management.

[0046] In the knowledge graph knowledge base related to diabetes rehabilitation management, the entities contained in the knowledge graph include but are not limited to: diabetes type, physical indicators, symptoms, treatment methods, optional drugs, dietary management recommendations, monitoring methods, nursing measures, exercise plans, psychological support measures, etc.

[0047] The relationships included in the knowledge graph include, but are not limited to: the inclusion relationship between diabetes types and physical indicators, the inclusion relationship between diabetes types and symptoms, the correlation relationship between physical indicators and symptoms, the applicability relationship of treatment methods to diabetes types, the applicability relationship of treatment methods to symptoms, the correlation relationship between treatment methods and optional drugs, the applicability relationship of dietary management suggestions to diabetes types, the applicability relationship of dietary management suggestions to physical indicators, the applicability relationship of dietary management suggestions to symptoms, the correlation relationship between dietary management suggestions and treatment methods, the applicability relationship of monitoring methods to diabetes types, the applicability relationship of monitoring methods to physical indicators, the applicability relationship of monitoring methods to symptoms, the correlation relationship between monitoring methods and treatment methods, the applicability relationship of nursing measures to diabetes types, the applicability relationship of nursing measures to physical indicators, the applicability relationship of nursing measures to symptoms, the correlation relationship between nursing measures and treatment methods, the applicability relationship of exercise plans to diabetes types, the applicability relationship of exercise plans to physical indicators, the applicability relationship of exercise plans to symptoms, the correlation relationship between exercise plans and treatment methods, the applicability relationship of psychological support measures to symptoms, and the correlation relationship between psychological support measures and treatment methods.

[0048] Therefore, through this knowledge graph knowledge base, multi-dimensional information related to diabetes rehabilitation management is reflected.

[0049] S3. Query the knowledge graph knowledge base related to diabetes rehabilitation management, and generate a personalized rehabilitation management plan based on the patient's relevant physical indicators.

[0050] In this step, based on the relevant physical indicators of the diabetic patients, a query statement for the knowledge graph knowledge base related to the diabetes rehabilitation management is automatically generated.

[0051] Based on the entity and relationship triples in the knowledge graph knowledge base related to diabetes rehabilitation management, respond to the query statement and feedback the rehabilitation management related query results.

[0052] The query results related to rehabilitation management include: at least one of: dietary management suggestions, monitoring methods, nursing measures, exercise plans, and psychological support measures.

[0053] Furthermore, according to the relevant physical indicators of the patient, quantitative execution indicators are set for the rehabilitation management related query results to form the personalized rehabilitation management plan.

[0054] S4. Use reinforcement learning algorithms to monitor and optimize the implementation progress of rehabilitation programs.

[0055] In this step, the rules of the Proximal Policy Optimization reinforcement learning algorithm (hereinafter referred to as the PPO algorithm) are adopted. First, the patient's relevant physical indicators are selected and defined as the patient's representation state, and the execution action is defined as the correction amount of the execution indicator of the personalized rehabilitation management plan. For example, the patient's representation state is defined as:

[0056]

[0057] Among them, the symbol In turn, they represent fasting blood sugar level, postprandial blood sugar level, body mass index, heart rate, and electrocardiogram results. The execution action is defined as:

[0058]

[0059] Among them, the symbol They represent the correction amounts for dietary management recommendations, monitoring methods, nursing measures, exercise plans, and psychological support measures, respectively. The reward function is defined according to the changes in the patient's rehabilitation status, and the goal is to maximize the patient's rehabilitation improvement. For example, the reward function is defined as:

[0060]

[0061] in , , , , Respectively represent the differences between the fasting blood glucose level, postprandial blood glucose level, body mass index, heart rate and electrocardiogram results and the ideal values; to Represents the weight of each difference value in the reward function.

[0062] During the execution of the PPO algorithm, the strategy network is initialized Network parameters , and initialize the value network Network parameters . According to the time step , collect trajectories of execution actions and representation states , and then by calculating the advantage function:

[0063]

[0064] The advantage function represents the time step Take action on The degree of advantage relative to the average level is is the discount factor, is the time step Instant rewards, Status The value function estimate under .

[0065] Update the network parameters of the policy network using the clipping objective function of PPO :

[0066]

[0067] Among them, the function Represents the time step The expected value on Indicates status The current network parameters are used Select Execute Action The probability of Indicates status Use the old network parameters Select the action to perform The probability of That is, the shear function, which means The ratio is limited to In the range of is a hyperparameter that controls the magnitude of policy updates. By maximizing the expected clipping objective function , to achieve optimal network parameters .

[0068] Then, through the value function loss , Update the value network Network parameters .function Represents the time step The expected value on Indicates status The network parameters are adopted The state value obtained, is the time step Minimizing the value function loss makes the gap between the state value estimated by the value network and the actual return as small as possible.

[0069] Finally, during the execution of the PPO algorithm, the updated policy network , it is possible to obtain the correction amount defined by the execution action according to the representation state of the patient's relevant physical indicators, and then to correct the execution indicators of the personalized rehabilitation management plan to provide the patient with a corrected personalized rehabilitation management plan.

[0070] Through the monitoring and evaluation of the PPO algorithm, real-time optimization of personalized rehabilitation plans can be achieved to ensure that the patient's rehabilitation process is more scientific and effective.

[0071] Through the user interface, personalized rehabilitation plans are provided to patients, and detailed monitoring and evaluation reports are generated to assist patients in effective rehabilitation management.

[0072] See also Figure 2 The present invention provides a personalized rehabilitation management system for diabetic patients based on knowledge graph, comprising:

[0073] A central database is used to obtain relevant physical indicators of diabetic patients, and pre-process and store the relevant physical indicators of diabetic patients; the relevant physical indicators include: fasting blood glucose level (FBG), postprandial blood glucose level (PBG), glycosylated hemoglobin (HbA1c), glucose level, sweat pH value, microbial population, blood pressure, heart rate, electrocardiogram (EGG), weight value, body mass index (BMI), exercise calorie consumption value, blood ketone body level, serum creatinine, urine microalbumin, fundus examination record, foot examination record, neuropathy examination record;

[0074] A knowledge graph construction unit is used to convert medical literature, expert advice and clinical guideline data resources on diabetes rehabilitation into structured entity and relationship triples to build a knowledge graph knowledge base related to diabetes rehabilitation management;

[0075] A personalized rehabilitation management program generation unit is used to query the knowledge graph knowledge base related to diabetes rehabilitation management and generate a personalized rehabilitation management program based on the patient's relevant physical indicators;

[0076] The optimization execution unit is used to select patient-related physical indicators and define them as the patient's representative state, and define the execution action as the correction amount of the execution indicator of the personalized rehabilitation management plan, define the reward function according to the change of the patient's rehabilitation status, and use the PPO reinforcement learning algorithm to monitor and optimize the execution process of the rehabilitation plan.

[0077] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A personalized rehabilitation management method for diabetic patients based on knowledge graph, characterized in that: include: Obtaining relevant physical indicators of diabetic patients, and preprocessing and storing the relevant physical indicators of diabetic patients; The relevant physical indicators include: fasting blood glucose level (FBG), postprandial blood glucose level (PBG), glycosylated hemoglobin (HbA1c), glucose level, sweat pH value, microbial population, blood pressure, heart rate, electrocardiogram (EGG), weight, body mass index (BMI), exercise calorie consumption, blood ketone body level, serum creatinine, urine microalbumin, fundus examination record, foot examination record, neuropathy examination record; Based on the medical literature, expert advice and clinical guideline data resources on diabetes rehabilitation, they are converted into structured entity and relationship triples to build a knowledge graph knowledge base related to diabetes rehabilitation management; Query the knowledge graph knowledge base related to diabetes rehabilitation management, and generate personalized rehabilitation management plans based on the patient's relevant physical indicators; Patient-related physical indicators are selected and defined as the patient's representation state, and the execution action is defined as the correction amount of the execution indicator of the personalized rehabilitation management plan. The reward function is defined according to the changes in the patient's rehabilitation status, and the PPO reinforcement learning algorithm is used to monitor and optimize the execution process of the rehabilitation plan.

2. The personalized rehabilitation management method for diabetic patients based on knowledge graph according to claim 1, characterized in that: Specific methods for obtaining the relevant physical indicators of the diabetic patient include: connecting to the medical information system or laboratory information management system of a medical institution through the network, using the patient ID information to query and download the relevant physical indicators collected by the medical institution through medical examination and testing methods; collecting and uploading the relevant physical indicators of the diabetic patient through wearable monitoring devices; collecting and uploading the relevant physical indicators of the diabetic patient through equipment at home or at community health service sites.

3. The personalized rehabilitation management method for diabetic patients based on knowledge graph according to claim 2 is characterized in that: The entities included in the knowledge graph include, but are not limited to: diabetes type, physical indicators, symptoms, treatment methods, optional drugs, dietary management recommendations, monitoring methods, nursing measures, exercise plans, and psychological support measures.

4. The personalized rehabilitation management method for diabetic patients based on knowledge graph according to claim 1, characterized in that: The relationships included in the knowledge graph include, but are not limited to: the inclusion relationship between diabetes types and physical indicators, the inclusion relationship between diabetes types and symptoms, the correlation relationship between physical indicators and symptoms, the applicability relationship of treatment methods to diabetes types, the applicability relationship of treatment methods to symptoms, the correlation relationship between treatment methods and optional drugs, the applicability relationship of dietary management suggestions to diabetes types, the applicability relationship of dietary management suggestions to physical indicators, the applicability relationship of dietary management suggestions to symptoms, the correlation relationship between dietary management suggestions and treatment methods, the applicability relationship of monitoring means to diabetes types, the applicability relationship of monitoring means to physical indicators, the applicability relationship of monitoring means to symptoms, the correlation relationship between monitoring means and treatment methods, the applicability relationship of nursing measures to diabetes types, the applicability relationship of nursing measures to physical indicators, the applicability relationship of nursing measures to symptoms, the correlation relationship between nursing measures and treatment methods, the applicability relationship of exercise plans to diabetes types, the applicability relationship of exercise plans to physical indicators, the applicability relationship of exercise plans to symptoms, the correlation relationship between exercise plans and treatment methods, the applicability relationship of psychological support measures to symptoms, and the correlation relationship between psychological support measures and treatment methods.

5. The personalized rehabilitation management method for diabetic patients based on knowledge graph according to claim 4 is characterized in that: The method of querying the knowledge graph knowledge base related to diabetes rehabilitation management and generating a personalized rehabilitation management plan according to the relevant physical indicators of the patient specifically includes: automatically generating a query statement for the knowledge graph knowledge base related to diabetes rehabilitation management according to the relevant physical indicators of the diabetic patient; responding to the query statement based on the entity and relationship triples in the knowledge graph knowledge base related to diabetes rehabilitation management, and feeding back the rehabilitation management related query results, the rehabilitation management related query results including: at least one of dietary management suggestions, monitoring methods, nursing measures, exercise plans, and psychological support measures; setting quantitative execution indicators for the rehabilitation management related query results according to the relevant physical indicators of the patient to form the personalized rehabilitation management plan.

6. The personalized rehabilitation management method for diabetic patients based on knowledge graph according to claim 5, characterized in that: Patient representation status is defined as: Among them, the symbol In turn, they represent fasting blood sugar level, postprandial blood sugar level, body mass index, heart rate, and electrocardiogram results; and the execution action is defined as: Among them, the symbol They represent the correction amounts for dietary management suggestions, monitoring methods, nursing measures, exercise plans, and psychological support measures respectively; the reward function is defined as: in , , , , Respectively represent the differences between the fasting blood glucose level, postprandial blood glucose level, body mass index, heart rate and electrocardiogram results and the ideal values; to Represents the weight of each difference value in the reward function.

7. The personalized rehabilitation management method for diabetic patients based on knowledge graph according to claim 6, characterized in that: During the execution of the PPO algorithm, the strategy network is initialized Network parameters , and initialize the value network Network parameters ; According to the time step , collect trajectories of execution actions and representation states , and then by calculating the advantage function: The advantage function represents the time step Take action on The degree of advantage relative to the average level is is the discount factor, is the time step Instant rewards, Status Estimation of the value function under PPO; Update the network parameters of the policy network using the PPO cut objective function : Among them, the function Represents the time step The expected value on Indicates status The current network parameters are used Select Execute Action The probability of Indicates status Use the old network parameters Select the action to perform The probability of That is, the shear function, which means The ratio is limited to In the range of is a hyperparameter that controls the magnitude of policy updates; by maximizing the expected clipping objective function , to achieve optimal network parameters .

8. The method for personalized rehabilitation management of diabetic patients based on knowledge graph according to claim 7, characterized in that: Through the value function loss , Update the value network Network parameters ;function Represents the time step The expected value on Indicates status The network parameters are adopted The state value obtained, is the time step The return value; minimizing the value function loss makes the gap between the state value estimated by the value network and the actual return as small as possible.

9. The personalized rehabilitation management method for diabetic patients based on knowledge graph according to claim 8, characterized in that: During the execution of the PPO algorithm, the updated policy network , it is possible to obtain the correction amount defined by the execution action according to the representation state of the patient's relevant physical indicators, and then to correct the execution indicators of the personalized rehabilitation management plan to provide the patient with a corrected personalized rehabilitation management plan.

10. A personalized rehabilitation management system for diabetic patients based on knowledge graph, characterized in that: include: A central database is used to obtain relevant physical indicators of diabetic patients, and pre-process and store the relevant physical indicators of diabetic patients; the relevant physical indicators include: fasting blood glucose level (FBG), postprandial blood glucose level (PBG), glycosylated hemoglobin (HbA1c), glucose level, sweat pH value, microbial population, blood pressure, heart rate, electrocardiogram (EGG), weight value, body mass index (BMI), exercise calorie consumption value, blood ketone body level, serum creatinine, urine microalbumin, fundus examination record, foot examination record, neuropathy examination record; A knowledge graph construction unit is used to convert medical literature, expert advice and clinical guideline data resources on diabetes rehabilitation into structured entity and relationship triples to build a knowledge graph knowledge base related to diabetes rehabilitation management; A personalized rehabilitation management program generation unit is used to query the knowledge graph knowledge base related to diabetes rehabilitation management and generate a personalized rehabilitation management program based on the patient's relevant physical indicators; The optimization execution unit is used to select patient-related physical indicators and define them as the patient's representative state, and define the execution action as the correction amount of the execution indicator of the personalized rehabilitation management plan, define the reward function according to the change of the patient's rehabilitation status, and use the PPO reinforcement learning algorithm to monitor and optimize the execution process of the rehabilitation plan.

Images