Method and system for accurately adjusting insulin based on artificial intelligence

Abstract

The invention discloses an insulin accurate adjustment method and system based on artificial intelligence, and the method comprises the steps: collecting monitoring data and physiological parameter data of a preset target, and carrying out the preprocessing of the monitoring data and physiological parameters; constructing a time sequence blood glucose curve according to the monitoring data and the physiological parameters to obtain a personalized metabolic map, and performing time sequence change analysis according to the personalized metabolic map and the insulin concentration to obtain insulin sensitivity; calculating a target insulin dosage according to the monitoring data and the insulin sensitivity, performing personalized correction on the target insulin dosage according to personalized data to obtain a corrected dosage, and constructing a blood glucose prediction model based on machine learning according to the monitoring data to obtain predicted blood glucose data; and according to the corrected dose and the predicted blood glucose data, constructing a diabetes insulin accurate regulation model, inputting to-be-regulated data into the diabetes insulin accurate regulation model, and outputting a regulation result.

Description

Technical Field

[0001] This invention relates to the field of dose regulation, and more particularly to a method and system for precise insulin regulation based on artificial intelligence. Background Technology

[0002] In current diabetes management, traditional insulin regulation methods rely on fixed dosage regimens or physician experience, which are insufficient to address individual metabolic differences and dynamic physiological changes, resulting in problems such as insufficient precision in blood glucose control and a higher risk of hypoglycemia. On the one hand, blood glucose in diabetic patients is affected by multiple factors such as diet, exercise, sleep, and stress. Traditional methods adjust dosage based on only a single blood glucose indicator, failing to integrate key physiological parameters such as heart rate variability, insulin concentration, and sleep quality. This easily overlooks individual differences in metabolic patterns, leading to frequent occurrences of postprandial hyperglycemia or post-exercise hypoglycemia.

[0003] On the other hand, existing adjustment protocols lack real-time and personalization. Most protocols are based on static data, making it difficult to dynamically respond to changes in a patient's daily physiological rhythms (such as diurnal fluctuations in insulin sensitivity), and they do not fully utilize multi-source monitoring data to build individualized metabolic models, thus failing to accurately match the patient's real-time blood glucose regulation needs. Furthermore, manual adjustments rely on patient self-monitoring and medical intervention, resulting in response delays and making it difficult to achieve continuous and precise dose optimization, thereby impacting the long-term management of diabetes.

[0004] With the development of artificial intelligence and wearable monitoring technology, integrating multi-dimensional physiological data and building dynamic prediction and regulation models have become breakthrough directions. There is an urgent need for an AI-based precision regulation method to solve the pain points of traditional solutions, such as insufficient personalization and poor dynamic adaptability, and to improve the accuracy and safety of insulin regulation. Summary of the Invention

[0005] The purpose of this invention is to provide a method for precise regulation of insulin based on artificial intelligence.

[0006] To achieve the above objectives, the present invention is implemented according to the following technical solution: This invention includes the following steps: Collect monitoring data and physiological parameter data for preset targets, and preprocess the monitoring data and physiological parameters; the monitoring data includes blood glucose data, dietary intake tracking data, and insulin concentration; the physiological parameters include heart rate variability, physical activity level, sleep quality data, and stress state indicators; A personalized metabolic profile is obtained by constructing a time-series blood glucose curve based on the monitoring data and the physiological parameters, and insulin sensitivity is obtained by analyzing the time-series changes based on the personalized metabolic profile and the insulin concentration. The target insulin dose is calculated based on the monitoring data and the insulin sensitivity. The target insulin dose is then personalized and corrected based on the personalized data to obtain the corrected dose. A machine learning-based blood glucose prediction model is then constructed based on the monitoring data to obtain predicted blood glucose data. A precise insulin regulation model for diabetes is constructed based on the corrected dose and the predicted blood glucose data. The data to be regulated is input into the precise insulin regulation model for diabetes, and the regulation result is output.

[0007] Furthermore, the method for obtaining the personalized metabolic profile includes: A continuous blood glucose curve is plotted with time on the horizontal axis and blood glucose concentration on the vertical axis based on blood glucose data. The target blood glucose range, postprandial hyperglycemic events, hypoglycemic risk periods, and insulin action windows are marked. Blood glucose fluctuation indicators are calculated and compared with reference values ​​for the same age group to obtain comparative values. The blood glucose fluctuation indicators include blood glucose fluctuation amplitude, the proportion of target blood glucose time, and average blood glucose. Feature engineering was used to extract blood glucose-derived indicators and physiological parameter-derived indicators. Blood glucose-derived indicators included postprandial blood glucose peak, blood glucose decline rate, and nocturnal blood glucose fluctuations. Physiological parameter-derived indicators included activity intensity index, sleep quality score, and heart rate variability index. Metabolic features were also extracted, including blood glucose fluctuation amplitude, dietary sensitivity, activity regulation effect, sleep stability, and stress state. Establish a regression model between carbohydrate intake and postprandial blood glucose peak, calculate the individual carbohydrate coefficient, compare the individual carbohydrate coefficient with the reference value of the same age group, and assess the individual dietary sensitivity. At the same carbohydrate intake, high glycemic index foods resulted in an average increase in blood glucose peak and an earlier peak time compared to low glycemic index foods. The mitigating effect of high protein diets on the rate of blood glucose decline was analyzed to obtain the reduction in the rate of postprandial blood glucose decline. The dynamic correlation between activity intensity and blood glucose changes and the optimization of activity timing were used to obtain the regulatory patterns of exercise metabolism. The dynamic correlation between sleep structure and nighttime blood glucose stability and heart rate variability and insulin sensitivity were used to obtain the characteristics of neuroendocrine regulation. Personalized metabolic profiles were constructed using time-series fusion plots and metabolic feature heatmaps. The time-series fusion plots displayed blood glucose curves, dietary intake, activity intensity, and heart rate variability on a single axis, with key correlation nodes marked. The metabolic feature heatmaps used time as the horizontal axis and blood glucose fluctuation amplitude, dietary sensitivity, activity regulation effect, sleep stability, and stress state as the vertical axis to indicate the intensity of metabolic features, forming an individual metabolic fingerprint. Core metabolic tags are generated based on personalized metabolic profiles, including postprandial hyperresponsiveness, susceptibility to post-exercise hypoglycemia, and sleep quality sensitivity.

[0008] Furthermore, the method for obtaining said insulin sensitivity includes: Based on the different physiological states identified in the personalized metabolic profile, it is divided into multiple analysis segments, including postprandial segments, nocturnal fasting segments, post-exercise segments, and stress period segments; Acquire fasting blood glucose concentration, fasting insulin concentration, cortisol, melatonin, sleep structure, body temperature rhythm, continuous glucose concentration, glycemic index, physical activity and stress indicators; obtain baseline sensitivity based on fasting blood glucose concentration and fasting insulin concentration; obtain circadian rhythm factors based on cortisol, melatonin, sleep structure and body temperature rhythm; and obtain real-time adjustment factors based on continuous glucose concentration, glycemic index, physical activity and stress indicators. Calculate insulin sensitivity: in Based on sensitivity, Let t be the circadian rhythm factor at time t. For the real-time adjustment factor at time t, Let be the insulin sensitivity at time t. This refers to fasting blood glucose concentration. This refers to the fasting insulin concentration. For cortisol rhythm terms, For sleep structure items, For body temperature rhythm, For blood glucose fluctuations, The glycemic index of food. This is a physical activity item. For stress state items, , , These are the weighting coefficients for each component of the circadian rhythm factor. , , , These are the weight coefficients for each component of the real-time adjustment factor.

[0009] Further, the method for calculating the target insulin dose includes: Based on monitoring data, dietary records, and physiological parameter monitoring, the following parameters are obtained: nominal carbohydrate content, cooking correction factor, glycemic production correction factor, initial insulin-carbohydrate ratio, circadian rhythm factor, target blood glucose concentration, dynamic correction factor, blood glucose trend factor, insulin type factor, and absorption rate factor. Calculate the target insulin dose: in Let t be the target insulin dose. Let be the dynamic carbohydrate ratio at time t. For time t, a multi-dimensional correction factor. To trend the blood glucose correction dose, For the safety boundary coefficient, The dynamic active insulin dose at time t. The initial insulin-to-carbohydrate ratio, Let be the insulin sensitivity at time t. This is the physiological rhythm coefficient. This refers to the nominal amount of carbohydrates. For cooking correction factor, This is the blood glucose generation correction factor. For the target blood glucose concentration, Let be the blood glucose concentration at time t. Let be the dynamic correction factor at time t. This is the blood glucose trend coefficient. This refers to the insulin injection dose over the past 8 hours. Insulin type coefficient, Let be the absorption rate coefficient at time t.

[0010] Further, the method for obtaining the corrected dose includes: A dual-mechanism correction model for global exploration and local mining is constructed based on quantum particle swarm optimization and Harris Eagle algorithm. Chaotic mapping is used for quantum particle swarm initialization, probability density localization and multi-strategy predation are used as search mechanisms, weights are dynamically adjusted based on blood glucose fluctuation amplitude for adaptive adjustment, and a penalty function constraint processing mechanism is introduced. Core parameters are obtained based on personalized metabolic profiles. These core parameters include metabolic core parameters, physiological regulatory parameters, dynamic state parameters, safety constraint parameters, insulin sensitivity, carbohydrate intake, heart rate variability, cortisol level, physical activity level, sleep efficiency, history of hypoglycemia, and liver and kidney function. Among these, metabolic core parameters, physiological regulatory parameters, dynamic state parameters, and safety constraint parameters are primary influencing factors. The weights of the core parameters are determined using the analytic hierarchy process (AHP). Based on precise blood glucose control and safety constraints, the objective function is constructed, and its expression is: in This refers to the blood glucose concentration in the next phase after insulin injection. For the target blood glucose concentration, This is the current insulin dose. This is the previous insulin dose. For low blood sugar risk factor, , , These are the proportional coefficients for blood glucose control, dose stability, and safety constraints, respectively. The target insulin dose and personalized data are input into a dual-mechanism correction model for global exploration and local mining to obtain the target insulin dose. Multiple candidate doses are randomly generated, and the model simulates a Harris eagle searching for prey. Five optimal solutions are retained, and the optimal solutions are updated using quantum behavior. The process is repeated multiple times until convergence, and the corrected dose is output.

[0011] Furthermore, the method for obtaining the predicted blood glucose data includes: We acquired blood glucose data, insulin data, carbohydrate intake and physiological parameters, and constructed a blood glucose prediction model based on mechanism-data joint drive. The blood glucose prediction model adopts a three-channel parallel LSTM architecture, in which the three channels respectively process insulin infusion data, carbohydrate intake data and historical blood glucose data. The glucose and insulin metabolic kinetic equations are embedded as physical constraints into the loss function of the neural network to construct a hybrid loss function, the expression of which is: in For a mixed loss function, For data loss items, For physical loss items, Weights for data loss The weights for physical loss, The equation for insulin absorption rate is as follows: Let be the intermediate variable of insulin absorption at time t. The half-life of insulin, Let t be the rate of exogenous insulin delivery. This represents the effective volume for the initial distribution of insulin in the subcutaneous tissue. Optimize the hyperparameters of the blood glucose prediction model by inputting personalized data into the model until the root mean square error between the actual blood glucose data and the predicted blood glucose data is less than 0.328, at which point the predicted blood glucose data is output.

[0012] Furthermore, the method for constructing the aforementioned precise insulin regulation model for diabetes includes: A precise insulin regulation model for diabetes is constructed, integrating perception, decision-making, and execution. The perception layer continuously collects multiple parameters such as blood glucose concentration and heart rate variability. The decision-making layer solves for the optimal insulin dose based on a hybrid optimization algorithm combining particle swarm optimization and Newton's iteration, which integrates global optimization of particle swarm optimization with local convergence features of Newton's iteration. The execution layer delivers the precise dose through an intelligent insulin pump. A multi-level regulation strategy is introduced for optimization, which includes a first-level coarse adjustment and a second-level fine adjustment. The first-level coarse adjustment calculates the basal dose based on the total daily carbohydrate intake; the second-level fine adjustment introduces the metabolic resistance coefficient and corrects the individualized dose by adjusting the metabolic resistance coefficient. Set dynamic safety boundaries and blood glucose control thresholds, and automatically activate the protection mechanism when blood glucose is predicted to enter the risk range. The objective function for constructing a precise insulin regulation model for diabetes is expressed as follows: in The predicted blood glucose concentration at time t. For the target blood glucose concentration, The penalty coefficient is... This is a function indicating hypoglycemia. This is the upper limit of the observation time. The objective function for a precise insulin regulation model for diabetes; Each set of parameters is particle-encoded, and the fitness function is taken as the average blood glucose fluctuation amplitude being less than the blood glucose fluctuation threshold. Iterative updates are performed using nonlinear inertial weights. One set of parameters includes basal insulin dose and metabolic resistance coefficient. The optimal solution of the particle swarm optimization algorithm is optimized a second time, and the error is corrected to less than 0.051, resulting in an optimized precise insulin regulation model for diabetes.

[0013] Secondly, an artificial intelligence-based precise insulin regulation system includes: Data acquisition and processing module: used to collect monitoring data and physiological parameter data of preset targets, and to preprocess the monitoring data and physiological parameters; the monitoring data includes blood glucose data, dietary intake tracking data, and insulin concentration; the physiological parameters include heart rate variability, physical activity level, sleep quality data, and stress state indicators; Metabolic mapping and analysis module: used to construct time-series blood glucose curves based on the monitoring data and physiological parameters to obtain personalized metabolic maps, and to perform time-series change analysis based on the personalized metabolic maps and insulin concentrations to obtain insulin sensitivity; Dosage and blood glucose prediction module: used to calculate the target insulin dose based on the monitoring data and the insulin sensitivity, perform personalized correction on the target insulin dose based on personalized data to obtain the corrected dose, and construct a machine learning-based blood glucose prediction model based on the monitoring data to obtain predicted blood glucose data; Model building and output module: used to build a precise regulation model of diabetes insulin based on the corrected dose and the predicted blood glucose data, input the data to be regulated into the precise regulation model of diabetes insulin, and output the regulation results.

[0014] The beneficial effects of this invention are: This invention is a method and system for precise insulin regulation based on artificial intelligence. Compared with existing technologies, this invention has the following technical advantages: This invention integrates multi-dimensional monitoring data and physiological parameters through preprocessing, obtaining personalized metabolic profiles, obtaining insulin sensitivity, calculating target insulin dose, obtaining corrected dose, obtaining predicted blood glucose data, and model building steps. It constructs personalized metabolic profiles and blood glucose prediction models, and dynamically calculates doses based on insulin sensitivity, reducing the bias caused by single data points in traditional methods. It introduces dynamic safety boundaries and hypoglycemia risk constraints, optimizing doses through a dual-mechanism correction model to reduce risks such as hypoglycemia. It adapts to individual metabolic differences and circadian rhythms, achieving real-time dynamic adjustment, avoiding the limitations of traditional static methods, and improving diabetes management effectiveness. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating the steps of an artificial intelligence-based method for precise insulin regulation according to the present invention. Detailed Implementation

[0016] The present invention will be further described below through specific embodiments. The illustrative embodiments and descriptions herein are used to explain the present invention, but are not intended to limit the present invention.

[0017] The present invention provides a method and system for precise insulin regulation based on artificial intelligence, comprising the following steps: like Figure 1 As shown, this embodiment includes the following steps: Collect monitoring data and physiological parameter data for preset targets, and preprocess the monitoring data and physiological parameters; the monitoring data includes blood glucose data, dietary intake tracking data, and insulin concentration; the physiological parameters include heart rate variability, physical activity level, sleep quality data, and stress state indicators; In the actual assessment, a patient was selected as the research subject. The patient was 32 years old, weighed 68 kg, had type 1 diabetes, and was currently being treated with insulin aspart. The monitoring data are as follows: current blood glucose concentration 8.9 mmol / L, target blood glucose concentration 6.0 mmol / L, planned carbohydrate intake 65g, high GI breakfast (white bread + jam), cooking method: baking; Physiological parameters were as follows: fasting blood glucose 5.8 mmol / L, fasting insulin 6.2 mU / L, sleep quality score 82 (good), current activity level at rest, heart rate variability within normal range, and stress indicators within normal levels. A personalized metabolic profile is obtained by constructing a time-series blood glucose curve based on the monitoring data and the physiological parameters, and insulin sensitivity is obtained by analyzing the time-series changes based on the personalized metabolic profile and the insulin concentration. In actual assessments, the personalized metabolic profile showed the following core metabolic markers: postprandial hyperresponsiveness, individual carbohydrate coefficient of 12 g / U (standard range 10-15 g / U), moderate sensitivity to activity regulation, and significant impact on sleep stability. Insulin sensitivity: Basal sensitivity is 0.1646, circadian rhythm factor is 1, real-time adjustment factor is 1.15, and insulin sensitivity is 0.1893; The target insulin dose is calculated based on the monitoring data and the insulin sensitivity. The target insulin dose is then personalized and corrected based on the personalized data to obtain the corrected dose. A machine learning-based blood glucose prediction model is then constructed based on the monitoring data to obtain predicted blood glucose data. In the actual assessment, the effective carbohydrate intake was 92.95g, the dynamic carbohydrate ratio was 2.70g / U, the trend-corrected glucose dose was −2.784U, and the dynamic active insulin dose was 1.44U; the injection record for the past 8 hours: 4U at 6:00 (remaining activity: 4×0.3×0.8=0.96U); the target insulin dose was 28.59U; The target dose was 28.59U, with insulin sensitivity of 0.1893, carbohydrate intake of 65g, normal heart rate variability, 82% sleep efficiency, and no history of hypoglycemia. The corrected dose was 26.8U (a reduction of 6.3%), optimized to avoid the potential risk of hypoglycemia 3 hours after a meal. After inputting the current data and corrected dose: the predicted blood glucose level at 30 minutes is 7.2 mmol / L, at 60 minutes it is 6.8 mmol / L, at 120 minutes it is 6.1 mmol / L, and at 180 minutes it is 5.9 mmol / L; the root mean square error of the blood glucose prediction model is 0.31 mmol / L < 0.328 threshold; A precise insulin regulation model for diabetes is constructed based on the corrected dose and the predicted blood glucose data. The data to be regulated is input into the precise insulin regulation model for diabetes, and the regulation result is output. In actual assessments, the recommended dosage is: a basal dose of 26.8U of insulin aspart, injected 15 minutes before meals; the expected results are: a peak blood glucose level ≤9.0 mmol / L 2 hours after meals, blood glucose levels returning to 5.5-7.0 mmol / L 4 hours after meals, and a hypoglycemic risk probability <2%; Key monitoring period: 3-4 hours after meals; warning threshold: when blood glucose concentration is <4.5mmol / L, the protection mechanism is activated; backup measures: if postprandial activity increases, it is recommended to supplement with 15g of carbohydrates. The actual monitoring results were as follows: pre-meal blood glucose was 8.9 mmol / L, 1 hour post-meal blood glucose was 8.1 mmol / L, 2 hours post-meal blood glucose was 7.3 mmol / L, 3 hours post-meal blood glucose was 6.2 mmol / L, and 4 hours post-meal blood glucose was 5.8 mmol / L.

[0018] In this embodiment, the method for obtaining the personalized metabolic profile includes: A continuous blood glucose curve is plotted with time on the horizontal axis and blood glucose concentration on the vertical axis based on blood glucose data. The target blood glucose range, postprandial hyperglycemic events, hypoglycemic risk periods, and insulin action windows are marked. Blood glucose fluctuation indicators are calculated and compared with reference values ​​for the same age group to obtain comparative values. The blood glucose fluctuation indicators include blood glucose fluctuation amplitude, the proportion of target blood glucose time, and average blood glucose. Feature engineering was used to extract blood glucose-derived indicators and physiological parameter-derived indicators. Blood glucose-derived indicators included postprandial blood glucose peak, blood glucose decline rate, and nocturnal blood glucose fluctuations. Physiological parameter-derived indicators included activity intensity index, sleep quality score, and heart rate variability index. Metabolic features were also extracted, including blood glucose fluctuation amplitude, dietary sensitivity, activity regulation effect, sleep stability, and stress state. Establish a regression model between carbohydrate intake and postprandial blood glucose peak, calculate the individual carbohydrate coefficient, compare the individual carbohydrate coefficient with the reference value of the same age group, and assess the individual dietary sensitivity. At the same carbohydrate intake, high glycemic index foods resulted in an average increase in blood glucose peak and an earlier peak time compared to low glycemic index foods. The mitigating effect of high protein diets on the rate of blood glucose decline was analyzed to obtain the reduction in the rate of postprandial blood glucose decline. The dynamic correlation between activity intensity and blood glucose changes and the optimization of activity timing were used to obtain the regulatory patterns of exercise metabolism. The dynamic correlation between sleep structure and nighttime blood glucose stability and heart rate variability and insulin sensitivity were used to obtain the characteristics of neuroendocrine regulation. Personalized metabolic profiles were constructed using time-series fusion plots and metabolic feature heatmaps. The time-series fusion plots displayed blood glucose curves, dietary intake, activity intensity, and heart rate variability on a single axis, with key correlation nodes marked. The metabolic feature heatmaps used time as the horizontal axis and blood glucose fluctuation amplitude, dietary sensitivity, activity regulation effect, sleep stability, and stress state as the vertical axis to indicate the intensity of metabolic features, forming an individual metabolic fingerprint. Core metabolic tags are generated based on personalized metabolic profiles, including postprandial hyperresponsiveness, susceptibility to post-exercise hypoglycemia, and sleep quality sensitivity.

[0019] In this embodiment, the method for obtaining the insulin sensitivity includes: Based on the different physiological states identified in the personalized metabolic profile, it is divided into multiple analysis segments, including postprandial segments, nocturnal fasting segments, post-exercise segments, and stress period segments; Acquire fasting blood glucose concentration, fasting insulin concentration, cortisol, melatonin, sleep structure, body temperature rhythm, continuous glucose concentration, glycemic index, physical activity and stress indicators; obtain baseline sensitivity based on fasting blood glucose concentration and fasting insulin concentration; obtain circadian rhythm factors based on cortisol, melatonin, sleep structure and body temperature rhythm; and obtain real-time adjustment factors based on continuous glucose concentration, glycemic index, physical activity and stress indicators. Calculate insulin sensitivity: in Based on sensitivity, Let t be the circadian rhythm factor at time t. For the real-time adjustment factor at time t, Let be the insulin sensitivity at time t. This refers to fasting blood glucose concentration. This refers to the fasting insulin concentration. For cortisol rhythm terms, For sleep structure items, For body temperature rhythm, For blood glucose fluctuations, The glycemic index of food. This is a physical activity item. For stress state items, , , These are the weighting coefficients for each component of the circadian rhythm factor. , , , These are the weight coefficients for each component of the real-time adjustment factor.

[0020] In this embodiment, the method for calculating the target insulin dose includes: Based on monitoring data, dietary records, and physiological parameter monitoring, the following parameters are obtained: nominal carbohydrate content, cooking correction factor, glycemic production correction factor, initial insulin-carbohydrate ratio, circadian rhythm factor, target blood glucose concentration, dynamic correction factor, blood glucose trend factor, insulin type factor, and absorption rate factor. Calculate the target insulin dose: in Let t be the target insulin dose. Let be the dynamic carbohydrate ratio at time t. For time t, a multi-dimensional correction factor. To trend the blood glucose correction dose, For the safety boundary coefficient, The dynamic active insulin dose at time t. The initial insulin-to-carbohydrate ratio, Let be the insulin sensitivity at time t. This is the physiological rhythm coefficient. This refers to the nominal amount of carbohydrates. For cooking correction factor, This is the blood glucose generation correction factor. For the target blood glucose concentration, Let be the blood glucose concentration at time t. Let be the dynamic correction factor at time t. This is the blood glucose trend coefficient. This refers to the insulin injection dose over the past 8 hours. Insulin type coefficient, Let be the absorption rate coefficient at time t.

[0021] In this embodiment, the method for obtaining the corrected dose includes: A dual-mechanism correction model for global exploration and local mining is constructed based on quantum particle swarm optimization and Harris Eagle algorithm. Chaotic mapping is used for quantum particle swarm initialization, probability density localization and multi-strategy predation are used as search mechanisms, weights are dynamically adjusted based on blood glucose fluctuation amplitude for adaptive adjustment, and a penalty function constraint processing mechanism is introduced. Core parameters are obtained based on personalized metabolic profiles. These core parameters include metabolic core parameters, physiological regulatory parameters, dynamic state parameters, safety constraint parameters, insulin sensitivity, carbohydrate intake, heart rate variability, cortisol level, physical activity level, sleep efficiency, history of hypoglycemia, and liver and kidney function. Among these, metabolic core parameters, physiological regulatory parameters, dynamic state parameters, and safety constraint parameters are primary influencing factors. The weights of the core parameters are determined using the analytic hierarchy process (AHP). Based on precise blood glucose control and safety constraints, the objective function is constructed, and its expression is: in This refers to the blood glucose concentration in the next phase after insulin injection. For the target blood glucose concentration, This is the current insulin dose. This is the previous insulin dose. For low blood sugar risk factor, , , These are the proportional coefficients for blood glucose control, dose stability, and safety constraints, respectively. The target insulin dose and personalized data are input into a dual-mechanism correction model for global exploration and local mining to obtain the target insulin dose. Multiple candidate doses are randomly generated, and the model simulates a Harris eagle searching for prey. Five optimal solutions are retained, and the optimal solutions are updated using quantum behavior. The process is repeated multiple times until convergence, and the corrected dose is output.

[0022] In this embodiment, the method for obtaining the predicted blood glucose data includes: We acquired blood glucose data, insulin data, carbohydrate intake and physiological parameters, and constructed a blood glucose prediction model based on mechanism-data joint drive. The blood glucose prediction model adopts a three-channel parallel LSTM architecture, in which the three channels respectively process insulin infusion data, carbohydrate intake data and historical blood glucose data. The glucose and insulin metabolic kinetic equations are embedded as physical constraints into the loss function of the neural network to construct a hybrid loss function, the expression of which is: in For a mixed loss function, For data loss items, For physical loss items, Weights for data loss The weights for physical loss, The equation for insulin absorption rate is as follows: Let be the intermediate variable of insulin absorption at time t. The half-life of insulin, Let t be the rate of exogenous insulin delivery. This represents the effective volume for the initial distribution of insulin in the subcutaneous tissue. Optimize the hyperparameters of the blood glucose prediction model by inputting personalized data into the model until the root mean square error between the actual blood glucose data and the predicted blood glucose data is less than 0.328, at which point the predicted blood glucose data is output.

[0023] In this embodiment, the method for constructing the precise insulin regulation model for diabetes includes: A precise insulin regulation model for diabetes is constructed, integrating perception, decision-making, and execution. The perception layer continuously collects multiple parameters such as blood glucose concentration and heart rate variability. The decision-making layer solves for the optimal insulin dose based on a hybrid optimization algorithm combining particle swarm optimization and Newton's iteration, which integrates global optimization of particle swarm optimization with local convergence features of Newton's iteration. The execution layer delivers the precise dose through an intelligent insulin pump. A multi-level regulation strategy is introduced for optimization, which includes a first-level coarse adjustment and a second-level fine adjustment. The first-level coarse adjustment calculates the basal dose based on the total daily carbohydrate intake; the second-level fine adjustment introduces the metabolic resistance coefficient and corrects the individualized dose by adjusting the metabolic resistance coefficient. Set dynamic safety boundaries and blood glucose control thresholds, and automatically activate the protection mechanism when blood glucose is predicted to enter the risk range. The objective function for constructing a precise insulin regulation model for diabetes is expressed as follows: in The predicted blood glucose concentration at time t. For the target blood glucose concentration, The penalty coefficient is... This is a function indicating hypoglycemia. This is the upper limit of the observation time. The objective function for a precise insulin regulation model for diabetes; Each set of parameters is particle-encoded, and the fitness function is taken as the average blood glucose fluctuation amplitude being less than the blood glucose fluctuation threshold. Iterative updates are performed using nonlinear inertial weights. One set of parameters includes basal insulin dose and metabolic resistance coefficient. The optimal solution of the particle swarm optimization algorithm is optimized a second time, and the error is corrected to be less than 0.051, outputting the optimized diabetes insulin precision regulation model; In actual assessments, the blood glucose fluctuation threshold is 3.0 mmol / L, based on nonlinear inertial weights using the tansig function. Secondly, an artificial intelligence-based precise insulin regulation system includes: Data acquisition and processing module: used to collect monitoring data and physiological parameter data of preset targets, and to preprocess the monitoring data and physiological parameters; the monitoring data includes blood glucose data, dietary intake tracking data, and insulin concentration; the physiological parameters include heart rate variability, physical activity level, sleep quality data, and stress state indicators; Metabolic mapping and analysis module: used to construct time-series blood glucose curves based on the monitoring data and physiological parameters to obtain personalized metabolic maps, and to perform time-series change analysis based on the personalized metabolic maps and insulin concentrations to obtain insulin sensitivity; Dosage and blood glucose prediction module: used to calculate the target insulin dose based on the monitoring data and the insulin sensitivity, perform personalized correction on the target insulin dose based on personalized data to obtain the corrected dose, and construct a machine learning-based blood glucose prediction model based on the monitoring data to obtain predicted blood glucose data; Model building and output module: used to build a precise regulation model of diabetes insulin based on the corrected dose and the predicted blood glucose data, input the data to be regulated into the precise regulation model of diabetes insulin, and output the regulation results.

[0024] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for precise insulin regulation based on artificial intelligence, characterized in that, Includes the following steps: Collect monitoring data and physiological parameter data for preset targets, and preprocess the monitoring data and physiological parameters; the monitoring data includes blood glucose data, dietary intake tracking data, and insulin concentration; the physiological parameters include heart rate variability, physical activity level, sleep quality data, and stress state indicators; A personalized metabolic profile is obtained by constructing a time-series blood glucose curve based on the monitoring data and the physiological parameters, and insulin sensitivity is obtained by analyzing the time-series changes based on the personalized metabolic profile and the insulin concentration. The target insulin dose is calculated based on the monitoring data and the insulin sensitivity. The target insulin dose is then personalized and corrected based on the personalized data to obtain the corrected dose. A machine learning-based blood glucose prediction model is then constructed based on the monitoring data to obtain predicted blood glucose data. A precise insulin regulation model for diabetes is constructed based on the corrected dose and the predicted blood glucose data. The data to be regulated is input into the precise insulin regulation model for diabetes, and the regulation result is output.

2. The method for precise insulin regulation based on artificial intelligence according to claim 1, characterized in that, The method for obtaining the personalized metabolic profile includes: A continuous blood glucose curve is plotted with time on the horizontal axis and blood glucose concentration on the vertical axis based on blood glucose data. The target blood glucose range, postprandial hyperglycemic events, hypoglycemic risk periods, and insulin action windows are marked. Blood glucose fluctuation indicators are calculated and compared with reference values ​​for the same age group to obtain comparative values. The blood glucose fluctuation indicators include blood glucose fluctuation amplitude, the proportion of target blood glucose time, and average blood glucose. Feature engineering was used to extract blood glucose-derived indicators and physiological parameter-derived indicators. Blood glucose-derived indicators included postprandial blood glucose peak, blood glucose decline rate, and nocturnal blood glucose fluctuations. Physiological parameter-derived indicators included activity intensity index, sleep quality score, and heart rate variability index. Metabolic features were also extracted, including blood glucose fluctuation amplitude, dietary sensitivity, activity regulation effect, sleep stability, and stress state. Establish a regression model between carbohydrate intake and postprandial blood glucose peak, calculate the individual carbohydrate coefficient, compare the individual carbohydrate coefficient with the reference value of the same age group, and assess the individual dietary sensitivity. At the same carbohydrate intake, high glycemic index foods resulted in an average increase in blood glucose peak and an earlier peak time compared to low glycemic index foods. The mitigating effect of high protein diets on the rate of blood glucose decline was analyzed to obtain the reduction in the rate of postprandial blood glucose decline. The dynamic correlation between activity intensity and blood glucose changes and the optimization of activity timing were used to obtain the regulatory patterns of exercise metabolism. The dynamic correlation between sleep structure and nighttime blood glucose stability and heart rate variability and insulin sensitivity were used to obtain the characteristics of neuroendocrine regulation. Personalized metabolic profiles were constructed using time-series fusion plots and metabolic feature heatmaps. The time-series fusion plots displayed blood glucose curves, dietary intake, activity intensity, and heart rate variability on a single axis, with key correlation nodes marked. The metabolic feature heatmaps used time as the horizontal axis and blood glucose fluctuation amplitude, dietary sensitivity, activity regulation effect, sleep stability, and stress state as the vertical axis to indicate the intensity of metabolic features, forming an individual metabolic fingerprint. Core metabolic tags are generated based on personalized metabolic profiles, including postprandial hyperresponsiveness, susceptibility to post-exercise hypoglycemia, and sleep quality sensitivity.

3. The method for precise insulin regulation based on artificial intelligence according to claim 1, characterized in that, The method for obtaining said insulin sensitivity includes: Based on the different physiological states identified in the personalized metabolic profile, it is divided into multiple analysis segments, including postprandial segments, nocturnal fasting segments, post-exercise segments, and stress period segments; The system acquires fasting blood glucose concentration, fasting insulin concentration, cortisol, melatonin, sleep structure, body temperature rhythm, continuous glucose concentration, glycemic index, physical activity, and stress indicators. It obtains baseline sensitivity based on fasting blood glucose concentration and fasting insulin concentration, circadian rhythm factors based on cortisol, melatonin, sleep structure, and body temperature rhythm, and real-time adjustment factors based on continuous glucose concentration, glycemic index, physical activity, and stress indicators. Calculate insulin sensitivity: in Based on sensitivity, Let t be the circadian rhythm factor at time t. For the real-time adjustment factor at time t, Let be the insulin sensitivity at time t. This refers to fasting blood glucose concentration. This refers to the fasting insulin concentration. For cortisol rhythm terms, For sleep structure items, For body temperature rhythm, For blood glucose fluctuations, The glycemic index of food. This is a physical activity item. For stress state items, , , These are the weighting coefficients for each component of the circadian rhythm factor. , , , These are the weight coefficients for each component of the real-time adjustment factor.

4. The method for precise insulin regulation based on artificial intelligence according to claim 1, characterized in that, A method for calculating the target insulin dose includes: Based on monitoring data, dietary records, and physiological parameter monitoring, the following parameters are obtained: nominal carbohydrate content, cooking correction factor, glycemic production correction factor, initial insulin-carbohydrate ratio, circadian rhythm factor, target blood glucose concentration, dynamic correction factor, blood glucose trend factor, insulin type factor, and absorption rate factor. Calculate the target insulin dose: in Let t be the target insulin dose. The dynamic carbohydrate ratio at time t. For time t, a multi-dimensional correction factor. To trend the blood glucose correction dose, For the safety boundary coefficient, The dynamic active insulin dose at time t. The initial insulin-to-carbohydrate ratio, Let be the insulin sensitivity at time t. This is the physiological rhythm coefficient. This refers to the nominal amount of carbohydrates. For cooking correction factor, This is the blood glucose generation correction factor. For the target blood glucose concentration, Let be the blood glucose concentration at time t. Let be the dynamic correction factor at time t. This is the blood glucose trend coefficient. This refers to the insulin injection dose over the past 8 hours. Insulin type coefficient, Let be the absorption rate coefficient at time t.

5. The method for precise insulin regulation based on artificial intelligence according to claim 1, characterized in that, A method for obtaining the corrected dose includes: A dual-mechanism correction model for global exploration and local mining is constructed based on quantum particle swarm optimization and Harris Eagle algorithm. Chaotic mapping is used for quantum particle swarm initialization, probability density localization and multi-strategy predation are used as search mechanisms, weights are dynamically adjusted based on blood glucose fluctuation amplitude for adaptive adjustment, and a penalty function constraint processing mechanism is introduced. Core parameters are obtained based on personalized metabolic profiles. These core parameters include metabolic core parameters, physiological regulatory parameters, dynamic state parameters, safety constraint parameters, insulin sensitivity, carbohydrate intake, heart rate variability, cortisol level, physical activity level, sleep efficiency, history of hypoglycemia, and liver and kidney function. Among these, metabolic core parameters, physiological regulatory parameters, dynamic state parameters, and safety constraint parameters are primary influencing factors. The weights of the core parameters are determined using the analytic hierarchy process (AHP). Based on precise blood glucose control and safety constraints, the objective function is constructed, and its expression is: in This refers to the blood glucose concentration in the next phase after insulin injection. For the target blood glucose concentration, This is the current insulin dose. This is the previous insulin dose. For low blood sugar risk factor, , , These are the proportional coefficients for blood glucose control, dose stability, and safety constraints, respectively. The target insulin dose and personalized data are input into a dual-mechanism correction model for global exploration and local mining to obtain the target insulin dose. Multiple candidate doses are randomly generated, and the model simulates a Harris eagle searching for prey. Five optimal solutions are retained, and the optimal solutions are updated using quantum behavior. The process is repeated multiple times until convergence, and the corrected dose is output.

6. The method for precise insulin regulation based on artificial intelligence according to claim 1, characterized in that, The method for obtaining the predicted blood glucose data includes: We acquired blood glucose data, insulin data, carbohydrate intake and physiological parameters, and constructed a blood glucose prediction model based on mechanism-data joint drive. The blood glucose prediction model adopts a three-channel parallel LSTM architecture, in which the three channels respectively process insulin infusion data, carbohydrate intake data and historical blood glucose data. The glucose and insulin metabolic kinetic equations are embedded as physical constraints into the loss function of the neural network to construct a hybrid loss function, the expression of which is: in For a mixed loss function, For data loss items, For physical loss items, Weights for data loss The weights for physical loss, The equation for insulin absorption rate is as follows: Let be the intermediate variable of insulin absorption at time t. The half-life of insulin, Let t be the rate of exogenous insulin delivery. This represents the effective volume for the initial distribution of insulin in the subcutaneous tissue. Optimize the hyperparameters of the blood glucose prediction model by inputting personalized data into the model until the root mean square error between the actual blood glucose data and the predicted blood glucose data is less than 0.328, at which point the predicted blood glucose data is output.

7. The method for precise insulin regulation based on artificial intelligence according to claim 1, characterized in that, The method for constructing the precise insulin regulation model for diabetes includes: A precise insulin regulation model for diabetes is constructed, integrating perception, decision-making, and execution. The perception layer continuously collects multiple parameters such as blood glucose concentration and heart rate variability. The decision-making layer solves for the optimal insulin dose based on a hybrid optimization algorithm combining particle swarm optimization and Newton's iteration, which integrates global optimization of particle swarm optimization with local convergence features of Newton's iteration. The execution layer delivers the precise dose through an intelligent insulin pump. A multi-level regulation strategy is introduced for optimization, which includes a first-level coarse adjustment and a second-level fine adjustment. The first-level coarse adjustment calculates the basal dose based on the total daily carbohydrate intake; the second-level fine adjustment introduces the metabolic resistance coefficient and corrects the individualized dose by adjusting the metabolic resistance coefficient. Set dynamic safety boundaries and blood glucose control thresholds, and automatically activate the protection mechanism when blood glucose is predicted to enter the risk range. The objective function for constructing a precise insulin regulation model for diabetes is expressed as follows: in The predicted blood glucose concentration at time t. For the target blood glucose concentration, The penalty coefficient is... This is a function indicating hypoglycemia. This is the upper limit of the observation time. The objective function for a precise insulin regulation model for diabetes; Each set of parameters is particle-encoded, and the fitness function is taken as the average blood glucose fluctuation amplitude being less than the blood glucose fluctuation threshold. Iterative updates are performed using nonlinear inertial weights. One set of parameters includes basal insulin dose and metabolic resistance coefficient. The optimal solution of the particle swarm optimization algorithm is optimized a second time, and the error is corrected to less than 0.051, resulting in an optimized precise insulin regulation model for diabetes.

8. An artificial intelligence-based insulin precision regulation system for performing the method according to any one of claims 1-7, characterized in that, include: Data acquisition and processing module: used to acquire monitoring data and physiological parameter data of preset targets, and to preprocess the monitoring data and physiological parameters; The monitoring data includes blood glucose data, dietary intake tracking data, and insulin concentration; the physiological parameters include heart rate variability, physical activity level, sleep quality data, and stress state indicators. Metabolic mapping and analysis module: used to construct time-series blood glucose curves based on the monitoring data and physiological parameters to obtain personalized metabolic maps, and to perform time-series change analysis based on the personalized metabolic maps and insulin concentrations to obtain insulin sensitivity; Dosage and blood glucose prediction module: used to calculate the target insulin dose based on the monitoring data and the insulin sensitivity, perform personalized correction on the target insulin dose based on personalized data to obtain the corrected dose, and construct a machine learning-based blood glucose prediction model based on the monitoring data to obtain predicted blood glucose data; Model building and output module: used to build a precise regulation model of diabetes insulin based on the corrected dose and the predicted blood glucose data, input the data to be regulated into the precise regulation model of diabetes insulin, and output the regulation results.

Images