A digital twin-based method for monitoring the health status of insulin pump systems

Abstract

This invention pertains to the field of insulin pump group health status monitoring, specifically relating to a digital twin-based method for monitoring the health status of insulin pump groups. It employs a detection system comprising a dataset unit, a mechanism model set unit, a digital twin model integration unit, a digital twin model consistency evaluation unit, and an insulin pump group fault monitoring unit. Through a hierarchical digital twin construction framework, a high-fidelity blood glucose regulation digital twin model is constructed based on a virtual model mapping the insulin pump group entity to the human blood glucose mechanism and historical monitoring data. This high-fidelity digital twin model, combined with real-time blood glucose monitoring data, monitors the health status of the insulin pump group.

Description

Technical Field

[0001] This invention belongs to the field of insulin pump group health status monitoring, specifically relating to a method for monitoring the health status of insulin pump groups based on digital twins. Background Technology

[0002] Diabetes mellitus is a disease caused by an absolute or relative deficiency of insulin secretion and decreased sensitivity of target tissue cells to insulin, resulting in a series of metabolic disorders involving glucose, protein, fat, water, and electrolytes. Insulin pumps, as the most advanced insulin infusion devices, can continuously deliver insulin subcutaneously to the user at the required dose, maintaining stable blood glucose levels throughout the day to control diabetes. However, malfunctions in insulin pump systems can lead to serious consequences such as ketoacidosis. An insulin pump consists of a pump, a small syringe, and connected infusion tubing. Common malfunctions include pump mechanical jamming, and leaks and blockages in the tubing and needle. Currently, insulin pumps primarily rely on chamber pressure signals and infusion control signals for health status monitoring, but due to the low insulin infusion rate and slow pressure changes, this method is not very effective.

[0003] Digital twins are a technology that combines a digital model of a physical system with real-time data. Through an integrated simulation process involving multiple disciplines, physical quantities, scales, and probabilities, a mapping is completed in virtual space. The interaction and iteration between the virtual and physical worlds monitor the physical world, reflecting the entire lifecycle of the corresponding physical equipment. In recent years, digital twin technology has been applied to an increasing number of fields, including aerospace, manufacturing, healthcare, urban development, and energy. Particularly in fault diagnosis, digital twin technology can monitor and analyze the health status of equipment in dynamic environments in real time, helping to improve the efficiency and accuracy of fault diagnosis. Summary of the Invention

[0004] The purpose of this invention is to provide a method for monitoring the health status of insulin pump sets based on digital twins. This method constructs a high-fidelity digital twin model of blood glucose regulation based on a hierarchical digital twin construction framework, a virtual model that maps the insulin pump set entity to the human blood glucose mechanism, and historical monitoring data. By combining the high-fidelity digital twin model with real-time blood glucose monitoring data, the health status of the insulin pump set can be monitored.

[0005] A method for monitoring the health status of insulin pump sets based on digital twins employs a detection system comprising a dataset unit, a mechanism model set unit, a digital twin model integration unit, a digital twin model consistency evaluation unit, and an insulin pump set fault monitoring unit. The aforementioned dataset unit collects physiological data and stores it in a database table; The digital twin model integration unit is based on a unified interface integration mechanism model. It uses blood glucose monitoring data of insulin pump groups in a healthy state in the database and Bayesian estimation to obtain dynamic parameters in the integration model, thereby constructing a digital twin model of insulin pump groups in a healthy state. The aforementioned digital twin model consistency evaluation unit performs consistency evaluation on the mechanism model, then obtains the comprehensive consistency evaluation result of the digital twin model based on the analytic hierarchy process, and adjusts the digital twin model based on the consistency evaluation result to meet the consistency evaluation requirements, thereby obtaining a high-fidelity digital twin model. The insulin pump set fault monitoring unit described above is based on a high-fidelity digital twin model. It dynamically estimates fault parameters by monitoring blood glucose data in real time and using an unscented Kalman filter algorithm.

[0006] Furthermore, the dataset unit collects physiological data and stores it in a database table.

[0007] Furthermore, the aforementioned mechanism model set unit includes an insulin pump geometric model, an insulin hydrodynamic model, an insulin infusion rule model, and a human blood glucose metabolism model.

[0008] Furthermore, the digital twin model integration unit integrates the above-mentioned mechanism model based on the FMI interface, and accesses the MySQL database in the above-mentioned dataset unit using the ODBC application programming interface; using the human blood glucose monitoring data of the insulin pump group in a healthy state, the dynamic parameters in the integrated model are obtained by Bayesian estimation, and a digital twin model of the insulin pump group in a healthy state is constructed.

[0009] Furthermore, the digital twin model consistency evaluation unit: identifies and measures the feature elements to be tested in the geometric model of the insulin pump assembly, and also measures the feature elements of the insulin pump entity, and calculates the degree of consistency between the two; sets a physical consistency evaluation index Cd(PM) for the insulin fluid dynamics model and the human blood glucose metabolism model; compares the insulin dose output by the insulin pump entity and the insulin fluid dynamics model under the same conditions through simulation comparison experiments, where n1 is the total number of simulation experiments and m1 is the number of simulation experiments in which the error between the output insulin dose of the entity and the virtual model is less than 10%; and calculates the consistency evaluation index R of the insulin pump assembly fluid dynamics model. re11 =m1 / n1; Additionally, the blood glucose values ​​of type 1 diabetes patients wearing insulin pumps and the human blood glucose model established in this paper were compared 15 minutes after the same insulin infusion dose. If the error was within the allowable range, it was considered to have met the consistency requirements; n2 is the total number of simulation experiments, m2 is the number of simulation experiments required for the human blood glucose model to meet the consistency requirements, and the consistency evaluation result of the human blood glucose model is R. rel2 =m² / n²; Cd(PM) = 0.5 × R rel1+0.5×R rel2 The physical consistency assessment result is calculated; actual data is input into the insulin infusion rule model to stimulate the insulin infusion rule model to make a corresponding response, and the response of the insulin infusion rule model is compared with the actual response; if the response error is within the allowable range, the rule model is valid and the rule consistency result Cd(RM)=1; otherwise, the rule model is incorrect or inaccurate and Cd(RM)=0; finally, the overall digital twin model consistency assessment result is obtained through the analytic hierarchy process (AHP), and the digital twin model is iteratively adjusted based on the assessment result until it meets the consistency requirements to obtain a high-fidelity digital twin model.

[0010] Furthermore, the insulin pump assembly fault monitoring unit estimates the blockage layer thickness h in the flexible tubing of the insulin pump based on a high-fidelity digital twin model and real-time blood glucose monitoring data using an unscented Kalman filter algorithm. z The insulin pump system is monitored in real time to determine whether it is in a healthy state by using a threshold-based method.

[0011] The advantages and positive effects of this invention are as follows: 1. A hierarchical digital twin construction method is adopted; 2. Utilize digital twin technology to establish a high-fidelity blood glucose regulation model, and quickly identify insulin pump unit malfunctions based on monitoring data; 3. The use of the unscented Kalman filter algorithm achieves high real-time performance and accuracy. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the steps of the method of the present invention; Figure 2 This is a schematic diagram of a hierarchical digital twin framework. Detailed Implementation

[0013] The following detailed description of the digital twin-based insulin pump group health status monitoring method of the present invention, with reference to the accompanying drawings and embodiments, is provided in conjunction with the accompanying drawings.

[0014] like Figure 1 As shown, a digital twin-based insulin pump set health status monitoring system comprises five units: a dataset unit, a mechanism model set unit, a digital twin model integration unit, a digital twin model consistency evaluation unit, and an insulin pump set fault monitoring unit. These five units are sequentially connected to monitor the fault status of the insulin pump set.

[0015] 1. Data Set Unit: This unit collects and stores measurable physiological data for individual diabetic patients, including blood glucose monitoring data, insulin input data, and weight. Blood glucose monitoring data is obtained using a continuous glucose monitoring system, providing a real-time blood glucose value for each diabetic patient every 15 minutes. Insulin input data in this unit is determined based on exogenous insulin infusion volume, and data processing is performed to obtain the average basal insulin infusion volume over the 15-minute time interval, assuming a constant basal insulin infusion rate between measurements. Weight data is obtained through patient physical examinations, and a corresponding data table is created in a MySQL database, with Navicat software used to manipulate the database.

[0016] 2. Mechanism Model Set Unit: This unit constructs a geometric model of the insulin pump assembly, a hydrodynamic model of insulin, and a rule model of insulin infusion based on the insulin pump assembly entity. It studies the mechanism of human glucose metabolism and constructs a human glucose metabolism model.

[0017] 3. Digital Twin Model Integration Unit: Constructing a digital twin model integration unit such as... Figure 2 A digital twin framework is proposed, and a digital twin model integration method is designed based on the model layer and data layer in this framework structure.

[0018] 3.1 The geometric model of the insulin pump set, the fluid dynamics model of insulin, the infusion rule model of insulin, and the human blood glucose metabolism model are encapsulated in the unified interface of the FMI specification. They are connected through the input and output interaction of each sub-model to achieve the integration of models from different domains.

[0019] 3.2. Using the ODBC interface, access the physiological monitoring data of patients in the healthy state of the insulin pump group in the MySQL database. Based on the transmitted data and the integrated model obtained in 3.1, Bayesian estimation is used to determine the dynamic parameters of the model, and a digital twin model of the insulin pump group in the healthy state is obtained.

[0020] 4. Digital Twin Model Consistency Evaluation Unit. This unit performs geometric consistency evaluation on the insulin pump assembly geometric model, physical consistency evaluation on the insulin fluid dynamics model and the human blood glucose metabolism model, and rule consistency evaluation on the insulin infusion rules. A comprehensive consistency evaluation result is then obtained, and the digital twin model is adjusted based on this comprehensive result.

[0021] 4.1 Consistency assessment of the geometric model of the insulin pump system: Step 1: Determine the feature element f to be measured s In this embodiment, the three elements are the geometry of the insulin pump assembly, the elastic modulus of the insulin pump assembly, and the viscosity of insulin. GF s The set of elements to be tested is represented as: GF s =(f s1 ,fs2 ,f s3 ).

[0022] In actual measurement, the feature element to be measured, f s It can be further increased, expressed as: GF s =(f s1 ,f s2 ,f s3 ,…,f sN ).

[0023] Step 2: Obtain the feature element x of the insulin pump entity through measurement, and the feature element set x(f) of the insulin pump entity. s )=(x1,x2,x3,…,x N The model's geometric parameter set is y(f) s )=(y1,y2,y3,…,y N The geometric consistency of the model is represented as Cd(GM).

[0024]

[0025] In the formula, i is the element number.

[0026] 4.2 Physical consistency assessment between the insulin hydrodynamic model and the human glucose metabolism model: Under the same conditions, the insulin dose output by the physical insulin pump unit and the insulin hydrodynamic model were compared. n1 represents the total number of simulation experiments, and m1 represents the number of simulation experiments where the error between the output insulin dose of the physical and virtual models was less than 10%. The consistency evaluation result of the insulin pump unit hydrodynamic model was R. rel1 Then: R rel1 =m1 / n1.

[0027] The blood glucose values ​​of type 1 diabetes patients wearing insulin pumps and the human blood glucose model established in this paper were compared 15 minutes after the same insulin infusion dose. If the error was within the allowable range, it was considered to have met the consistency requirements. n² represents the total number of simulation experiments, m² represents the number of simulation experiments required for the human blood glucose model to meet the consistency requirements, and the consistency evaluation result of the human blood glucose model is R. rel2 Then: R rel2 =m2 / n2.

[0028] The physical model consistency assessment result Cd(PM) is: Cd(PM) = 0.5 × R rel1 +0.5×R rel2 .

[0029] 4.3 Consistency assessment of insulin infusion rule model: Step 1: Input actual data into the insulin infusion rule model to stimulate the insulin infusion rule model to make corresponding responses.

[0030] Step 2: Compare the response of the insulin infusion rule model with the actual response. If the response error is within the allowable range, the rule model is valid, and the rule consistency result Cd(RM) = 1. Otherwise, the rule model is incorrect or inaccurate, and Cd(RM) = 0.

[0031] 4.4. Using the Analytic Hierarchy Process (AHP) to comprehensively evaluate and adjust the consistency of the digital twin model: Step 1: Construct a judgment matrix A for the three evaluation indicators: geometric consistency Cd(GM), physical model consistency evaluation result Cd(PM), and rule consistency result Cd(RM). This matrix describes the importance of each indicator. Judgment matrix element a ij The scaling methods are as follows: Table 1: Scale Table

[0032] .

[0033] Step 2: λ max Let w represent the largest eigenvalue of matrix A. Let w represent the eigenvector, which is also the weight vector.

[0034] Aw=λ max w.

[0035] Step 3: Determine the degree of logical collapse of the matrix. CI represents the consistency index of matrix A. RI represents the average random consistency index, and its standard values ​​are shown in Table 2. Then obtain the consistency ratio CR. If CR < 0.1, the judgment matrix is ​​believed to meet the requirements. Otherwise, the judgment matrix should be adjusted.

[0036] .

[0037] Table 2: Standard values ​​of the average random consistency index RI RI 0 0 0.58 0.90 1.12 1.24 1.32 1.41 1.45 1.49 Step 4: Based on the weight vector w = (w1, w2, w3) obtained in Steps 2 and 3, assuming the weights corresponding to the geometric, physical, and rule consistency evaluation results Cd(GM), Cd(PM), and Cd(RM) are w1, w2, and w3 respectively, calculate the comprehensive digital twin model consistency evaluation result Cd: , If the consistency result is below 0.95, repeat step three.

[0038] 5. Insulin pump unit fault monitoring unit: Based on a high-fidelity digital twin model and real-time blood glucose monitoring data, fault parameters are dynamically estimated using an unscented Kalman filter algorithm. A fault threshold is set to determine the health status.

[0039] The method of this invention was used to monitor the health status of a certain type of insulin pump unit: 1. Collect 7 days of physiological monitoring data from individual diabetic patients and store the data in the corresponding database and tables in MySQL software. Use Navicat software for database visualization.

[0040] 2. Determine the mechanism model set. Taking the Medtronic 722 insulin pump set as an example, the geometric model of the insulin pump set is constructed in the modeling software. Based on bond graph theory, according to the connection method and energy conversion process of the insulin pump set, the power bond graph of the hydraulic system of the insulin pump set is obtained. According to the relevant rules of bond graph modeling, the fluid dynamics state equation of the insulin pump set can be obtained: Let the system state variable be X= Input variables but:

[0041] in:

[0042]

[0043] Where Q1 is the outlet flow rate of the drug reservoir interface; P1 is the outlet pressure of the drug reservoir interface; Q2 is the outlet flow rate of the flexible hose; P2 is the outlet pressure of the flexible hose; Q3 is the outlet flow rate of the rigid needle; P3 is the outlet pressure of the rigid needle; Q d To block insulin flow at the fault, P d To block the insulin pressure at the fault location; S f Input flow rate to the insulin pump; S e Back pressure provided to the subcutaneous space of the human body.

[0044] R1, R f1 R f2 R2 represents the static friction fluid resistance of the pipeline, which includes the drug reservoir interface, flexible pipe, and rigid needle.

[0045] In the formula: μ is the kinematic viscosity of insulin (m2•s-1); L is the length of the tube (m); d is the inner diameter of the tube (m).

[0046] R d1 R d2 For flexible pipes, the dynamic friction fluid resistance

[0047] In the formula: L is the length of the flexible pipeline segment (m); d is the inner diameter of the flexible pipeline (m).

[0048] , The equivalent hydraulic resistance for blockage under blockage fault conditions.

[0049] ,

[0050] In the formula: C d ρ is the flow rate coefficient of insulin through the blocked tubing; r is the radius (m) of the flexible tubing; h z To add thickness (m) to the inner wall of the blocked pipeline; ρ is the density of insulin (kg•m⁻³); P d The pressure at the point of blockage in the pipe; L is the length of the blocked pipe (m).

[0051] I1, I2, and I3 are the fluid sensing points for the drug reservoir interface, flexible tubing, and rigid needle, respectively.

[0052]

[0053] In the formula: A1 is the cross-sectional area of ​​the pipe at the medicine storage device interface.

[0054] C1, C2, and C3 are the liquid volumes of the drug reservoir interface, flexible pipe, and rigid needle, respectively.

[0055]

[0056] In the formula: B is the liquid volume modulus of insulin (N / m³). 2 E is the bulk modulus of elasticity of the drug reservoir interface pipeline (N / m). 2 ); δ is the wall thickness of the medicine reservoir connector.

[0057] Based on the human glucose metabolism mechanism, this model considers the complete glucose regulation process, which is divided into three sub-processes: insulin is infused from subcutaneous tissue via an insulin pump and absorbed into the plasma; the coupling effect between insulin, glucose, and food intake in the plasma; and the transfer of plasma glucose after coupling to the subcutaneous space and detection by a blood glucose meter. The complete glucose-insulin model is as follows:

[0058]

[0059]

[0060]

[0061]

[0062]

[0063] In the formula, u(t) represents the exogenous insulin infusion rate at time t, and postprandial high-dose insulin infusion is assumed to be pulsed; x1(t) and x2(t) represent the insulin volumes in the first and second insulin absorption chambers, respectively; x(t) represents the insulin concentration in the plasma; and t I The value represents the time (min) at which the effective insulin concentration reaches its maximum. M represents the effective insulin clearance rate, and W represents the patient's body weight. G(t) represents the plasma glucose concentration. i S* represents the insulin sensitivity coefficient. i G represents a set of factors related to insulin sensitivity. b This represents the baseline blood glucose level, K represents the glucose autoregulation rate, and kg represents the baseline blood glucose level. b G represents the amount of endogenous glucose produced, while D(t) represents the sum of interferences from external factors such as carbohydrate intake on glucose concentration. I U(t) represents the measurable glucose concentration in the subcutaneous interstitium, τ represents the time it takes for blood glucose to be transferred from the plasma to the subcutaneous interstitial space, and U(t) represents the lumped uncertainty.

[0064] 3. Digital twin model integration: Based on a unified interface, the geometric model of the insulin pump group, the insulin fluid dynamics model, the insulin infusion rule model, and the human blood glucose metabolism model are integrated.

[0065] Historical physiological monitoring data of patients wearing insulin pumps under healthy conditions were accessed through the ODBC interface in the MySQL database, and model parameters were determined using Bayesian estimation.

[0066] Based on historical monitoring data of patients wearing insulin pumps in a healthy state within the dataset, Bayesian estimation was used to determine the model parameters. The glucose-insulin model parameter θ... k =[x1,x2,G,G I ,t I ,S* i The joint posterior distribution P(θ) of [K,τ,U] k |D T It can be obtained from the following formula:

[0067] P ( θ k ) is the model parameter θ k The prior probability, L(θ) k | D T ) is based on data D TThe calculated likelihood function, where D T This refers to data within a time window T, including continuous blood glucose monitoring data, insulin infusion data, and food intake data. The likelihood function L(θ) k | D T Related to model parameters and observation data, it can be expressed as:

[0068] m represents the number of data samples within the time window T, and D Ti Let θ represent the i-th data point within T, and the model parameter set θ k The posterior probability can be expressed as (θ) k | D T If the mean of the posterior distribution of the parameters is taken as the point estimate of the parameters, then the parameter estimates of the model are... It can be calculated using the joint posterior probability formula: .

[0069] 4. Consistency Evaluation of Digital Twin Models. Based on simulation, the geometric consistency evaluation result of the insulin pump group geometric model is Cd(GM)=1, the physical consistency evaluation result of the insulin pump group fluid dynamics model is Cd(PM)=1, and the consistency evaluation result of the insulin pump group infusion rule model is effective Cd(RM)=1.

[0070] The following table shows the order of importance for geometry, physics, and rules: Table 3: Importance Assessment Table Geometric consistency 1 1 / 5 1 Physical consistency 5 1 5 Rule Consistency 1 1 / 5 1 Based on the table above, the judgment matrix A is:

[0071] λ is calculated max =3, because CI=CR=0, the judgment matrix A meets the requirements, W=(1,5,1), the comprehensive evaluation result of the consistency of the digital twin model. The digital twin model is judged to meet the consistency requirements.

[0072] 5. Based on real-time blood glucose monitoring data and a high-fidelity digital twin model, the thickness h of the blockage layer in the flexible tubing of an insulin pump is determined. z Set as the fault parameter, h is estimated inversely using the unscented Kalman filter algorithm. z The value of . The specific implementation steps of the unscented Kalman filter algorithm are as follows: Generate k Sigma points and their corresponding weights. It is used to store the state transition matrix and calculate the single-step prediction value for all sigma points. :

[0073] Calculate state variables Single-step prediction:

[0074] Calculate the covariance matrix ,in Let V be the variance of the state equation.

[0075]

[0076] Based on the single-step prediction results, the UT transformation is used again to generate a new Sigma point set, where n is related to the state transition matrix and λ is the scaling parameter.

[0077] Substituting the new Sigma point set into the observation equation H, we obtain the output. :

[0078] The system output value is obtained by weighted summation. mean and covariance and , For observation noise:

[0079] Calculate the Kalman gain matrix K(k): Update system status: Update the covariance matrix:

[0080] Based on real-time blood glucose data of patients and an unscented Kalman filter algorithm, fault parameters If the value exceeds the safe limit, it indicates that the flexible tubing of the insulin pump unit is blocked.

Claims

1. A method for monitoring the health status of an insulin pump system based on digital twins, characterized in that, A detection system is adopted, which includes a dataset unit, a mechanism model set unit, a digital twin model integration unit, a digital twin model consistency evaluation unit, and an insulin pump set fault monitoring unit; The aforementioned dataset unit collects physiological data and stores it in a database table; The digital twin model integration unit is based on a unified interface integration mechanism model. It uses blood glucose monitoring data of insulin pump groups in a healthy state in the database and Bayesian estimation to obtain dynamic parameters in the integration model, thereby constructing a digital twin model of insulin pump groups in a healthy state. The aforementioned digital twin model consistency evaluation unit performs consistency evaluation on the mechanism model, then obtains the comprehensive consistency evaluation result of the digital twin model based on the analytic hierarchy process, and adjusts the digital twin model based on the consistency evaluation result to meet the consistency evaluation requirements, thereby obtaining a high-fidelity digital twin model. The insulin pump assembly fault monitoring unit, based on a high-fidelity digital twin model and real-time blood glucose monitoring data, determines the thickness h of the blockage layer in the flexible tubing of the insulin pump as the cause of the blockage. z The thickness h of the blockage layer is estimated in reverse using an unscented Kalman filter algorithm, assuming the fault parameters are set as parameters. z The specific implementation steps include: generating Sigma points and their corresponding weights, calculating the single-step predicted values ​​of all Sigma points; calculating the single-step predicted values ​​of the state variables; calculating the covariance matrix; generating a new set of Sigma points again using UT transformation based on the single-step prediction results; substituting the new set of Sigma points into the observation equation to obtain the output; obtaining the mean and covariance of the system output values ​​through weighted summation; calculating the Kalman gain matrix; updating the system state; updating the covariance matrix; and, based on real-time blood glucose data and the unscented Kalman filter algorithm, determining that the flexible tubing of the insulin pump group is blocked when the fault parameter is greater than the safe value. The aforementioned mechanism model set unit includes an insulin pump geometric model, an insulin hydrodynamic model, an insulin infusion rule model, and a human blood glucose metabolism model.

2. The method for monitoring the health status of an insulin pump system based on digital twins according to claim 1, characterized in that, The dataset unit collects physiological data and stores it in a database table.

3. The method for monitoring the health status of an insulin pump system based on digital twins according to claim 1, characterized in that, The digital twin model integration unit integrates the above-mentioned mechanism model based on the FMI interface and accesses the MySQL database in the above-mentioned dataset unit using the ODBC application programming interface; it uses Bayesian estimation to obtain the dynamic parameters in the integrated model using human blood glucose monitoring data of insulin pump group in healthy state, and constructs a digital twin model of insulin pump group in healthy state.

4. The method for monitoring the health status of an insulin pump system based on digital twins according to claim 1, characterized in that, The digital twin model consistency evaluation unit identifies and measures the key features of the insulin pump assembly geometric model, and also measures the key features of the insulin pump entity, calculating the degree of consistency between the two. It sets a physical consistency evaluation index Cd(PM) for the insulin fluid dynamics model and the human blood glucose metabolism model. Through simulation comparison experiments, the insulin dose output by the insulin pump entity and the insulin fluid dynamics model is compared under the same conditions. n1 is the total number of simulation experiments, and m1 is the number of simulation experiments where the error between the output insulin dose of the entity and the virtual model is less than 10%. The consistency evaluation index R of the insulin pump assembly fluid dynamics model is calculated. re11 =m1 / n1; Additionally, the blood glucose values ​​of type 1 diabetes patients wearing insulin pumps and the human blood glucose model established in this paper were compared 15 minutes after the same insulin infusion dose. If the error was within the allowable range, it was considered to have met the consistency requirements; n2 is the total number of simulation experiments, m2 is the number of simulation experiments required for the human blood glucose model to meet the consistency requirements, and the consistency evaluation result of the human blood glucose model is R. rel2 =m² / n²; Cd(PM) = 0.5 × R rel1 +0.5×R rel2 The physical consistency assessment results are calculated; the actual data is input into the insulin infusion rule model to stimulate the insulin infusion rule model to make a corresponding response, and the response of the insulin infusion rule model is compared with the actual response; If the response error is within the allowable range, the rule model is valid, and the rule consistency result Cd(RM)=1; otherwise, the rule model is incorrect or inaccurate, and Cd(RM)=0. Finally, the overall digital twin model consistency evaluation result is obtained through the analytic hierarchy process. Based on the evaluation result, the digital twin model is iteratively adjusted until it meets the consistency requirements, and a high-fidelity digital twin model is obtained.

Images