Information processing methods, devices, equipment, and storage media for insulin delivery control

Abstract

This application discloses an information processing method, apparatus, device, and medium for insulin delivery control. The method includes: acquiring blood glucose monitoring data, insulin infusion data, patient model parameters, and meal declaration data; establishing a discrete state-space model of the patient; estimating the equivalent meal perturbation online using an augmented state estimator; calculating a meal correction factor and making an overall correction to the meal input sequence within the prediction domain of the model predictive control; solving the model predictive control optimization problem based on the corrected meal input sequence to obtain and output insulin infusion control data. This application solves the problem of how to estimate the equivalent meal perturbation online based on blood glucose feedback when there is a deviation in the meal declaration amount, and how to convert this single-step estimation result into meal input sequence correction information that can be used in the future prediction time domain. This application improves prediction accuracy, shortens the duration of postprandial hyperglycemia, and reduces the risk of hypoglycemia due to overcompensation.

Description

Technical Field

[0001] This invention belongs to the field of medical and health monitoring and automatic control technology. Specifically, it relates to an information processing method for insulin delivery control based on online estimation of meal disturbances, an insulin delivery control device, a computer device, and a computer-readable storage medium. Background Technology

[0002] Patients with type 1 diabetes mellitus (T1DM) require long-term reliance on exogenous insulin to maintain stable blood glucose levels due to pancreatic β-cell dysfunction. With the development of continuous glucose monitoring (CGM) devices, insulin pumps, and control algorithms, closed-loop or semi-closed-loop blood glucose control based on artificial pancreas has become possible. Model predictive control (MPC) is a commonly used control method in artificial pancreas systems. Based on the patient's physiological model, current blood glucose status, predicted meal intake, and insulin infusion constraints, it continuously optimizes future insulin delivery within a finite prediction time domain, explicitly handling information such as infusion limits, rate of change constraints, and future meal intake.

[0003] However, in practical applications, the amount of carbohydrates declared by patients often deviates from their actual intake, and the rate of food absorption and glycemic response exhibit individual differences and time variability. Traditional predictive control models typically feed the declared food intake directly into the predictive model as a known external disturbance; once the declared amount does not match the actual food intake, the predicted glycemic trajectory will be systematically biased, leading to insufficient or excessive insulin delivery. Simply relying on fixed safety constraints to limit insulin delivery, while preventing hypoglycemia to some extent, is insufficient to compensate for postprandial hyperglycemia in a timely manner, and may also increase the risk of hypoglycemia due to delayed compensation. Furthermore, although some existing solutions use state estimators to estimate the patient's internal physiological state (such as insulin absorption state, glucose absorption state, or glycemic state), they usually do not treat the food declaration error as an independent, online-estimated equivalent disturbance state, nor do they convert the disturbance estimation result into food input correction information that can be used in the MPC prediction domain.

[0004] Therefore, in model predictive control-based insulin delivery, the following technical problems urgently need to be solved: when there is a deviation between the patient's reported meal amount and the actual meal effect, how to estimate the equivalent meal perturbation online based on continuous blood glucose feedback, and how to transform the single-step perturbation estimation result into meal input sequence correction information that can be used in the future prediction time domain, so that the model predictive controller can optimize the insulin delivery strategy in advance based on the corrected meal input throughout the prediction window, rather than just compensating for the current control amount at a single point, thereby reducing the risk of postprandial hyperglycemia and hypoglycemia caused by overcompensation while meeting the upper and lower limits of insulin infusion, the rate of change limit, and the hypoglycemia protection constraints. Summary of the Invention

[0005] To address the technical problems existing in the prior art, the present invention provides an information processing method for insulin delivery control, an insulin delivery control device, a computer device, and a computer-readable storage medium.

[0006] As one aspect of the present invention, an information processing method for insulin delivery control is provided, which is executed by a computer device. The information processing method includes: acquiring current blood glucose monitoring data, previous insulin infusion data, patient model parameter data, and meal reporting data; establishing a discrete state space model of the patient, the discrete state space model including at least insulin absorption state, glucose absorption state, and blood glucose state; estimating the equivalent perturbation of the meal online using an augmented state estimator based on the current blood glucose monitoring data and the previous insulin infusion data, wherein the equivalent perturbation of the meal is incorporated into the discrete state space model as an augmented state; calculating a meal correction factor based on the equivalent perturbation of the meal, and performing an overall correction on the meal input sequence within the prediction domain of the model predictive control; solving the model predictive control optimization problem based on the corrected meal input sequence, the current patient state determined by the patient discrete state space model, the target blood glucose, and the insulin infusion constraints, to obtain the insulin infusion control quantity data for the next control cycle; and outputting the insulin infusion control quantity data.

[0007] In some implementations, the meal correction factor is adaptively and dynamically adjusted based on at least one of the duration of the current time relative to the meal reporting time, the current rate of change in blood glucose, and the historical perturbation estimate, to generate a time-varying correction factor, which is then used to correct the meal input sequence at different sampling steps within the prediction domain.

[0008] In some implementations, after estimating the equivalent perturbation of the meal online, the equivalent perturbation of the meal is subjected to amplitude limiting, and the amplitude-limited equivalent perturbation of the meal is used to calculate the meal correction factor; after calculating the meal correction factor, the meal correction factor is subjected to amplitude limiting, and the amplitude-limited meal correction factor is used to correct the meal input sequence in the prediction domain.

[0009] In some implementations, prior to the output, the insulin infusion control data is subject to a joint safety limit based on a comparison of the current blood glucose monitoring data with the hypoglycemia threshold and the target blood glucose level, and a comparison of the residual insulin level in the body with the in vivo protection threshold. Specifically, when the current blood glucose monitoring data is below the hypoglycemia warning threshold and no meal declaration data is input, or when the residual insulin level in the body is above the in vivo protection threshold and the current blood glucose monitoring data is below the target blood glucose level, the upward adjustment of the insulin infusion control data is restricted.

[0010] In another aspect of the present invention, an insulin delivery control device is provided, comprising: a data acquisition unit for acquiring current blood glucose monitoring data, previous insulin infusion data, patient model parameter data, and meal reporting data; a model building unit for building a discrete state space model of the patient, the discrete state space model including at least insulin absorption state, glucose absorption state, and blood glucose state; a perturbation estimation unit, configured with an augmented state estimator, for incorporating the meal equivalent perturbation as an augmented state into the discrete state space model based on the current blood glucose monitoring data and the previous insulin infusion data, and estimating the meal equivalent perturbation online; a prediction domain correction unit for calculating a meal correction factor based on the meal equivalent perturbation and performing overall correction on the meal input sequence within the prediction domain of the model predictive control; an optimization solution unit for solving the model predictive control optimization problem based on the corrected meal input sequence, the current patient state determined by the patient discrete state space model, the target blood glucose, and insulin infusion constraints, to obtain insulin infusion control data for the next control cycle; and an output unit for outputting the insulin infusion control data.

[0011] In some implementations, the perturbation estimation unit includes a recursive filter module, which is communicatively connected to the model building unit. The recursive filter module is used to obtain the augmented state transition matrix and the augmented state estimate of the previous time step of the augmented patient discrete state space model, and output a posterior estimate containing the meal equivalent perturbation based on the deviation between the current blood glucose monitoring data and the prior predicted blood glucose output.

[0012] In some implementations, the data acquisition unit includes a data synchronization and validity verification module, which is communicatively connected to the user input interface of an external continuous blood glucose monitoring device and apparatus. The module is used to perform time alignment and outlier removal on the received real-time blood glucose data stream and meal declaration data, and output the verified data to the model building unit and the disturbance estimation unit, respectively.

[0013] In some embodiments, the device further includes a historical record unit, which is communicatively connected to the data acquisition unit, the disturbance estimation unit, the prediction domain correction unit, and the optimization solution unit, respectively. The historical record unit is used to receive and associate the blood glucose monitoring data, the meal equivalent disturbance estimate, the corrected meal input sequence, and the insulin infusion control data of the current control cycle, and output the corresponding historical data to the disturbance estimation unit and the optimization solution unit in subsequent control cycles.

[0014] In another aspect of the present invention, a computer device is provided, the computer device including a processor and a memory connected to the processor, wherein the memory stores program instructions; the processor is configured to execute the program instructions stored in the memory to implement the information processing method described above.

[0015] In another aspect, a computer-readable storage medium is provided, the computer-readable storage medium storing program instructions that, when executed, implement the information processing method described above.

[0016] Beneficial Effects: This invention incorporates the equivalent meal perturbation corresponding to meal reporting errors as an augmented state into the patient's discrete state space model. An augmented state estimator is used to estimate this perturbation online and convert it into a meal correction factor. This factor then comprehensively corrects the meal input sequence within the prediction domain of the model's predictive control, enabling the controller to optimize insulin delivery strategies in advance based on the corrected meal input within the future prediction window. Furthermore, an optimization problem is solved and control variables are output under constraints of insulin infusion upper and lower limits, rate of change limits, and hypoglycemia protection. Thus, when there is a deviation in the meal reporting amount, the equivalent meal perturbation can be estimated online based on blood glucose feedback, and this single-step estimation result can be converted into meal input sequence correction information that can be used in the future prediction time domain. Therefore, this invention improves the accuracy of predicting future blood glucose trajectories, enabling more timely insulin regulation during the postprandial blood glucose rise phase, thereby shortening the duration of hyperglycemia. Simultaneously, the combined use of perturbation estimation limits, correction factor limits, and infusion constraints reduces the risk of overcompensation and hypoglycemia, and supports individualized parameter configuration to suit patient groups with different insulin sensitivities, absorption times, and baseline blood glucose levels. Attached Figure Description

[0017] The above and other aspects, features, and advantages of embodiments of the present invention will become clearer from the following description taken in conjunction with the accompanying drawings, in which: Figure 1 This is a flowchart of an information processing method for insulin delivery control according to Embodiment 1 of the present invention; Figure 2 This is a schematic diagram illustrating the relationship between the declared meal input, disturbance estimation, and prediction domain correction according to Embodiment 1 of the present invention. Figure 3 This is an input-output relationship diagram of the model predictive control optimizer according to Embodiment 1 of the present invention; Figure 4 This is a unit diagram of an insulin delivery control device according to Embodiment 5 of the present invention; Figure 5 This is a schematic diagram of the structure of a computer device according to Embodiment 9 of the present invention; Figure 6This is a schematic diagram of the structure of a computer storage medium according to Embodiment 10 of the present invention; Figure 7 This is a diagram showing the actual closed-loop control effect of the technical solution of this invention in a simulation platform. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0019] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0020] As described in the background section, in existing model-predictive control-based insulin delivery protocols, the amount of carbohydrates declared by the patient is typically fed directly into the prediction model as a known external disturbance. When there is a discrepancy between the declared amount and the actual effect of the meal, the predicted blood glucose trajectory will exhibit systematic bias, leading to insufficient or excessive insulin delivery. Although some existing protocols utilize state estimators to estimate the patient's internal physiological state, they typically do not treat the meal declaration error as an independent, online-estimateable equivalent disturbance state, nor do they convert the disturbance estimation result into meal input correction information that can be used in the prediction domain of model-predictive control. Consequently, the controller cannot optimize the insulin delivery strategy in advance based on the corrected meal input throughout the prediction window and can only rely on fixed safety constraints for hysteresis compensation.

[0021] Therefore, to solve the above-mentioned technical problems, the overall technical approach of this invention is as follows: The equivalent meal disturbance corresponding to the meal reporting error is incorporated as an augmented state into the patient's discrete state space model. An augmented state estimator is used to estimate this disturbance online based on continuous blood glucose feedback. The estimated equivalent meal disturbance is converted into a meal correction factor, and the meal input sequence within the prediction domain of the model predictive control is corrected as a whole. This allows the model predictive controller to calculate the insulin infusion control amount based on the corrected meal input within the future prediction window. Finally, the optimization problem is solved and the control amount is output under the constraints of insulin infusion upper and lower limits, rate of change limits, and hypoglycemia protection, thereby achieving a leap from single-step compensation to overall correction of the prediction domain.

[0022] The following description, in conjunction with the accompanying drawings, details specific embodiments to more clearly and completely demonstrate the overall solution of the present invention.

[0023] <Example 1>

[0024] Figure 1 This is a flowchart of an information processing method for insulin delivery control according to Embodiment 1 of the present invention. All steps of the information processing method for insulin delivery control according to Embodiment 1 of the present invention are executed by a computer device. (Refer to...) Figure 1 The information processing method for insulin delivery control according to Embodiment 1 of the present invention includes steps S101 to S106.

[0025] In step S101, the current blood glucose monitoring data, the previous insulin infusion data, the patient model parameter data, and the meal declaration data are obtained.

[0026] In this embodiment, the current blood glucose monitoring data can be automatically uploaded by a continuous glucose monitoring (CGM) device at a fixed sampling period (e.g., 5 minutes), or it can be manually entered by the user via a mobile terminal or web interface. The insulin infusion data at the previous moment includes the actual output or executed insulin infusion control amount data of the previous control cycle. The patient model parameter data includes individualized physiological parameters such as sampling period, insulin sensitivity coefficient, basal blood glucose, basal insulin concentration, insulin absorption time constant, glucose absorption time constant, glucose bioavailability, glucose clearance rate, and patient weight. These parameters can be obtained through historical data identification, population model selection, or manual input. The meal declaration data includes the number of grams of carbohydrates in the meal entered by the patient and the meal start time.

[0027] In step S102, a discrete state-space model of the patient is established.

[0028] The patient discrete state-space model is a discrete-time mathematical model describing the dynamic changes in insulin absorption, glucose absorption, and blood glucose concentration within the patient's body. It includes at least the insulin absorption state, glucose absorption state, and blood glucose state. In one embodiment, the patient discrete state-space model employs a five-state discrete-time model: ; Where k is the discrete time step number; x(k) The state vector; u I (k) represents insulin infusion input; u G y(k) represents glucose input from the meal; y(k) represents glucose output. This is the state transition matrix; For insulin input matrix; Input the food matrix; A constant offset vector; This is the output matrix.

[0029] After establishing a discrete state-space model of the patient, the controller constructs a predicted meal input based on the patient's reported meal time and carbohydrate content. The meal input can be discretized using a half-sine absorption curve to ensure that the total meal quantity is conserved over discrete time. ; Where D represents the total carbohydrates in the meal; N G α is the number of sampling steps corresponding to the duration of food absorption; α is the normalization coefficient; n is the discrete sampling number in the food absorption process; when time step k is within the food absorption window, n increases from 1 to N. G Otherwise u G (k)=0.

[0030] When there is a discrepancy between the patient's reported meal quantity and their actual intake, the meal input constructed based on the reported quantity is recorded as follows: The actual food input is recorded as Both conditions are met: , Wherein, d(k) is the equivalent difference between the actual meal effect and the reported meal effect, and this difference is used as the meal equivalence estimated online by the subsequent augmented state estimator.

[0031] In step S103, based on the current blood glucose monitoring data and the insulin infusion data at the previous moment, the augmented state estimator is used to estimate the equivalent perturbation of the meal online.

[0032] The augmented state estimator refers to a recursive estimator that estimates the expanded state vector online after adding a meal-equivalent perturbation state to the original patient state. The meal-equivalent perturbation d(k) is the equivalent difference between the actual meal effect and the reported meal effect as defined by the above-mentioned meal input construction relationship.

[0033] To estimate the error in meal reporting, this embodiment incorporates the equivalent perturbation of the meal as an augmented state into the state-space model. The augmented model can be represented as: ; in, For the augmented state vector, The first state x1(k) and the second state x2(k) represent insulin absorption-related states; the first state x3(k) and the second state x4(k) represent glucose absorption-related states; the fifth state G(k) represents blood glucose concentration; and d(k) is the meal equivalent perturbation. The augmented state transition matrix is ​​constructed by augmenting A and the perturbation state transition parameters; To augment the insulin input matrix; To augment the food input matrix; The augmented constant offset vector; To augment the output matrix; For the declaration of food input.

[0034] In one implementation, a recursive filter is used to predict and update the augmented state. Here, a pointed-cap symbol represents the estimated value of the corresponding state or parameter; a superscript minus sign "−" represents the prior estimate, i.e., the current estimate obtained based on the posterior estimate from the previous time step and the model prediction; a pointed-cap symbol without a minus sign represents the posterior estimate, i.e., the estimate corrected by incorporating the current measurement; the superscript T represents matrix transpose; and the superscript −1 represents matrix inversion.

[0035] predict: ; renew: ; ; in, Let be the error covariance matrix at step k; Let be the prior error covariance matrix for the (k+1)th step; The process noise covariance matrix; To measure the noise covariance (a scalar, since blood glucose is a single output); The Kalman gain matrix; It is the identity matrix; This is the blood glucose measurement value at step k+1.

[0036] In step S104, a meal correction factor is calculated based on the equivalent perturbation of the meal, and the meal input sequence within the prediction domain of the model prediction control is corrected as a whole.

[0037] The meal correction factor refers to a coefficient calculated based on equivalent meal perturbations, used to proportionally correct the meal input sequence within the prediction domain. The prediction domain (Prediction Horizon) refers to the time window used in model predictive control to predict future blood glucose trajectories and optimize the control sequence, containing N... p One future sampling step, N p It is a positive integer.

[0038] This embodiment does not simply use the disturbance estimate as an immediate compensation term, but rather converts it into a correction factor for the food input in the prediction domain. Specifically, the correction ratio is calculated based on the current disturbance estimate and the current declared food input: , in, For meal correction factors; This is an estimate of the equivalent perturbation of the meal. To prevent small positive numbers from being divided by zero.

[0039] Subsequently, the meal input sequence in the prediction domain of the model predictive control was corrected as follows: ; in, The corrected meal input sequence; i represents the future sampling step within the prediction domain.

[0040] Figure 2 This is a schematic diagram illustrating the relationship between the reported meal input, disturbance estimation, and prediction domain correction according to Embodiment 1 of the present invention. Figure 2 As shown, a meal correction factor is calculated based on the current declared meal input and the estimated meal disturbance, and this meal correction factor is used to make an overall correction to the declared meal input sequence within the prediction domain of the model prediction control. Figure 2 The upper part is a signal processing block diagram, showing the input for food declaration. Compared with the disturbance estimate The meal correction factor was obtained through the calculation unit. This leads to the output of the corrected prediction domain meal input. The process. Figure 2 The lower part is a curve comparison chart, with the horizontal axis representing the prediction step and the vertical axis representing the meal input. The solid line represents the reported meal input curve, and the dashed line represents the corrected meal input curve. The corrected meal input sequence is generally higher than the reported meal input sequence. The corrected meal input sequence is used for future blood glucose prediction, enabling the controller to adjust the insulin delivery strategy in advance when meal reporting errors exist.

[0041] It should be noted that the food input model can be constructed using a half-sine curve, an exponential absorption curve, a bimodal absorption curve, or a data-driven absorption model, as long as the total amount of food is conserved in discrete time.

[0042] In step S105, based on the corrected meal input sequence, the patient's current state determined by the patient's discrete state space model, the target blood glucose level, and insulin infusion constraints, the model predictive control optimization problem is solved to obtain the insulin infusion control amount data for the next control cycle.

[0043] Figure 3 This is an input-output diagram of the model predictive control optimizer according to Embodiment 1 of the present invention. Figure 3As shown, the inputs to the model predictive control optimizer include the current patient state, the corrected meal input sequence, the target blood glucose level, and insulin infusion constraints. The optimizer aims to minimize the predicted blood glucose bias and the variation in insulin infusion, while simultaneously satisfying the upper and lower limits of insulin infusion and the rate of change constraints.

[0044] Control objectives may include: ; The constraints include: ; Where J is the optimization objective function; For the predicted blood glucose output at step k+i; y ref Target blood glucose; Q y The output bias weight matrix (or scalar weight coefficients); R △u To control the incremental weight matrix (or scalar weight coefficients); △u I (k+j) represents the change in insulin infusion at step k+j; N u To control the time domain length (positive integer, N) u ≤N p ); u min and u max The lower and upper limits of insulin infusion volume; △u min and △u max These represent the lower and upper limits of the rate of change in insulin infusion.

[0045] In an alternative implementation, the objective function may be further supplemented with a hypoglycemia penalty term, an insulin in vivo penalty term, a control smoothing term, or an individualized risk weight to achieve additional suppression of hypoglycemia risk and insulin accumulation in the body.

[0046] After solving the optimization problem, the first control variable is taken as the insulin infusion control variable data for the current control cycle: ; Among them, the superscript " This indicates that the control increment is the optimal value obtained by solving the optimization problem. It is the first element of the optimal control increment sequence.

[0047] In step S106, the insulin infusion control data is output.

[0048] The insulin infusion control data can serve as automatic control commands for the insulin pump, or as recommended values ​​displayed to users or healthcare professionals by an auxiliary decision-making system. In another implementation, the control output can be the basal rate, provisional basal rate, corrected dose, recommended pre-meal dose, or a combination thereof, to adapt to different clinical application scenarios.

[0049] The above steps are repeated in each sampling cycle to form a periodic closed-loop update, so as to continuously track changes in the patient's blood glucose and dynamically adjust the insulin delivery strategy.

[0050] <Example 2>

[0051] It should be noted that Embodiment 2 includes all the contents of Embodiment 1. The repeated contents will not be repeated. Only the different contents will be described below.

[0052] As is well known, food absorption exhibits significant time-varying nonlinear characteristics. Applying a fixed correction factor across the entire prediction domain can lead to insufficient correction in the early stages and excessive correction in the later stages. To address this technical problem, this embodiment employs the following solution: based on at least one of the following factors—the duration of the current time relative to the food reporting time, the current rate of change in blood glucose, and historical perturbation estimates—the food correction factor is adaptively and dynamically adjusted to generate a time-varying correction factor. This time-varying correction factor is then used to correct the food input sequences at different sampling steps within the prediction domain.

[0053] Specifically, in the early stages of meal absorption (when the duration is relatively short), the meal correction factor can be set to a small value to avoid premature overcompensation. In the peak absorption phase (when the duration is close to half of the absorption period and the current blood glucose rate is high), the correction factor can be increased to enhance compensation. In the late stages of meal absorption (when the duration is close to the full absorption period), the correction factor can be decreased again. Historical perturbation estimates can be used to determine if the patient currently has a systematic meal reporting bias. If the historical perturbation estimates are consistently positive, a historical weighting term can be added to the current correction factor to make the correction more forward-looking.

[0054] Through the above-described technical solution in this embodiment, the correction factor is no longer a fixed value, but can be dynamically adjusted according to the meal absorption process and individual blood glucose response, thereby more accurately matching the time distribution characteristics of the actual meal effect, reducing blood glucose prediction deviation caused by improper correction magnitude within the prediction domain, and thus improving the accuracy of model prediction control in tracking postprandial blood glucose trajectory.

[0055] <Example 3>

[0056] It should be noted that Embodiment 3 includes all the contents of Embodiment 1 and / or Embodiment 2. The repeated contents will not be repeated. Only the different contents will be described below.

[0057] During application, it was found that under continuous blood glucose monitoring noise or sudden changes in meal reporting, the augmented state estimator generates abnormal disturbance estimates. These abnormal values ​​are amplified in the prediction domain correction stage, causing drastic fluctuations in control commands and affecting the robustness of the control system. Therefore, to solve this technical problem, in this embodiment, after estimating the equivalent meal disturbance online, the equivalent meal disturbance is amplitude-limited, and the amplitude-limited equivalent meal disturbance is used to calculate the meal correction factor; after calculating the meal correction factor, the meal correction factor is amplitude-limited, and the amplitude-limited meal correction factor is used to correct the meal input sequence within the prediction domain.

[0058] Specifically, to avoid overly aggressive control due to abnormal estimates, the disturbance estimate can be limited: ; Where, d min d is the lower bound of the estimated value of the equivalent disturbance of the meal. max This represents the upper limit of the estimated equivalent disturbance value for the meal.

[0059] After calculating the meal correction factor, it is further limited: ; Where, ρ min ρ is the lower limit of the meal correction factor. max This represents the upper limit of the meal correction factor.

[0060] Through the above dual limiting process, the interference of outlier estimates on the model predictive control optimization problem is suppressed. Therefore, even if the augmented state estimator produces outlier estimates due to measurement noise or sudden disturbances, these outliers will be gradually restrained within a safe range before entering the prediction domain for correction and MPC optimization. This avoids drastic fluctuations in insulin infusion, significantly reduces the risk of hypoglycemia or hyperglycemia runaway due to estimation anomalies, and improves the robustness of the closed-loop control system.

[0061] <Example 4>

[0062] It should be noted that Embodiment 4 includes all the contents of Embodiment 1 and / or Embodiment 2 and / or Embodiment 3. The repeated contents will not be repeated. Only the different contents will be described below.

[0063] In the process of model predictive control adjusting insulin infusion based on meal disturbance estimation, ignoring the patient's current blood glucose level and residual insulin levels may lead to overcompensation risks when active insulin has not been fully metabolized or blood glucose is already low. Therefore, to address this technical problem, this embodiment employs the following solution: before the output, the insulin infusion control data is jointly limited for safety based on comparisons between the current blood glucose monitoring data and the hypoglycemic threshold and target blood glucose level, as well as comparisons between the residual insulin level and the in vivo protection threshold.

[0064] The residual insulin level refers to the total amount of active insulin that has not yet been metabolized and cleared from the body after previous infusions. The in vivo protection threshold is a safe upper limit set to prevent excessively high residual insulin levels. The hypoglycemia warning threshold is a blood glucose concentration threshold set to provide early warning of hypoglycemia risk. The hypoglycemia pausing threshold is a lower blood glucose concentration threshold used to trigger the pausing or reduction of insulin infusion.

[0065] Specifically, when the current blood glucose monitoring data is below the hypoglycemia warning threshold (e.g., 70 mg / dL) and no meal declaration data is entered, or when the residual insulin level in the body is higher than the in vivo protection threshold and the current blood glucose monitoring data is lower than the target blood glucose level, the upward adjustment of the insulin infusion control data is restricted.

[0066] In addition, if the current blood glucose level is below the hypoglycemic pausing threshold (e.g., 60 mg / dL), the basal insulin infusion is paused or reduced. Upper and lower limits and rate-of-change limits are set for the final insulin infusion volume. In semi-closed-loop or auxiliary decision-making modes, suggested values ​​are output and user confirmation is required.

[0067] By employing the above technical solutions, while meal disturbance correction drives insulin upregulation, the output control quantity is jointly constrained by the dual safety thresholds of blood glucose threshold and residual insulin level in the body. This enables the establishment of a dynamic balance between postprandial hyperglycemia compensation needs and hypoglycemia protection, effectively preventing hypoglycemia events caused by overcompensation and improving the safety of closed-loop control.

[0068] <Example 5>

[0069] This embodiment provides an insulin delivery control device that can be used to execute the information processing methods described in Embodiments 1 to 4. Figure 4 This is a unit diagram of an insulin delivery control device according to Embodiment 5 of the present invention. Figure 4 As shown, the device includes: a data acquisition unit, a model building unit, a disturbance estimation unit, a prediction domain correction unit, an optimization solution unit, and an output unit.

[0070] The data acquisition unit is used to acquire current blood glucose monitoring data, previous insulin infusion data, patient model parameter data, and meal declaration data.

[0071] The model building unit is used to build a discrete state space model of the patient, which includes at least insulin absorption state, glucose absorption state and blood glucose state; the model building unit is also used to incorporate meal equivalent perturbation as an augmented state into the discrete state space model to form an augmented discrete state space model of the patient.

[0072] The perturbation estimation unit is equipped with an augmented state estimator, which is used to incorporate the meal equivalent perturbation as an augmented state into the discrete state space model based on the current blood glucose monitoring data and the insulin infusion data at the previous moment, and to estimate the meal equivalent perturbation online.

[0073] The prediction domain correction unit is used to calculate the meal correction factor based on the equivalent perturbation of the meal and to make overall corrections to the meal input sequence within the prediction domain of the model prediction control.

[0074] The optimization solution unit is used to solve the model predictive control optimization problem based on the corrected meal input sequence, the patient's current state determined by the patient's discrete state space model, the target blood glucose, and insulin infusion constraints, so as to obtain the insulin infusion control amount data for the next control cycle.

[0075] The output unit is used to output the insulin infusion control data.

[0076] <Example 6>

[0077] It should be noted that Example 6 includes all the contents of Example 5. The repeated contents will not be repeated. Only the different contents will be described below.

[0078] Augmented state estimation involves a joint high-dimensional recursive calculation of the original patient state and the state affected by the meal disturbance. Therefore, it is difficult to achieve real-time, low-latency online estimation in resource-constrained embedded hardware or mobile terminals. To address this issue, this embodiment includes a recursive filter module in the disturbance estimation unit. This recursive filter module is communicatively connected to the model building unit and is used to obtain the augmented state transition matrix and the previous time-series estimate of the augmented patient discrete state space model. Based on the deviation between the current blood glucose monitoring data and the prior predicted blood glucose output, it outputs a posterior estimate containing the equivalent meal disturbance.

[0079] The prior estimate refers to the state estimate at the current moment obtained based on the previous moment's estimate and the system model prediction before the current moment's measurement value is obtained. The posterior estimate refers to the state estimate at the current moment obtained after the current moment's measurement value is obtained, combining the prior prediction value and measurement deviation correction.

[0080] The recursive filter module performs prediction and update steps: the prediction step calculates the prior state based on the augmented state transition matrix and the previous time-series estimate; the update step calculates the posterior estimate based on the deviation between the current blood glucose monitoring data and the prior predicted blood glucose output. The posterior estimate includes the original patient state component and the meal equivalent perturbation component, wherein the meal equivalent perturbation component is extracted and output to the prediction domain correction unit.

[0081] Without departing from the core idea of ​​this invention, the above-mentioned recursive filter module can also be implemented using an extended Kalman filter, an unscented Kalman filter, a moving window estimator, or other recursive estimation methods.

[0082] Therefore, in this embodiment, by encapsulating the high-dimensional augmented state estimation task into a recursive filter module and establishing a direct parameter communication link with the model building unit, the state estimation process can be completed with a fixed computational overhead in each sampling period, meeting the strict requirements of embedded controllers or mobile terminals for real-time performance and resource consumption, while ensuring the continuity and accuracy of disturbance estimation.

[0083] <Example 7>

[0084] It should be noted that Embodiment Seven includes all the contents of Embodiment Five and / or Embodiment Six. The repeated contents will not be repeated. Only the different contents will be described below.

[0085] The device simultaneously receives automatically uploaded blood glucose monitoring data streams and manually entered meal declaration data. These two sources exhibit heterogeneity in sampling frequency, transmission delay, and data format. Therefore, ensuring the time consistency and quality validity of the multi-source input data before the start of the control cycle is difficult, leading to timescale misalignment and distorted disturbance estimation. To address this issue, this embodiment includes a data synchronization and validity verification module in the data acquisition unit. This module communicates with both an external continuous blood glucose monitoring device and the device's built-in user input interface. It performs time alignment and outlier removal on the received real-time blood glucose data streams and meal declaration data, and outputs the verified data to the model building unit and the disturbance estimation unit, respectively.

[0086] Specifically, since the real-time blood glucose data stream uploaded by the continuous glucose monitoring device has a fixed sampling frequency (e.g., one data point every 5 minutes), while the meal declaration data manually entered by the user through the user input interface may have arbitrary timestamps, the data synchronization and validity verification module first performs time alignment on the two types of data to ensure that the blood glucose data and meal data used in the same control cycle correspond to the same or adjacent times. Then, it removes or marks outlier values ​​such as obviously abnormal blood glucose jump values ​​(e.g., the deviation between adjacent sampling points exceeds a preset threshold) and incorrectly formatted meal declaration data, and only transmits the data that passes the verification to the downstream unit.

[0087] Therefore, this embodiment eliminates disturbance estimation bias and MPC prediction distortion caused by inconsistent time scales of multi-source data or the mixing of abnormal measurement values, ensuring the quality of the input data of the control algorithm from the source and improving the reliability and safety of the closed-loop control system.

[0088] <Example 8>

[0089] It should be noted that Example 8 includes all the contents of Example 5 and / or Example 6 and / or Example 7. The repeated contents will not be repeated. Only the different contents will be described below.

[0090] If the control algorithm relies solely on instantaneous measurements between adjacent cycles without a historical estimation context, it can easily lead to initialization bias in the augmented state estimator or drift in the prediction model parameters, thus failing to achieve cycle-level data closure and making it difficult to maintain long-term control stability and consistency. Therefore, to address this technical problem, this embodiment adds a historical data recording unit to the device. This unit is communicatively connected to the data acquisition unit, the disturbance estimation unit, the prediction domain correction unit, and the optimization solution unit. It receives and associates the blood glucose monitoring data, meal equivalent disturbance estimates, corrected meal input sequences, and insulin infusion control data for the current control cycle. In subsequent control cycles, it outputs historical meal equivalent disturbance estimates to the disturbance estimation unit as prior information and historical insulin infusion control data to the optimization solution unit for calculating infusion rate constraints.

[0091] Through the historical record unit, the augmented state estimator can use the disturbance estimate of the previous cycle as prior information in the next control cycle, and the optimization solution unit can use historical control data to calculate the injection rate constraint, thereby achieving cross-cycle data closure.

[0092] Therefore, in this embodiment, the continuous accumulation and feedback of historical estimates and control variables make the state estimation process memory-based and continuous, avoiding initialization errors that start from zero in each cycle. At the same time, it enables the controller to identify the patient's long-term metabolic trends and systematic reporting deviations, thereby significantly improving the stability and individualized adaptation capability of multi-day closed-loop control.

[0093] <Example 9>

[0094] To implement the information processing methods described in Embodiments 1 to 4 above, this embodiment provides a computer device 300, for details please refer to... Figure 5 The computer device 300 in this embodiment includes a processor 31, a memory 32, an input / output device 33, and a bus 34.

[0095] The processor 31, memory 32, and input / output device 33 are respectively connected to the bus 34. The memory 32 stores program data, and the processor 31 is used to execute the program data to implement the information processing methods described in Embodiments 1 to 4 above.

[0096] In this embodiment, processor 31 can also be referred to as a CPU (Central Processing Unit). Processor 31 may be an integrated voltage control system chip with signal processing capabilities. Processor 31 can also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The general-purpose processor can be a microprocessor, or processor 31 can be any conventional processor.

[0097] <Example 10>

[0098] Figure 6 This is a schematic diagram of the structure of a computer storage medium according to Embodiment 10 of the present invention. Please continue reading. Figure 6 The computer storage medium 40 stores program data 41, which, when executed by the processor, is used to implement the information processing methods described in Embodiments 1 to 4 above.

[0099] When the embodiments of this application are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0100] Simulation Verification

[0101] The feasibility of the above technical solution was demonstrated through simulation. The simulation platform used was the University of Virginia / Padova Type 1 Diabetes Metabolic Simulation Platform (UVA / Padova T1DMS) v3.2.1. The simulation conditions are shown in the table below:

[0102]

[0103] In a series of realistic closed-loop simulation experiments, simulation platform version 3.2.1 was used, the patient was adult#002.mat, the simulation duration was 24 hours, the sampling period was 5 minutes, and the controller was a model predictive controller. In this experiment, the meal error was 0%, meaning the actual meal amount was consistent with the reported meal amount; the three meals were 45 grams, 70 grams, and 60 grams of carbohydrates, respectively. This result can be used to demonstrate that the technical solution of this invention (i.e., the constructed controller) can operate stably under standard closed-loop simulation conditions and can maintain blood glucose within the target range for most of the time.

[0104] For the corresponding actual control effect, see Figure 7 .like Figure 7 The diagram shown illustrates the actual closed-loop control effect of the technical solution of this invention (i.e., the constructed controller) in the University of Virginia / Padova Type 1 Diabetes Metabolic Simulation Platform. Figure 7 The title is MPC Closed-Loop Simulation—UVA / Padova T1DMS v3.2.1, Patient: adult#002.mat, Duration: 24h, Dt=5min.

[0105] Continue to refer to Figure 7The upper left area displays the true blood glucose curve and the continuous glucose monitoring curve. The solid blue line represents BG (Browser Glucose), the solid cyan line represents the CGM (Continuous Glucose Monitoring) sensor, the dashed green line indicates the Target 120 mg / dL, the red dotted lines indicate High 180 and Low 70 respectively, and the dashed yellow lines indicate the times and sizes of the three meals: 45g / 45g, 70g / 70g, and 60g / 60g. The blood glucose curve shows that blood glucose rises after meals, but the controller adjusts promptly to bring blood glucose down, maintaining it within the target range for most of the time.

[0106] The upper right area is a Time-In-Range statistical bar chart, showing that the time percentage within the target range (In Range 70-180) is 92.4%, represented in green; the time percentage within the high blood sugar range (High 181-250) is 7.6%, represented in orange; and the percentages for VeryHigh (greater than 250), Low (54-69), and Very Low (less than 54) are all zero.

[0107] The left-central area displays the insulin delivery curve. The red step line represents the MPC insulin rate, and the gray dashed line represents the patient basal level of 1.35 U / h. The vertical axis represents the insulin rate in U / h, ranging from 0 to 3.0. The controller dynamically adjusts the infusion rate based on blood glucose changes after meals and at night, fluctuating around the basal rate.

[0108] The right-center area contains the BG Statistics text box, which displays the mean blood glucose (Mean) as 141.1 mg / dL, the standard deviation (Std) as 26.6 mg / dL, the coefficient of variation (CV) as 18.8%, the minimum blood glucose (Min) as 91.4 mg / dL, the maximum blood glucose (Max) as 185.7 mg / dL, the total insulin dosage as 36.99 U / day, and the root mean square error (CGM) of continuous blood glucose monitoring as 18.42 mg / dL.

[0109] The bottom left area is a bar chart of Meal Intake: Actual vs Announced. The orange Actual meal and the blue Announced meal completely overlap. Since the error of the meal is positive 0%, the three meals are approximately 45 grams at 7 hours, 70 grams at 12 hours, and 60 grams at 18 hours.

[0110] The bottom right area contains the Simulation Info text box, displaying the platform as UVA / Padova T1DMSv3.2.1, the controller as MPC, prediction time domain 30, control time domain 12, output weight 1.0, control increment weight 1.0, the patient as adult#002.mat, weight 80.34394770443093kg, basal rate 1.35 U / h, sensor as guardianRT, pump as Generic 1, meal error positive 0%, and simulation duration 1958 seconds.

[0111] The key metrics derived from the simulation results are shown in the table below:

[0112]

[0113] The results show that during the 24-hour closed-loop control process, the target range was maintained for 92.4% of the time, with no hypoglycemia levels below 70 mg / dL. The highest blood glucose level was 185.7 mg / dL, only slightly above the hyperglycemia threshold of 180 mg / dL. These results demonstrate that the model predictive controller of this invention can achieve relatively stable blood glucose control under conditions of meal intake and insulin infusion constraints.

[0114] <Decision Support and Cloud Deployment>

[0115] In one implementation, the system runs on a mobile or web application. The system receives user input of the current continuous blood glucose level, target blood glucose level, previous basal insulin infusion rate, insulin volume in body, and grams of carbohydrates from the meal. After calculating the recommended basal insulin infusion rate, the system does not directly control the insulin pump but instead displays the suggested value, recommended duration, reason for safety limitations, and whether confirmation is required to the user or healthcare professional. If the current blood glucose level is below the hypoglycemic pausing threshold, the system outputs a suggestion to pause or reduce insulin infusion. If the calculation result exceeds the safety boundary, the system limits the recommended value and displays the reason for the limitation. This embodiment is suitable for research, education, decision support, and non-automatic closed-loop scenarios.

[0116] In another implementation, the control algorithm is deployed on a cloud server or a local hospital server. Clients submit blood glucose levels, meal information, patient parameters, and control constraints via an application programming interface (API). The server performs meal disturbance estimation and model predictive control calculations, and returns insulin recommendations. The system can incorporate user authentication, authorization management, simulation history recording, and data export functions to achieve multi-user management and commercial deployment. This system can connect to continuous glucose monitoring devices, insulin pumps, smartphones, wearable devices, or cloud-based health management platforms, forming a complete closed-loop or semi-closed-loop glucose management ecosystem.

[0117] In summary, this invention incorporates the equivalent meal disturbance corresponding to the meal reporting error as an augmented state into the patient's discrete state space model. An augmented state estimator is used to estimate this disturbance online and convert it into a meal correction factor, which is then used to comprehensively correct the meal input sequence within the prediction domain of the model's predictive control. Based on this, the model predictive control optimization problem is solved under constraints of insulin infusion upper and lower limits, rate of change limits, and hypoglycemia protection, outputting the insulin infusion control amount for the next control cycle. This achieves the technical effects of improving prediction accuracy, reducing the duration of postprandial hyperglycemia, and reducing overcompensation and the risk of hypoglycemia. It also supports individualized parameter configuration for patients, thus adapting to patient groups with different insulin sensitivities, absorption times, and baseline blood glucose levels.

[0118] The above description is merely an embodiment of this application and does not limit the patent scope of this application. Equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. An information processing method for insulin delivery control, characterized in that, It is executed by a computer device, wherein the information processing method includes: Acquire current blood glucose monitoring data, previous insulin infusion data, patient model parameter data, and meal declaration data; Establish a discrete state space model of the patient, which includes at least insulin absorption state, glucose absorption state, and blood glucose state. Based on the current blood glucose monitoring data and the insulin infusion data at the previous moment, the equivalent perturbation of the meal is estimated online using an augmented state estimator, wherein the equivalent perturbation of the meal is incorporated into the discrete state space model as an augmented state. The meal correction factor is calculated based on the equivalent perturbation of the meal, and the meal input sequence in the prediction domain of the model predictive control is corrected as a whole. Based on the corrected meal input sequence, the patient's current state determined by the patient's discrete state space model, the target blood glucose, and insulin infusion constraints, the model predictive control optimization problem is solved to obtain the insulin infusion control amount data for the next control cycle. Output the insulin infusion control data.

2. The information processing method according to claim 1, characterized in that, The meal correction factor is adaptively and dynamically adjusted based on at least one of the following: the duration of the current time relative to the meal reporting time, the current rate of change in blood glucose, and the historical perturbation estimate, to generate a time-varying correction factor. The time-varying correction factor is then used to correct the meal input sequence at different sampling steps within the prediction domain.

3. The information processing method according to claim 1, characterized in that, After estimating the equivalent perturbation of the meal online, the equivalent perturbation of the meal is subjected to amplitude limiting, and the amplitude-limited equivalent perturbation of the meal is used to calculate the meal correction factor; After calculating the meal correction factor, the meal correction factor is subjected to amplitude limiting, and the amplitude-limited meal correction factor is used to correct the meal input sequence in the prediction domain.

4. The information processing method according to claim 1, characterized in that, Before the output, the insulin infusion control data is subject to a joint safety limit based on the comparison between the current blood glucose monitoring data and the hypoglycemia threshold and target blood glucose, as well as the comparison between the residual insulin level in the body and the in vivo protection threshold. Specifically, when the current blood glucose monitoring data is lower than the hypoglycemia warning threshold and no meal declaration data is input, or when the residual insulin level in the body is higher than the in vivo protection threshold and the current blood glucose monitoring data is lower than the target blood glucose, the upward adjustment of the insulin infusion control data is restricted.

5. An insulin delivery control device, characterized in that, include: The data acquisition unit is used to acquire current blood glucose monitoring data, previous insulin infusion data, patient model parameter data, and meal declaration data. The model building unit is used to build a discrete state space model of the patient, which includes at least insulin absorption state, glucose absorption state and blood glucose state. The perturbation estimation unit is equipped with an augmented state estimator, which is used to incorporate the meal equivalent perturbation as an augmented state into the discrete state space model based on the current blood glucose monitoring data and the insulin infusion data at the previous moment, and to estimate the meal equivalent perturbation online. The prediction domain correction unit is used to calculate the meal correction factor based on the equivalent perturbation of the meal and to make overall corrections to the meal input sequence within the prediction domain of the model prediction control. The optimization solution unit is used to solve the model predictive control optimization problem based on the corrected meal input sequence, the patient's current state determined by the patient's discrete state space model, the target blood glucose and insulin infusion constraints, so as to obtain the insulin infusion control amount data for the next control cycle; The output unit is used to output the insulin infusion control data.

6. The insulin delivery control device according to claim 5, characterized in that, The perturbation estimation unit includes a recursive filter module, which is communicatively connected to the augmented patient discrete state space model establishment unit. The recursive filter module is used to obtain the augmented state transition matrix and the previous time-estimation value of the augmented discrete state space model, and outputs a posterior estimate containing the meal equivalent perturbation based on the deviation between the current blood glucose monitoring data and the prior predicted blood glucose output.

7. The insulin delivery control device according to claim 5, characterized in that, The data acquisition unit includes a data synchronization and validity verification module, which is connected to the user input interface of the external continuous blood glucose monitoring device and apparatus. The module is used to perform time alignment and outlier removal on the received real-time blood glucose data stream and meal declaration data, and output the verified data to the model building unit and the disturbance estimation unit, respectively.

8. The insulin delivery control device according to claim 5, characterized in that, The device further includes a historical record unit, which is communicatively connected to the data acquisition unit, the disturbance estimation unit, the prediction domain correction unit, and the optimization solution unit. The historical record unit is used to receive and associate the blood glucose monitoring data, the meal equivalent disturbance estimate, the corrected meal input sequence, and the insulin infusion control data of the current control cycle, and output the corresponding historical data to the disturbance estimation unit and the optimization solution unit in subsequent control cycles.

9. A computer device, characterized in that, The computer device includes a processor and a memory connected to the processor, wherein... The memory stores program instructions; The processor is used to execute program instructions stored in the memory to implement the information processing method as described in any one of claims 1 to 4.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores program instructions that, when executed, implement the information processing method as described in any one of claims 1 to 4.

Images