CN111341451A - Model-free adaptive predictive control algorithm with interference compensation for glycemic control - Google Patents

Model-free adaptive predictive control algorithm with interference compensation for glycemic control Download PDF

Info

Publication number
CN111341451A
CN111341451A CN202010153532.4A CN202010153532A CN111341451A CN 111341451 A CN111341451 A CN 111341451A CN 202010153532 A CN202010153532 A CN 202010153532A CN 111341451 A CN111341451 A CN 111341451A
Authority
CN
China
Prior art keywords
interference
equation
control
time
disturbance
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202010153532.4A
Other languages
Chinese (zh)
Other versions
CN111341451B (en
Inventor
王浩平
胡麦汀
田杨
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing University of Science and Technology
Original Assignee
Nanjing University of Science and Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Nanjing University of Science and Technology filed Critical Nanjing University of Science and Technology
Priority to CN202010153532.4A priority Critical patent/CN111341451B/en
Publication of CN111341451A publication Critical patent/CN111341451A/en
Application granted granted Critical
Publication of CN111341451B publication Critical patent/CN111341451B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/50—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for simulation or modelling of medical disorders
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/30—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for calculating health indices; for individual health risk assessment

Landscapes

  • Health & Medical Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Medical Informatics (AREA)
  • Public Health (AREA)
  • Biomedical Technology (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • Pathology (AREA)
  • Epidemiology (AREA)
  • General Health & Medical Sciences (AREA)
  • Primary Health Care (AREA)
  • Feedback Control In General (AREA)

Abstract

The invention discloses a model-free adaptive predictive control algorithm with interference compensation for blood sugar control, which comprises the steps of firstly establishing a very local data model with interference compensation, and deducing the structure of a prediction equation according to the model for describing the input-output relationship of a controlled system in a certain period of time in the future; carrying out online estimation on dynamic linearization parameters in a prediction equation by using historical input and output data; calculating a future optimal control input sequence by using a projection algorithm, and solving the control input at the next moment according to a rolling optimization principle; and estimating the interference of the system by using an adaptive extended state observer, and calculating a feeding interference sequence which possibly appears in the future by using a least square method for interference compensation. The control method can effectively deal with the hysteresis of the insulin action process, can inhibit the influence of feeding interference on the blood sugar system of the human body, has the safety and the rapidity of the control effect, and is very suitable for realizing the blood sugar control of the artificial pancreas.

Description

Model-free adaptive predictive control algorithm with interference compensation for glycemic control
Technical Field
The invention relates to an artificial pancreas control algorithm, in particular to a model-free self-adaptive prediction control algorithm with interference compensation for blood sugar control.
Background
The artificial pancreas has wide application space in the population of diabetics. The artificial pancreas is mainly composed of three parts, namely a continuous blood sugar monitoring device, an insulin pump and a closed-loop control algorithm, and can adjust the injection dosage of insulin according to the blood sugar level of a patient so as to keep the blood sugar level of the patient within a normal range. The closed-loop control algorithm is used as the core of the artificial pancreas, and is a difficult problem in the development center of gravity of the current artificial pancreas, and the problems of the hysteresis of insulin effect, the nonlinearity and uncertainty of a human blood sugar physiological system and the like in blood sugar control need to be solved.
Most of the current common artificial pancreas control algorithms are model prediction control. An event triggered model predictive control algorithm is used, for example, in the literature (Chakrabarty A, Zavidisanou S, Doyle FJ, Dassau E. event-triggered model predictive control for embedded specific systems, IEEE Transactions on biomedicalengineering.2018,65: 575-. The method depends on a human blood sugar-insulin system model, and has a plurality of parameters needing to be adjusted, thereby requiring longer training time. And the method has limitation on anti-interference performance, and is difficult to ensure a good control effect when large blood sugar fluctuation caused by human body eating is faced.
Disclosure of Invention
The invention aims to provide a model-free adaptive prediction control algorithm with interference compensation for blood sugar control, which combines interference estimation and compensation with the model-free adaptive prediction control algorithm, thereby taking rapidness, safety and anti-interference of a control effect into consideration.
The technical solution for realizing the purpose of the invention is as follows: a model-free adaptive predictive control algorithm with interference compensation for glycemic control, comprising the steps of:
step 1, establishing a prediction equation: firstly, establishing a very local data model for predicting single-step output, and then deriving a prediction equation capable of realizing multi-step prediction output as a control frame main body of an algorithm;
step 2, dynamic linearization parameter online estimation: estimating dynamic linearization parameters in a prediction equation according to the obtained input and output data;
step 3, calculating control input: on the basis of the established prediction equation, calculating the control input of the current moment by adopting a projection algorithm;
and 4, estimating the interference based on the adaptive extended observer: estimating the disturbance of feeding interference possibly occurring to the system by using an adaptive extended observer;
and 5, interference prediction and compensation based on a least square method: the possible future occurrence of the interference sequence is predicted using the least squares method and used to compensate for the interference term in the prediction equation at the next time instant.
Compared with the prior art, the invention has the following remarkable advantages:
(1) the model-free adaptive predictive control algorithm is not based on a model, and the control effect is ensured to still have the safety and effectiveness of model predictive control on the premise of avoiding model parameter identification; (2) the dynamic linearization parameter estimation method used by the invention can adjust the parameters of the controller on line through the input and output data, so that the controller has self-adaptability, does not need manual adjustment, and can cope with the conditions of performance change and the like of a controlled system; (3) the interference compensation method used by the invention estimates, predicts and compensates the possible food intake interference, reduces the influence of the food intake interference on the control effect, enhances the safety of the algorithm, which is not available in the traditional model-free adaptive control; (4) the invention adopts a discrete form and has larger sampling interval, reduces the calculated amount on the premise of ensuring the control effect and ensures the practicability of the algorithm.
The present invention is described in further detail below with reference to the attached drawing figures.
Drawings
FIG. 1 is a general flow chart of the control algorithm of the present invention.
FIG. 2 is a block diagram of a model-free adaptive predictive control algorithm with interference compensation.
FIG. 3 is a flow chart of the steps of an implementation of the control algorithm of the present invention.
FIG. 4(a) is a plot of simulated blood glucose levels for a closed-loop glycemic control system of the control algorithm of the present invention.
FIG. 4(b) is a graph of insulin blood glucose infusion rate simulated by the closed-loop glycemic control system of the control algorithm of the present invention.
Detailed Description
Referring to fig. 1, the model-free adaptive predictive control algorithm with interference compensation for glycemic control according to the present invention comprises the following steps:
step 1, establishing a prediction equation. The predictive equation can be used for describing the relation of the input and the output of the controlled system in the future, and is the main body of the control algorithm structure of the invention, as shown in FIG. 2. Firstly, designing a very local data model with interference compensation, which is used for describing the influence of input variation and interference on single-step output variation; and deducing a prediction equation which can be used for multi-step prediction based on the extremely local data model, and finding the optimal future control input sequence to enable the output of the controlled system to be as close to the reference trajectory as possible.
Step 1.1, designing a single-step local data model with interference compensation:
Δy(k+1)＝f(k)+AT(k)ΔU(k) (1)
wherein Δ u (k) ═ Δ u (k),.. DELTA.u (k-L +1)]TU (k) -u (k-1), u (k) and y (k) are input and output in the k-th sampling interval, respectively, f (k) includes interference and other fast time-varying parts in the system, a (k) α1(k),...,αL(k)]TIs a dynamic linearization parameter; l is the length of the vector, and is 36 according to the acting time of the insulin, namely, the influence of the change of the insulin infusion in 3 hours on the change of the current blood sugar level is described.
Step 1.2, deducing a multistep extremely local data model:
the form of the multi-step extremely local data model can be deduced by gradually deducing from the single-step extremely local data model. For the convenience of derivation, it can be assumed that:
Figure BDA0002403249320000031
with P and Q, the past inputs can be separated from the inputs in the future control time domain, and the multi-step extremely local data model derivation process is as follows:
Figure BDA0002403249320000032
the law of the multi-step extremely local data model can be summarized according to the formula (3):
Figure BDA0002403249320000033
step 1.3, deducing a prediction equation of multiple steps:
designed to facilitate the formulation of the prediction equations, the following vectors may be defined:
YP(k+1)＝[y(k+1),...,y(k+Hp)]T(5)
Figure BDA0002403249320000034
Figure BDA0002403249320000035
wherein, YP(k +1) is the system output sequence, Δ U, at each sampling instant in the prediction time domainP(k) Inputting a differential sequence for controlling the control in each sampling time in the control time domain, wherein F (k) is an influence sequence of interference on system output at each sampling time in the prediction time domain; huIs Delta UP(k) Dimension of (C), HpIs YPThe dimension of (k + 1). The prediction equation can be compiled from equations (4) to (7):
YP(k+1)＝Ey(k)+F(k)+A(k)ΔU(k-1)+B(k)ΔUP(k) (8)
wherein E ═ 1, 1.., 1]T(ii) a A and B are time-varying coefficient matrixes, and are shown as the following two formulas:
Figure BDA0002403249320000041
Figure BDA0002403249320000042
and 2, carrying out on-line estimation on the dynamic linearization parameters. The dynamic linearization parameters a (k + i), i ═ 0., H in equations (9) (10)P1 is an important parameter for embodying the dynamic characteristics of the system, describes the relationship between the system input and the system output in the future, but cannot directly acquire the value of the system input and the system output. Therefore, these parameters will be replaced by their estimated values, by
Figure BDA0002403249320000043
i＝0,...,HP-1 and will be estimated from pharmacokinetics by means of establishing a cost function.
Step 2.1, designing dynamic linearization parameter vectors based on insulin action characteristics
According to pharmacokinetics, the diffusion process and hypoglycemic effect of the injected insulin in the human body can be approximated by a high-order inertia link, which also indicates that the variation of the system input and the variation of the system output can be approximated by the high-order inertia link. The estimate of the dynamic linearization parameter vector a (k) can therefore be represented by a set of sequences:
Figure BDA0002403249320000051
wherein v is1(k) And v2(k) Are two parameters to be estimated. M is defined as:
M＝-[g(T),g(2T),...,g(HpT)]T(12)
wherein g (T) is a 2 nd order inertia element impulse response function, and T is a sampling interval. Since insulin has a lowering-promoting effect on blood glucose, the value of M is negative. In formula (11), H (v)2) Is a parameter matrix used to adjust the response characteristics, which can be expressed as:
Figure BDA0002403249320000052
such as
Figure BDA0002403249320000053
Is similar to the (fine Impulse Response) filter. Unlike FIR filters, however, this sequence of parameters describes the dynamics between the system input and output differentials. Using the form of equation (11), the number of parameters to be adjusted can be reduced, and the parameter sequence can be modified
Figure BDA0002403249320000054
The description of the dynamic properties can be close to the actual pharmacokinetics.
Step 2.2, in order to find v1(k)、v2(k) Establishes a cost function:
Figure BDA0002403249320000055
the second term on the right of equation (14) is the soft constraint set to prevent the parameter estimation from being too sensitive to noise, with μ as its weight. Thus v1(k)、v2(k) Can be expressed as:
Figure BDA0002403249320000056
for type 2 diabetic patients, the blood glucose physiological system has the characteristic of slow time variation, so that the value of A (k) in the prediction time domain can be assumed to be kept unchanged, and the influence on the accuracy of the prediction equation in practical application is negligible. Therefore, the temperature of the molten metal is controlled,
Figure BDA0002403249320000057
i＝1,...,HP-1 can be set to
Figure BDA0002403249320000058
Are equal.
And 3, calculating the control input of the current moment by adopting a projection algorithm on the basis of the established prediction equation of the formula (8).
3.1, in order to find out the future optimal control input difference sequence, establishing a cost function based on a projection algorithm theory:
Figure BDA0002403249320000061
wherein y is*(k) For the reference output at time k, the second term on the right of the equation is a soft constraint that limits the amount of change in the control input, λi+1The weight of the variation is input for each time control.
And 3.2, converting the cost function in the formula (16) into a binomial form:
J＝(YP *(k+1)-YP(k+1))T(YP *(k+1)-YP(k+1))+ΔUP(k)TλΔUP(k) (17)
wherein Y isP *(k+1)＝[yP *(k+1),...,yP *(k+HP)]T，λ＝diag(λ1,...,λHu)T。
Step 3.3, make
Figure BDA0002403249320000062
And substituting equation (8) for equation (17) to calculate the future optimal control input differential sequence:
Figure BDA0002403249320000063
and 3.4, taking the first value in the sequence as a control input according to a rolling optimization strategy, wherein the first value is represented by the following formula:
u(k)＝u(k-1)+dTΔUP(k) (19)
wherein d ═ 1,0,. 0, 0]TI.e. U (k) takes only Δ UP(k) As the control input increment.
And 4, estimating the interference based on the adaptive extended observer. And (3) estimating the disturbance of the feeding disturbance on the system, namely f (k) in the formula (1), which possibly occurs in the next sampling interval by using an adaptive extended observer.
Step 4.1, in order to estimate the disturbance value f (k), a discrete type self-adaptive extended state observer is designed, and the structure of the discrete type extended state observer is as follows:
Figure BDA0002403249320000064
Figure BDA0002403249320000071
wherein the content of the first and second substances,
Figure BDA0002403249320000072
the state quantity of the discrete extended state observer is an estimated value of the output y (k) of the system.
Figure BDA0002403249320000073
And
Figure BDA0002403249320000074
is two extended states of a discrete extended state observer, wherein
Figure BDA0002403249320000075
As an estimate of the interference term f (k), i.e. the interference estimate
Figure BDA0002403249320000076
In order to improve the tracking performance of the observer, two extended state pairs are taken through practice
Figure BDA0002403249320000077
And observing to form a three-order discrete extended state observer. A. theOAnd BOIs a matrix of coefficients. L isD(k) Is an adaptive gain.
Step 4.2, based on the Kalman filtering theory, updating the adaptive gain L at each sampling moment by the following formulaD(k)：
Figure BDA0002403249320000078
Figure BDA0002403249320000079
Wherein C isO＝[1,0,0]T。PO(k) I.e. the prior estimated covariance in the kalman filter, the initial value of which can be selected according to the initial condition error. ROTo measure the noise covariance, it is a known quantity. θ is a coefficient for accelerating convergence, and is usually taken as:
Figure BDA00024032493200000710
QOto describe by
Figure BDA00024032493200000711
The covariance of the introduced error can be summarized in the form:
Figure BDA00024032493200000712
wherein
Figure BDA00024032493200000713
Can be based on
Figure BDA00024032493200000714
The range of values that may occur within a sampling interval T is selected.
Thus initially passing through pair PO(0)、QOAnd ROCan realize the self-adaptive setting of the gain of the extended state observer, and
Figure BDA0002403249320000081
and (6) carrying out observation.
And 5, interference prediction and compensation based on a least square method. Firstly, whether food feeding occurs or not is detected, if so, a least square method is used for predicting interference sequences which possibly occur in the future, and the interference sequences are used for compensating interference terms in a prediction equation at the next moment.
Step 5.1, whether food intake occurs or not is detected, and a food intake interference possibility index is established by adopting an evanescent memory method:
Figure BDA0002403249320000082
wherein Im(k) 0 is the index of the possibility of food disturbance and is 0 < sm< 1 forgetting factor. Once I has beenm(k) Exceeds a set specific threshold value IthThat is, it is considered that the eating disturbance is detected at this time, and this time is kd. Assuming that eating disturbances are detected only during an action time, the forgetting factor s in equation (26) can be usedmAnd I of threshold valuethSetting an offset koAnd will kd-koThe time is the time at which the eating disturbance starts. In addition, a threshold I is also establishedter＜IthAnd is used for judging whether the eating disturbance duration period is over or not. If it was at the previous moment ImGreater than and equal to or less than I at the momentterThis time can be set as the end of the eating disturbance at time kter. And it may be assumed that there is only one meal during the period when eating disturbances are detected, and that meals in this period with eating disturbances are all considered this meal at the beginning of the period.
Step 5.2, after eating disturbance is detected, f (k + i), i ═ 0, H in the prediction equation is neededP-1, i.e. predicting and compensating for eating disturbances in the time domain. Establishing a high-order inertia link close to the time constant of a blood sugar physiological system, and sampling the unit impulse response of the high-order inertia link, wherein the unit impulse response is represented by the following formula:
Ga＝[ga(1·T),ga(2·T),...,ga(jT)]T j×1(27)
wherein g isa(t) is the impulse response function, j > HP。
Step 5.3, defining a parameter omega containing the food intake size information*(k) And assume ω*(k) Is 0 when no eating disturbance is detected, and is represented by the following formula：
Figure BDA0002403249320000083
Wherein
Figure BDA0002403249320000084
Is a parameter that is estimated after the detection of eating disturbances. When the food intake is disturbed at kdThe time is detected, and the least square method is used to correct
Figure BDA0002403249320000085
An online estimation is performed, represented by equation (29):
Figure BDA0002403249320000091
wherein
Figure BDA0002403249320000092
A sequence of historical disturbance values estimated for an adaptive extended state observer having dimensions k-kd+ko+1 and will increase with increasing sampling time until the end of the eating disturbance period. And is
Figure BDA0002403249320000093
Is k-kd+koA unit impulse response differential sequence of +1 dimension, where Δ ga(kT)＝ga(kT)-ga((k-1)T)。
Step 5.4, predict the vector [ f (k), ] in the equation, f (k + H)p-1)]TCan be replaced by its predicted value, represented by the following equation:
Figure BDA0002403249320000094
the interference compensation method can acquire more data along with the increase of sampling time when the eating interference occurs, so that the estimation result is more accurate, and the interference compensation in the prediction equation can be continuously updated by the on-line estimation and prediction of the eating interference, so that the control result of the control input can be optimized in a rolling way.
FIG. 3 is a flow chart of the steps of an algorithm implementation of the present invention. In order to verify the effectiveness of the control algorithm, MATLAB/Simulink is used for carrying out numerical simulation on a closed-loop blood sugar control system. The reference blood glucose level was given as 90mg/dL and a meal containing 75g of carbohydrate was given at 300 min. The safe blood glucose level interval is 70mg/dL to 180 mg/dL.
FIG. 4(a) is a simulated blood glucose level curve for a closed-loop glycemic control system of the control algorithm of the present invention; FIG. 4(b) is an insulin blood glucose infusion rate curve simulated by the closed-loop glycemic control system of the control algorithm of the present invention. The simulation result shows that the control algorithm can quickly control the blood sugar level to the preset reference value, cannot exceed the safe blood sugar level interval, and has rapidity and safety.

Claims (6)

1. A model-free adaptive predictive control algorithm with interference compensation for glycemic control, comprising the steps of:
step 1, establishing a prediction equation: firstly, establishing a very local data model for predicting single-step output, and then deriving a prediction equation capable of realizing multi-step prediction output as a control frame main body of an algorithm;
step 2, dynamic linearization parameter online estimation: estimating dynamic linearization parameters in a prediction equation according to the obtained input and output data;
step 3, calculating control input: on the basis of the established prediction equation, calculating the control input of the current moment by adopting a projection algorithm;
and 4, estimating the interference based on the adaptive extended observer: estimating the disturbance of feeding interference possibly occurring to the system by using an adaptive extended observer;
and 5, interference prediction and compensation based on a least square method: the possible future occurrence of the interference sequence is predicted using the least squares method and used to compensate for the interference term in the prediction equation at the next time instant.
2. The model-free adaptive predictive control algorithm with interference compensation for glycemic control of claim 1, wherein step 1 builds a predictive equation, specifically:
step 1.1, designing a single-step local data model with interference compensation:
Δy(k+1)＝f(k)+AT(k)ΔU(k) (1)
wherein Δ u (k) ═ Δ u (k),.. DELTA.u (k-L +1)]TU (k) -u (k-1), u (k) and y (k) are input and output in the k-th sampling interval, respectively, f (k) includes interference and other fast time-varying parts in the system, a (k) α1(k),...,αL(k)]TIs a dynamic linearization parameter; l is the length of the vector;
step 1.2, deducing a multistep extremely local data model:
gradually deducing from the single-step local data model, deducing the form of the multi-step local data model, and setting:
Figure FDA0002403249310000011
and separating the past input from the input in the future control time domain by utilizing P and Q, and summarizing the rule of the multi-step extremely local data model:
Figure FDA0002403249310000021
step 1.3, deriving a prediction equation for multiple steps
The following vectors are defined:
YP(k+1)＝[y(k+1),...,y(k+Hp)]T(5)
Figure FDA0002403249310000022
Figure FDA0002403249310000023
wherein, YP(k +1) is the system output sequence, Δ U, at each sampling instant in the prediction time domainP(k) Inputting a differential sequence for controlling the control in each sampling time in the control time domain, wherein F (k) is an influence sequence of interference on system output at each sampling time in the prediction time domain; huIs Delta UP(k) Dimension of (C), HpIs YPThe dimension of (k + 1); the prediction equation can be compiled from equations (4) to (7):
YP(k+1)＝Ey(k)+F(k)+A(k)ΔU(k-1)+B(k)ΔUP(k) (8)
wherein E ═ 1, 1.., 1]T(ii) a A and B are time-varying coefficient matrixes, and are shown as the following two formulas:
Figure FDA0002403249310000024
Figure FDA0002403249310000031
3. the model-free adaptive predictive control algorithm with interference compensation for glycemic control of claim 2, wherein the step 2 estimates the dynamic linearization parameters in the predictive equation based on the established predictive equation and the obtained input and output data, and comprises the following steps:
step 2.1, designing dynamic linearization parameter vectors based on insulin action characteristics
The estimated value of the dynamic linearization parameter vector a (k) can be represented by a set of sequences:
Figure FDA0002403249310000032
wherein v is1(k) And v2(k) Two parameters to be estimated; m is defined as:
M＝-[g(T),g(2T),...,g(HpT)]T(12)
wherein g (T) is a 2-order inertia element impulse response function, and T is a sampling interval; since insulin has a function of promoting the decrease of blood sugar, the value of M is negative; in formula (11), H (v)2) Is a parameter matrix used to adjust the response characteristics, which can be expressed as:
Figure FDA0002403249310000033
step 2.2, in order to find v1(k)、v2(k) Establishes a cost function:
Figure FDA0002403249310000034
the second term on the right of equation (14) is the set soft constraint to prevent the parameter estimation from being too sensitive to noise, with μ being its weight;
thus v1(k)、v2(k) Can be expressed as:
Figure FDA0002403249310000041
for the type 2 diabetes patients, the blood sugar physiological system has the characteristic of slow time variation, so that the influence on the accuracy of the prediction equation in practical application is negligible on the assumption that the value of A (k) in the prediction time domain is kept unchanged; therefore, the temperature of the molten metal is controlled,
Figure FDA0002403249310000042
can be set as
Figure FDA0002403249310000043
Are equal.
4. The model-free adaptive predictive control algorithm with interference compensation for glycemic control of claim 3, wherein the step 3 calculates the control input at the current time by using a projection algorithm based on the established prediction equation, specifically:
3.1, in order to find out the future optimal control input difference sequence, establishing a cost function based on a projection algorithm theory:
Figure FDA0002403249310000044
wherein y is*(k) For the reference output at time k, the second term on the right of the equation is a soft constraint that limits the amount of change in the control input, λi+1Controlling the weight of the input variable quantity for each moment;
and 3.2, converting the cost function in the formula (16) into a binomial form:
J＝(YP *(k+1)-YP(k+1))T(YP *(k+1)-YP(k+1))+ΔUP(k)TλΔUP(k) (17)
wherein Y isP *(k+1)＝[yP *(k+1),...,yP *(k+HP)]T，
Figure FDA0002403249310000045
Step 3.3, make
Figure FDA0002403249310000046
And substituting equation (8) for equation (17) to calculate the future optimal control input differential sequence:
Figure FDA0002403249310000047
and 3.4, taking the first value in the sequence as a control input according to a rolling optimization strategy, wherein the first value is represented by the following formula:
u(k)＝u(k-1)+dTΔUP(k) (19)
wherein d ═ 1,0,. 0, 0]TI.e. U (k) takes only Δ UP(k) As the control input increment.
5. The model-free adaptive predictive control algorithm with disturbance compensation for glycemic control of claim 4, wherein the step 4 estimates the disturbance of the system caused by the feeding disturbance that may occur by using an adaptive extended observer, and specifically comprises:
step 4.1, in order to estimate the disturbance value f (k), a discrete type self-adaptive extended state observer is designed, and the structure of the discrete type extended state observer is as follows:
Figure FDA0002403249310000051
Figure FDA0002403249310000052
wherein the content of the first and second substances,
Figure FDA0002403249310000053
the state quantity of the discrete extended state observer is an estimated value of the output y (k) of the system;
Figure FDA0002403249310000054
and
Figure FDA0002403249310000055
is two extended states of a discrete extended state observer, wherein
Figure FDA0002403249310000056
As an estimate of the interference term f (k), i.e. the interference estimate
Figure FDA0002403249310000057
To improve the tracking performance of the observer, two extended state pairs are taken
Figure FDA0002403249310000058
Observing to form a three-order discrete extended state observer; a. theOAnd BOIs a coefficient matrix; l isD(k) Is an adaptive gain;
step 4.2, based on the Kalman filtering theory, updating the adaptive gain L at each sampling moment by the following formulaD(k)：
Figure FDA0002403249310000059
Figure FDA00024032493100000510
Wherein C isO＝[1,0,0]T；PO(k) Corresponding to the prior estimation covariance in Kalman filtering, the initial value can be selected according to the initial condition error; rOTo measure the noise covariance, as a known quantity; θ is a coefficient for accelerating convergence, and is taken as:
Figure FDA00024032493100000511
QOto describe by
Figure FDA00024032493100000512
Covariance of introduced error:
Figure FDA0002403249310000061
wherein
Figure FDA0002403249310000062
Can be based on
Figure FDA0002403249310000063
The range of values that may occur within a sampling interval T is selected.
6. The model-free adaptive predictive control algorithm with interference compensation for glycemic control of claim 5, wherein the step 5 uses a least squares method to predict the interference sequences that may occur in the future and to use them to compensate the interference terms in the prediction equation at the next time, specifically:
step 5.1, detecting whether food intake occurs, and establishing a food intake interference possibility index by adopting an evanescent memory method:
Figure FDA0002403249310000064
wherein Im(k) 0 is the index of the possibility of food disturbance and is 0 < sm< 1 forgetting factor; once I has beenm(k) Exceeds a set specific threshold value IthThat is, it is considered that the eating disturbance is detected at this time, and this time is kd(ii) a Assuming that eating disturbances are detected only during an action time, the forgetting factor s in equation (26) can be usedmAnd I of threshold valuethSetting an offset koAnd will kd-koSetting the time as the time of starting the eating disturbance; in addition, a threshold I is also establishedter＜IthFor judging whether the eating disturbance duration period is over; if it was at the previous moment ImGreater than and equal to or less than I at the momentterThis time can be set as the end of the eating disturbance at time kter(ii) a And assuming that there is only one meal during the period when eating disturbances are detected, and that meals in the period with eating disturbances are all considered this meal at the beginning of the period;
step 5.2, after eating disturbance is detected, f (k + i), i ═ 0, H in the prediction equation is neededP-1, predicting and compensating for eating disturbances in the prediction time domain; establishing a high-order inertia link, and sampling a unit impulse response of the high-order inertia link, wherein the unit impulse response is represented by the following formula:
Ga＝[ga(1·T),ga(2·T),...,ga(jT)]T j×1(27)
wherein g isa(t) is the impulse response function, j > HP；
Step 5.3, defining a parameter omega containing the food intake size information*(k) And assume ω*(k) 0 when no eating disturbance is detected, represented by the following formula:
Figure FDA0002403249310000071
wherein
Figure FDA0002403249310000072
Is a parameter that is estimated after the detection of eating disturbances; when the food intake is disturbed at kdThe time is detected and is paired by the least square method
Figure FDA0002403249310000073
An online estimation is performed, represented by equation (29):
Figure FDA0002403249310000074
wherein
Figure FDA0002403249310000075
A sequence of historical disturbance values estimated for an adaptive extended state observer having dimensions k-kd+ko+1 and will increase with increasing sampling instant until the end of the eating disturbance period; and is
Figure FDA0002403249310000076
Is k-kd+koA unit impulse response differential sequence of +1 dimension, where Δ ga(kT)＝ga(kT)-ga((k-1)T)；
Step 5.4, predict the vector [ f (k), ] in the equation, f (k + H)p-1)]TCan be replaced by its predicted value, represented by the following equation:
Figure FDA0002403249310000077
。
CN202010153532.4A 2020-03-06 2020-03-06 Model-free adaptive predictive control method with interference compensation for glycemic control Active CN111341451B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010153532.4A CN111341451B (en) 2020-03-06 2020-03-06 Model-free adaptive predictive control method with interference compensation for glycemic control

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010153532.4A CN111341451B (en) 2020-03-06 2020-03-06 Model-free adaptive predictive control method with interference compensation for glycemic control

Publications (2)

Publication Number Publication Date
CN111341451A true CN111341451A (en) 2020-06-26
CN111341451B CN111341451B (en) 2022-09-20

Family

ID=71187239

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010153532.4A Active CN111341451B (en) 2020-03-06 2020-03-06 Model-free adaptive predictive control method with interference compensation for glycemic control

Country Status (1)

Country Link
CN (1) CN111341451B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112402731A (en) * 2020-10-10 2021-02-26 广东食品药品职业学院 Closed-loop insulin infusion system for preventing hypoglycemia phenomenon
CN116560220A (en) * 2022-09-27 2023-08-08 东北大学 Artificial pancreas self-adaptive model prediction control system with variable priority

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101500475A (en) * 2006-08-08 2009-08-05 皇家飞利浦电子股份有限公司 Method and device for monitoring a physiological parameter
CN103907116A (en) * 2011-08-26 2014-07-02 弗吉尼亚大学专利基金会 Method, system and computer readable medium for adaptive advisory control of diabetes
CN105339943A (en) * 2013-06-21 2016-02-17 费森尤斯维尔公司 Method and control device for controlling administration of insulin to patient

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101500475A (en) * 2006-08-08 2009-08-05 皇家飞利浦电子股份有限公司 Method and device for monitoring a physiological parameter
CN103907116A (en) * 2011-08-26 2014-07-02 弗吉尼亚大学专利基金会 Method, system and computer readable medium for adaptive advisory control of diabetes
CN106326651A (en) * 2011-08-26 2017-01-11 弗吉尼亚大学专利基金会 Method and system for adaptive advisory control of diabetes
CN105339943A (en) * 2013-06-21 2016-02-17 费森尤斯维尔公司 Method and control device for controlling administration of insulin to patient

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112402731A (en) * 2020-10-10 2021-02-26 广东食品药品职业学院 Closed-loop insulin infusion system for preventing hypoglycemia phenomenon
CN116560220A (en) * 2022-09-27 2023-08-08 东北大学 Artificial pancreas self-adaptive model prediction control system with variable priority
CN116560220B (en) * 2022-09-27 2024-03-01 东北大学 Artificial pancreas self-adaptive model prediction control system with variable priority

Also Published As

Publication number Publication date
CN111341451B (en) 2022-09-20

Similar Documents

Publication Publication Date Title
Chee et al. Optimal H/sup/spl infin//insulin injection control for blood glucose regulation in diabetic patients
CN105050539B (en) For the daily periodic object intervals control in the Model Predictive Control problem of the artificial pancreas of type 1 diabetes application
CN111341451B (en) Model-free adaptive predictive control method with interference compensation for glycemic control
Hajizadeh et al. Plasma-insulin-cognizant adaptive model predictive control for artificial pancreas systems
Beschi et al. Characterization of symmetric send-on-delta PI controllers
Chee et al. Expert PID control system for blood glucose control in critically ill patients
Marchetti et al. An improved PID switching control strategy for type 1 diabetes
Eren-Oruklu et al. Adaptive control strategy for regulation of blood glucose levels in patients with type 1 diabetes
US8690820B2 (en) Automatic insulin pumps using recursive multivariable models and adaptive control algorithms
JP7042808B2 (en) An automated system to control a patient&#39;s blood glucose level
Ngo et al. Control of blood glucose for type‐1 diabetes by using reinforcement learning with feedforward algorithm
Diop et al. Preserving stability/performance when facing an unknown time-delay
CN110124150A (en) It is a kind of based on adaptive softening because of the generalized predictive control closed-loop insulin infusion system of substrategy
JP4901486B2 (en) PID control method and PID control apparatus
Boiroux et al. An ensemble nonlinear model predictive control algorithm in an artificial pancreas for people with type 1 diabetes
Magni et al. Model predictive control of glucose concentration in subjects with type 1 diabetes: an in silico trial
US20200016337A1 (en) Feedback predictive control approach for processes with time delay in the manipulated variable
Haverbeke et al. Nonlinear model predictive control with moving horizon state and disturbance estimation-application to the normalization of blood glucose in the critically ill
Birjandi et al. Insulin infusion rate control in type 1 diabetes patients using information-theoretic model predictive control
Bhattacharjee et al. Evaluation of an artificial pancreas in in silico patients with online-tuned internal model control
CN109925568A (en) It is a kind of based on adaptive softening because of the generalized predictive control infusion of insulin amount calculation method of substrategy
Cassany et al. A robust control solution for glycaemia regulation of type-1 diabetes mellitus
Boiroux et al. Meal estimation in nonlinear model predictive control for type 1 diabetes
Dodek et al. Optimal state estimation for the artificial pancreas
Cetin et al. Simultaneus estimation and control of glucose-insulin dynamics in type-1 diabetes

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant